The Singapore Economy
Singapore’s phenomenal transformation from Third World to First World status has been of great i...
65 downloads
1667 Views
2MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
The Singapore Economy
Singapore’s phenomenal transformation from Third World to First World status has been of great interest to economists around the world yet there has been little quantitative research done on its economy and institutions. This innovative new research monograph fills the lacunae by presenting the Singapore economy through a macroeconometric model and laying foundations for further research. Using formal econometric analysis and novel modelling techniques, Abeysinghe and Choy offer rare insights into how the Singapore economy works. Each of the major chapters discusses the implications of the empirical findings for current policy and an entire chapter has been devoted to macroeconomic policy simulations. This book is a unique introduction to the Singapore economy and would be of interest to econometric modellers and policy makers in Singapore as well as advanced undergraduates and graduate researchers interested in modelling small open economies. Tilak Abeysinghe is Associate Professor of Economics at the National University of Singapore. Keen Meng Choy is Assistant Professor of Economics at the Nanyang Technological University, Singapore.
Routledge Studies in the Growth Economics of Asia
1
The Changing Capital Markets of East Asia Edited by Ky Cao
2
Financial Reform in China Edited by On Kit Tom
3
Women and Industrialization in Asia Edited by Susan Horton
4
5
6
7
8
9
The Politics of Economic Development in Indonesia Contending perspectives Edited by Ian Chalmers and Vedi Hadiz
10
Studies in the Economic History of the Pacific Rim Edited by Sally M. Miller, A. J. H. Latham and Dennis O. Flynn
Japan’s Trade Policy Action or reaction? Junichiro Wada
11
Workers and the State in New Order Indonesia Vedi R. Hadiz
The Japanese Election System Three analaytical perspectives Junichiro Wada
12
The Japanese Foreign Exchange Market Beate Reszat
13
Exchange Rate Policies in Emerging Asian Countries Edited by Stefan Collignon, Jean Pisani-Ferry and Yung Chul Park
14
Chinese Firms and Technology in the Reform Era Yizheng Shi
15
Japanese Views on Economic Development Diverse paths to the market Kenichi Ohno and Izumi Ohno
The Economics of the Latecomers Catching-up, technology transfer and institutions in Germany, Japan and South Korea Jang-Sup Shin Industrialization in Malaysia Import substitution and infant industry performance Rokiah Alavi Economic Development in Twentieth Century East Asia The international context Edited by Aiko Ikeo
16
Technological Capabilities and Export Success in Asia Edited by Dieter Ernst, Tom Ganiatsos and Lynn Mytelka
26
Capital and Labour in Japan The functions of two factor markets Toshiaki Tachibanaki and Atsuhiro Taki
17
Trade and Investment in China The European experience Edited by Roger Strange, Jim Slater and Limin Wang
27
Asia Pacific Dynamism 1550–2000 Edited by A. J. H Latham and Heita Kawakatsu
28 18
Technology and Innovation in Japan Policy and management for the twenty-first century Edited by Martin Hemmert and Christian Oberländer
The Political Economy of Development and Environment in Korea Jae-Yong Chung and Richard J Kirkby
29
Japanese Economics and Economists since 1945 Edited by Aiko Ikeo
30
China’s Entry into the World Trade Organisation Edited by Peter Drysdale and Ligang Song
31
Hong Kong as an International Financial Centre Emergence and development 1945–1965 Catherine R. Schenk
19
20
Trade Policy Issues in Asian Development Prema-chandra Athukorala Economic Integration in the Asia Pacific Region Ippei Yamazawa
21
Japan’s War Economy Edited by Erich Pauer
22
Industrial Technology Development 32 in Malaysia Industry and firm studies Edited by K. S. Jomo, Greg Felker and Rajah Rasiah
23
Technology, Competitiveness and 33 the State Malaysia’s industrial technology policies Edited by K. S. Jomo and Greg Felker
24
Corporatism and Korean Capitalism Edited by Dennis L. McNamara
25
Japanese Science Samuel Coleman
34
Impediments to Trade in Services Measurement and policy implication Edited by Christoper Findlay and Tony Warren The Japanese Industrial Economy Late development and cultural causation Ian Inkster China and the Long March to Global Trade The accession of China to the World Trade Organization Edited by Alan S. Alexandroff, Sylvia Ostry and Rafael Gomez
35
Capitalist Development and Economism in East Asia The rise of Hong Kong, Singapore, Taiwan, and South Korea Kui-Wai Li
45
State Collaboration and Development Strategies in China The case of the China-Singapore Suzhou Industrial Park (1992–2002) Alexius Pereira
36
Women and Work in Globalizing Asia Edited by Dong-Sook S. Gills and Nicola Piper
46
Capital and Knowledge in Asia Changing power relations Edited by Heidi Dahles and Otto van den Muijzenberg
37
Financial Markets and Policies in East Asia Gordon de Brouwer
47
Southeast Asian Paper Tigers? From miracle to debacle and beyond Edited by Jomo K.S.
38
Developmentalism and Dependency 48 in Southeast Asia The case of the automotive industry Jason P. Abbott
39
40
41
42
Manufacturing Competitiveness in Asia How internationally competitive national firms and industries developed in east Asia Edited by Jomo K.S.
Law and Labour Market Regulation in East Asia Edited by Sean Cooney, Tim Lindsey, Richard Mitchell and Ying Zhu
49
The Economy of the Philippines Elites, inequalities and economic restructuring Peter Krinks
The Korean Economy at the Crossroads Edited by MoonJoong Tcha and Chung-Sok Suh
50
China’s Third Economic Transformation The rise of the private economy Edited by Ross Garnaut and Ligang Song
Ethnic Business Chinese capitalism in southeast Asia Edited by Jomo K.S. and Brian C. Folk
51
The Vietnamese Economy Awakening the dormant dragon Edited by Binh Tran-Nam and Chi Do Pham
Exchange Rate Regimes in East Asia Edited by Gordon De Brouwer and Masahiro Kawai
52
Financial Governance in East Asia Policy dialogue, surveillance and cooperation Edited by Gordon De Brouwer and Yunjong Wang
53
Designing Financial Systems in East Asia and Japan Edited by Joseph P.H. Fan, Masaharu Hanazaki and Juro Teranishi
43
Restructuring Korea Inc. Jang-Sup Shin & Ha-Joon Chang
44
Development and Structural Change in the Asia-Pacific Globalising miracles or end of a model? Edited by Martin Andersson and Christer Gunnarsson
54 State Competence and Economic Growth in Japan Yoshiro Miwa
63
East Asia, Globalization and the New Economy F. Gerard Adams
55 Understanding Japanese Saving Does population aging matter? Robert Dekle
64
China as a World Factory Edited by Kevin Honglin Zhang
65
China’s State Owned Enterprise Reforms An industrial and CEO approach Juan Antonio Fernandez and Leila Fernandez-Stembridge
56 The Rise and Fall of the East Asian Growth System, 1951–2000 International competitiveness and rapid economic growth Xiaoming Huang 66 57 Service Industries and Asia-Pacific Cities New development trajectories Edited by P. W. Daniels, K. C. Ho and 67 T. A. Hutton 58 Unemployment in Asia Edited by John Benson and Ying Zhu
61 Chinese Capitalism and the Modernist Vision Satyananda J. Gabriel 62 Japanese Telecommunications Edited by Ruth Taplin and Masako Wakui
Innovation and Business Partnering in Japan, Europe and the United States Edited by Ruth Taplin
68
Asian Informal Workers Global risks local protection Santosh Mehrotra and Mario Biggeri
69
The Rise of the Corporate Economy in Southeast Asia Rajeswary Ampalavanar Brown
70
The Singapore Economy An econometric perspective Tilak Abeysinghe and Keen Meng Choy
59 Risk Management and Innovation in Japan, Britain and the USA Edited by Ruth Taplin 60 Japan’s Development Aid to China The long-running foreign policy of engagement Tsukasa Takamine
China and India A tale of two economies Dilip K. Das
The Singapore Economy An econometric perspective
Tilak Abeysinghe and Keen Meng Choy
First published 2007 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN Simultaneously published in the USA and Canada by Routledge 270 Madison Avenue, New York, NY 10016 Routledge is an imprint of the Taylor & Francis Group, an informa business
This edition published in the Taylor & Francis e-Library, 2007. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” © 2007 Tilak Abeysinghe and Keen Meng Choy All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data Abeysinghe, Tilak, 1952The Singapore economy: an econometric perspective/Tilak Abeysinghe and Keen Meng Choy.–1st ed. p. cm, – (Routledge studies in the growth economies of Asia; 70) Includes bibliographical references and index. ISBN-13: 978-0-415-41821-8 (hb) 1. Singapore–Economic conditions–Econometric models. I. Choy, Keen Meng. II. Title. HC445.8.A62 2007 330.95657001 5195–dc22 2006026521
ISBN 0-203-01124-4 Master e-book ISBN
ISBN 10: 0-415-41821-6 (hbk) ISBN 10: 0-203-01124-4 (ebk) ISBN 13: 978-0-415-41821-8 (hbk) ISBN 13: 978-0-203-01124-9 (ebk)
To our parents for demonstrating what sacrifice means for happiness
Contents
List of figures List of tables Foreword Preface
xiii xv xvi xviii
1
Introduction 1.1 A primer on the Singapore economy 1 1.2 The ESU01 model of Singapore: History, philosophy and methodology 4 1.3 Overview of the book 12
1
2
The aggregate consumption function 2.1 Introduction 16 2.2 The consumption puzzle 16 2.3 Traditional consumption functions 19 2.4 Explaining the puzzle 22 2.5 The Singaporean consumption function 24 2.6 Policy recommendations 28
3
Modelling investment expenditures 31 3.1 Introduction 31 3.2 Net investment commitments and international competitiveness 32 3.3 Modelling investment commitments 34 3.4 Investment in machinery and transport equipment 36 3.5 Construction investment in Singapore 40 3.6 Policy options 44
4
The trade sector 4.1 Introduction 46 4.2 Literature review 46 4.3 Testing export hypotheses 48 4.4 A theoretical model of export determination 52
16
46
xii
Contents 4.5 4.6 4.7 4.8
The NODX function 54 Other export components 59 Import functions for Singapore 63 Policy implications 67
5
The labour market 5.1 Introduction 70 5.2 The methodological framework 71 5.3 Labour market equations 74 5.4 Unemployment in Singapore 81 5.5 Policy conclusions 84
70
6
Sectoral production 6.1 Introduction 87 6.2 Supply-side modelling 87 6.3 Sectoral equations 90
87
7
Ancillaries and identities 7.1 Introduction 95 7.2 Consumer prices and the Balassa-Samuelson effect 95 7.3 Producer prices 100 7.4 Import and export prices 101 7.5 Tax and CPF equations 105 7.6 Bridging equations 109 7.7 Identities in the ESU01 model 111
95
8
Multiplier analysis 8.1 Introduction 115 8.2 Model validation 115 8.3 Foreign trade multipliers 118 8.4 Fiscal multipliers 126 8.5 Policy lessons 129
115
9
Policy simulations 9.1 Introduction 133 9.2 The mechanics of model simulation 134 9.3 A closed-door foreign worker policy 135 9.4 The conduct of monetary policy 136 9.5 The quest for growth: An exercise in optimal control 140
133
Appendix A: Computational methods of variables Appendix B: Listing of equations and variables Bibliography Index
145 155 171 177
List of figures
1.1 2.1 2.2 2.3 2.4 2.5 2.6 2.7 3.1 3.2 3.3 3.4 3.5 3.6 3.7 4.1 4.2 4.3 4.4 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 6.1 6.2 6.3 7.1 7.2 7.3
Flowchart of the ESU01 model The APC in Singapore Private consumption, real GDP and disposable income Measures of household wealth in Singapore The nexus between house prices, financial assets and consumption The APC and key ratios Forecasts of consumption The adjusted APC Variability of consumption and investment growth (%) Value-added content in manufacturing output (%) User cost of capital Forecasts of machinery and transport equipment investment Public share of construction investment GDP growth cycles and public share of construction Impulse response for construction investment Prices of exports and competing goods Forecasts of non-oil domestic exports Forecasts of retained imports The determinants of NODX The labour market Impulse response for employment Labour force participation rates (%) Actual wages versus market-clearing wages (S$) Growth rates of resident and foreign populations (%) Forecasts of unemployment Structural shifts in real GDP Structural shifts in the unemployment rate Change in inventories ($ million) Sectoral value-added (solid line) and final demand (dashed line) Actual (solid line) and predicted (dashed line) sectoral growth rates Wages, productivity and the CPI Consumer, producer and import prices Relationship between taxes and income
13 17 19 20 22 27 28 29 32 34 38 39 41 42 44 51 57 66 68 72 76 77 79 81 82 83 84 89 92 94 97 98 106
xiv 7.4 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 9.1
List of Figures
Relationship between government fees and income Static simulation (solid line = actual, dashed line = simulated) Dynamic simulation (solid line = actual, dashed line = simulated) Macroeconomic impact of an increase in foreign income (%) Sectoral impact of an increase in foreign income (%) Macroeconomic impact of an increase in chip sales (%) Sectoral impact of an increase in chip sales (%) Macroeconomic impact of government consumption ($) Macroeconomic impact of government investment ($) Effects of a contractionary monetary policy (% deviation from baseline) 9.2 The policy frontier
108 119 120 122 123 124 125 128 130 138 139
List of tables
2.1 2.2 2.3 3.1 3.2 3.3 4.1 4.2 4.3 4.4 4.5 6.1 6.2 8.1 8.2 8.3 8.4 8.5 9.1 9.2 9.3
APC in selected economies (real consumption/GDP ratio) APC regression results Dynamic elasticities for consumption Average annual rate of change of RULC and RCPI Elasticity estimates for NIC in manufacturing Dynamic elasticities for investment (IMT p ) Co-integration test results Export hypotheses tests Dynamic elasticities for non-oil domestic exports Dynamic elasticities for service exports Dynamic elasticities for retained imports Sectoral value-added composition of final demand Error correction models of sectoral value-added Tracking performance of the ESU01 model Dynamic multipliers for an increase in foreign income Dynamic multipliers for an increase in chip sales Dynamic multipliers for government consumption Dynamic multipliers for government investment Foreign worker policy simulation (% deviation from baseline) Assumptions on exogenous variables Projected growth rates of endogenous variables
17 25 28 33 35 40 50 51 58 63 66 90 93 117 123 125 129 131 136 141 143
Foreword
Econometric modelling has long been an integral part of the academic and policymaking scene in Singapore and has contributed in important ways to enhancing our understanding of key relationships and interlinkages in the economy. This in turn has formed the basis for deeper and more valuable discussions of macroeconomic policy issues. Indeed, the practice of econometric modelling has evolved into an indispensable skill for students, academics, forecasters and policy makers alike. This book fills an important gap by providing a formalized, interpretive framework of Singapore’s economic development, structure and policies. The ESU model in its fully developed version documented here represents an important addition to our stock of knowledge on the workings of the Singapore economy and the leverage effects of key policy instruments. The model itself incorporates the latest thinking in economic and econometric techniques. For example, the authors adopt the error correction model (ECM) estimation technique, for efficiently capturing the dynamics, while imposing long-run constraints through empirically validated restrictions. Nevertheless, the model’s architects have acclimatized its design to the local setting by taking careful account of the unique features of the Singapore economic landscape. Sometimes these present particular challenges, but I believe that the authors have managed to attain a satisfactory compromise, while maintaining overall model consistency. This is most evident in the modelling of the labour market, which in Singapore is characterized by fairly complex interactions among market forces, foreign labour supply and other institution-specific factors. The specification of labour market relationships highlights another of the book’s strengths, namely the careful attention that has been paid to data issues. The systematic record of the construction of selected key labour market variables, including various “synthetic” series, will be of immense value to researchers working in this area. The ESU model continues the recent trend in modelling by combining both supply (sectoral) and demand sides of the Singapore economy, an approach that I have found to provide helpful perspectives in our own work at the Monetary Authority of Singapore (MAS). The authors have used a novel method to bring together the two sides of the economy, thereby demonstrating the value of pluralism in modelling approaches.
Foreword xvii As I have been personally involved in the estimation of macroeconomic policy effects, I found the authors’ work in this area to be of particular interest. For example, the ESU model simulation results confirm the relatively limited impact that activist monetary policy management has on real activity in Singapore, thus reinforcing the role of the exchange rate as an anchor of stability in our small open economy setting. I believe this book will establish itself as a key reference for any serious student interested in the Singapore economy. I am particularly appreciative of the efforts at bringing transparency to the practice of econometric modelling, which has too often been plagued by references to models as “black boxes”. The authors have made an important contribution in this book. I applaud their endeavour of introducing greater rigour and empirical relevance to the analysis of macroeconomic dynamics in the Singapore economy. Edward S. Robinson Executive Director Economic Policy & Macroeconomic Surveillance Departments Monetary Authority of Singapore August 2006
Preface
Through a unique mix of market and state intervention, Singapore has carved out a “market-driven guided economy” that propelled the country from Third World to First World status in one generation. This achievement has stimulated keen interest from the rest of the world, especially the developing world, to learn more about Singapore and its institutional structures. Presenting the Singapore economy through a macroeconometric model and laying the foundation for further research in this area, where the literature is rather scanty, was the driving force behind this book. The model fleshed out in the book is the ESU01 model we constructed in 2001 for the Econometric Studies Unit (ESU) of the Department of Economics at the National University of Singapore. Policy and event simulation exercises led to further refinements of the model and its present form was finalized in 2005. The book is innovative in a couple of respects. Due to the peculiar characteristics and features of Singapore’s economic history and development, we were sometimes forced to approach the econometric modelling in unconventional ways. As a result, the book offers a novel approach to macro-modelling of a small open economy. At the same time and unlike books in this genre, we have taken great pains to discuss the economics behind the econometrics and to present rare insights into the Singapore economy gleaned from formal econometric analysis. The implications of the findings are also emphasized, with policy sections included in the major chapters and an entire chapter devoted to policy simulations. Overall, the objective of the book is to introduce readers to how the Singapore economy works and to trace out the macroeconomic consequences of various government policies. The ESU01 model is also the first macroeconometric model of the Singapore economy to be released to the public in its complete form. Despite the long history of econometric modelling at three major institutions in Singapore (National University of Singapore, Nanyang Technological University and Monetary Authority of Singapore) and the wide media coverage given to forecasts and policy evaluations released by these institutions, the detailed structures of their models have never been released to the public. And notwithstanding the great mathematician Norbert Wiener’s assertion that “Economics is a one- or two-digit science”, our book represents the first attempt to open up this black box and foster an open debate on econometric modelling, forecasting and policy evaluation in Singapore.
Preface
xix
Through publication of this book, we hope to elicit constructive suggestions for improvement from practitioners in the field and also stimulate further quantitative research into the Singapore economy. Although the book is a research monograph, the technical material is exposited in a readable and lucid form. It is aimed at a wide audience of econometric modellers, policy makers in Singapore and beyond, and researchers engaged in applied work, particularly those who are interested in modelling small open economies. Furthermore, it can be used as a required or supplementary text in graduatelevel econometrics and macroeconomics courses, or as supplementary reading for Honours (i.e. advanced) undergraduate students studying the Singapore economy. Since the mid-1980s, many people in the Department of Economics have participated in ESU research activities and their contributions directly or indirectly paved the way for us to produce this book. We thank all of them for their support and encouragement. In particular, we would like to express our gratitude to: Toh Mun Heng, Linda Low, Koh Lin Ji, Soon Teck Wong, Tse Yiu Kuen, Chen Kang, Tan Khee Giap, Basant Kapur, Ngiam Kee Jin, Tan Lin Yeok, Ng Hock Guan, David Owyong, Phang Sock Yong, Amina Tyabji, Chow Hwee Kwan, Christopher Lee, Isaac Koh, Lim Tan San, Peter Wilson, Anthony Tay, Reza Siregar and Rajaguru Gulasekaran. Special thanks are due to Enrico Tanuwidjaja for giving us an excellent helping hand in running the model in EViews and tidying up loose ends. We must also thank the National University of Singapore for two research grants (R-122-000-005-112 and R-122-000-010-112) given to one of us (Tilak Abeysinghe), and the Department of Economics and the Singapore Center for Applied and Policy Economics (SCAPE) for the logistical support rendered to this book project. Last but not least, we are very grateful to Edward Robinson for writing the Foreword. All said and done, the book was a product of our personal research efforts and the usual disclaimer applies. Tilak Abeysinghe and Keen Meng Choy Singapore June 2006
1
Introduction
This is a book about the Singapore economy as seen through an econometric lens. Accordingly, we introduce the reader in this chapter first to the economy of Singapore, then to the ESU01 model from which we obtain our econometric perspective, and finally to the book itself. The aim is to provide the economic and technical background for the subsequent material. At the end of the chapter, we have included a summary flowchart of the model which will be useful for keeping track of the big picture.
1.1
A primer on the Singapore economy
Founded in 1819 by Sir Stamford Raffles as a British trading post on a laissez-passer ideology of free trade, Singapore is justifiably labelled as a small open economy. Combining a land area of just 699 km2 and a labour force of 2.3 million workers with an ample capital stock and state-of-the-art technology, it produced a real gross domestic product (GDP) of US$181 billion in 2004. Even though Singapore is certainly one of the top league nations in per capita income rankings, the absolute size of its gross national income (GNI) in purchasing power parity terms of US$113 billion also placed it amongst the largest 40 economies in the world (World Development Indicators Database). It is more useful, therefore, to characterize Singapore’s “smallness” as its inability to influence global market conditions, with the corollary that local traders cannot really dictate the terms on which exports and imports are exchanged, since prices are effectively set in world markets. As for the “openness” of the economy, total merchandise trade amounted to nearly three times the size of GNI. Perhaps a better measure of trade openness than this ratio in view of the voluminous entrepôt trade that passes through Singapore’s ports is the ratio of domestic exports to GDP, which stands at well over one. Even in terms of more comprehensive measures of openness, Singapore stands out as a totally open economy (see Sachs and Warner, 1995; Gwartney et al., 2000). All this can be explained by the tiny domestic market, as constituted by an initial population of about 2 million people when independence was achieved
2
Introduction
in 1965 (compared with 4.35 million now), which compels the Republic to export the bulk of what is produced domestically. A lack of natural resources, except for its excellent geographical location, also meant that Singapore could not have afforded the protectionism so prevalent in larger developing economies – import duties are essentially confined to a select few consumer items. Thus, the overwhelming dependence on exports is matched by a similarly heavy reliance on imports, which constitute a high proportion of both final consumption goods and intermediate inputs. Mirroring the exposure to foreign trade, Singapore has also been completely open to financial flows since the last vestiges of capital controls were removed in 1978. Historically, foreign direct investment (FDI) has been the preponderant form of capital inflows since the late 1960s, even as short-term movements of monetary capital have increased significantly in recent times. This underlines the importance of multinational corporations (MNCs) to the economy, particularly in the manufacturing sector. In 2003, foreign firms (wholly and majority foreign-owned) accounted for 71% of net fixed assets in the sector, 73% of value-added and 44% of employment. The dependence on MNCs with their technological prowess, managerial skills and marketing networks could well have stemmed inadvertently from the dearth of indigenous entrepreneurs at the onset of the industrialization drive in the sixties, but is also a consequence of a deliberate government policy of cultivating foreign enterprise with a wide range of tax incentives and export benefits. As a result, the manufacturing sector grew steadily in importance during the last four decades so that by 2004, its value-added made up a quarter of real GDP. Manufacturing has always been the country’s leading sector in a service-dominated urban economy and will continue to be so in view of the government’s plan to double its output by 2018. Within the sector, the electronics industry is dominant as it attracts more than half of inward FDI and churns out about a third of value-added. Refined oil, petrochemicals, pharmaceuticals and newly emerging biomedical products constitute the other pillars of the manufacturing sector. In terms of trade values, electronic products are even more critical as they account for two-fifths of Singapore’s merchandise exports to the rest of the world. These statistics, moreover, underestimate the true importance of electronics because the health of ancillary industries in precision engineering and the vibrancy of the seaport and airport are directly dependent upon it. In part because of liberal financial policies, Singapore has successfully developed into an international financial centre, with the attendant activities contributing another quarter of GDP. Much of the growth in the sector over the three decades from 1968 to 1997 before the Asian financial crisis struck the region was also actively promoted by fiscal measures and spurred by the economic development of Malaysia, Indonesia and Thailand. As in olden days, Singapore has acted as the financial hub of a thriving South-East Asia. Apart from the traditional niches in banking and insurance, the financial and business services sector now consists of stockbroking, foreign exchange trading, fund management, real estate, advertising, consultancies and all kinds of professional services. A major source of
Introduction 3 service export earnings also comes from the more than 8 million tourists who visit the country each year. Though not sufficiently recognized by casual observers, Singapore’s economic circumstances have also been instrumental in shaping the framework and modus operandi of macroeconomic policies. The most obvious instance of this influence is in the conduct of monetary policy, where the large import content of domestic consumption and production has led the Monetary Authority of Singapore (MAS) since 1981 to adopt a policy of targeting the exchange rate as the most efficacious way of maintaining price stability. Once the central bank opted to do this, it does not have much leeway to manipulate domestic interest rates because of near-perfect capital mobility. Local interest rates are usually quoted at a discount to US rates because of market expectations of a persistent appreciation of the Singapore dollar. In such circumstances, any attempt by the MAS to exercise independent control over monetary targets would entail volatile fluctuations in the exchange rate as capital flows respond to the resulting cross-country interest rate differentials. Consequently, the domestic money supply expands and contracts passively to accommodate changes in economic activity, including the fluctuations caused by festive seasons. In operational terms, the managed float of the exchange rate is carried out by changing the value of the Singapore dollar against a weighted basket of currencies of Singapore’s main trading partners and competitors – in short, the nominal effective exchange rate (NEER). The MAS intervenes in the foreign exchange market to keep the NEER within an undisclosed band, which is itself a moving target that reflects the central bank’s assessment of short-term risks to the economy and longer-term economic fundamentals. On the one hand, the aim is to neutralize imported inflation by gradually appreciating the NEER over time but on the other hand, a band that is set too high will have adverse effects on export competitiveness and hence, economic growth. The MAS therefore has to steer a delicate course between the Scylla of inflation and the Charybdis of recession, making its achievement of a low average inflation rate of 2% in the period after the second world oil shock all the more outstanding. Another subtle implication of economic openness concerns fiscal policy in Singapore. The conventional wisdom holds that the high marginal propensity to import reduces the multiplier effects on domestic income of fiscal stabilization measures, although “crowding out” effects are limited because local interest rates are pinned down by foreign rates. As W.G. Huff (1994, p. 21) has noted in his excellent economic history of Singapore, this handicap was inherited from the colonial era. Nonetheless, public fixed capital formation in the form of construction works has consistently been used for counter-cyclical purposes. On the revenue side of the budget picture, the authorities have demonstrated their willingness to cut taxes and charges so as to reduce business costs when the need arises, in particular during the three classical recessions experienced by independent Singapore: the 1985–86 slump, the Asian financial crisis of 1997–98 and the recent electronics downturn in 2001. The overall fiscal stance, however, is one of extreme prudence and conservatism, with the consequence that the government routinely accumulates huge budget surpluses in non-recessionary years.
4
Introduction
The last set of macroeconomic policies that need to be discussed has to do with the state’s ubiquitous regulation of the labour market. Such intervention can be traced all the way back to the legislation introduced in the late 1960s to bring about a disciplined workforce and peaceful industrial relations. By the early 1970s, these measures had secured full employment and incipient labour shortages began to develop, making the importation of workers necessary. Today, Singapore prides itself as being one of the most cosmopolitan cities of the world with a liberal immigration policy on foreign talent. The long-term goal of the policy is to augment the stock of the working-age population and thereby ensure that the supply of labour keeps pace with demand. Concurrently, the influx of migrant workers into the country is allowed on a short-term basis to alleviate labour shortages in specific industries. An important aspect of labour market control concerns wages. Since the formation of the National Wages Council (NWC) in 1972, a unique variant of incomes policy has been practised in Singapore. The NWC is a tripartite body of employers, trade unions and government representatives who collectively set annual wage guidelines in the light of macroeconomic conditions (see Lim, 1997 for a review by a former chairman of the Council). Even though compliance is not mandatory, these recommendations were widely implemented in both the public and private sectors at least up to 1985 (Chew, 1988), after which date quantitative guidelines were discontinued because wage increases that were out of line with productivity growth were identified to be one of the causes of the mid-eighties downturn. In their place, the authorities have repeatedly cut contributions to the Central Provident Fund (CPF)1 by employers, which are part of the total wage package, and exhorted wage restraint by employees. Starting in 1986, the CPF rate was slashed by 15 percentage points; at the height of the Asian financial crisis in 1999, it was reduced by 10 percentage points; in the wake of the SARS pandemic of 2003, the rate was again cut by 3 percentage points. Thus, it appears that discretionary wage policy has been used as the extra instrument that is needed to achieve the dual objectives of price stability and export competitiveness, with the exchange rate being optimally assigned to the first of these objectives.
1.2
The ESU01 model of Singapore: History, philosophy and methodology
We will begin this section by tracing the historical antecedents to the ESU01 macroeconometric model of the Singapore economy, which is the subject of this book.2 Next, we lay bare our modelling philosophy that is motivated by Singapore’s circumstances. Lastly, the methodology followed in the econometric
1 The CPF is a fully funded pension scheme into which employers and employees are required to contribute a proportion of monthly salaries. 2 For the early history of economic modelling in Singapore, see Peebles and Wilson (1996, Chapter 8).
Introduction 5 specification, estimation and testing of individual equations, including the issues raised by the “Lucas critique”, are addressed. History of ESU models The ESU was established in 1981 as a unit within the Department of Economics at the National University of Singapore with a private endowment fund donated by the late Tan Sri Khoo Teck Puat with the objective of undertaking applied research in economics. From its inception, the primary task of the ESU has been to build and maintain a structural macroeconometric model of the Singapore economy. The first prototype model which came into operation in 1988 was formulated by Toh Mun Heng, consisted of 24 equations, including identities, and used annual data (see Chapter 16, Lim and Associates, 1988). Short-term forecasts and policy analyses based on the model were initially released at in-house seminars. Soon the seminars were turned into a major one-day public conference at the beginning of the year and another public seminar in the middle of the year. This became a regular feature of ESU activities and the media provided wide coverage to the unit’s research output. In 1992, one of us (Tilak Abeysinghe) took over the ESU macroeconometric model. The annual model did not stay stagnant under his watch; by 1994, it had expanded to 70 equations. From very early on, Abeysinghe was concerned with constructing small-scale quarterly models for the Singapore economy. His first attempt in this direction was a 3-equation quarterly model (Abeysinghe, 1992) which provided baseline GDP growth forecasts for the annual model for a number of years. In 1996, he developed a mixed-frequency regression procedure to forecast quarterly GDP growth based on monthly external trade data (Abeysinghe, 1998). Later, he constructed yet another single-equation model for the same purpose, which has as its explanatory variables foreign GDP, global chip sales and relative unit business costs. One weakness of the early ESU models was that the incomes of Singapore’s major trading partners were assumed to be given exogenously. The Asian financial crisis, however, shattered this presumption and showed how interdependent Singapore and its trading partners are, especially within the region. To accommodate this fact as well as to generate growth projections for Singapore and the major global economies endogenously, Abeysinghe developed a novel structural vector autoregressive (VAR) methodology in 1998 and used it to construct a 12-equation model, with one equation for each country (or region) covered. Subsequently, he put this quarterly multi-country VAR model to several applications (Abeysinghe and Forbes, 2005). In a further attempt to replace the ESU annual model with a higher frequency counterpart for the purpose of conducting policy simulations, Abeysinghe and Lee (1997) built a quarterly 32-equation model in 1996. This model provided reliable growth forecasts but it was not comprehensive enough for policy analysis. Consequently, the present authors constructed a new model in 2001, named it ESU01, and further refined it in 2005. The ESU01 model is a disaggregated,
6
Introduction
quarterly representation of the Singapore economy that can be utilized for economic analysis, policy simulation and ex ante forecasting. Modelling philosophy Being a state-of-the-art descendant of the earlier models, the ESU01 “vintage” emerged as a response to many imperatives, chief among them being the unique features of the Singapore economy. Due to the singularities of Singapore’s economic development, we were forced to approach the econometric modelling in unconventional ways. It is worth highlighting five key considerations that underlie the model construction process: 1.
2.
3.
Prediction and policy analysis. A basic objective of macroeconometric models is to provide reliable policy and event analyses. Subjecting a policy to thorough analysis before it is implemented is highly desirable and necessary to avoid a trial-and-error approach. However, there is a misperception among some modellers that a model designed for policy analysis need not be a good predictive model. Even though this may be true for short-term extrapolation, as long as a model provides reasonably good predictions over medium to long time horizons, it can be used for policy analysis. In fact, policy evaluation and prediction are inseparable concepts because the former is a form of conditional forecasting. Therefore, parameter constancy and predictive capability are important criteria in the formulation of the ESU01 model. Non-standard theory and models. The importance of economic theory in the formulation of meaningful econometric models cannot be overemphasized. In certain cases, we were pleasantly surprised to see how theory makes a huge difference to the empirics. A good example is our model for CPI inflation: incorporating the Balassa-Samuelson effect resulted in a quantum leap in the model’s goodness-of-fit, and no amount of experimentation would have driven us to this formulation. Occasionally, we had to go even further and deviate from orthodox views on small open economies to come up with our own theoretical specifications. A case in point is the export equation. We find that traditional export models that rely on the demand-supply dichotomy simply do not fit the Singapore experience – world demand is observed to be the most robust determinant of exports. Under the price-taker assumption, we would have been forced to formulate a supply function and ignore the role of demand. The reality, however, is that with a limited domestic market, a pricetaking MNC operating in Singapore has to consider its marketing channels in making production decisions. This led to the formulation of a novel export function. Long-run restrictions. Another issue that we had to decide on at the outset was whether to impose steady-state restrictions on the model based on the corpus of macroeconomic theory. From a practical point of view, the imposition of long-run properties – except for empirically tested co-integration
Introduction 7 restrictions – may result in misleading inferences from policy simulations. As an illustration, our estimates as well as those of others show that the foreign income elasticity of Singapore’s exports is much higher than the domestic income elasticity of imports. This implies persistent trade surpluses for Singapore even if the economy were to grow at the same rate as her trading partners. However, given that world imbalances will continue into the foreseeable future, there is no reason why Singapore cannot continue to enjoy trade surpluses, especially if she succeeds in penetrating into newly emerging markets such as China and India. Therefore, constraining the two trade elasticities to be identical will distort model estimates and hence its predictions. 4. Data constraints. Data constraints strongly influenced the way we formulated the ESU01 model. It is somewhat sad that the statistical authorities in Singapore do not publish some time series although they use the data in their own work. For instance, constant dollar GDP expenditure categories are available, but their nominal counterparts and the corresponding deflators are not. Labour productivity growth and unit labour costs are available at quarterly intervals but not the level of employment or the market wage rate, despite the fact that the sources of information used to compute these variables are the same. Because of missing data, we sometimes had to settle for “second-best” specifications and also ended up spending a lot of time re-constructing time series. Appendix A details some of the innovative methods we used to compile and refine the required data. We hope that modellers in Singapore and elsewhere will find this appendix useful. 5. Small is beautiful. It has been the general experience that small models predict better. Under certain circumstances, a univariate ARIMA model may provide the best short-run forecasts. Needless to say, such a model is of little use for policy evaluation. As a model becomes larger, not only does its forecasts tend to deteriorate in quality, its medium to long-term predictive ability might also be compromised. For these reasons, we have deliberately tried to keep the size of the core ESU01 model relatively small, with the option of building satellite models for existing or new variables if the need arises in future. The equation format The time series used in the ESU01 model are of the quarterly frequency and are primarily sourced from the Singstat Time Series (STS) online database maintained by the Singapore Department of Statistics (DOS). Where applicable and available, we downloaded the seasonally adjusted time series; if these are not provided, we adjust the raw data ourselves using the X-12 ARIMA programme. Seasonal adjustment is not necessarily innocuous, but although modelling seasonality is a better option that enriches the dynamic specification of structural relationships, the current state of seasonal modelling methodologies is far from satisfactory, especially within the context of large models. As for estimation, data over 25 years from 1978 to 2003 are used in many cases. Due to data unavailability, other equations are derived from shorter sample periods starting from the 1980s.
8
Introduction
Prior to model estimation, tests for unit roots in macroeconomic time series are usually performed. In our view, these tests often cannot differentiate between a unit root and a near-unit root, and are helpful only in distinguishing clearly stationary series. Therefore, unless otherwise stated and with the exception of policy variables such as tax rates, it is assumed that all time series are generated from non-stationary I (1) random processes. We establish co-integration amongst such I (1) variables through various means, namely Johansen’s trace test, the Dickey-Fuller residualbased test, the error correction model (ECM) unit root test (see below), graphical methods and of course, informed judgement. We also use Johansen’s maximum-likelihood (ML) method to examine the weak exogeneity of the explanatory variables in a co-integrating relation (Johansen, 1995), which is a prerequisite for contemporaneous conditioning with explanatory variables in a regression equation. If a dependent variable of interest is found to be co-integrated with its explanatory variables, our operating procedures entail running two regressions – one in levels and another in first differences. The first regression establishes the equilibrium relationship that binds the dependent and independent variables together in the long run. These equilibrium relationships are normally culled from the propositions of economic theory and are the only type of restrictions that we ensure the ESU01 model respects, but although they serve as a basis for specifying the behavioural equations of the model, economic theory is largely silent on the nature of the dynamic adjustment process towards equilibrium. Even if it provides some guidance, the complexities in the observed data may not be fully captured. Dynamic specification, therefore, is more of an empirical rather than a theoretical exercise. In this regard, ECMs specified in first differences which combine long-run co-integration relationships with empirically fitted dynamics provide a rich class of models for describing the short-run evolution of macroeconomic variables in the single equation context. They are so called because the temporary deviations from the long-run equilibrium relationships trigger dynamic adjustment processes by which such “errors” are corrected.3 ECMs are now widely accepted in the profession as the best available procedure for econometric modelling and forecasting, as they ensure that both the short-run and long-run aspects of the behaviour of economic agents are taken into consideration. To exposit our modelling methodology, and in the interest of avoiding notational clutter, consider the following bivariate autoregressive distributed lag (ADL) model: yt = φ0 +
p i=1
φi∗ yt−i +
p
ϕi∗ xt−i + εt
(1.1)
i=0
3 David Hendry, who popularized the term “error correction model”, now refers to them as “equilibrium correction models” (Hendry, 1995). The latter term, however, does not carry the connotation of “correction towards equilibrium” whereas the former does. Our preference, therefore, is to use the original term.
Introduction 9 where xt is assumed to be weakly exogenous for the parameters of the model and E(εt ) = 0 and Var(εt ) = σ 2 . An equivalent transformation of (1.1) is given by the ECM representation: yt = φ0 +
p−1
φi yt−i + ϕ0 xt +
p−1
ϕi xt−i + α ( yt−1 − βxt−1 ) + εt
i=1
i=1
(1.2) ∗ + φ ∗ + · · · + φ ∗ ), ∗ + ϕ∗ + · · · + ϕ0 = ϕ0∗ , ϕi = −(ϕi+1 where φi = −(φi+1 i+2 p−1 i+2 ∗ ∗ ∗ ∗ ∗ ϕp−1 ) for i = 1, 2, . . . , p, and β = (ϕ0 + ϕ1 + · · · + ϕp )/(1 − φ1 − φ2∗ − · · · − φp∗ ) is the long-run co-integrating coefficient. The term ECt−1 = yt−1 − βxt−1 is the long-run equilibrium error while α = −(1 − φ1∗ − φ2∗ − · · · − φp∗ ) is the adjustment coefficient which measures the speed of adjustment of yt to non-zero equilibrium errors, or the extent to which deviations from equilibrium arising in the previous quarter will be corrected during the current quarter. Through the first differenced terms, the specification in (1.2) captures the short-run dynamics as well as the inertia due to agents’ expectations and adjustment costs, both pecuniary and nonpecuniary. With the passing of time, the actual values of yt will converge to its long-run equilibrium values given by βxt . A distinct advantage of estimating (1.2) is a substantial amelioration of the multicollinearity problem that engulfs the unrestricted distributed lag model in (1.1). However, estimation of the former requires a prior estimate of β. A couple of alternative re-parameterizations of (1.2) that do not require an estimate of β are:
yt = φ0 +
p−1
φi yt−i + ϕ0 xt +
i=1
p−1
ϕi xt−i + αyt−1 + γ xt−1 + εt
i=1
(1.3) yt = φ0 +
p−1
φi yt−i + ϕ0 xt
i=1
+
p−1
ϕi xt−i + α( yt−1 − xt−1 ) + δxt−1 + εt
(1.4)
i=1
where γ = −αβ and δ = α(1 − β). The formulation in (1.4) is quite appealing if economic theory suggests that it is the ratio of two variables (or their logarithmic differences) that is stable in the long run. However, these alternative formulations loose their appeal compared to (1.2) when the number of level variables in the model increases and collinearity builds up again. Another advantage of the twostep method in (1.2) is that the long-run component can be estimated using data over a longer time span since co-integrating relationships are more robust to data deficiencies, omitted stationary variables and structural breaks. If the short-run
10
Introduction
component appears to have changed, perhaps due to the emergence of new relevant variables, then the estimation of the ECM model could be confined to a more recent period. Furthermore, since data updating tends to affect the constant term, we drop the constant term from the co-integrating regression when it enters the ECM formulation. This way we have to re-estimate only the ECMs when data are updated. Except for a few variables, we have estimated all the equations in the book in the ECM format of (1.2). To do this, we usually arrive at an estimate of β after comparing the outcomes of several methods such as static ordinary least squares (OLS), dynamic OLS (Stock and Watson, 1993), the ADL regression in (1.1), Johansen’s ML procedure and more rarely, the implied estimates of β from (1.3) or (1.4). Obviously, some judgement is needed in the selection of an appropriate method for each equation. Most of the time, we use the ADL estimate of β (i.e. the static solution to (1.1)) which is more accurately estimated than the dynamic coefficients. Furthermore, Pesaran and Shin (1998) have shown that the t-statistic for β from the ADL regression has an asymptotic normal distribution. The ADLECM formulation also conveniently provides an alternative test of co-integration based on the speed of adjustment coefficient. This is the ECM unit root test of the null hypothesis of no co-integration (α = 0) proposed by Banerjee et al. (1998).4 After estimating the co-integrating part of the behavioural equation, we follow through with a general-to-specific search procedure (Hendry, 1995) to arrive at a parsimonious and economically meaningful ECM. The approach begins with very general specifications which are then progressively simplified via statistical testing. Each model is then put through a rigorous diagnostic checking process. In addition to the standard error, R2 and Durbin-Watson statistic, a battery of misspecification tests are conducted for each stochastic equation in the ESU01 model. These are the Breusch-Godfrey LM (Lagrange multiplier) test for autocorrelation up to five lags, Engle’s test for ARCH effects, a small sample version of the Bowman-Shenton test for normality, the White test of heteroscedasticity, and Ramsey’s RESET test for functional form (see Hendry and Doornik, 2001). Exact or approximate F-test results are reported, except for the normality test where the asymptotic χ 2 value is shown. As emphasized above, parameter invariance is a very important requirement for employing a model in policy analysis. A structural break is a serious matter because it implies that the substantive conclusions about the economy being modelled are vitiated. Hence, we also carry out a formal test of parameter stability in the form of the Chow test of forecast performance in the post-sample period from 2001 to 2003. This is complemented by visual inspection of the recursive parameter estimates of each equation. Even where dynamic error-correcting models were not estimated, as in the case of some ancillary equations, the same testing strategy was applied to the search for congruent and adequate specifications.
4 Under the null hypothesis, the distribution of the t-statistic is non-standard and critical values are given in the paper. When co-integration holds, however, the statistic follows the usual distribution.
Introduction 11 Finally, we use the estimated ECM coefficients to generate dynamic elasticities. This is done by converting the model back to its level form as in (1.1). Using the lag operator L (Li zt = zt−i ), we can write (1.1) as: φ ∗ (L)yt = φ0 + ϕ ∗ (L)xt + εt
(1.5)
where φ ∗ (L) = 1 − φ1∗ L − φ2∗ L2 − · · · − φp∗ Lp and ϕ ∗ (L) = ϕ0 + ϕ1∗ L + ϕ2∗ L2 + · · · + ϕp∗ Lp . The time profile of the impact of a unit change in xt on yt can be obtained from the infinite sequence φ ∗ (L)−1 ϕ ∗ (L) = π0 + π1 L + π2 L2 + · · · . A plot of the π coefficients against the time lag (k = 0, 1, 2, . . .) provides the familiar impulse response function. If variables are measured in logarithms, as in all our ECMs, then π0 gives the impact elasticity sums of πi and the cumulative ∗ (1)/φ ∗ (1) = β is the yield the dynamic elasticities. In the limit, ∞ π = ϕ i=0 i long-run elasticity. The Lucas critique At this stage, we need to draw attention to the famous Lucas critique that has cast a shadow of doubt on the usefulness of macroeconometric models for forecasting and policy analysis. Lucas (1976) argued that the parameters of traditional econometric models estimated on the basis of past experience do not stay constant when policy rules and people’s expectations change. As a result, such models cannot be used for policy evaluation. In reality, the debate over the issue is essentially an empirical question: are there regime switches that are sufficiently large in magnitude to alter the structural parameters drastically? Eckstein (1983) found no convincing evidence in the affirmative. Instead, the major sources of forecasting errors seem to be unusual events such as wars, strikes, embargoes and the like. Notwithstanding the critique, it is often possible to formulate regression models that demonstrate remarkable parameter constancy. Observed instabilities typically result from mundane reasons such as model misspecification. After an extensive survey and analysis of the Lucas critique, Ericsson and Irons (1995) conclude that “the Lucas critique is a possibility theorem, not an existence theorem”. To understand the point made by these authors, consider the following simple example illustrating the employment decision of a firm operating in Singapore: Nt+1 = αE(Wt+1 |It ) + εt
(1.6)
Wt = (1 + λt )Wtm
(1.7)
m Wtm = δ + Wt−1 + ut
(1.8)
The first equation shows that the firm decides on the next period’s level of employment Nt+1 , based on the expected real wage cost E(Wt+1 |It ), given the information available at time t. α is the structural parameter of interest. The second equation
12
Introduction
is an identity stating that the wage cost to the firm is the market wage rate W m , plus statutory payments such as CPF contributions and foreign worker levies, which are represented by the policy parameter λ. The third equation assumes that W m is a unit root process with drift δ and ut is a zero mean stationary process. If the current wage is observed by economic agents, then It = Wtm , E(Wt+1 |It ) = (1 + λt )δ + (1 + λt )Wtm , and (1.6) becomes Nt+1 = β0 + β1 Wt + εt
(1.9)
where β0 = αδ(1 + λt ) and β1 = α. Equation (1.9) takes the form of a standard econometric model that can be estimated and used for policy evaluation by varying the value of λ. However the model is vulnerable to the Lucas critique because β0 does not stay constant when λ changes. This problem, however, can easily be remedied by adding (1+λt ) as an additional explanatory variable into the equation. Thus, it is possible in general to specify models with constant parameters that satisfy the “super-exogeneity” property required for legitimate policy or intervention analysis (Hendry, 1995). In the ESU01 model, expectations are not explicitly modelled but they are treated implicitly instead by the inclusion of current and lagged variables and the use of proxies for agents’ expectations, especially those of investors. As noted earlier, unanticipated events may still induce temporary instabilities into the parameter estimates of the model and throw forecasts completely out of line. Such forecast failures, however, do not diminish the value of the model as a tool for policy simulations.
1.3
Overview of the book
We will now take the reader through a helicopter tour of the book’s chapters. In the process, we shall explain how the different parts of the ESU01 model hang together. This we do with the help of a schematic flowchart that shows the behavioural relationships binding the key macroeconomic variables in the model (Figure 1.1). There are altogether 62 endogenous variables and equations in the model: 36 behavioural equations and 26 identities, supplemented by 35 exogenous variables. (See Appendix B for a quick perusal of the variables and their notations.) They are divided among four blocks of equations – domestic demand, trade, labour market and sectors. Our exercise in model building begins with the domestic demand block. It consists of equations that are estimated for the domestic components of aggregate demand i.e. private consumption and investment (government spending is not modelled, but taken to be an exogenous policy variable instead). Chapter 2 describes the specification and estimation of the Singapore aggregate consumption function, a particularly challenging task due to an apparent puzzle: the average propensity to consume (APC) has declined steadily over the years to just two-fifths of income. After a thorough search for possible explanations, the falling APC in the long run
Figure 1.1 Flowchart of the ESU01 model.
14
Introduction
is found to be explicable by secular movements in real disposable income, financial wealth, household indebtedness and visitor expenditures. Chapter 3 starts by discussing a regression model for net investment commitments in Singapore’s manufacturing sector. These commitments, together with the user cost of capital and GDP, are found to be the main drivers of machinery and transport equipment investment. We then turn to the property market and analyze its links to local construction activity. Using the real value of contracts awarded as an explanatory variable, we present a dynamic equation for private construction investment. Despite the difficulty of modelling highly volatile capital expenditures, the econometric models estimated in this chapter provide a good fit to the historical data. A major component of the ESU01 model, in view of Singapore’s total dependence on the world economy, is the trade sector dealt with in Chapter 4. This cluster of equations functions as the external demand block of the model. In the chapter, we first test various export hypotheses on Singapore data. The standard hypotheses are all rejected, motivating us to formulate a theoretical model to guide the specification of disaggregated export equations. Behavioural equations are then estimated for non-oil domestic exports, oil exports, re-exports and service exports, taking into account the key determinants of the republic’s trade with the rest of the world. In arriving at empirical import functions, we distinguish between retained merchandise imports and service imports and allow each of them to be determined by final demand and relative prices. As a result, important links are established between the domestic demand and trade blocks in the model. The labour market is analyzed in Chapter 5. This econometric investigation is useful in its own right because of the insights it reveals on how employment, the labour force, wages and the unemployment rate are simultaneously determined in conjunction with output and the price level. Of more practical relevance, the labour demand, supply and wage equations estimated in a disequilibrium framework enable us to generate reliable short-term predictions of these variables – the subject of intense policy and public interest in Singapore for good reasons. Labour market variables in turn affect the trade sector and prices through their impact on unit labour costs and productivity, providing good examples of the feedback relationships present in the ESU01 model. We end the chapter with a careful examination of unemployment in Singapore, touching on measurement issues and the recent phenomenon of structural unemployment. The sectoral block is the subject of Chapter 6. Instead of explicitly estimating production functions or relying solely on input-output (IO) tables, we expound a novel approach to modelling and forecasting the outputs of the various industries in the economy. This method makes the value-added of each sector dependent on the final demand for the sector’s output, thus creating linkages between the sectoral block on the one hand, and the domestic demand and trade blocks on the other. Since the sectoral block also represents the production side of the economy, its role is to close the ESU01 model by providing a buffer, in the form of unintended inventory investment, to short-term disequilibria between the economy’s aggregate demand and supply.
Introduction 15 Chapter 7 is concerned with a number of ancillary equations and identities in the ESU01 model. Despite being labelled as “ancillary”, these equations are nonetheless crucial to the working of the model in its entirety and involve important policy variables such as taxes and CPF contributions. They include, inter alia, stochastic equations capturing the pass-through effects from the exogenously given NEER and foreign prices to import prices, and henceforward to domestic producer and consumer prices. In the consumer price equation, we incorporated the wellknown Balassa-Samuelson effect operating in the non-tradable goods sector. The definitional identities built into the ESU01 model are also listed in this chapter and the more important ones are explicated to highlight their role in tying the various parts of the model together. The last two chapters of the book turn to an examination of the systemic properties of the ESU01 model. In Chapter 8, we finally solve the different blocks of the model simultaneously and validate it in two ways. Initially, we compute measures of predictive accuracy to formally evaluate the in-sample tracking ability of the model. Following this, impact and dynamic multipliers are derived for unit shocks to foreign income, chip sales and government spending variables. By giving us an idea of the empirical effects of external and fiscal shocks on aggregate output and other variables, the multiplier estimates can provide guidance on the formulation and implementation of macroeconomic stabilization policy in Singapore. Finally, Chapter 9 focuses on hypothetical and counterfactual simulations of the ESU01 model. We pose interesting scenarios such as: what would happen to the Singapore economy if the floodgates to foreign workers were to be closed and if exchange rate policy had followed a binding rule instead of a discretionary approach? The quantitative answers given to these questions by the model illustrate an important prescriptive use of macroeconometric models in practice – the production of economic inputs for policymakers. The book concludes with an optimal control exercise that provides solutions to key exogenous and policy variables that are required to ensure that a certain GDP growth target is achieved. Appendix A details the computational methods of some of the time series used in the estimation of the ESU01 model. Appendix B provides a complete equation and variable listing of the estimated short-run and long-run relationships found in the book.
2
2.1
The aggregate consumption function
Introduction
The aggregate consumption function is perhaps the most basic building block of a macroeconometric model.1 It is also the logical starting point of Keynes’ theory of effective demand, which manifests itself in households spending their income on consumption of goods and services. Together with gross investment expenditures, private consumption spending contributes to domestic demand in Singapore and an understanding of the factors that affect it is essential for both analytical and prediction purposes. Empirical consumption functions estimated for most countries have traditionally employed a log-linear functional specification with measures of income and wealth as basic explanatory variables. In simple consumption functions without a wealth variable, a constant long-run average propensity to consume (APC) requires the income elasticity of consumption expenditures to be unity. However, the handful of researchers who have estimated such consumption functions for Singapore in the context of macroeconometric models obtained elasticities that are well below one, thus implicitly allowing the APC to fall as income increases2 (Wong, 1974; Lim and Associates, 1988, Chapter 16; Toh and Ramstetter, 1994). We have dubbed this curious phenomenon of a falling APC the “aggregate consumption puzzle” in Singapore (Abeysinghe and Choy, 2004). Our aim in this chapter is to resolve the puzzle and at the same time, present the equilibrium and short-run consumption functions that are built into the ESU01 model.
2.2
The consumption puzzle
A rough way to calculate the APC is to take the ratio of real private consumption expenditures to real GDP. In Singapore, this ratio has fallen steadily over time from 0.80 in 1960 to 0.42 in 2003 (Figure 2.1), producing the lowest share of
1 Hadjimatheou (1987) provides a comprehensive survey of the major developments pertaining to the consumption function. 2 Given the simple consumption function C = AY α where C is consumption and Y is income, the APC is C/Y = A(1/Y 1−α ). If 0 < α < 1, the APC will fall as income rises.
The aggregate consumption function 17
Figure 2.1 The APC in Singapore.
private consumption in output in the free world. Even amongst the former centrally planned economies, Ermisch and Huff (1999) noted that the lowest share of private consumption reached in the Soviet Union was 55%. The steady decline in Singapore’s APC is puzzling and anomalous because since Simon Kuznets’ (1946) pioneering work, it has been observed that the APC is relatively stable in the long run. The post-war evidence for selected economies in the world bear out this remarkable fact: Table 2.1 shows that the shares of private consumption in GDP of the United States and United Kingdom have hovered around two-thirds to three-quarters from 1960 to 2000. Switzerland, a mature industrialized economy, is perhaps the best comparison with Singapore in view of its smallness and openness. As Singapore reaches a stage of development that is comparable to Switzerland’s, one would have expected her APC to stabilize. This, however, has not happened thus far. In Asia, Japan saw its APC fall during the 1960s and 1970s but the ratio has stabilized since 1980; this trend was mimicked by Taiwan and Hong Kong. The initial Table 2.1 APC in selected economies (real consumption/GDP ratio)
Singapore USA UK Japan Taiwan Hong Kong Switzerland
1960
1970
1980
1990
2000
0.80 0.68 0.73 0.60 0.64 0.69 0.57
0.61 0.69 0.69 0.53 0.62 0.65 0.55
0.52 0.66 0.72 0.61 0.57 0.64 0.60
0.46 0.68 0.75 0.58 0.61 0.65 0.57
0.41 0.68 0.77 0.61 0.60 0.64 0.58
Sources: Singapore – authors’ calculation; Taiwan in 2000 – Bureau of Statistics; other countries – Penn World Table 6.1.
18
The aggregate consumption function
declines experienced by these countries are likely to have been caused by demographic shifts affecting their working populations – an explanation emphasized by the proponents of the life-cycle theory (Ando and Modigliani, 1963). Demographic change has in fact been cited as an important factor explaining the upward trend in Singapore’s private saving rate and hence the downward trend in the APC, since consumption is the flip side of savings (Monetary Authority of Singapore, 1993; Husain, 1995). It was argued that a significant fall in the number of young and old dependents in the population led to a higher savings rate. However, official statistics suggest that if this had any role to play at all in explaining the APC’s secular decline, it cannot extend beyond the mid-1980s, when the ratio of the working-age population to the total population stabilized at 70%. Yet the APC has continued its precipitous fall during the decade of the 1990s. Some writers such as Husain (1995) have argued that another factor behind the dramatic increase in the private saving rate and by implication, the APC’s decline, is the strong growth in per capita income during the last three decades.3 In view of Kuznets’ findings, however, this argument holds water only if we restrict the time period to the “short run”. Regardless of how one defines that, we think the 40 years over which the APC has been halved in Singapore is too long an interval to satisfy the requirement. Ermisch and Huff (1999) claimed in a provocative article that forced saving in Singapore, extracted through compulsory CPF contributions levied on companies and workers and the manipulation of the internal terms of trade by the major statutory boards providing utilities and telecommunication services, have been responsible for the high savings rate and hence for “spectacular drops in consumption as a share of GDP” (p. 30). However, it will do well to note that private consumption as a share of disposable income, and not just GDP, has also exhibited a pronounced downward trend (see the next section). This suggests that rising CPF contributions and increases in government fees and charges, all of which were deducted to arrive at disposable income, could not have been the reasons behind the APC’s decline. By the same token, high personal income taxes are not a feasible explanation; as a matter of fact, marginal tax rates have been progressively reduced over the years. The Singapore authorities have maintained that household savings are not excessively high and that there is no consumption puzzle in Singapore (Tan and Thia, 2004a). Using unpublished data on personal household disposable income (PDI), they argue that the APC, defined as the ratio of private consumption expenditures to PDI, is only slightly lower than those in other countries. According to them, what explains the low consumption-GDP ratio is the low share of PDI in GDP due to the large foreign participation in the Singapore economy. Even though there must be some truth in these arguments, the official study did not address the time
3 Chapter 4 of Peebles and Wilson (2002) contains a thorough review of the literature on savings in Singapore.
The aggregate consumption function 19 series behaviour of consumption directly. It is the secular decline in the APC, rather than its level, that constitutes the consumption puzzle in Singapore.
2.3
Traditional consumption functions
Historically, the life-cycle and permanent income theories of consumption were formulated by economists to explain the stylized fact of a constant APC, as the simple Keynesian consumption function had implied that the APC will decrease over time with increases in income. Essentially, these explanations pointed out the key role played by wealth and other human assets in maintaining a stable ratio of consumption to GDP in the long run. Previous studies of the Singapore consumption function have excluded a proper wealth variable, casting doubt on the veracity of the estimated income elasticities. As the first step in modelling the consumption function, we shall re-examine the traditional relationship between aggregate private consumption, disposable income and wealth. Singapore’s statistical data on private consumption expenditures are not broken down into spending on consumer durables, non-durable goods, and services; hence, we are forced to model aggregate expenditures without differentiating between these major categories.4 Moreover, official time series for disposable income and wealth are not made publicly available in Singapore and therefore had to be constructed. The sources of data and methods of construction are described in Appendix A. Figure 2.2 plots real private consumption expenditures, real GDP
Figure 2.2 Private consumption, real GDP and disposable income.
4 Another drawback of using total consumption is that it includes expenditures on housing services in the form of imputed rents on owner-occupied dwellings, so that an increase in house prices will cause consumption to rise independently of any wealth effect.
20
The aggregate consumption function
and disposable income using data beginning in 1978 and ending in 2003. Even though the business cycles found in real GDP and disposable income over this period are also evident in private consumption, the paths of these variables have tended to diverge in the long run, reflecting the phenomenon of a falling APC in Singapore. The nominal wealth variable that we constructed for Singapore has two major components, gross housing wealth (HWN ) and financial wealth (FWN ). These were deflated by the Consumer Price Index (CPI) to arrive at their real counterparts, HW and FW. Figure 2.3 plots each of these series and their sum W . The figure shows that household wealth grew relatively slowly in the 1980s, picked up sharply in the 1990s with regional economic prosperity, and then fell during the Asian financial crisis of 1997–98 as stock and property prices deflated. In fact, house prices had started to head down after the Singapore government introduced measures to curb rampant property speculation in May 1996. Aggregate wealth rebounded strongly after the crisis, only to decline again with the onset of economic recession in 2001. Using the data plotted in Figures 2.2 and 2.3 for the period 1978Q1–2003Q4, ordinary least squares (OLS) log-linear regressions of aggregate consumption expenditure on disposable income (Yd ) and the different wealth variables are reported below: ln Ct = 1.06 + 0.64 ln Ydt + 0.14 ln Wt (11.9) (13.7) (3.86) R 2 = 0.988
SE = 0.049
DW = 0.28
ADF = −2.21
Figure 2.3 Measures of household wealth in Singapore.
(2.1)
The aggregate consumption function 21 ln Ct = 1.22 + 0.78 ln Ydt + 0.03 ln HWt (12.6) (26.8) (1.63) R 2 = 0.987
SE = 0.052
DW = 0.29
ADF = −1.96
ln Ct = 0.88 + 0.49 ln Ydt + 0.28 ln FWt (9.54) (8.44) (5.85) R 2 = 0.990
SE = 0.046
DW = 0.34
(2.2)
(2.3)
ADF = −2.93
The numbers in parentheses are the t-statistics, DW is the (co-integrating regression) Durbin-Watson statistic and ADF is the residual-based augmented Dickey-Fuller test statistic for co-integration computed from an AR(2) regression. The 5% critical value for the ADF test is −3.84 (MacKinnon, 1991). The estimated income elasticities seem to be rather low and vary substantially depending on the wealth variable employed. If regression (2.1) with aggregate wealth is used, the income elasticity of 0.64 translates into a marginal propensity to consume (MPC) of about 35¢ out of an additional $1 of disposable income. The implied marginal propensity to consume with respect to wealth is even smaller, at merely 3¢ per $100 of wealth. The reason for this low estimate is found in (2.2): the indivisibility of the housing stock and limited avenues for realizing capital gains in the local property market mean that consumption expenditure is largely unaffected by housing wealth. In contrast, financial wealth in regression (2.3) exerts a far greater influence on consumption, though the results are far from satisfactory. None of the above regressions represent co-integrating relations, implying that the OLS estimates are inconsistent. When we imposed the homogeneity restriction that the sum of the income and wealth elasticities equals unity, the resulting logarithmic regression of the consumption-income ratio on the wealth-income ratio not only failed to co-integrate, but also produced a negative wealth coefficient in every case. This finding is yet another manifestation of the non-stationary APC ratio. The absence of co-integration between consumption and income even after allowing for wealth suggests that textbook macroeconomic theories of consumption cannot provide an explanation for the anomalous spending behaviour of Singaporeans. A potentially important determinant of savings behaviour is the rate of interest (Blinder and Deaton, 1985). In theory, the after-tax real interest rate is the relative price that influences inter-temporal substitution between present and future consumption. Hence, consumption is expected to be negatively related to movements in the real interest rate. Deaton (1977) has also hypothesized that unexpected inflation could lead to decreases in consumer spending because households mistake nominal price increases for real price increases i.e. an absolute rise in the price of all goods and services is confused with a relative price rise for the good that the consumer is considering buying.
22
The aggregate consumption function
Since it is difficult to calculate the effective tax rate on interest income in Singapore, we investigated the sensitivity of consumption to changes in the pre-tax real interest rate. The latter is computed as the difference between the prime lending rate and the annual rate of change of the CPI – a crude, albeit convenient, proxy for expected inflation. We simultaneously perform a simple test of the “Deaton effect” by adding both the real interest rate and the actual inflation rate as arguments to the traditional consumption function. Even though the real interest rate appears to be significant, inflation clearly is not, thus rejecting Deaton’s hypothesis. Furthermore, no evidence of co-integration could be found.
2.4
Explaining the puzzle
The analyses in the previous section suggest that the quest for an explanation of the APC’s long-term decline in Singapore must extend beyond the traditional variables used in consumption functions. It requires an examination of the peculiar circumstances in which Singaporean households find themselves, in particular the dramatic increases in house and car prices seen in the 1980s and the first half of the 1990s. In those days, even the most affordable public apartments in Singapore could cost five to 10 times the average annual household income while car prices stood as the highest in the world. House and car price inflation in Singapore can in turn be traced to the limited land space, the rising aspirations of the population to upgrade to better housing, and the demand for cars outstripping the supply of quotas for car ownership, which drove up the prices of the Certificates of Entitlements (COE) needed for purchasing automobiles. Since house prices appear to be the key to unlocking the APC puzzle in Singapore, it is worth looking into their links with consumption expenditures on non-housing goods and services in detail. Figure 2.4 shows this nexus. A young couple in Singapore typically starts a family by initially buying a medium-sized Housing and Development Board (HDB) flat and then upgrading later to a larger
Figure 2.4 The nexus between house prices, financial assets and consumption.
The aggregate consumption function 23 apartment or to private housing as their income and financial assets grow.5 Even though housing wealth grows in line with increases in house prices, the savings locked up in a house cannot be easily converted into the consumption of nonhousing goods and services. Coupled with a lack of financial instruments like reverse mortgages in Singapore, this means that housing assets are effectively illiquid (as confirmed by our regression results). For this reason, we have crossed the link from these assets to consumption in Figure 2.4. Unlike in bigger countries, there are also no cheap suburbs in Singapore where one could buy a similar or better house and enjoy the capital gains from the sale of the existing one. The sole option available to Singaporeans, apart from emigration to other countries, is to downgrade to a smaller unit. Direct downgrading from private housing to HDB apartments appears to be uncommon because of psychological resistance and a segmented housing market, especially in terms of the amenities provided by private estates. However, downgrading does take place in a more subtle way. The fact that about 85% of elderly parents live with their children suggests that parents might move into a smaller flat with their children in their old age. In this case, the parental house or the proceeds from its sale are passed on to descendants in the form of bequests. As house prices go up, the increase in the value of housing assets is accompanied by a concurrent rise in the financial liabilities of households, in the form of higher down payments for the purchase of residential properties and burgeoning housing loans. Due to the limited avenues for liquidating property assets, households have to build up sufficient financial assets to smooth the profiles of their lifetime consumption of non-housing goods and services. In Figure 2.4, the dotted arrow from housing loans to financial assets shows this indirect link. While an increase in household financial wealth is likely to have a positive impact on consumption, part of the build-up in financial assets occurs because of the illiquid nature of house assets. The implication is that, as house prices rise over time in Singapore, the consumption profiles chosen by later generations of households do not increase as fast as their income. Therefore, one has to control for this effect in a regression to avoid the problem of omitted variable bias. The negative effect on consumption of an increase in house prices working through the loans channel is also indicated in Figure 2.4. There is some empirical evidence for this direct link: Phang (2004) found that anticipated house price increases have a dampening effect on aggregate consumption in Singapore, although the impact is statistically insignificant. She attributes the finding to what she called the “negative wealth effect” of price increases on those seeking to enter the housing market or to upgrade to better housing (Ludwig and Sløk, 2002 refer
5 The HDB is the statutory board responsible for the provision of subsidized public housing in Singapore. About four-fifths of Singapore residents live in HDB flats and the rest in private housing (condominiums, semi-detached houses and bungalows). Singapore citizens are eligible to buy new HDB flats at subsidized rates from the government though in the past, this usually entailed a long waiting period.
24
The aggregate consumption function
to this as a “substitution effect”). Since both wealth and substitution effects have to be non-negative for a normal good and for lack of a better name, we have called the negative impact of a rise in house prices the “price effect”. The price effect is likely to be well captured by a loans variable. Unfortunately, a time series on housing loans is not available in Singapore. A composite variable made up of loans for buildings, construction and housing was found to be less appropriate. The best proxy for housing loans we can find is withdrawals from the CPF to finance the purchases of houses and mortgages (CPFHOUSEN ).6 To capture other liabilities, including car loans, we used data on bank lending to professional and private individuals (BANK ). The two loan proxies, after deflating by the CPI, are found in preliminary investigations to have negative effects on consumption which are roughly equal in magnitude. We therefore combined them to form a total loans variable (LOAN ). Apart from disposable income, financial wealth and loans, we needed another variable to fully explain the fall in the APC. This variable, again peculiar to Singapore, is total visitor expenditures in real terms (V ).7 In principle, visitor expenditures should be irrelevant as private consumption is defined as expenditures by the resident population on the purchases of final goods and services. It is obtained by netting out the expenditures of tourists from the estimates of total consumption in the domestic market. However, the fact that the number of visitors who visit Singapore each year is nearly twice the size of the resident population means that any errors-in-variables problem created by visitor spending would be amplified.
2.5
The Singaporean consumption function
The long-run regression that we estimate using the variables discussed above is as follows: ln
Ct FWt LOANt Vt = β0 + β2 ln + β3 ln + β4 ln + εt Ydt Ydt Ydt Ydt
(2.4)
with an implied income elasticity of β1 = 1 − β2 − β3 − β4 . This regression has a nice interpretation – the dependent variable is the APC while the independent
6 Withdrawals from the CPF to pay for publicly built flats have been allowed since 1968; in 1981, the rules were liberalized to cover private residential properties. The increase in such withdrawals have been particularly rapid since the early 1980s, having grown from 2.8% of disposable income in 1977 to 6.4% by 2000; as a proportion of CPF contributions, they have averaged about 40% during the 1990s. 7 Total visitor expenditures are proxied by the total number of visitor days spent by tourists in Singapore. We resorted to this measure because expenditure data are only available on an annual basis from 1985 and interpolation of these annual statistics to obtain quarterly figures may introduce further measurement errors. Over the period concerned, average visitor expenditure per day has gone up from $190 in 1985 to $314 in 1990 and then fallen steadily to $203 in 1999. Somewhat reluctantly, we have to ignore these fluctuations.
The aggregate consumption function 25 variables are the ratios of financial wealth, loans and visitor expenditures to disposable income. If the specification were to lead to a co-integrating regression, we would have obtained a constant APC in the long run or put differently, a unitary income elasticity conditional on constant wealth-income, loan-income and visitor expenditure-income ratios. The regression results for Equation (2.4) are shown in column (3) of Table 2.2 (the estimate of the constant term is not reported). The residual-based ADF t-statistic of –4.68 is significant at the 1% level according to the MacKinnon (1991) critical value; hence, this test renders strong support for co-integration. For comparison, we show in columns (1) and (2) regressions that either include the housing wealth variable or omit the visitor expenditures variable. The insignificance of the housing wealth coefficient reinforces the earlier regression findings and vindicates our perception that housing assets in Singapore are illiquid. The ADF t-statistics associated with these two regressions indicate that they do not represent co-integrating relationships. The fact that visitor expenditures turned out to be critical for co-integration in column (3) points to a residual element of tourist spending in the aggregate consumption data. To check the robustness of our results, we turned to the Johansen ML procedure for testing co-integration. The trace test, based on a VAR(2) specification of the four variables in Equation (2.4), supports the existence of a single co-integrating vector. The last column of Table 2.2 shows that the estimated long-run coefficients in the co-integrating vector are not too different from those in
Table 2.2 APC regression results OLS (1) Financial wealth Housing wealth Loans Visitor expenditures R2 SE DW ADF
0.09 (1.99) 0.02 (0.88) −0.18 (−7.74) 0.18 (7.91) 0.868 0.041 0.74 −3.44
DOLS (2) 0.13 (2.38) – −0.26 (−13.7) – 0.741 0.057 0.42 −2.51
Johansen
(3) 0.11 (2.81) –
0.16 (2.25) –
0.17 (2.11) –
−0.17 (−9.94) 0.19 (9.72)
−0.22 (−5.71) 0.19 (6.50)
−0.17 (−5.04) 0.16 (4.12)
0.867 0.041 0.73 −4.68*
– – – –
– – – –
Notes: The figures in parentheses are t-statistics; the corrected statistics for the DOLS estimates are computed from an AR(1) regression on the residuals. The ADF statistics in (1) and (2) are based on AR(2) regressions of residuals and the one in (3) is based on an AR(1) regression. *indicates significance at the 1% level.
26
The aggregate consumption function
the OLS regression. Further testing shows that the wealth, loan and visitor expenditure ratios are weakly exogenous for the parameters of interest. This offers a justification for estimating a single equation dynamic consumption function below. Lastly, in order to correct for possible endogeneity and measurement error biases of the OLS estimates and their standard errors in small samples, we employed the dynamic OLS (DOLS) procedure developed by Stock and Watson (1993) to re-estimate Equation (2.4) (see Hayashi, 2000, pp. 654–7). We set the lead and lag lengths in the dynamic regression to four quarters, corresponding roughly to the T 1/3 rule, and ran an AR(1) regression on the residuals to compute the longrun variance. Table 2.2 shows that the DOLS coefficient estimates on financial wealth and loans are larger than those in the OLS case and statistically significant according to the corrected t-statistics. The income elasticity of consumption implied by the DOLS estimates is 0.87, somewhat lower than the usual elasticities reported in the literature but higher than the estimates from regressions (2.1) to (2.3) and those found by previous Singapore researchers. Using these results, we are able to derive an ex post explanation of the APC’s decline. The estimated long-run elasticities suggest that a 1% increase in the financial wealth ratio raises the APC by 0.16%, a similar increase in the loan ratio reduces it by 0.22%, while a 1% decrease in the visitor expenditure ratio reduces it by 0.19%. Perhaps not surprisingly, the price effect dominates the wealth effect, thereby explaining the secular fall in the APC. Even if the wealth and loan coefficients are taken to be of the same magnitude as in the Johansen estimates in Table 2.2, the observed steeper trend in the loan ratio has more than offset the wealth effect and thereby exerted a downward pressure on the APC. Figure 2.5 charts the historical evolution of the three key ratios and their impact on the APC. The superimposed regression trend lines show that the ratio of household financial wealth to disposable income has an upward trend, but this increase in the wealth ratio has not been sufficient to stem the decline in Singapore’s APC, which has been driven by a sharply rising loan ratio and a declining visitor expenditure ratio. Since the number of visitors hosted by Singapore has risen steadily over time, the falling visitor expenditure ratio is a direct reflection of the drop in average visitor expenditures. This structural trend has inadvertently found its way into the official estimates of private consumption, as we explained earlier. However, it is evident from Figure 2.5 that the declining visitor expenditure ratio is not the sole, or even the most critical, reason for the secular decline in the APC. We have thus far concentrated on modelling the long-term behaviour of private consumption expenditures in Singapore, but do short-run fluctuations in consumption also respond to the long-run factors that are found to be important? To answer this question and close our analysis, we present below the ECM estimated for aggregate consumption over 1978Q1–2003Q4, in which the error-correction term denoted by EC is simply the residuals computed from the DOLS estimates in
The aggregate consumption function 27
Figure 2.5 The APC and key ratios.
Table 2.2 without the constant term: ln Ct = −0.03 −0.26 ln Ct−1 + 0.25 ln Ydt + 0.08 ln FWt (4.05) (2.36) (−4.12) (−3.15) −0.03 ln CPFHOUSEt−1 − 0.24 ECt−1 (−5.94) (−2.06)
(2.5)
R2 = 0.377 SE = 0.018 DW = 1.96 LM(5) = 0.41 (0.84) ARCH(4) = 1.25 (0.30) Chow = 0.61 (0.83) Normality = 0.73 (0.70) Heteroscedasticity = 1.05 (0.41) RESET = 0.30 (0.59) In this regression, we tried the total loans variable and its components (CPF housing withdrawals and loans to individuals) separately, but only CPF housing withdrawals turned out to be statistically significant. Unlike consumption functions estimated for most countries, Singapore’s consumption expenditure is subject to more short-term variation as reflected by the negative coefficient on ln Ct−1 . The estimated error-correction coefficient implies a moderate pace of adjustment to the long run, with about 24% of the short-term disequilibrium in last period’s consumption expenditures being eliminated in the current quarter. The diagnostic
28
The aggregate consumption function
Figure 2.6 Forecasts of consumption. Table 2.3 Dynamic elasticities for consumption Lag (quarters)
Yd
FW
CPFHOUSE
V
0 4 8 12
0.25 0.66 0.85 0.93
0.08 0.09 0.10 0.11
0 −0.05 −0.08 −0.10
0 0.11 0.15 0.17
test statistics (p-values in parentheses) indicate that the ECM formulation is satisfactory while the recursive parameter estimates of the model are remarkably stable. Finally, Figure 2.6 shows that the ex ante predictions of consumption for the last 3 years of the sample are able to capture actual trends in spending, including the declines registered in 2003 as a result of the loss in confidence caused by the US-Iraq war and the outbreak of the SARS disease. We use the estimated ECM to compute the first set of dynamic elasticities in this book for consumption and report them in Table 2.3. A 1% increase in disposable income raises consumption expenditures by a quarter of a percent instantaneously and two-thirds of a percent in a year’s time. This should be contrasted with the impact of visitor expenditures, which is discernable only after a year. Similarly, the price effect as proxied by housing withdrawals is more of a medium-term to long-run effect.
2.6
Policy recommendations
We conclude this chapter by making some policy recommendations. First of all, it should be pointed out that due to “measurement error” caused by tourist spending,
The aggregate consumption function 29 the APC in Singapore is underestimated. Even though we do not want to engage in second guessing the published data, we could not resist performing the following counterfactual experiment: how much higher would the APC have been if the residual effect of visitor expenditures was purged from the data? The easiest way to answer the question is to divide Ct /Ydt by (Vt /Ydt )0.19 to obtain an adjusted APC. However, the series adjusted in this way needs to be rescaled and we did this by setting the adjusted APC to be the same as the unadjusted one in the first quarter of 1978. Figure 2.7 plots both the unadjusted and adjusted APCs (4-quarter moving averages are used to smooth out the fluctuations). The discrepancy between the two average propensities to consume is small in the early years but it widened in the last decade. It appears therefore that the decline in the visitor expenditures to income ratio has been a particularly important reason behind the fall in the measured APC during the 1990s. Why should the falling APC be a cause for concern? It is a fact in most countries, Singapore not excluded, that private consumption expenditures constitute the most stable component of the final demand for goods and services. With a declining APC, the more variable components in aggregate demand such as investment and exports will come to dominate cyclical fluctuations, as the following two chapters show. In particular, being a small open economy, Singapore is highly vulnerable to external shocks affecting her domestic exports and resulting in more volatile GDP growth. During such times, measures to stimulate consumer spending will not be very potent given that private consumption expenditures currently account for less than half of GDP. In other words, consumption spending cannot serve as a built-in automatic stabilizer for the economy unless the government takes countervailing measures to raise the share of such expenditures in output.
Figure 2.7 The adjusted APC. Note: Lines are centred moving averages over 4 quarters.
30
The aggregate consumption function
The main result of this chapter is that rising loans and withdrawals from the CPF to finance house and car purchases have played a crucial role in the evolution of Singapore’s APC. It is observable that a fall in house prices is accompanied by a rise in APC with a lag of 4–5 years. This leads us to an important policy recommendation: to prevent further declines in the APC, the government should ensure that residential property and other household assets remain affordable. This it could do through direct policy measures or indirectly, through its vast ownership of land in Singapore. At the least, the government should try its best to prevent a recurrence of the bubble-like increases in residential property prices experienced in the early 1980s and the 1990s, periods during which the APC lost much ground. More specifically, we recommend that any increase in property prices that exceeds the trend growth rate of disposable income be mitigated by policy interventions. If the wealth-income ratio grows faster than the loan-income ratio, the adjusted APC should eventually recover to levels that could support a more stable GDP growth rate. This observation suggests another policy action: the government, working hand-in-hand with the private sector, should seek to introduce new avenues for disposing of currently illiquid housing wealth, such as reverse mortgages, and to make it feasible to downsize to smaller but nonetheless attractive apartments within the physical confines of Singapore. The availability of these extra options would go a long way towards resolving the “asset-rich, cash-poor” syndrome that apparently blights many Singaporeans.
3
3.1
Modelling investment expenditures
Introduction There is no generally agreed theoretical framework for analysing and explaining long-run trends in capital accumulation …. unfortunately the theory of the investment function, especially in the long run, is not one of the strongest points in economic theory. Mathews, 1982, pp. 326–7; quoted by Scott, 1989, p. 216
Investment spending is the source of fixed capital formation. Physical capital with embodied technology and human capital with embodied technical know-how are the keys to Singapore’s phenomenal ascendance to high-income status within such a short time-span. Investment expenditures accounted for 26.7% of real GDP in 2003, compared to an average of 34% during the 1990s. Even though this is lower than private consumption expenditures, gross fixed capital formation (GFCF) is far more volatile. As Figure 3.1 shows, the variability of the growth rates of consumption expenditure gravitated in the neighbourhood of 5% whereas that of investment spending has overshot 15% at times. These fluctuations, arising from less predictable factors, make it a daunting task to find investment equations that provide a good fit to the historical data and also forecast well. To some extent, the problem lies in economic theory itself – the quotation above succinctly highlights the unsettled state of investment modelling even today. Despite this, a voluminous literature exists on the specification and estimation of aggregate investment functions (refer to Chirinko, 1993 and Kopcke, 1993 for surveys). We do not dwell on these conventional investment models because Singapore’s unique economic circumstances – in particular, the heavy reliance on foreign direct investment (FDI) – requires a different approach. Our modelling efforts in this chapter are greatly facilitated by the availability of data on two forward-looking variables, net investment commitments in the manufacturing sector and building contracts awarded in the construction sector. Since fixed capital formation is predominantly driven by expected demand and profitability, the use of these expectations-laden variables simplifies the task of incorporating agents’ anticipations into investment equations for machinery and transport equipment, and construction. In addition, even though these two variables are essentially
32
Modelling investment expenditures
Figure 3.1 Variability of consumption and investment growth (%). Note: Lines are centred rolling standard deviations over 16 quarters.
exogenous in the ESU01 model, we also present single equation models for them to enhance our understanding of their determinants and to generate forecasts for them if the need should arise.
3.2
Net investment commitments and international competitiveness
Net investment commitments in the manufacturing sector (NIC) refer to actions taken by firms “to begin work on the ground e.g. [a] company has entered into a contract for factory construction or purchase of machinery” (Singapore Economic Development Board website, 2005). In 2003, such commitments amounted to $7.51 billion, with the lion’s share going to the electronics sector, followed by the petrochemical and biomedical industries. Reflecting the strong foreign participation in the economy, 83% of commitments were made by outside investors, mostly from the USA, Europe and Japan. While investment decisions in the real world depend on a myriad of tangible and intangible factors – including political events – and are subject to waves of optimism and pessimism, it is intuitively evident that the readiness of investors to pour money into a country is crucially affected by its international competitiveness. How to measure international competitiveness is more contentious because it may mean different things to different people. Boltho (1996), who provided an assessment of seven papers contributed to the Oxford Review of Economic Policy (Vol. 12, No. 3, 1996) on the subject of international competitiveness, points out that in the short run, international competitiveness may be measured by movements of the real exchange rate while in the longer run, what matters is productivity growth.
Modelling investment expenditures
33
Common measures of the real exchange rate are relative prices or relative costs of production expressed in the same currency. Relative CPI (RCPI) is the most popular indicator primarily because of the ready availability of data. However, the presence of too many non-traded goods and services in CPI is regarded as a drawback of RCPI as a measure of international competitiveness. A widely preferred measure of the real exchange rate is relative unit labour costs (RULC). If RULC is expressed in US dollars, then the rate of change of the RULC of country i against country j can be written as: ˙i −W ˙ j ) − (LP˙ i − LP˙ j ) − (˙ei − e˙ j ) RU˙ LCi = (W where a dot over the variable indicates its growth rate, W is the nominal wage rate of country i or j in the national currency, LP is labour productivity and e is the exchange rate expressed in domestic currency units per US dollar. In practice, short-run fluctuations in RULC tend to be dominated by movements in the nominal exchange rate and wages, while the long-term trend is essentially determined by productivity growth. The RULC, therefore, is designed to capture explicitly the short-run and long-run aspects of international competitiveness that Boltho (1996) was alluding to. Despite the desirable qualities of RULC, it is hardly used in empirical work because of a lack of data. Using interpolation techniques, however, we were able to compile the RULCs for the NIE4 and ASEAN4 countries in Table 3.1. As Abeysinghe (2001a) shows, a plot of the RCPIs and RULCs for these eight countries reveals that the two alternative measures of competitiveness share common turning points. However, they tell very different stories depending on which time period one is focusing on. To provide a further grasp of the two measures, Table 3.1 shows the appreciation rates over two periods: the pre-Asian financial crisis period 1993Q1–1997Q2 and the crisis period 1997Q3–1998Q2. The figures indicate that, according to RULC, all the NIE4 countries gained economic competitiveness during the pre-crisis period at the expense of the ASEAN4 countries. The RCPI tells a different and mixed story. During the crisis period,
Table 3.1 Average annual rate of change of RULC and RCPI ASEAN4
NIE4
Malaysia Indonesia Thailand Philippines Singapore HK 93Q1–97Q2 RULC RCPI
S. Korea Taiwan
8.0 −0.7
0.9 0.3
3.0 −0.4
5.6 2.4
−2.5 0.3
−3.4 3.0
−0.4 −2.3
−4.4 −5.1
97Q3–98Q2 RULC −6.7 RCPI −16.4
−37.0 −41.6
−16.3 −18.1
−9.3 −9.2
15.3 20.2
47.0 −27.3 39.8 −13.6
0.3 −2.5
Note: Both RCPI and RULC are export-weighted geometric averages against other countries.
34
Modelling investment expenditures
both indicators were roughly in agreement although the extent of appreciation or depreciation suggested by them could vary substantially. For example, the RCPI indicates a depreciation of about 16% for Malaysia whereas the RULC shows a much milder depreciation of about 7%. The opposite is true for South Korea. For Singapore, however, both measures unambiguously point to a loss in competitiveness vis-à-vis its regional rivals (except for Hong Kong) as a result of the sharp devaluation of the latter’s currencies. This happened even as the MAS allowed the Singapore dollar to depreciate against the US dollar after the financial crisis broke out.
3.3
Modelling investment commitments
For a land-scarce country like Singapore, rentals constitute an important expense item, in addition to the wage bill and the costs of internationally mobile factors of production such as capital and raw materials. This suggests that relative unit business costs (RUBC), which include both labour costs as well as nonlabour costs, may be an even better indicator of international competitiveness than RULC (Abeysinghe, 2001a). Due to the paucity of good quality quarterly data, we proxy RUBC by the ratio of Singapore’s manufacturing UBC index to Malaysia’s manufacturing ULC index, both expressed in local currency. In addition to RUBC, general business conditions can also be expected to have some influence on investment decisions, which may be captured by any one of a number of variables measuring the recent performance of the economy such as GDP, manufacturing output or non-oil domestic exports. It is also important to take into account the composition and quality of FDI. As wage and business costs rose relative to her competitors, Singapore has increasingly attracted higher value-added investments. Within the context of a regression, this effect can be accounted for by including the value-added content of manufacturing output as an additional explanatory variable. Figure 3.2 shows the ratio
Figure 3.2 Value-added content in manufacturing output (%).
Modelling investment expenditures
35
of nominal manufacturing value-added to gross output over the last 30 years. The graph reveals a rather unexpected and interesting observation. Even though the value-added content fluctuates, the annual average between 1975 and 1985 was 23%. Instead of a gradual upward movement, however, the value-added ratio jumps to an average of 26% after the 1985–86 recession. In the aftermath of the Asian financial crisis, there seems to be yet another step increase. In other words, every economic crisis precipitated a restructuring of the economy which in turn resulted in a rise in the value-added contribution of local manufacturing production. Unfortunately, quarterly data on the value-added ratio is lacking and we have therefore left this restructuring effect out. The best-fitting OLS regression model for nominal net investment commitments (NIC N ) turned out to be of the form1 : ln NICtN = β0 + β1 ln RUBCt + β2 ln VAMANt−1 + εt
(3.1)
where VAMANt−1 is real manufacturing output in the previous quarter. Table 3.2 presents the estimates of the parameters appearing in Equation (3.1) for total investment commitments and their foreign/local breakdown during the period 1985Q1–2003Q4 (data on commitments were not collected prior to 1985). In spite of the highly unpredictable nature of investment commitments, the R2 statistics
Table 3.2 Elasticity estimates for NIC in manufacturing
Constant RUBC VAMANt−1 R2 SE DW ADF
Local firms
Foreign firms
All firms
−15.50 (−6.74) −1.47 (−2.00) 2.50 (8.70)
−9.20 (−8.85) −0.90 (−2.69) 1.89 (14.6)
−9.75 (−10.8) −0.92 (−3.20) 1.99 (17.60)
0.66 0.616 1.63 −5.21*
0.86 0.279 2.40 −11.03*
0.90 0.242 1.92 −4.15*
Notes: Variables are in logarithms and figures in parentheses are t-statistics. The ADF test statistics for co-integration are based on AR(1) regressions of residuals in the case of total and local investment commitments; that for foreign commitments does not include any lags. *indicates significance at the 1% level.
1 We included a step dummy variable after 1997Q2 to account for the substantial drop in investment commitments since the outbreak of the Asian financial crisis. The inclusion of the dummy did not significantly alter the values of the other coefficients in the equation but it improves the overall fit.
36
Modelling investment expenditures
show that the regressions explain a substantial proportion of their variation. What is even more remarkable is that these are co-integrating regressions, as indicated by the significant values of the ADF t-ratios. Since the DW statistics in Table 3.2 indicate that no residual autocorrelation is present, the regression estimates can be interpreted as both short-run and longrun elasticities, rendering the estimation of ECMs unnecessary. The elasticity with respect to RUBC shows that Singapore’s business costs relative to her competitors figure prominently in investment decisions. In particular, a 1% increase in this proxy for international competitiveness can deter or drive out 0.9% of foreign commitments and almost 1.5% of local investments. Table 3.2 also shows that the recent performance of the economy is another significant determinant of net investment commitments: a 1% increase in manufacturing output leads to twice as much an increase in overall investment.
3.4
Investment in machinery and transport equipment
In this section, we present econometric equations for the combined total of two major categories of investment spending in Singapore: machinery and equipment, and transport equipment (henceforth, machinery and transport equipment, denoted IMT p ). These two types of investment outlays constituted 40% and 11%, respectively, of real gross fixed capital formation in 2003. We only model private expenditures and treat the public components as exogenous variables in the ESU01 model. Public sector expenditures on the two categories are rather miniscule anyway – about 5% of machinery and equipment expenditures and 2% of transport equipment spending. In Singapore, data constraints not only dictated the level of aggregation of the investment models that can be estimated, but also restricted the choice of variables to use. Previous attempts to model capital spending typically utilized output and cost measures as explanatory variables. An example is the equation estimated by Toh and Low (1990) for the ASEAN Link model (t-statistics are in parentheses): ln IMACPt = 2.44 + 0.63 ln IMACPt−1 + 0.000075 GDPRt (1.91) (3.30) (2.20) + 0.14 (WC/CPI )t−1 (1.44) where IMACP is investment in machinery and equipment (excluding transport equipment), GDPR refers to Singapore’s real GDP and the WC/CPI term represents real wage costs. This equation seems to be based on an accelerator theory of investment, with tagged on dynamics induced by a partial adjustment process. The change in GDP has the right sign and its t-statistic is significant by conventional standards. However, the same cannot be said for real wages, which plays a dubious role in the regression. In any event, its coefficient is statistically insignificant.
Modelling investment expenditures
37
By far the most notable attempt to estimate Singapore investment functions has been made by Tan and Thia (2004b). They put in considerable effort to compute capital stock measures and the user cost of capital (UCC ) for three categories of investment (machinery and equipment, transport equipment, and non-residential buildings) and estimated models in the ECM format using the Hall-Jorgenson (1967) model as a guide. What is rather puzzling about these models is that the dependent variable in their specifications is gross investment, which is typically an I (1) variable. Moreover, the lag structure of their equations seems to be determined by statistical rather than economic significance. Due to these methodological weaknesses, the results they obtained are somewhat questionable. At this stage, what we propose de novo is essentially an empirical model for gross investment that fits the data well, deferring a formal theoretical formulation to the future. The approach is similar in spirit to the standard demand for capital specification with the assumption of an infinitely elastic supply of investment goods, which is not unrealistic for a small open economy like Singapore. Scott (1989, 1992), who is quite critical of this production function-based approach, lays a heavy emphasis on modelling gross investment or its share in output directly. Even though we do not follow Scott’s methodology here, we see a lot of merit in focusing on gross investment because of the almost insurmountable difficulties encountered in the computation of the net capital stock, making econometric implementation of the production function approach less than straightforward. Jorgenson’s (1963) neoclassical capital-stock adjustment model basically relates gross investment (I ) to the change in output (Y ), the change in the user cost of capital (UCC ) and the capital stock in the previous period (Kt−1 ). This specification, however, does not guarantee a balanced regression since the variables involved are of different orders of integration that may or may not reduce to an I (0) linear combination. Tobin’s (1962) “q model” and its variants, on the other hand, relates the gross investment-capital ratio (I /K, the rate of change of the capital stock) to q, the ratio of the market value of capital assets to their replacement value – a forward-looking variable that should embody the unobserved expectations deemed to play a critical role in investment decisions. The model, however, has not shown much empirical success probably because measured q has been a poor proxy for its theoretical counterpart. The empirical model we have in mind is a hybrid of these two types of models. It relates I /Y (instead of I /K) to a variable z to form an I (0) linear combination that can be modelled together with other I (0) variables. The variable z may have to be extracted or constructed from a large vector of variables that influence investment decisions. In this respect, our exercise is greatly simplified by the availability of data on net investment commitments, which are well explained by Singapore’s cost competitiveness and economic performance, as seen in the previous section. Like q, this expectations-laden variable contains information on the long-run determinants of gross investment. Therefore, the ratio IMT p /NIC, where NIC = NIC N /P K is real investment commitments, must form a stationary variable
38
Modelling investment expenditures
in our model.2 Both the Johansen and OLS residual-based tests provide strong support for co-integration between ln IMT p and ln NIC with a co-integrating coefficient that is smaller than unity, reflecting the fact that the data on machinery and transport equipment investment is for the aggregate economy while investment commitments pertain to the manufacturing sector. Since we can find an equivalent transformation of the model to account for the non-unitary coefficient, we proceeded straightaway to formulate an ECM with the IMT p /NIC ratio as the error-correction variable. Before presenting the estimated model, it is worth drawing attention to the user cost of capital variable that we computed. The most common finding in the applied literature is that UCC forms the weakest link in the investment function. As this could partly be a result of data problems, we paid considerable attention to its construction (see Appendix A for details). Figure 3.3 shows UCC and its three major components, UCC r , UCC tax and UCC price , with their base set to 100 in 1980Q1. The long-term trend of the overall UCC over the last two decades has been downward. This is because a continuous appreciation of the Singapore dollar has brought down the relative price of capital (P K ) and hence UCC price ; at the same time, lavish depreciation allowances (in present value terms) and a falling corporate tax rate contributed to a small decline in UCC tax . The real interest component UCC r , on the other hand, is largely stationary but responsible for the volatile fluctuations in the UCC. In particular, the sharp drop of UCC between 1986 and 1987 can be traced to unusual movements in our proxy for P K namely the import price index for machinery and transport equipment. The aberration affected the
Figure 3.3 User cost of capital.
2 Our previous argument about relating I /Y to z implies that IMT p /NIC is the ratio to focus on because both variables have to be rescaled by dividing through with Y .
Modelling investment expenditures
39
estimates of user cost elasticity and led us to use a shorter sample from 1990Q1 to 2003Q4. Based on this sample period, the final model specification and estimates that we arrived at are given below: p
p
ln IMTt = 2.56 −0.23 ln IMTt−1 + 2.10 ln GDPt−1 (3.03) (3.53) (−1.67) −0.28 ln UCCt −0.20 ln (NIC N /P k )t−1 (−1.75) (−3.16) p −0.35 ln IMTt−1 /(NIC N /P k )t−1 (−3.63)
(3.2)
R 2 = 0.40 SE = 0.09 DW = 2.37 LM(5) = 1.73 (0.16) ARCH(4) = 1.87 (0.13) Chow = 0.48 (0.91) Normality = 0.81 (0.67) Heteroscedasticity = 1.32 (0.26) RESET = 0.67 (0.41) The simple and parsimonious model we obtained in (3.2) is quite satisfactory. For a first differences specification and a variable as volatile as investment, the R2 statistic indicates that a reasonably good fit has been obtained. The model passes all the diagnostic tests and the predicted values follow the turning points of IMT p growth rates appreciably well (see Figure 3.4). The recursive parameter estimates are also highly stable. The negative coefficient of the lagged dependent variable highlights the volatility of investment expenditures. Though empirically chosen, it is interesting to note that ln GDP enters the model with a one-quarter lag, indicating the delayed response of investment expenditures to GDP growth. The lag also takes care of the endogeneity issue that is often singled out as a serious problem in estimating investment functions.
Figure 3.4 Forecasts of machinery and transport equipment investment.
40
Modelling investment expenditures
Table 3.3 Dynamic elasticities for investment (IMT p ) Lag (quarters)
NIC N
GDP
UCC
0 1 4 8 12
0.00 0.16 0.33 0.42 0.44
0.00 2.08 0.64 0.22 0.08
−0.28 −0.13 −0.07 −0.02 −0.01
The most noteworthy estimate is the short-run UCC elasticity. Not only does this coefficient carry the correct sign, it also comes with a much larger magnitude than usually observed. Leaving statistical significance to one side (the variable is significant at the 10% but not the 5% level), the magnitude of the coefficient seems to be robust to variations in the sample size. The speed of adjustment to the long run is quite fast, with 35% of the deviation from equilibrium being eliminated in any quarter. The dynamic responses of machinery and transport equipment investment to changes in its determinants are shown in Table 3.3. Several observations are worth highlighting. First and foremost, the effects of an increase in net investment commitments take more than 3 years to materialize fully – an estimate of the gestation lag, or the time that elapses between the initial decision to invest and the subsequent bulge in capital formation. Since Equation (3.2) provides the co-integrating relation ln IMT p − 0.48 ln NIC, a 1% increase in real investment commitments to the manufacturing sector translates approximately into a 0.5% rise in realized investment spending for the whole economy in the long run. By contrast, both GDP and UCC growth capture the shorter-term responses of investors to economic and cost conditions. In line with findings elsewhere (see for example Chirinko, 1993), GDP growth is the stronger determinant of investment spending, with a one percentage point increase in the economic growth rate leading to a 2.1% increase in fixed asset formation after one quarter. The positive effect of GDP growth lasts well over 2 years. In contrast, the response to a change in the user cost of capital is more immediate and short-lived. Overall, these estimates seem quite reasonable and confirm that expectations and output have larger effects on fixed investment in Singapore than the cost of funds, not surprisingly given that the bulk of capital spending is committed and financed by large multinationals and conglomerates.
3.5
Construction investment in Singapore
Investment expenditures on construction and works account for the rest of capital spending in Singapore, or as much as half of GFCF in recent years. More than 80% of the expenditures on construction and works consists of residential and non-residential buildings. Unlike machinery and transport equipment, the government’s share in this category is much larger and has fluctuated substantially over
Modelling investment expenditures
41
Figure 3.5 Public share of construction investment.
the years. The public share of residential building construction peaked at 56% in 1978 and declined to 16% by 2004 while the share of non-residential buildings stood at 41% (Figure 3.5). The figure also shows that activity in the “Other” category, which consists primarily of infrastructure development, is concentrated in public hands. These facts aside, construction investment is worthy of detailed study because of the traditionally important role of the construction sector in Singapore and the dramatic boom-and-bust cycles it has gone through over the last three decades. These include two large housing booms, the first in the early 1980s and the second from 1990 to 1996. As we have seen in the last chapter, these episodes coincided with huge run ups in property prices in Singapore. Since the Asian financial crisis, however, asset prices have retreated, forcing the construction industry to downsize and restructure. The construction investment cycle in Singapore is the product of two distinct forces. The basic cycle is dictated by demand and supply conditions in the private property market. Superimposed onto the fundamental cycle, however, are the pump-priming actions of the Singapore government. As seen in Figure 3.6, public expenditures on construction exhibit a clear counter-cyclical element. Just like government machinery and transport equipment investment, however, we treat public investment spending on construction and works as an exogenous variable in the ESU01 model. In a simplified framework where the stock of existing residential buildings is fixed, real estate prices are primarily determined by demand forces that include demographic factors and household incomes. These prices in turn determine the supply of new dwelling units. However, real estate prices are subject to large disequilibrium forces created by demand-supply imbalances that feed into buyers’ and sellers’ expectations. Similar intricacies arise in the determination of spending
42
Modelling investment expenditures
Figure 3.6 GDP growth cycles and public share of construction.
on non-residential buildings. These intricate dynamics can be captured to a large extent by construction contracts awarded to private firms, which play an analogous role to investment commitments. The availability of data on contracts awarded enables us to model construction investment in a manner similar to machinery and transport equipment spending – we regard the amount of contracts awarded as a forward-looking variable that can potentially explain construction investment expenditures. It must be remarked that this sequential approach to modelling Singapore’s construction activity is not novel. Lim and Associates (1988, Chapter 16) adopted such a strategy in the first ESU model of the Singapore economy, except that they employed new building commencements as the anticipations variable instead of contract awards. In a similar vein, Chow and Choy (1995) used building commencements as a leading indicator to predict the growth of construction output. By contrast, Abeysinghe and Wilson (2001) used contracts awarded to forecast construction value-added based on a transfer function model. As in the case of net investment commitments, we first present a model for nominal construction contracts awarded by the private sector (CAN ), and then model the dynamic relationship between construction investment spending and contracts awarded. Our arguments above show that the current value of buildings should be a good predictor of contracts given out and hence, of construction investment. The relevant price variable in this case is an index of the average prices of residential and office properties in Singapore (PPI ).3 Another predictor
3 We experimented with the prices of retail and industrial properties but these turned out not to be useful for explaining contracts awarded. Besides, they are highly collinear with the price of office properties.
Modelling investment expenditures
43
variable we consider is the cost of funds, proxied by the prime lending rate (PLR) of the major banks. The forecasting equation for CAN estimated using data over the period 1981Q1–2003Q4 is: N ln CAN t = −0.03 −0.41 ln CAt−1 + 2.0 ln PPIt (2.84) (−0.79) (−4.45)
+ 0.23 ln PPIt−1 + 1.27 ln PPIt−2 −0.24 PLRt−1 (0.29) (2.19) (−3.86) (3.3) R 2 = 0.364 SE = 0.298 DW = 2.16 LM(5) = 0.52 (0.76) ARCH(4) = 1.11 (0.36) Chow = 0.56 (0.87) Normality = 1.73 (0.42) Heteroscedasticity = 0.88 (0.56) RESET = 0.40 (0.53) The signs and magnitudes of the estimated coefficients are not far from our prior beliefs. Property prices affect construction investment decisions both contemporaneously and with lags of up to 6 months, reflecting the immediate and delayed responses of building developers to a rise in valuations. Like machinery and equipment investment, the nominal value of contracts awarded is curtailed by an increase in bank borrowing costs in the previous period. We surmise that an increase in the nominal interest rate exerts a dual impact on the property market in Singapore: by raising mortgage costs for potential house buyers and construction costs for property developers, it acts to dampen demand and supply simultaneously. To explain private construction investment (ICON p ), we first deflate contracts awarded by the building materials price index (P bm ) into a series expressed in real terms. A data plot shows the close co-movement of CAN /P bm and ICON p , with the former leading the latter. Johansen’s trace test supplies strong evidence of co-integration between these series and ML estimation produces a co-integrating coefficient of less than one. We therefore fitted an ECM for construction investment which is similar to Equation (3.2), with the ICON p /(CAN /P bm ) ratio acting as the error-correction term: p ln ICONt = 0.513 + 0.045 ln CAN /P bm t − 0.038 ln CAN /P bm t−1 (4.82) (2.55) (−1.88) N bm − 0.052 ln CA /P t−1 (−3.97) p (3.4) − 0.155 ln ICONt−1 /(CAN /P bm )t−1 (−7.38) R 2 = 0.467 SE = 0.058 DW = 2.14 LM(5) = 1.24 (0.30) ARCH(4) = 0.21 (0.93) Chow = 0.50 (0.85) Normality = 0.49 (0.78) Heteroscedasticity = 1.56 (0.10) RESET = 1.94 (0.17)
44
Modelling investment expenditures
Figure 3.7 Impulse response for construction investment.
This equation shows that the growth in contracts awarded over the current and past quarters is translated into gross capital formation in residential and non-residential structures. The estimates of the equation yield the co-integrating relation ln ICON − 0.7 ln CAN /P bm . The coefficient on the error-correction variable is very significant but relatively small, implying a tardy pace of adjustment to the long-run equilibrium. All the diagnostic and forecasting tests on the equation are comfortably passed. Figure 3.7 plots the impulse response pattern for construction investment. The graph reveals that the impact of contracts awarded builds up gradually and reaches a peak half a year later. Thereafter, the impact peters out exponentially, although the dynamic effects are still felt after 4 years. The long-run elasticity calculated from the model confirms that the effect of a 1% increase in real contracts awarded is a rise in construction spending of 0.7%.
3.6
Policy options
Our study of investment expenditures in this chapter confirmed that GFCF in machinery and transportation equipment is a function of the net amount of investment commitments that Singapore is able to attract. The latter variable captures investors’ expectations of both the country’s short-term international competitiveness and longer-term growth prospects relative to its regional rivals. At first glance, this seems to imply that Singapore will lose out to cheaper destinations like China and India in the race for foreign investment. Still, it needs to be pointed out that Singapore is currently attracting high-technology investments that require specialized skills not available elsewhere in the region (witness the continually rising value-added content of manufacturing production). The impact of business costs should therefore diminish as Singapore climbs further up the technology ladder and restructures the economy towards the so-called “knowledge-based” industries.
Modelling investment expenditures
45
Investor confidence in the Singapore economy, or what is sometimes referred to as “business sentiment”, has also been demonstrated to be very critical. The upshot is that good times can feed on themselves but equally, bad times tend to get worse. Policy interventions can correct a vicious cycle, however; this happens for instance when the government indirectly stimulates investment by manipulating domestic business costs. Our estimates suggest that a 5% drop in manufacturing output requires a more than 10% compensating reduction in RUBC in order to achieve positive growth in investment commitments by foreign firms. However, a reduction in the corporate income tax rate by itself is found to be an ineffective investment incentive. The boost given to machinery and equipment spending from a cut in the headline rate is negligible. Needless to say, this comparative static exercise does not internalize the psychological effects of a tax reduction, if any, which are likely to be confounded with investor expectations. Regardless of this, the quantitative evidence leans towards the view that further tax cuts beyond those already announced are not a viable policy option, especially in view of their detrimental impact on government finances.
4
4.1
The trade sector
Introduction
Singapore is one of the most open free trade economies in the world. With a ratio of merchandise trade to GNP of 2.9, it stands as an archetypal example of what Lloyd and Sandilands (1986) have called a “re-export economy”, or an economy in which imported inputs are used intensively in the production of exports for world markets. Given Singapore’s small physical size and the dearth of natural resources except for its strategic location at the crossroads of trade, the high import content of exports is unsurprising. Indeed, Singapore began its economic history as an entrepôt for South-East Asia, importing commodities from the regional hinterland and then re-exporting them to the other countries in the world, and vice versa. Like its counterpart in East Asia, Hong Kong, this role has given rise to supporting service industries such as shipping, port facilities, insurance, logistics and transport, all of which continue to be important up to the present day. Whilst economic openness has brought Singapore immense benefits, the huge foreign trade sector also meant that external shocks are a perennial factor behind domestic macroeconomic fluctuations. The city-state is extremely vulnerable to vagaries in the global demand for its goods and services, be they electronic products, refined oil or financial services. In this chapter, we develop a theoretical model of exports that embeds this and other salient features of the Singapore economy. Guided by this model, we estimate the disaggregated export equations for goods and services that constitute the trade block of the ESU01 model. We then do the same for two major categories of imports; retained merchandise imports and service imports.
4.2
Literature review
The predominant approach to modelling exports is to estimate a single export demand function, which is often justified on the grounds of the so-called “Armington trade model” (Armington, 1969). This model underlies the assumption that manufactured goods produced in different countries are imperfect substitutes for each other. Modelling export demand in this way entails a number
The trade sector
47
of problems, however. First, the demand model often produces low price elasticity estimates, which have been a concern of researchers for a long time (Orcutt, 1950; Harberger, 1953; Leamer, 1981; Alston et al., 1990). Alston et al. in particular show that the Armington model understates the price elasticity estimates for US cotton exports by about 50% compared to an almost ideal demand system (AIDS). Second, the demand model fails to explain the rapid expansion of exports experienced by fast-growing economies such as the Asian newly industrializing economies (NIEs) without a secular decline in their terms of trade. Third, although a downward sloping demand curve with an infinitely elastic supply is quite plausible for imports, such an assumption for export demand is at odds with reality. An importing firm in Singapore for example could import a commodity or its close substitutes from different countries to satisfy a given demand. However, an importing firm in the USA may not be able to import all that it needs from Singapore alone to meet demand at the same price. In other words, export supply from a single country cannot be infinitely elastic. An obvious improvement over the single equation demand model is the simultaneous estimation of both export demand and supply equations, pioneered by Goldstein and Khan (1978). However, if a country is small relative to other exporters, the price of the exported good is determined in the world market and given exogenously. In other words, the country is a price-taker and faces a perfectly elastic demand curve. What needs to be estimated then is an export supply function. Browne (1982) consequently argued that the Goldstein-Khan formulation was an inadequate representation for testing the price-taker hypothesis. Following this, Riedel (1988) proposed a model formulation for testing the price-taker assumption explicitly. He specified for the small open economy of Hong Kong an export demand equation with the export price as the dependent variable, an export supply equation with quantity as the dependent variable, and a wage equation. Estimated as a simultaneous system of equations, Riedel observed that the twostage least squares (2SLS) estimates of the coefficients of the quantity and world income variables in the export demand equation were not significantly different from zero while the coefficient for the price of competing goods in importing countries was non-zero. He argued that this constitutes prima facie evidence of an infinitely elastic export demand, a characteristic of a price-taking country in international markets. Riedel’s study led to a vigorous debate with Nguyen (1989), Faini et al. (1992), Muscatelli (1995a, 1995b) and Muscatelli et al. (1992, 1994, 1995a, 1995b), who all contested his model formulation and results. One of the issues of contention – which came to be known as the “normalization paradox” – concerns the interpretation of the price equation as an inverse demand function. With regard to this apparent paradox, two points are noteworthy. First, it is well-known that normalization cannot convert a non-zero estimate to a zero coefficient if the conditioning variable set remains the same. Second, a regression with price as the dependent variable suffers from endogeneity bias because of the correlation between exports and the error term.
48
The trade sector
Even though the debate and subsequent responses in Athukorala and Riedel (1991, 1994, 1996) did not produce a consensus, Riedel’s exercise alerted researchers to the inherent theoretical inconsistency in estimating an export demand equation under the assumption of price-taking behaviour. Notwithstanding this, the practice of estimating export demand functions abounds (see the macroeconometric models published in Economic Modelling). Empirical export functions estimated for Singapore, including those found in the forerunners of the ESU01 model, are no exceptions to this rule (Toh and Low, 1990; Abeysinghe and Tan, 1998). Not only are these based on a demand function, which contradicts the small country assumption, they often rely on simple log-linear specifications, sometimes with explanatory variables and dynamic restrictions (such as partial adjustment) imposed arbitrarily.
4.3
Testing export hypotheses
To assess the validity of alternative empirical specifications of the export function for Singapore, we shall test in this section a number of hypotheses commonly encountered in the applied literature on export modelling. For this purpose, consider the following stylized export demand and supply equations: Demand: Supply:
ln X = α0 + α1 ln Ptx + α2 ln Ptw + α3 ln Ytw ln X = β0 + β1 ln Ptx + β2 ln Ptrm + β3 ln Ptd + β4 ln Kt
(4.1) (4.2)
In these two equations, X is the export volume, P x is the price of exports, P w is the price of competing goods in the importing countries, Y w is the aggregate real income of the importing countries, P rm is the price of imported raw materials, P d is the price of domestic inputs, and K represents production capacity. All the variables are in Singapore dollars. If P w , Y w , P rm , P d and K are assumed to be exogenously given, demand and supply jointly determine X and P x . This is the standard model adopted for countries with some price-setting power. When a country is a price-taker, P x is also exogenously determined. In this case, one has to choose between estimating (4.1) or (4.2). Many applied researchers have chosen the export demand function due to the ready availability of data rather than the theoretical reasons given in the previous section. In contrast, estimation of the export supply equation is often hampered by data constraints. Let Zt be the (7 × 1) vector of variables that enter the demand and supply equations in (4.1) and (4.2): Zt = (ln Xt , ln Ptx , ln Ptw , ln Ytw , ln Ptrm , ln Ptd , ln Kt )
(4.3)
The trade sector
49
The vector autoregression (VAR) for the variables in Zt can be written in vector error-correction model (VECM) format as1 Zt = αβ Zt−1 +
p−1
j Zt−j + 0 + εt
j=1
where β Zt−1 consists of r co-integrating relationships. For r = 2, we have: β11 β21 β31 β41 β51 β61 β71 β = β12 β22 β32 β42 β52 β62 β72 We shall impose restrictions on the co-integrating relationships to test the hypotheses listed below. Hypothesis 1 The country has an influence on price and hence, both demand and supply equations need to be estimated: 1 β21 β31 β41 0 0 0 β = 0 β52 β62 β72 1 β22 0 Under this hypothesis, the first co-integrating vector corresponds to the demand equation and the second to the supply equation. Hypothesis 2 The country is a price-taker and has an infinitely elastic supply curve: 1 β21 β31 β41 0 0 0 β = 0 1 β32 0 0 0 0 In this case, the first co-integrating vector represents demand and the second assumes that P x is proportional to P w . Hypothesis 3 The country is a price-taker and faces an infinitely elastic demand curve: 1 β21 0 0 β51 β61 β71 β = 0 1 β32 0 0 0 0 The first vector here represents the supply function while the second vector assumes that P x is proportional to P w , as in Hypothesis 2.
1 Unless an ambiguity arises, we shall henceforth refer to the variables in (4.3) without the natural logarithm prefix.
50
The trade sector
The validity of these hypotheses will be tested using Johansen’s (1995) likelihood ratio (LR) test. The constraints imposed on the co-integrating vectors under each hypothesis are all of the exclusion type and they represent over-identifying restrictions, as can be verified by checking the order condition, which states that at least r − 1 restrictions must be imposed on the parameters for a co-integrating vector to be identified. The LR test statistic has an asymptotic χ 2 distribution with degrees of freedom equal to the number of over-identifying restrictions imposed. We confine our attention for the time being to the main export category in Singapore – non-oil domestic exports (NODX), which comprise 55% of Singapore’s total exports of manufactured goods (the rest is made up of oil exports and re-exports). For the price of competing goods P w , we created a composite index based on selected items from the US producer price index.2 For P d , we use the domestic UBC index. The estimation period is 1981Q1–2003Q4, with pre-sample values coming from 1980. Standard ADF tests of unit roots show that the seven variables in (4.3) are best characterized as I (1) processes. The optimal model specifications chosen by the Schwarz Bayesian Criterion (SBC) and Akaike Information Criterion (AIC) are VAR(1) and VAR(2), respectively. We decided to proceed with the VAR(2) model because the Johansen ML methodology is well known to be very sensitive to the choice of the lag order; under-specification in particular is known to produce misleading results (Harris, 1997; Maddala and Kim, 1998). Table 4.1 presents the results of Johansen’s trace test for co-integration rank and Table 4.2 shows the LR test outcomes for alternative demand and supply specifications. The results indicate the presence of two co-integrating vectors at the 5% level and one vector at the 1% level. Under the assumption of two co-integrating relationships, the traditional demand-supply model in Hypothesis 1 faces a marginal rejection at the 5% level. Even though one may be tempted to proceed further with this hypothesis, the coefficients of P x and P w are wrongly signed in the demand equation, and so is the coefficient of P x in the supply equation. Taken together, the results suggest that the demand-supply model that jointly determines export price and quantity is not suitable for Singapore. Table 4.1 Co-integration test results Rank
0
1
2
3
4
5
6
Trace test p-value
144.9** 0.002
97.3* 0.037
59.4 0.254
36.6 0.372
19.3 0.481
5.9 0.712
0.1 0.758
Note: ** and * indicate statistical significance at the 1% and 5% levels, respectively.
2 Following a common practice, we also considered an index based on the total producer price indices of major trading partners of Singapore. This proved to be inferior to the US index. A detailed discussion on the choice of variables is given in Appendix A.
The trade sector
51
Table 4.2 Export hypotheses tests Two co-integrating vectors H1 D
H2 S
D
One co-integrating vector H3
Px ∝ Pw S
X 1 1 1 – Px 1.23 −0.69 1.31 1 Pw −0.16 – −0.39 0.43 Yw 3.85 – 3.71 – P rm – −6.29 – – – −2.67 – – Pd K – 0.99 – – χ2 8.08 39.45 p-value 0.045 0.000
H4 Px ∝ Pw D
1 – 18.04 1 – −0.37 – – 18.96 – 4.96 – 4.57 – 29.59 0.000
H5
H6
S
D&S
1 1 1 1.05 −3.90 0.85 −0.30 – – 3.70 – 3.20 – −95.28 −1.04 – −64.12 −0.17 – −11.95 0.17 10.91 10.96 0.54 0.012 0.004 0.464
Notes: D represents a demand equation and S represents a supply equation. H(i) stands for Hypothesis i = 1, . . . , 6. A dash in a cell indicates a zero restriction imposed on the co-integrating vector. Variables are in logarithms and non-normalized coefficients are multiplied by −1 for easy comparison with the expected sign, as in a regression model.
Hypotheses 2 and 3 are also soundly rejected by the LR test, the main reason for this being the lack of co-integration between P x and P w . Notwithstanding the tendency for these two price series to co-move together (Figure 4.1), both the Johansen trace test and the residual-based ADF test clearly indicate that the two variables are not co-integrated. Still, a dynamic regression of P x on all the variables in the model shows that only P w and lagged P x are statistically significant. When we carried out the same exercises using first differences, the only variable that remained significant was P w . Despite the absence of co-integration between P x and P w , these results strongly suggest that the price of exports is
Figure 4.1 Prices of exports and competing goods.
52
The trade sector
exogenously determined. The results in Chapter 7 (Section 7.4) further show that Singapore is in fact a price-taker in the export market. We are thus led to consider two further hypotheses based on the assumption of a single co-integrating vector, labelled as H4 and H5 in Table 4.2. Hypothesis 4 is a pure demand specification while Hypothesis 5 is a pure supply specification. As with the preceding hypotheses, however, both H4 and H5 are rejected by the LR test while P x and P w carried the wrong signs in the demand equation. Note also the unusual magnitudes of the supply equation’s coefficients in Hypothesis 5, a common problem resulting from misspecification in the Johansen procedure. In summary, we conclude that neither a standard demand or supply equation, nor a simultaneous demand-supply system, provides an adequate model for Singapore’s exports.
4.4
A theoretical model of export determination
In view of the foregoing rejections of the standard export hypotheses, we shall depart from the traditional demand-supply specifications in modelling the export sector in Singapore. We present here a theoretical model of export production which is inspired by the pioneering work of Holt et al. (1960) on the dynamic optimizing behaviour of a firm that takes the demand for its product, the product price and input prices as given. Imagine a multinational corporation (MNC) that has established a manufacturing facility in Singapore for the purpose of exporting to world markets. Even though such an enterprise is compelled to adopt price-taking behaviour on account of the small size of the Singapore economy, its ability to export to global markets depends vitally on the width of firm-specific marketing channels, which are in turn influenced by external demand conditions at any one time.3 To keep the exposition simple, it is assumed that the firm produces solely to meet new export orders and does not maintain finished goods inventories. Furthermore, the firm can predict orders with certainty and maximizes accordingly the discounted sum of present and future profits given by: ∞
2 (1 + ρ)−i Pt+i Xt+i − c1 Xt+i − c2 (Ut+i − c3 Nt+i )2
(4.4)
i=0
subject to: Ut = Ut−1 + Nt − Xt
(4.5)
where ρ is the fixed discount rate, X is exports, P is the price, N denotes new orders received and U is the quantity of unfilled orders. c1 and c2 are cost parameters that
3 The arguments in this paragraph are due to Kapur (1983), though he did not derive his postulated export function from firms’ optimizing behaviour, which we do here.
The trade sector
53
depend on, inter alia, prevailing input prices, wages, hiring costs, and the tangible and intangible costs associated with unfulfilled orders (for notational simplicity, we do not attach time subscripts to c1 and c2 ). The first squared term in (4.4) represents a standard quadratic cost function. The second squared term specifies a target level of unfilled orders Ut∗ = c3 Nt , where c3 is a firm-specific constant, with departures from this target entailing additional costs to the firm in terms of lost sales and service deficiencies. The constraint in (4.5) states that the change in unfilled orders from one period to the next must equal the difference between new orders received and current export production. Owing to the constraint on demand, the export-oriented firm in this model confronts a slightly different optimization problem from the usual one faced by a profit-maximizing firm. This requires the firm to make a decision every period for all time periods on the amount of output to produce (and implicitly on the level of unfilled orders) by balancing at the margin the costs of changing output against those of letting orders go unfilled, in the face of demand fluctuations. Substituting the constraint (4.5) into the cost function (4.4) and differentiating the resulting expression with respect to Ut , we obtain the following first-order conditions: (c1 + c2 )Xt − c2 Ut−1 − c2 (1 − c3 )Nt =
Pt 2
(4.6)
c2 c2 c3 Ut − (1 + ρ)−1 Ut+1 − 1 + Nt − Ut−1 + 1 + (1 + ρ)−1 + c1 c1 + (1 + ρ)−1 Nt+1 = 0
(4.7)
The solution to the difference equation in (4.7) is: Ut = λUt−1 +
∞
λ(1 + ρ)
−1
i+1
c 2 c3 (1 + ρ) 1 + c1
i=0
Nt+i − Nt+1+i
where λ is the smallest root of the characteristic equation given by λ2 − [2 + ρ + (1 + ρ)(c2 /c1 )] λ + (1 + ρ) = 0. Using this expression for unfilled orders and (4.6), we can solve for the optimal level of export output as follows: Xt = λXt−1 +
Pt λPt−1 − 2(c1 + c2 ) 2(c1 + c2 )
+
c2 {1 − c3 − λ(1 − ρ)−1 [1 + λ(1 + c2 c3 /c1 )]} Nt c1 + c2
+
c2 λc3 (1 + c2 /c1 ) Nt−1 c1 + c 2
+
i+1 c2 [λ(1 + c2 c3 /c1 ) − 1] λ(1 + ρ)−1 Nt+i c1 + c2 ∞
i=1
(4.8)
54
The trade sector
This equation implies that the firm’s output level is positively correlated with the product price and past, current and future new orders, and negatively related to input prices and other costs. If c2 is zero or negligibly small, (4.8) reverts back to the standard long-run supply function with quantity as a function of product and input prices. Otherwise, demand forces enter into the production decision process of a price-taking firm. For instance, when there is a narrowing of marketing channels, new orders N received by the firm declines and this in turn increases the costs of holding the original level of U . As a result, the firm’s production falls. The opposite occurs when there is an enlargement of distribution channels through, for example, the discovery of new export markets. At the firm level, if c1 and c2 move over time in such a way that c2 /(c1 + c2 ) is approximately constant and data on new orders are available, (4.8) can be estimated as a dynamic linear regression model of X on P/TC and N , where TC represents a composite cost variable. Since information on new orders is not collected in Singapore, foreign income (Y w ) has to be used as a proxy for current and past N . The last term in (4.8) involving the discounted present value of future N can then be represented by the capital stock (K) because firms are likely to expand their production capacity only if they expect a permanent increase in future demand. At the aggregate level, therefore, we get an export function of the form Xt = f (Pt , TCt , Ytw , Kt ), combining demand and supply influences. The most appropriate empirical representation of this function is best determined by the data.
4.5
The NODX function
The model formulated in the previous section makes it clear that it is legitimate to estimate an export function for Singapore that has the usual price and cost arguments found in the supply function (i.e. P x , P rm and P d ), but also depends on variables representing current and expected future demand (Y w and K, respectively). In terms of Johansen’s multivariate framework laid down in Section 4.3, we specify the following co-integrating vector to represent such an export function: β = (1
β21
0
β41
β51
β61
β71 )
∗ , β ∗ , β ∗ > 0 and β ∗ , β ∗ < 0, where β ∗ = −β . with the expected signs β21 j1 41 71 51 61 j1 The parameter β31 on the coefficient of P w is set to zero on account of pricetaking behaviour on the part of firms. The estimated co-integrating vector is shown in Table 4.2 under the column denoted H6. In stark contrast to the hypotheses considered earlier, the estimated coefficients have the correct signs and the LR test strongly supports the zero restriction on β31 .
The trade sector
55
Having established a unique co-integrating vector, we can revert to OLS estimation and proceed to obtain a more refined export function.4 We found that the coefficient of K is very sensitive to variations in the sample size as a result of its high collinearity with Y w , but a robust estimate can be obtained by using ln K in the regression instead.5 Replacing ln K with ln K, P x with P nodx , P d with UBC, and imposing the homogeneity restriction with respect to output and input prices given by β21 + β51 + β61 = 0, the NODX function can be written as6 : ln NODXt = β0 + β1 ln Ytw + β2 ln Kt + β4 ln P rm − ln P nodx t + β5 ln UBC − ln P nodx t + ut
(4.9)
This formulation was basically what Kapur (1983) had postulated for Singapore’s exports, although his idea went largely unnoticed.7 Kapur went on to argue that an appreciation of the domestic currency would hurt exports, as can be seen by plugging into (4.9) the equation ln Pti = ln Et + ln PtiF (i = nodx, rm), where E is the effective exchange rate (Singapore dollars per unit of foreign currency) and P iF is a relevant foreign price index expressed in foreign currency units. A currency appreciation (fall in E) has the effect of lowering imported raw material prices measured in domestic currency and pari passu, the domestic price of exports (which leave export volumes unaffected), but it also increases the real cost in foreign currency terms of domestic inputs employed in the export sector, thereby curtailing exports. Abeysinghe and Tan (1998) have shown, however, that the effect of a currency appreciation on exports depends on the import content of exports. To incorporate this idea, we reformulate the export function in (4.9) as: ln NODXt = β0 + β1 ln Ytw + β2 ln Kt + γ TCt + ut
(4.10)
where TCt = θt (ln P rm − ln P nodx )t + (1 − θt )(ln UBC − ln P nodx )t represents the weighted sum of the costs of imported and domestic inputs, relative to the
4 OLS estimates are known to be more stable than Johansen’s maximum likelihood estimates (see Maddala and Kim, 1998, Chapters 5 and 6). Phillips (1994) has also shown that the Johansen estimator has a Cauchy-type distribution which tends to produce outliers. 5 With the short sample period we have, it is difficult to detect whether ln K is I (0) or I (1) although the ADF test supports the former. We repeated the tests in Tables 4.1 and 4.2 with ln K in place of ln K. This does not affect the findings we reported earlier except that an extra co-integrating vector has to be allowed for to account for this apparently stationary variable. 6 Estimating the function in dynamic form with one lag of each variable and testing the homogeneity restriction yields a Wald statistic of 1.02 with a chi-square p-value of 0.313. Thus, the restriction is not rejected by the data. Without a dynamic specification, the Wald test may be invalid (Sims et al., 1990). 7 To our knowledge, the only paper that has implemented Kapur’s model is Disney and Ho (1990), which is concerned with estimating a model of employment.
56
The trade sector
export price, and the weight θt (0 ≤ θt ≤ 1) measures the imported input content of exports. It should be noted that this formulation allows β4 and β5 in (4.9) to vary over time. Given that γ < 0, (4.10) shows that the impact of exchange rate changes on exports depends critically on import content. When domestic inputs are negligible (θ is close to unity), a currency appreciation has virtually no effect; conversely, when domestic inputs are dominant (θ is close to zero), appreciation has its full impact as measured by γ . The specification in (4.10) therefore results in a very parsimonious model with a concomitant reduction in multicollinearity between regressors. The key question is how to estimate θ . Since more than 60 percent of Singapore’s manufactured products are exported, one minus the ratio of value-added to gross output of the manufacturing sector should yield a reasonable estimate of the proportion of intermediate imports in exports. Reflecting Singapore’s “re-export economy” status, θ was 77% in the early 1980s by our calculation, but this figure fell to about 74% after the severe recession in 1985–86 as the value-added ratio jumped up (refer to Figure 3.2). In the aftermath of the 1997 Asian financial crisis, the import content dropped further to 73%. To account for these shifts and allow θ to change gradually, we converted the annual estimates of import content to quarterly figures through a univariate interpolation method (the cubic spline method) and took a 12-quarter moving average. We report below the long-run solution to a dynamic specification of (4.10) with a single lag of each variable over the period 1981Q1–2003Q4: ln NODXt = −7.1 + 3.55 ln Ytw + 2.70 ln Kt − 0.99 TCt (4.38) (−21.5) (50.7) (−6.11)
(4.11)
The coefficients in (4.11) are virtually the same as those from a static OLS regression. Moreover, the recursive estimates of these coefficients are highly stable over time, a virtue due in no small part to the allowance for time-varying import content. Even though we cannot test the validity of the import content restriction directly because of a changing θ , an F-test based on the R2 values for dynamic regressions of (4.9) and (4.10) produces a statistic of 3.276 with a p-value of 0.074, hence indirectly supporting the restriction at the 5% level. For the purpose of estimating a short-run export function, we removed the constant in (4.11) and rounded off the error-correction term as ECt = ln NODXt − 3.5 ln Ytw − 2.7 ln Kt + TCt . Equation (4.11) predicts the long-run movements of exports very tightly but it does not fully capture the short-run turning points. The missing variable is the world demand for electronics, fluctuations in which have translated into dramatic expansions and contractions in the exports of Singapore and other Asian countries who depend heavily on the semiconductor trade (Abeysinghe, 2001b; Ng et al., 2004). The best proxy indicator of the electronics cycle is global chip sales (CHIPN , in US$), downloaded from the Semiconductor Industry Association (SIA) website. Unfortunately, a consistent data series on this variable begins only from 1989, so the estimation period had to be truncated to 1989Q1–2003Q4, the last eight quarters in the sample period being retained for dynamic forecasting.
The trade sector
57
A general-to-specific search led us to the following ECM for explaining the short-run fluctuations in exports: N ln NODXt = −2.3 + 0.31 ln CHIPtN + 0.15 ln CHIPt−1 (1.61) (−3.56) (3.43)
− 0.35 ECt−1 (−3.59)
(4.12)
R 2 = 0.491 SE = 0.033 DW = 2.26 LM(5) = 3.62 (0.01) ARCH(4) = 0.28 (0.89) Chow = 1.35 (0.23) Normality = 6.80 (0.03) Heteroscedasticity = 0.43 (0.85) RESET = 1.51 (0.22) The estimated model suggests that the transitory dynamics in Singapore’s NODX are strongly influenced by the world electronics cycle and deviations from the long-run equilibrium. The adjustment coefficient on the error-correction term is statistically significant at the 1% level and the model passes most of the diagnostic and misspecification tests.8 We retained the insignificant lag of chip sales in (4.12) because recursive estimates show that this coefficient is very stable and helps to improve the performance of dynamic forecasts of exports. Figure 4.2 plots these forecasts together with the actual values, which suggest that the model is able to track exports well.
Figure 4.2 Forecasts of non-oil domestic exports.
8 Curiously, despite using seasonally adjusted data, the residuals contain a seasonal effect at lag four. This appears to be noise rather than a signal.
58
The trade sector
Table 4.3 Dynamic elasticities for non-oil domestic exports Lag (quarters) Y w
CHIP N K P rm − P nodx UBC − P nodx ULC − P nodx NLC − P nodx
0 1 4 8 12
0.33 0.35 0.11 0.02 0
0 1.89 2.87 3.45 3.57
0 0.99 2.40 2.88 2.98
0 −0.24 −0.59 −0.71 −0.74
0 −0.09 −0.21 −0.25 −0.26
0 −0.04 −0.10 −0.12 −0.12
0 −0.05 −0.11 −0.14 −0.14
Note: NLC is non-labour costs.
The dynamic elasticities extracted from the ECM are shown in Table 4.3. There are two main determinants of export performance on the demand side – the income of Singapore’s major trading partners and global chip sales. A 1% growth in foreign income boosts export growth by 1.89% after one quarter and achieves its full impact of 3.6% after 12 quarters. These estimates not only are statistically significant but are also robust to alternative specifications of the NODX function, and serve to highlight the strong dependence of Singapore on the world economy. In keeping with the city’s historical evolution, trade with the rest of the world has once again proven to be the “engine of growth” in its modernization phase. By contrast, global chip sales merely create a transitory effect, peaking after one quarter at 0.35% and then tapering off over about eight quarters. Despite these smaller elasticities, the ups and downs in the electronics cycle do produce large gyrations in exports. Recall that the capital stock in our model is a proxy variable for expected future demand. The manufacturing sector’s capital stock has had a salutary effect on Singapore’s NODX, with the dynamic elasticity analysis suggesting that a 1 percentage point increase in the growth rate of the capital stock leads to an increase in export volumes of 3% in the long run. The conclusion is that the phenomenal expansion of Singapore’s exports over the last three decades has been partly driven by an increase in production capacity which is spearheaded by foreign firms attracted to the prospects of selling to world markets. Turning to the supply-side variables, the long-run export price elasticity entering the ECM via the weighted costs variable is unity. By fixing the value of θ at 0.74, we are able to compute the elasticities for the components of this variable, namely, imported raw material costs (ln P rm − ln P nodx ) and domestic input costs (ln UBC−ln P nodx ). The econometric estimates in Table 4.3 indicate that escalating raw material prices have a more adverse impact on export production than the rising costs of doing business in Singapore. The immediate effect of a 1% increase in the relative price of imported inputs is −0.24% and this increases to −0.74% in the long run. Correspondingly, the short-run impact of domestic input costs is only −0.09% while the long-run effect is −0.26%. Breaking the latter down into their labour costs (ln ULC − ln P nodx ) and non-labour costs (ln NLC − ln P nodx ) components shows that, quite contrary to the conventional wisdom, the effect of
The trade sector
59
rising real wage costs on Singapore’s exports – due either to rising nominal wages or currency appreciation – has been rather mild.9
4.6
Other export components
We proceed in this section to estimate behavioural equations for the other major categories of exports: oil domestic exports (ODX), re-exports (RX) and service exports (SX), constituting 7.6%, 33.4% and 17.9%, respectively, of Singapore’s direct shipments of goods and services.10 The testing and estimation procedures adopted for these disaggregated export classifications parallel the treatment of NODX. Oil exports For a country with no natural oil resources, Singapore has ironically become the third largest oil refining centre and trading hub in the world (see Horsnell, 1997 for a comprehensive account of the development of Singapore’s oil market). Reflecting the rise of the electronics and chemicals industries, however, Singapore’s oil production as a share of total industrial output has fallen from 35% in 1983 to 13% in 1993, although its share in value-added has remained roughly constant at around 7–9%. In 1995, Singapore imported about 84% of its crude oil from the Middle East and exported refined oil products mostly to Asian countries – Hong Kong, Malaysia, Japan, Thailand, Vietnam, China, Taiwan, Australia and Korea. In addition, Singapore supplies fuel to the ships and aircrafts calling at her ports as part of the bunkering trade. Being a major supplier of refined oil to the region, one would expect a standard demand-supply model to be applicable to Singapore’s oil exports. A proper specification of the demand equation requires a good price index for P w . Due to the non-availability of data, we use Japan’s wholesale price index for petroleum and coal products as a proxy for this variable. We again use unit business costs to represent P d and the export and import price indices for mineral fuels (P ox and P om ) to represent the prices of oil exports and imports, respectively. The Johansen trace test on the variable vector Zt = (ln ODXt , ln Ptox , ln Ptw , ln Ytw , ln Ptom , ln Ptd , ln Kt ) indicates the presence of at most one
9 The statistical authorities derive domestic unit business costs as an arithmetic average of ULC and NLC, with the latter being made up of services costs and government rates and fees. Assuming that this formula holds approximately for a geometric average, we can write ln (P d /P nodx ) = α ln (ULC/P nodx ) + (1 − α) ln (NLC/P nodx ), where α = 0.464 is the weight. Domestic input costs rose because of higher unit business costs and currency appreciation prior to 1997 and a fall in the foreign price of exports after 1997. To assess whether the exchange rate has an independent impact on exports, we re-ran the regressions with ln NEER or ln NEER as an additional explanatory variable but found that the effect was statistically insignificant. 10 ODX does not include oil re-exports, which are lumped together with other re-exports. Officially, re-exports are defined as trade in commodities which are subject to only minor handling, packaging and processing.
60
The trade sector
co-integrating vector. Further investigation of the co-integrating vector shows that a pure supply specification is rejected by the LR test while a pure demand specification gains mild support, albeit with a wrongly estimated sign for P w . Additional LR tests of zero restrictions on the coefficients of P w and P d are not rejected. The irrelevance of P d for oil exports is explicable largely as a result of the high proportion of raw crude oil used in refinery operations. A dynamic regression of the oil export price on the other variables shows that the only statistically significant determinant of P ox is P om , not surprisingly since a data plot shows that these variables share very similar trends and turning points. We have to conclude that Singapore is a price-taker in the global oil market despite being a major oil supplier to Asia. The co-integrating export equation from ADL estimation with one lag of each variable for the period 1978Q1–2003Q4 is: ln ODXt = 3.36 + 0.57 ln Ytw + 0.32 ln Kt − 0.65 ln Ptom + 0.42 ln Ptox (6.75) (4.77) (5.4) (1.93) (−3.23) (4.13) Unlike the case of NODX, the homogeneity restriction on the coefficients of P ox and P om is strongly rejected. The relative magnitudes of the two coefficients suggest that a permanent rise in the import price of crude oil is not fully matched by an equiproportionate increase in the export price of refined oil. This finding implies that a sustained increase in the crude oil price will depress Singapore’s oil exports in the long term.11 In the short run, fluctuations in the imported oil price also have a significant impact, as the ECM below shows: ln ODXt = 2.76 + 0.75 ln Kt − 0.45 ln Ptom − 0.82 ECt−1 (9.09) (2.76) (−7.55) (−9.11) (4.14) R 2 = 0.561 SE = 0.064 DW = 1.93 LM(5) = 0.81 (0.55) ARCH(4) = 0.10 (0.98) Chow = 1.22 (0.29) Normality = 0.36 (0.83) Heteroscedasticity = 0.58 (0.75) RESET = 0.01 (0.94) Reflecting the effect of the large loading coefficient in (4.14), the dynamic elasticities for oil exports settle down to their long-run values within a year. Re-exports Compared to oil refining, re-exports contain even higher import content and less domestic value-added. Moreover, the flows of re-exports into and out of Singapore
11 In contrast to the NODX equation, the capital stock enters the ODX equation in levels as it is insignificant in differences. We do not have a good explanation for this difference in the absence of disaggregated data on the oil sector’s capital stock.
The trade sector
61
are increasingly being driven by trade in intermediate components and parts, particularly of electronics, as production becomes fragmented across the regional supply chain. We may therefore expect the volume of re-exports to be swayed by the global electronics cycle, just like non-oil domestic exports. Consequently, the demand variables relevant to re-exports must include both Y w and CHIPN . Preliminary estimates show that P d can be omitted from the vector for testing co-integration on account of the high import content of re-exports. Similarly, K can also be left out since it represents the production capacity of the manufacturing sector. For the price variables, we include the re-export price index (P rx ), the overall import price index (P m ) to serve as a proxy for the prices of commodities destined to be re-exported, and the weighted average of the producer price indices (PPIs) of Singapore’s major trading partners (P w ) to represent competing prices in the importing countries. The Johansen trace test revealed that a single co-integrating vector exists among the above set of variables, and subsequent LR tests indicated that this vector is compatible with the export function specification used for NODX. Since the high import content implies that θ ≈ 1, the weighted costs variable reduces to the price ratio P m /P rx , with the implication that the exchange rate has no effect on re-exports. When single equation techniques were used to estimate a long-run re-export function for the period 1989Q1–2003Q4, entrepôt exports were found to be unresponsive to relative price changes. We therefore estimated a function without the price variables and obtained the following results: ln RXt = −0.66 + 1.87 ln Ytw + 0.41 ln CHIPtN (5.04) (−0.75) (8.08)
(4.15)
Making use of the residuals from the above regression as the error-correction term, the short-run dynamic model for re-exports is given by: ln RXt = −0.15 + 0.27 ln CHIPtN + 0.13 D_94Q2t − 0.23 ECt−1 (3.85) (−3.20) (4.19) (−3.46) (4.16) R 2 = 0.536 SE = 0.034 DW = 2.11 LM(5) = 2.23 (0.08) ARCH(4) = 0.21 (0.93) Chow = 1.23 (0.30) Normality = 2.55 (0.28) Heteroscedasticity = 1.86 (0.12) RESET = 6.26 (0.02) A dummy variable was also included in the model to remove an outlier in 1994Q2. Dynamic elasticities calculated from (4.15) and (4.16) show that a sustained 1% increase in global chip sales increases re-exports by 0.41% after about 12 quarters whereas a similar increase in world income raises them by a much larger 1.78% in the same time period. Notwithstanding the smaller magnitudes of
62
The trade sector
the CHIP elasticities, swings in electronics demand induce cycles in Singapore’s re-exports too. Service exports Transportation and travel services are the major components of Singapore’s service exports to the rest of the world. Between 2000 and 2003, these two types of receipts accounted for 40% and 15% of total service exports, respectively. This means that more than half of service earnings is tourism related, so total visitor days (V ) is included as an additional explanatory variable in the export function for services. In stark contrast to re-exports, the import content of service exports is very low. Hence, the price of imported inputs can be ignored in the model for service exports and we can set θ = 0, which implies that the cost variable becomes P d /P sx . Whilst P d continues to be represented by the UBC index, a price index for service exports is not available. The implicit price deflator for the service industries in Singapore is used as a proxy instead. We also do not consider the capital stock and the price of competing services because of data problems. Due to parameter instability surfacing in the data after 2000, we carried out the co-integration analysis for the restricted sample from 1986Q1 to 2000Q4. The trace test results again hinted at the existence of a single co-integrating relationship binding the variables together, but pure demand and pure supply interpretations of the co-integration vector were rejected by the LR tests. However, the export function specification could not be tested explicitly due to a lack of over-identifying restrictions. After running several static OLS and dynamic ADL regressions, we decided to use the former set of estimates as the long-run parameters of the co-integration relation, partly because the recursive estimates from the OLS regressions are highly stable. Even though we cannot test the homogeneity restriction given the nonstandard distributions involved, the coefficient estimates for UBC and P sx emerge with the expected signs and roughly equal magnitudes. Imposing the restriction results in the estimates below12 : ln SXt = −0.08 + 1.50 ln Ytw + 0.50 ln Vt − 0.89 ln(UBC/P sx )t (4.17) (12.7) (−0.55) (29.2) (−6.10) In order to estimate the ECM for the full sample period up to 2003Q4, we had to introduce a step dummy variable that jumps from zero to unity from 2001Q1 onwards to account for the parameter instability mentioned above. Even with the dummy, we observe a slight shrinkage of the magnitudes of the coefficient estimates for V and the ratio UBC/Psx . It appears that the cause of this lies in the
12 As in the case of NODX, the effect of the exchange rate enters the model indirectly through P sx . We tried again to assess whether NEER has any direct effect on service exports and found that none is present.
The trade sector
63
Table 4.4 Dynamic elasticities for service exports Lag (quarters)
Yw
V
UBC − P sx
0 1 4 8
2.31 1.90 1.55 1.50
0.30 0.40 0.49 0.50
−0.30 −0.60 −0.86 −0.89
volatile growth that followed the electronics boom-and-bust cycle in 2000–01. The ECM estimates for the full sample are given below: ln SXt = −0.17 + 2.31 ln Ytw + 0.30 ln Vt − 0.30 ln (UBC/P sx )t (7.08) (−4.22) (3.85) (− 1.57) − 0.51 ECt−1 + 0.02 D_01Q1t (2.08) (−4.30)
(4.18)
R 2 = 0.56 SE = 0.031 DW = 2.11 LM(5) = 0.50 (0.77) ARCH(4) = 0.50 (0.73) Chow = 1.24 (0.30) Normality = 1.17 (0.56) Heteroscedasticity = 0.96 (0.48) RESET = 2.14 (0.15) The dynamic elasticities for service exports are reported in Table 4.4. Like the other export categories, world income is the main determinant of Singapore’s invisible exports. However, the short-run impact on service exports is much larger and it also overshoots the long-run effect. Visitor expenditures are a second key variable that exerts a positive effect on service exports. Interestingly, this variable is able to fully account for the deep plunge and subsequent rebound in service earnings during the SARS episode in 2003. Finally, domestic cost inflation, relative to export prices, has significant negative effects that increase over time, suggesting that the exchange rate has a role to play in maintaining the competitiveness of service exports.
4.7
Import functions for Singapore
Compared to exports, empirical modelling of imports into Singapore is relatively straightforward. The econometric literature typically estimates log-linear import demand equations which depend on an activity variable and the relative price of imports (see Lim et al., 1996 for such a study on Singapore). As pointed out in Section 4.2, the supply of imports from the rest of the world can be safely assumed to be infinitely elastic at world prices, thereby obviating the need to model it. Therefore, we only estimate separate demand equations for retained imports of goods (RM) and service imports (SM). While the behaviour of imported consumption and intermediate goods are likely to differ, data non-availability precludes us from analyzing imports according to their end uses.
64
The trade sector
Retained imports Retained imports are defined as the total quantity of merchandise imports minus re-exports.13 It is natural for us to focus on retained imports rather than aggregate imports as we have modelled re-exports in the previous section. Moreover, Lim and Associates (1988, Chapter 16) have pointed out that the marginal propensity to import based on total imports exceeds one due to the large entrepôt trade. Since this is implausible, the component of imports that is pertinent to the computation of multiplier effects in Chapter 8 is retained imports. We tried alternative measures of domestic economic activity such as real GDP, disposable income and domestic demand (or absorption) to see which of them is essential to achieving co-integration. As we have seen earlier, Singapore is heavily dependent on imported raw materials and intermediate inputs for the production of goods and services destined for export. Hence, we also considered domestic merchandise exports in real terms (i.e. excluding re-exports) as a driver of retained imports, either on their own or combined with another activity variable. Assuming homogeneity, the relative price of imports is computed as the ratio of the import price index to the Singapore Manufactured Products Price Index (P m /SMPI), the latter being a weighted average of the prices of goods produced by local manufacturers. Co-integration is obtained when derived expenditures on imports are used as the composite activity variable. This sum has a convenient economic interpretation as a measure of the final demand for imports (FDM ). The weights used in the computation of final demand are the import contents of various expenditure components, as reported in the Singapore Input-Output Tables 1988 (Table XVI) FDMt ≡ 0.4088Ct + 0.6644It + 0.2866Gt + 0.5788TDXt + 0.8492INVTt (4.19) where C is consumption, I is investment, G is government consumption expenditures, TDX is total domestic exports and INVT is the change in inventories, to be discussed further in Chapters 6 and 7. The import content of TDX is not published and so it is assumed to be the same as that of total exports. As suspected, the hypothesis of price homogeneity is not rejected by the Wald test, so we went ahead to obtain the following ADL regression estimates over the period 1986Q1–2003Q4: ln RMt = −2.95 + 1.21 ln FDMt − 0.87 ln (P m /SMPI )t (−6.12) (25.4) (−4.41)
(4.20)
The long-run elasticity of imports with respect to final demand is larger than unity while the relative price elasticity is below one in absolute value.
13 This way of deriving retained imports is subject to a minor problem. Entrepôt imports are valued at “cif ” prices whereas re-exports are valued at “fob” prices. Unfortunately, we do not have access to the cif component of import values.
The trade sector
65
Much has been made of the theoretical significance of a unitary income elasticity of imports (see Hong, 1999), but we should not expect to observe this in (4.20) since domestic exports are included in our measure of final demand. Given Singapore’s huge reliance on imported inputs, moreover, an income elasticity bigger than one makes sense, especially when juxtaposed against the even larger foreign income elasticities for exports reported in the previous sections. The corresponding short-run model for retained goods imports is: ln RMt = −0.68 − 0.25 ln RMt−1 + 1.09 ln FDMt (13.0) (−2.52) (−2.29) + 0.52 ln FDMt−1 − 0.23 ECt−1 (3.50) (−2.50)
(4.21)
R 2 = 0.765 SE = 0.023 DW = 2.00 LM(5) = 0.29 (0.92) ARCH(4) = 2.21 (0.08) Chow = 0.65 (0.79) Normality = 3.08 (0.21) Heteroscedasticity = 1.40 (0.22) RESET = 0.79 (0.38) The ECM estimates point to an impact income elasticity of 1.1 while the price elasticity is effectively zero in the short run as relative prices enter the model only through the long-run error-correction term. Thus, import volumes react sluggishly to price movements – a restatement of the “J-curve” effect in the empirical trade literature. In the short term, contractual obligations on the part of importers and habit persistence on the part of consumers hinder a quicker pace of adjustment towards the long-run equilibrium. The diagnostic statistics for the model are good; in particular, there is no evidence of the autocorrelation that so often plagues the residuals of import equations. Even though import elasticities are prone to abrupt change arising from fundamental shifts in economic structure, tests of parameter constancy also do not indicate any instability. Finally, Figure 4.3 testifies to the excellent fit of the equation, both in and out of sample. Table 4.5 compiles the dynamic elasticities derived from the ECM. As noted above, the price elasticity adjusts towards its long-run value slowly. The total expenditure elasticity peaks after one quarter and returns to its long-run value after about 12 quarters. Decomposing final expenditures into its components using their 2003 shares, we find the same pattern repeated. Given that NODX accounts for about 92% of TDX, the dynamic elasticities also reveal the predominant role that non-oil exports play as a determinant of Singapore’s import demand. Service imports As in the case of service exports, the major contributors to service import payments are expenses on transportation and travel. Between 2000 and 2003, these two components accounted for 45% and 18% of total service imports, respectively. The third largest single category was royalty payments, which accounted for 12% over the same period. Unlike merchandise imports, we observe that the income
66
The trade sector
Figure 4.3 Forecasts of retained imports.
elasticity of demand for service imports is very robustly estimated regardless of whether real GDP or real disposable income is used. Real GDP is chosen as it is a broader measure covering items such as royalties that are conceptually not tied to disposable income. Since a deflator for service imports is unavailable, we constructed one based on the price index for all goods and services (see Appendix A). This index is denoted by P sm . We experimented with alternative indices, including the CPI, as proxies for the prices of domestic competing services but none of them proved to be meaningful. In fact, given that the bulk of service imports are incurred on travel outside of Singapore, there are hardly any close substitutes for them locally. We therefore proceeded without a price for substitutes in the model. Johansen’s test for co-integration rank does not reject the existence of a longrun equilibrium relationship linking the logarithms of SM, GDP and P sm over the period 1986Q1–2003Q4. Regression estimates based on an ADL specification with a one-quarter lag on each variable are as follows: ln SMt = −2.86 + 1.56 ln GDPt − 0.83 ln Ptsm (1.15) (−3.01) (11.3)
(4.22)
Table 4.5 Dynamic elasticities for retained imports Lag (quarters)
C
I
G
TDX
Total expenditure
Pm /SMPI
0 1 4 8 12
0.21 0.26 0.24 0.24 0.24
0.20 0.26 0.24 0.23 0.23
0.04 0.05 0.05 0.05 0.05
0.73 0.91 0.84 0.82 0.81
1.09 1.37 1.26 1.23 1.22
0 −0.20 −0.49 −0.69 −0.79
Note: TDX is total domestic exports (NODX + ODX + SX).
The trade sector
67
Even though this regression produced a coefficient for P sm that is roughly equal to its standard error, the t-statistic on its OLS counterpart is significantly larger than the conventional critical values. The corresponding ECM with two (1,0) dummies added to account for the SARS quarters of 2003 is: ln SMt = −0.44 + 0.81 ln GDPt − 0.83 ln Ptsm (−2.24) (2.56) (−6.14) − 0.075 D_03Q2t + 0.084 D_03Q3t − 0.13 ECt−1 (1.98) (−1.64) (−1.74)
(4.23)
R 2 = 0.52 SE = 0.043 DW = 1.96 LM(5) = 0.54 (0.75) ARCH(4) = 0.16 (0.96) Chow = 0.93 (0.50) Normality = 34.7 (0.00) Heteroscedasticity = 1.41 (0.22) RESET = 0.02 (0.90) All the explanatory variables in (4.23) are significant and yielded the signs predicted by theory. Echoing the findings for many industrialized economies (see Hooper et al., 2000), the short-run elasticity of service imports with respect to domestic income, estimated at 0.8, is half the long-run elasticity of 1.6. On the other hand, the own-price elasticity is the same in the short and long runs, falling in the inelastic range. In view of possible deficiencies in the price index we used, it is inadvisable to draw strong conclusions from these estimates. Nonetheless, the finding that service imports are price-inelastic and income-elastic at the same time suggests that holidays abroad are a luxury good which is affordable only to high income earners who are less sensitive to price variations. This is consistent with the conclusion in Urata and Kiyota (2003) that richer countries in Asia tend to spend more on travel.
4.8
Policy implications
In this final section, we highlight the practical implications for policymaking of the export functions estimated in this chapter. To facilitate the discussion, the explanatory variables in the NODX equation are graphed in Figure 4.4. Figure 4.4a plots the terms of trade (P rm /P x ), whose marked deterioration since about 1999 – resulting primarily from a sharp fall in the export price of electronics – is seen to have been a major drag on export growth. Unfortunately, just as in the case of foreign demand, this price ratio is beyond the control of policymakers. The only variables they can influence are domestic input prices (P d ), the exchange rate (NEER) and production capacity (K). We will discuss these in turn. Insofar as domestic business and wage costs are concerned, the main policies the Singapore government has deployed in the past were direct cuts in the employer’s share of CPF contributions and reductions in public utility charges, fees and rentals (see Section 1.1 and Figure 4.4c). Our elasticity estimates show, however, that this cost-cutting policy entails more pain than gain because non-oil export demand is not too responsive to reductions in domestic costs due to the relatively low proportion of local inputs used, while oil exports and re-exports are totally unresponsive.
68
The trade sector
Figure 4.4 The determinants of NODX. Notes: The quarterly growth of K is in percentages while all other variables are in index form with 1990 as the base year. A rise in NEER indicates an appreciation of the domestic currency. P xf is derived by removing the effect of NEER from P x .
For the former, to offset the effect of a 1% contraction in world income growth, real wage costs have to be brought down by 26%! That said, such cost-cutting measures might help to boost business confidence and hence ameliorate a fall in investment and job retrenchments. Standard textbooks hold that a currency depreciation stimulates exports, discourages imports and improves the trade balance. In Singapore, however, the exchange rate is a double-edged sword because, as our theoretical and empirical analyses show, a depreciation of the currency not only lowers the foreign prices of domestic exports but also raises the domestic costs of imported inputs. The MAS is emphatic that manipulation of the nominal and real exchange rate would not be effective in fostering the export trade, especially in the long run (Teh and Shanmugaratnam, 1992). This chapter’s empirical results confirm that the exchange rate is an inefficacious tool for bolstering the foreign demand for Singapore-made goods. Nonetheless, a comparison of the results in Tables 4.3 and 4.4 highlights the much stronger negative impact that domestic cost inflation and currency appreciation have on the exports of invisibles. Thus, a currency depreciation should benefit
The trade sector
69
service exports given their high domestic value-added and low import content – at least in the short run. Paradoxically, the need for an appreciating Singapore dollar to control imported inflation and a depreciating dollar to boost service exports are two conflicting objectives that one policy instrument cannot handle. Luckily, we shall see in Chapter 9 that the trade-off is not so stark in practice. Fortunately for Singapore too, a strong foreign reserves position has enabled the MAS to dictate the path of the local dollar in order to stimulate exports at critical junctures. Figure 4.4d suggests that this happened in the mid-eighties and late nineties. Our estimated equation for service exports also suggests that exchange rate policy is not sufficient to smooth earnings from invisibles because unpredictable growth rates in tourist arrivals add to volatility.14 Stable growth of tourist arrivals, therefore, is also important, as are efforts to lengthen visitor stay and encourage them to spend more. Given the physical constraints faced by Singapore, 5 million visitors in a year who stay for 3 days each are arguably better than 8 million visitors who stay for 2 days. In this regard, the planned integrated resorts (IRs) at Marina Bay and Sentosa will increase not just the length of stay, but the volume of tourist traffic as well. Throughout the last four decades, the Singapore government has been extremely pro-active in attracting FDI through tax concessions and numerous incentive schemes. Our econometric estimates bear testimony to their success: increments to the national capital stock have clearly played an important role in Singapore’s export-led economic development. However, Figure 4.4b shows that the growth of production capacity has slowed down in recent years as competition for FDI from China and other newly emerging economies has made it more difficult for Singapore to attract investments. The implication is that the importance of this source of growth will diminish in the future.
14 The visitor days proxy for tourist expenditures has seen heady growth rates of 20% year-on-year. At the other extreme, the SARS pandemic brought visitor days down by 46% in 2003Q2.
5
5.1
The labour market
Introduction
The labour market attracts a tremendous amount of attention from academics, policymakers and laymen alike. Many writers have commented on the policy issues surrounding employment, wages, productivity and the unemployment rate in Singapore (see for example Krause et al., 1987 and Lim and Associates, 1988). Yet few have analyzed their subtle links. We provide a fresh effort to investigate the interdependent relationships between labour market variables in this chapter. A glance at Figure 1.1 reveals that our intention is to relate developments in the macroeconomy – especially fluctuations in aggregate output and the overall price level – to the job market. By now, the reader would have realized that the Singapore economy is a small open economy with unique features. Nowhere is this more evident than in the labour market. Any attempt to describe it qualitatively or quantitatively must come to terms with the following characteristics of employment and wage determination that are peculiar to the country: •
•
•
The presence of a large pool of foreign workers in the labour force, with attendant implications for the demand and supply of labour. As at the end of 2004, the number of foreigners stands at 621,400 or 28% of the total workforce. The role of the National Wages Council (NWC) in affecting aggregate wage settlements. The economy-wide pay increases recommended by the NWC every year are widely perceived to have a significant influence on wage and CPF rates. Government intervention in the labour market has been particularly important. As explained in Chapter 1, this has taken the form of foreign labour, wage and CPF policies.
In the next section, the methodological framework employed to address the above features is explained. Based on it, we specify and estimate behavioural equations for employment, the labour force and real wages. Following that, our econometric model of the labour market is used to predict the unemployment rate
The labour market 71 in Singapore. We also discuss issues pertaining to the measurement of, and trends in, natural unemployment. Lastly, we draw policy conclusions from our empirical findings.
5.2
The methodological framework
The modelling of the labour market is the subject of a good deal of controversy. One approach, strongly advocated by Nickell (1988), uses the wage-price mechanism of macroeconometric models to define a “supply side”, which arises as the long-run solution to dynamic wage and price equations.1 An application in this spirit is found in the old MAS macroeconomic model described in Low (1994), also known as Singmod.2 In essence, this involved the estimation of an expectations-augmented Phillips curve to explain nominal wage growth on the basis of past consumer price inflation, the level of unemployment and changes in labour productivity, from which the non-accelerating inflation rate of unemployment (NAIRU) can be inferred. To accommodate the institutional features of the Singapore labour market, however, we shall instead adopt a structural approach that entails specifying and estimating demand, supply and wage equations for the labour market. This framework is depicted in Figure 5.1. Consistent with competitive firms maximizing profits subject to a Cobb-Douglas production function, the demand for labour is specified as a negative function of the real wage rate and a positive function of the rental price of capital and aggregate output: Nt = f1
p
Wt p , rt , Yt Pt
(5.1)
where N is labour demand or employment. W p /P p is the labour costs facing producers i.e. nominal wages inclusive of CPF contributions deflated by producer prices. r is the rental cost of capital in real terms and is meant to capture factor substitution; as labour becomes more expensive, firms may resort to more capitalintensive operations. The supply of labour, defined simply as the labour force (L), is drawn as an upward-sloping curve in Figure 5.1.3 The key determinant of the labour force is the total working population aged 15 years and above (POP), consisting of the resident population (Singapore citizens and permanent residents) and non-resident population (primarily foreign workers). The workforce also depends positively on the real consumption wage W c /P c , calculated as nominal wages plus CPF
1 Chapter 4 of Whitley (1994) provides a detailed exposition of the approach. 2 This model has been superceded by the Monetary Model of Singapore (MMS). See Enzler et al. (2005). 3 In the absence of statistical data on hours worked in Singapore, labour demand and supply are specified in terms of the number of persons engaged in productive activities.
72
The labour market
Figure 5.1 The labour market.
contributions less income taxes, deflated by the CPI. The aggregate labour supply function can be written symbolically as: Lt = f2
Wtc POPt , Xt Ptc
(5.2)
where X represents other relevant variables. After eliminating the differences between producer wages and consumer wages, Figure 5.1 plots labour demand and supply against a measure of the real wage that is expressed in terms of a common price index. As shown in the figure, the labour market is assumed to clear in the long run, subject to a “natural” amount of unemployment U ∗ that is given exogenously i.e. L∗ = N ∗ + U ∗ , where an asterisk denotes an equilibrium quantity. Due to frictions and mismatches even in this classical labour market, there are workers in the labour force who are unemployed at the market-clearing real wage (W /P)∗ , which from (5.1) and (5.2) works out to be:
W P
∗
= f3 (Ut∗ , rt , Yt , POPt , Xt , Tt )
(5.3)
t
where T (for tax wedge) is a set of CPF, tax and price variables responsible for inserting a wedge between producer and consumer wages. In the short term however, the labour market is in disequilibrium due to the presence of real rigidities – a presumption of New Keynesian macroeconomic theories.
The labour market 73 As a consequence, in response to the downward shift in the labour demand schedule, the real wage only adjusts slowly towards the market-clearing equilibrium as follows:
W P
= f4 t
W − P
W P
∗
, Ut∗ , rt , Yt , POPt , t−1
Xt , Tt , NWCt
(5.4)
The quantum of adjustment in the short run is hypothesized to be proportional to the gap one quarter ago between the actual and equilibrium wage levels, as well as to changes in the determinants of the demand and supply of labour. NWC wage guidelines, which have a distinctly Singaporean flavour, are also added as an explanatory variable to account for the NWC’s short-term influence on the wage-setting process. It is worth highlighting a number of important features entailed by Equation (5.4). First, though it resembles a short-run Phillips curve, we have actually modelled changes in the real wage, and not the nominal wage. Second, the disequilibrium term is [W /P − (W /P)∗ ] rather than the usual unemployment rate. The latter does not fully account for the disequilibrium forces acting on wages because foreign workers have to leave the country if they are unemployed (cf. Section 5.4). The exodus of these workers, however, has an impact on wages which is captured by our disequilibrium term.4 Third, far from being an ad hoc formulation, Equation 5.4 is embedded in an integrated model of labour demand and supply, and acts as the dynamic adjustment mechanism within our disequilibrium framework. Notice that the nominal wage can be determined by jointly solving with a price equation, to be discussed in Chapter 7. Finally, the unemployment rate is derived from the usual identity for labour market flows: ut =
Ut Lt
100 =
Lt − Nt Lt
100
(5.5)
This measure of unemployment is the discrepancy between the demand for and supply of labour at the “sticky” real wage of W /P in Figure 5.1. As this wage gradually returns to its market-clearing value in the long run, employment increases to its equilibrium level and the unemployment rate converges to the natural rate. The increase in employment in turn has a feedback effect on prices through productivity changes in the economy, as we shall see later.
4 Another advantage of our approach is that it does not require data on foreign workers; until recently, this was one of the best kept secrets in Singapore.
74
5.3
The labour market
Labour market equations
In this section, we describe the estimation of the labour demand-supply system with disequilibrium dynamics summarized by Equations (5.1)–(5.5). The linearized specification of the market-clearing model is given by: emp
ln Nt = α0 + α1 ln
Wt (1 + CPFt p Pt
ln Lt = β0 + β1 ln
Wt (1 + CPFt Ptc
emp
)
+ α2 rt + α3 ln Yt
(5.6)
y
− τt )
+ β2 ln POPt + β3 ln Xt
ln(1 − ut∗ ) = ln Nt − ln Lt
(5.7) (5.8)
where CPF emp is the employers’ contribution rate to the provident fund, a variable under policy control. In view of the fact that such contributions are exempt from tax, the nominal wage package to the worker is W (1 − τ y ) + W · CPF emp = W (1 + CPF emp − τ y ), where τ y is the average tax rate for individual income earners constructed in Appendix A. While (5.6) and (5.7) represent labour demand and supply, respectively, Equation (5.8) states the equilibrium condition (u∗ is now expressed as a fraction of the labour force). Inserting this condition into either of the other equations leads to the following solution for the market-clearing real wage:
W ln Pc
∗ t
=
1 [α0 − β0 − ln(1 − ut∗ ) + α2 rt + α3 ln Yt β1 − α 1 − β2 ln POPt + β3 ln Xt ] emp
+
P c (1 + CPFt ) α1 y emp − ln(1 + CPFt − τt ) ln p t β1 − α1 Pt (1 + CPFtemp − τty ) (5.9)
The dependent variable in this formulation is the real wage expressed in consumer prices and excluding CPF contributions and taxes. The last two variables on the right-hand side of the equation define the tax wedge, which is potentially relevant because employers and employees focus on different wage concepts. For estimation purposes, the last term can be merged with the dependent variable. Using the approximation ln(1 − u∗ ) ≈ −u∗ ,− ln(1 − u∗ ) in (5.9) can also be replaced with u∗ for ease of interpretation. As we have already indicated, the market-clearing wage rate given in Equation (5.9) plays a crucial role in our analysis. Unfortunately, we find it difficult to obtain a robust estimate of the wage elasticity of labour supply (β1 ) by estimating (5.7). Ideally, one should estimate a relationship for labour supply in terms of the hours workers are willing to supply at a given wage rate. This is again bedevilled by the absence of information on hours worked in Singapore. We decided to
The labour market 75 bypass (5.7) and derive the market-clearing wage rate by estimating (5.9) directly. Even though this creates an over-identification problem, we found in the empirical implementation that the problem is not serious.5 Labour demand Preliminary unit root tests revealed that in the labour demand equation (5.6), the rental cost of capital is I (0) while the other variables are well characterized as I (1) processes (r is taken to be the real prime lending rate, P p is the domestic supply price index and Y is replaced by its correlate, real GDP). Moreover, the Johansen trace test supports co-integration among these variables. An ADL model with one lag on each variable produces the following demand for labour during the period 1981Q1–2003Q4: ln Nt = 1.19 −0.41 ln W (1+CPF)emp /DSPI t + 0.002 rt + 0.91 ln GDPt (4.69) (−6.76) (1.95) (13.8) (5.10) The ECM test confirms the presence of co-integration with a t-statistic of −4.90. The high t-ratios attached to the coefficients for the real wage rate and GDP indicate that they are estimated with considerable precision. These labour demand elasticities stand in contrast to estimates by Singapore’s Ministry of Trade and Industry (Tan et al., 2002), who reported an output elasticity of 0.61 and a real wage elasticity of −0.16 from a regression that does not include a cost-of-capital variable. Our findings imply a tighter link between job creation and GDP growth, although we will see that labour market adjustment is slow. The t-statistic for the rental cost of capital is not always significant in recursive runs but it has the right positive sign suggesting a substitution away from labour as investment goods become relatively cheaper. A better measure of the cost of capital is needed to capture the full effect of factor substitution in the long run. By writing the composite wage term in (5.10) as ln(W /CPI )∗t + ln [CPI (1 + CPF emp )/DSPI ]t and substituting for the market-clearing wage ln(W /CPI )∗t from (5.13) below, we obtain estimates of ln Nt∗ , the equilibrium employment level. These estimates are incorporated into the ECM: ln Nt = 0.0001 + 0.57 ln Nt−1 + 0.11 ln GDPt + 0.11 ln GDPt−1 (0.14) (8.71) (3.74) (3.69) + 0.07 ln GDPt−2 −0.06 (ln N −ln N ∗ )t−1 (2.17) (−3.71)
(5.11)
5 Non-unique solutions for equilibrium quantities may arise because, given estimates from the employment and wage equations and ignoring the intercept terms, three values each of β1 and β2 can be worked out. Nevertheless, the values appear in a narrow range despite this over-identification problem.
76
The labour market
R 2 = 0.82 SE = 0.004 DW = 2.03 LM(5) = 1.39 (0.24) ARCH(4) = 1.38 (0.25) Chow = 1.24 (0.30) Normality = 4.01 (0.13) Heteroscedasticity = 1.19 (0.31) RESET = 7.86 (0.01) As expected, changes in the price of capital have no effect on employment in the short run. Equation (5.11) confirms a well-known fact about the Singapore labour market: employment growth tends to trail behind GDP growth by 3–6 months, thereby making the number of jobs a lagging indicator of economic activity. The small, albeit statistically significant, error-correction coefficient indicates that there are substantial costs to changing employment levels which stem from recruiting, training and sacking workers. As a result, a 1% increase in real GDP causes businesses to raise employment by a mere 0.11% in the current quarter (Figure 5.2). Only two quarters later does the impulse response achieve a maximum effect of 0.23%. After four quarters, about 90% of the adjustment to long-run equilibrium is completed, with the balance spread out over another year. Labour supply As stated earlier, obtaining a reliable estimate of the wage elasticity of labour supply through estimation of (5.7) is problematic. If the labour force participation rate is roughly constant, then changes in the size of the workforce will be determined essentially by the trend increase in the working age population, implying that β1 = β3 = 0 and β2 = 1. Figure 5.3 plots annual participation rates in Singapore. The total participation rate published by DOS has remained roughly the same between 1980 and 2003. In spite of a substantial trend increase in the female participation rate (22%), this is not fully reflected in the total rate due to a drop in the male participation rate (7%). The male participation rate has declined because with an aging population, the proportion over 65 years old who are no longer in the labour force has grown correspondingly. In the case of females, the secular rise in their activity rate is
Figure 5.2 Impulse response for employment.
The labour market 77
Figure 5.3 Labour force participation rates (%).
explained by the strong economic growth experienced by Singapore over the past three decades and the concomitant increase in real wages caused by a shortage of labour. As elsewhere, changes in socio-economic norms and institutions that affected women’s behaviour – such as higher educational attainment, equality of employment opportunities and smaller families – doubtless also played an important role. These observations suggest that separate modelling of male and female labour supplies is likely to be fruitful. Sadly, the lack of the required quarterly data prevents us from engaging in such an exercise. Even at the aggregate level, our own estimate of the participation rate in Figure 5.3 contains a more pronounced cyclical component compared to the DOS estimate. As explained in Appendix A, we derived our estimates from published data on changes in employment and unemployment rates. The employment figures include foreign workers at construction sites and daily commuters from neighbouring Malaysia. Since the DOS estimate excludes these workers, who move in and out of the labour force depending on their employment status, our participation rate tends to fluctuate more than theirs. To capture the effect of increasing female labour force participation, we tried a number of additional variables (X ) in the aggregate labour supply equation, including the residential property price index, the wealth and loans variables used in Chapter 2, and an interpolated series on the female education level.6 None of them performed well. Consequently, we have to be content with a first differences
6 An increase in wealth is expected to have a negative effect on labour supply, but this could be more than offset by a rise in loans precipitated by ballooning house and car prices.
78
The labour market
regression for explaining labour force movements in the short run. Our preferred specification captures the “discouraged worker” effect with the lagged change in the unemployment rate: ln Lt = 0.80 ln Lt−1 + 0.24 ln POPt + 0.58 2 ln POPt (14.7) (2.8) (7.52) −0.003 ut−1 (−5.37)
(5.12)
R 2 = 0.906 SE = 0.003 DW = 1.47 LM(5) = 2.66 (0.03) ARCH(4) = 0.24 (0.91) Chow = 2.04 (0.03) Normality = 44.5 (0.00) Heteroscedasticity = 0.72 (0.69) RESET = 0.68 (0.41) Notice that we have included an extra term for the working age population to capture second-order effects generated by accelerating or decelerating population growth rates. We find here that the recent trend in unemployment has a statistically significant effect on current labour supply via pro-cyclical changes in the participation rate. During a severe downturn, unemployed workers might stop looking for a job after a while and quit the labour force. The R2 statistic indicates that the equation fits extremely well, notwithstanding autocorrelated and non-normal residuals. The Chow test statistic is misleading in this regard, biased as it is by the outlier in 2003Q3 caused by the SARS crisis, which we have adjusted for with an impulse dummy variable in estimating (5.12). Wage determination The focus of this section is the derivation of the market-clearing wage rate in (5.9), which is required for estimating equilibrium employment and the wage adjustment Equation (5.4). To do this, we need extraneous estimates of u∗ , the natural unemployment rate in Singapore. Under our structural shift hypothesis elucidated in Section 5.4, we can describe the evolution of the natural rate with a step function defined by u∗ = f5 (D1, D2, D3), where D1, D2 and D3 are (1,0) dummy variables defined over the three sub-periods shown in Figure 5.8 (D1 = 1 over 1980Q1–1988Q1 and 0 otherwise; D2 = 1 over 1988Q2–1998Q2 and 0 otherwise; D3 = 1 over 1998Q3–2003Q4 and 0 otherwise). After several attempts to estimate (5.9), however, we realized that the above dummies as well as other possible determinants of u∗ are superfluous in the regression because the structural shifts of the natural rate are mirrored by analogous shifts in real GDP (see Figure 5.7). We also experimented with the X variables that we had earlier considered in the labour supply equation, but not one of them yielded a significant coefficient in the wage equation. As for the remaining variables, the coefficients of ln GDP and ln POP are found to be significant and can be restricted to 1 and −1, respectively, thus providing a very plausible unitary-elastic relationship between the real wage rate and the per capita income of the working age population in the long run. The restricted regression estimates for the period
The labour market 79 1982Q1–2003Q4 are given below: ln
W CPI
∗ t
GDP = 5.27 + 0.001 rt +ln POP t (287.0) (0.33) emp CPIt (1+CPFt ) y emp −0.20 ln −τt ) y −ln(1+CPFt emp DSPIt (1+CPFt −τt ) (−3.23) (5.13)
Equation (5.13) forms a very robust co-integrating relationship with a highly significant ECM t-statistic of −4.19. Despite the statistical insignificance of r, we retained it in the regression because of its important role in the labour demand equation. The last two terms can be combined approximately into the empirical tax wedge variable [−0.2 ln(CPI /DSPI ) − CPF emp + 0.8τ y ]. This means that a 1% increase in the relative price ratio lowers real wages by 0.2% in the long run. Furthermore, a one percentage point increase in the CPF rate shifts the labour demand curve to the left and the supply curve to the right, leading to a 1% fall in the real wage. Finally, the same rise in the income tax rate shifts only the labour supply curve up and raises the long-run real wage by 0.8%. Figure 5.4 plots the observed real wage rate against the market-clearing wage rate predicted by Equation (5.13). It is instructive to observe the persistence of the actual wage despite market pressures for it to fall during the 1985–86 recession, the Asian financial crisis in the late 1990s and again at the height of the SARS crisis of 2003. Interestingly, the real wage is only rigid in the downward direction – a conclusion gleaned from the observation that the solid and dashed lines in Figure 5.4 stay close to each other during the economic boom of the 1990s.
Figure 5.4 Actual wages versus market-clearing wages (S$).
80
The labour market
In the same figure, we show the real wage rate inclusive of employers’ CPF contributions. What is interesting about this graph is that the timing of the government-induced CPF corrections to wages during downturns did not coincide with the declines in the market-clearing wage rate. In fact, CPF adjustments lagged behind the market rate: the latter started to fall in 1985 as the economy entered into recession but the CPF rate was only cut in 1986; similarly, the market-clearing wage fell in 1997 while the CPF adjustment was belatedly implemented two years later. Once good estimates of the market-clearing real wage have been obtained, a proxy for NWC wage guidelines needs to be found for inclusion in Equation (5.4). Modelling NWC and government intervention in the labour market econometrically is a difficult problem, however. Cognizant of the fact that the NWC has been issuing qualitative wage recommendations since 1986, we decided to introduce an indicator variable to account for its possible influence. This NWC variable takes on the value of 1 when wage increases are recommended across the board, 0 when wage restraint is urged, and −1 if a wage cut is exhorted.7 Upon estimating the disequilibrium wage equation with this indicator variable entered contemporaneously or with a one quarter lag, we found that its coefficient is totally insignificant. We may therefore surmise that, quite contrary to the popular perception, the Council’s guidelines are no longer as relevant to wage determination in the Singapore context as findings pertaining to an earlier period suggest (Chew, 1988). The final wage adjustment equation that we settled for is reported below:
W ln CPI
emp
t
= 0.016 −0.57 CPFt −0.74 ln POPt−1 (7.0) (−4.92) (−2.83) W W ∗ −0.23 ln − ln CPI CPI (−7.05) t−1
(5.14)
R 2 = 0.61 SE = 0.014 DW = 2.34 LM(4) = 1.04 (0.40) ARCH(4) = 1.37 (0.25) Chow = 1.05 (0.35) Normality = 0.15 (0.93) Heteroscedasticity = 1.68 (0.13) RESET = 1.62 (0.21) The simple form of the wage adjustment equation that emerged in (5.14), which is seen to be in error-correction form, underlies our departure from the Phillips curve format. Both the fit and diagnostics of the equation are surprisingly good, notwithstanding the difficulties encountered in modelling wages in practice.8 In contrast, a standard Phillips curve specification with the growth in
7 The sources of information on the NWC’s wage guidelines are the Singapore Yearbook of Labour Statistics and the Singapore National Employers Federation (SNEF) website. 8 An impulse dummy was used in (5.14) to account for a large fall in the real wage rate in 2001Q1. Even without this dummy, the model fits quite well with highly stable recursive estimates, though it fails some of the diagnostic tests.
The labour market 81
Figure 5.5 Growth rates of resident and foreign populations (%).
real wages as the regressand (i.e. with static price homogeneity imposed) and the unemployment rate and productivity growth as regressors, has an R2 of only 34%, compared to 61% for our disequilibrium equation. When the CPF rate is held constant in Equation (5.14), short-run wage adjustments occur primarily as a response to changes in the labour force brought about by the working-age population and excess demand (or supply) pressures. Figure 5.5 demonstrates that it is the foreign component of the population that fluctuates wildly and hence contributes to the short-run movements of the real wage rate. In other words, foreign workers who constitute close to a third of the workforce are now a major determining factor of earnings growth in Singapore. At the same time, the highly significant coefficient of the disequilibrium term shows that the labour market also reacts strongly to excess demand pressures. Construed as the speed of convergence, the estimated value of 0.23 means that nearly twothirds of any existing disequilibrium is removed in 1 year. While it takes time, the real wage eventually adjusts to the long-run equilibrium level determined by the macroeconomic, demographic and policy variables in (5.13).
5.4
Unemployment in Singapore
We turn now to examine unemployment in Singapore. For a start, Figure 5.6 plots the predicted rates of unemployment obtained from Equations (5.5), (5.11) and (5.12). The graph shows that our econometric model of the labour market produces respectable forecasts of the unemployment rate, especially in recent years (actual unemployment rates prior to 1986 were backcast using data on employment and the labour force). Putting aside the fact that they are somewhat more volatile, the predicted rates are able to replicate the turning points in unemployment cycles.
82
The labour market
Figure 5.6 Forecasts of unemployment.
A comment on the actual and fitted unemployment rates shown in Figure 5.6 is warranted. As long as the government sets a limit on the number of foreign workers in the country such that their supply does not exceed demand and all of them are gainfully employed (i.e. their participation rate is 100%), these overall rates underestimate the extent of unemployment amongst domestic residents.9 Such an upper bound on supply is indeed assured by the foreign worker quota system that is in place, whereby employers have to apply for work permits or employment passes to be issued to imported labourers and also pay a monthly levy to the government. Official figures released for the period 1997–2003 show that the difference between the overall and the resident unemployment rates has hovered around half a percentage point. By the same argument, the natural rate of unemployment is understated in Singapore. The empirical estimate of this rate in the Singmod model is about 3% (Low, 1994), while Ministry of Manpower researchers have lately put it at 4% based on the estimation of a Beveridge curve for the Singapore economy (Teo et al., 2004).10 New research by the central bank allowing for a time-varying natural rate of unemployment suggests that equilibrium unemployment rose from
9 Using our notation N for employment, L for labour force, U for unemployment and letting R stand for residents and F for foreigners, we have U = L − N = LR + LF − N R − N F = LR − N R = U R because LF = N F . Therefore, u = (L − N )/L = U R /L underestimates the resident unemployment rate. 10 The Beveridge curve is an inverse relationship between the unemployment rate and the vacancy rate. In steady-state equilibrium, equality between these two rates yields an equilibrium unemployment rate that is often interpreted as the natural rate.
The labour market 83 2.9% during the early 1980s to 3.5% at the peak of the 1985 recession (Monetary Authority of Singapore, 2004). It then fell to 2.3% between 1990 and 1997 – an unprecedented period of economic prosperity which led many economists to think that the natural rate was fixed. Since the onslaught of the Asian financial crisis, however, the MAS found that it has again risen to around 3–3.5%. When estimating (5.9), we required estimates of the natural rate of unemployment u∗ . If observed unemployment u is a pure I (0) process, then we could simply take the average unemployment rate to be the natural rate. However, an augmented Dickey-Fuller test delivers the conclusion that u is an I (1) process. Since the unemployment rate is typically a stationary variable, this must be the result either of local trends or structural shifts. Singapore is neither a welfare state nor was there any major demographic transition during our sample period, so structural economic change appears to be the more likely cause of the unit root present in the unemployment rate.11 We have already alluded to the structural shift hypothesis for manufacturing value-added content in Chapter 3, Figure 3.2. This phenomenon is also observed in other key macroeconomic series such as real GDP and the unemployment rate. Figure 5.7 plots the logarithm of the GDP series with superimposed trend lines from different periods. After the 1985–86 recession, the path of output was thrown off its original trend line and it settled down eventually to an almost parallel new trend rate of growth. In the wake of the Asian financial crisis, the trend of GDP shifted down again and this was followed by a further decline after the global IT industry went bust in early 2001.
Figure 5.7 Structural shifts in real GDP.
11 We came to this conclusion after trying different methods of estimating a natural rate that changes only slowly. One such estimate we obtained was from a regression of the unemployment rate on the trend components of the vacancy rate and the female share of the labour force. Even though this estimate appears quite plausible, it did not play a significant role in our wage equation.
84
The labour market
Figure 5.8 Structural shifts in the unemployment rate.
Similar breaks can be discerned in the unemployment rate as well (Figure 5.8). Before the onset of the 1985–86 recession, Singapore experienced a higher average unemployment rate (3.2% between 1979 and 1984) than that observed between 1988Q2 and 1998Q2 (1.9%). In retrospect, the jump in the unemployment rate to 6% in 1986 demarcates a clear structural discontinuity in the post-independence Singapore economy. The next big shift came after the turbulent Asian financial crisis. In 1998Q3, the unemployment rate rose to 4% and ever since then, it has stayed high on average (3.8% if one excludes the SARS-ridden third quarter of 2003 when unemployment rose to 5.7%). In Figure 5.8, the deviations of the unemployment rate from the sub-period averages turn out unambiguously to be I (0) – as they should be. We take this finding to mean that both the observed and natural rates of unemployment in Singapore are stationary variables which were subject to discrete structural shifts from time to time.
5.5
Policy conclusions
The policy conclusions arising from our modelling of the labour market block in the ESU01 model pertain to four major issues: foreign workers, the role of the NWC, flexibility of the labour market, and the natural unemployment rate. First, foreign workers act as a volatile buffer against aggregate demand shocks, alleviating the latter’s effects on the recorded unemployment rate through systematic changes in the labour force. This is demonstrated by our labour force equation – when the economy slows, working-age population growth follows suit as foreign workers are repatriated, leading to more moderate labour force growth and a lower measured unemployment rate than would be the case. Accordingly, we recommend
The labour market 85 that the relevant authorities and the public pay more attention to the resident unemployment rate as an indicator of joblessness. During better times, foreigners also play a key role in the wage adjustment process through their impact on labour supply. By supplementing the indigenous reservoir of workers and helping to ease labour shortages, they have held down wage increases. Based on the recent strength of the foreign workforce, it is estimated that a 1% increase in their numbers dampens average real wages by slightly less than one-third of a percent (see also the policy simulation results in Chapter 9). The implication is that the Singapore government still has at its disposal a potent instrument for intervening in the labour market, namely adjustments to the foreign worker quotas and levies. By fine-tuning these, the government can indirectly control labour supply, wage costs and ultimately, Singapore’s international competitiveness. Second, the role of the NWC in determining wage settlements in Singapore seems to have diminished due to the reforms implemented in the labour market since the late 1980s, most notably the move towards a flexible wage system. By encouraging companies to pay both a base wage and a variable component that depends on enterprise profitability and productivity, the flexi-wage system assigns a greater role to market forces in shaping earnings. The NWC recommendations could also have lost their bite after the switch to the issuance of general guidelines in 1986. Whatever the cause, we did not find econometric evidence to indicate that actual real wage growth was significantly influenced by the Council. The government ought therefore to re-evaluate the objectives of this tripartite forum for the future, perhaps by focusing on how it can contribute to spotting trends in wages in a more timely manner. Third, institutional inertia and real rigidities are present in Singapore but do not appear to be prevalent, as reflected by the responsiveness of the labour market to disequilibria. Still, changes in employment and real wages take some time to be effected, as we saw from the historical persistence of these variables. The faster wages respond, the less time the unemployment rate spends away from the natural rate, the lighter the burden falling on quantity adjustment through changes in employment and the foreign workforce, and the lesser the need for direct and painful cuts in the employers’ CPF contribution rate. We therefore support renewed efforts at wage reform and also welcome the move announced in August 2003 to introduce greater downward flexibility into workers’ total wage package by allowing the CPF rate to vary counter-cyclically with economic conditions – going up in good years and coming down in bad times (although how this will be implemented remains to be seen). In this connection, an econometric model of the labour market such as the one we estimated in this chapter can provide useful guidelines on the quantum of the CPF or wage cuts required to minimize market distortion whenever adverse shocks hit the economy. Fourth, the natural rate of unemployment appears to have gone up in spite of increased wage flexibility. None of the usual explanations for an increase in the natural rate – unemployment benefits, trade union power, government regulations and minimum wage laws – apply to the upward trend in Singapore’s equilibrium
86
The labour market
unemployment rate. Instead, a more plausible reason is the accelerated pace of economic and technological change, which in turn stemmed from globalization and intensified competition from emerging economic giants like China and India. The result is increased structural unemployment caused by skill mismatch between the demand for and supply of various types of workers and higher labour market “churn”. Despite the apparent rise in the natural unemployment rate during recent years, there is no reason to conclude that actual unemployment will be permanently higher. Indeed, empirical evidence tends to suggest that, over long spans of time, the unemployment rate is mean-reverting. We can safely assume that natural unemployment in Singapore is made up of frictional and structural components. It is unlikely that frictional unemployment will rise noticeably in the future in light of expected advancements in communication technologies. In addition, the ongoing restructuring process of the Singapore economy in response to the changing external environment and intense efforts at retraining low-skilled and older workers might yet bring the structural unemployment rate down again.
6
6.1
Sectoral production
Introduction
Policymakers and private analysts in Singapore closely monitor the growth performances of the major sectors in the economy. To accommodate this popular demand for sectoral forecasts and enrich the results of simulation exercises, we extend our model to cover production by the sectors. Once we decided to do this, the sectoral block of the ESU01 model becomes responsible for churning out the economy’s aggregate supply of goods and services, which interact with aggregate demand to determine inventories endogenously and thereby close the model (see Figure 1.1 and also Chapter 7). The most common approach to modelling the supply side of the economy is to estimate production functions. When we created the capital stock for the manufacturing sector (see Appendix A), we had expressed our reservations on the use of the perpetual inventory method. However, to even use this method to estimate sectoral production functions, detailed investment data broken down by sector are required. Due to the lack of such time series, Abeysinghe and Lee (1997) modelled Singapore’s sectoral value-added by specifying reduced form regressions derived from supply and demand equations for each sector. In the present chapter, we adopt a different and novel method that fits neatly into the demand-driven structure of the ESU01 model.
6.2
Supply-side modelling
We start by postulating the following relationship for a business firm that holds inventory stocks with the motive of smoothing production1 : Qt∗ = Ste + (INVTt∗ − INVTt−1 )
(6.1)
where Q∗ is the planned output level, S e is the expected level of sales and INVT ∗ is the desired stock of inventories. We assume that firms do not produce the optimal
1 The basic assumptions in our model are similar to those in Fair (1984, Chapter 4) but the final formulation is very different.
88
Sectoral production
level of output or adjust inventories to the desired level immediately because of adjustment costs (cf. the export model in Chapter 4).2 If partial adjustments are made in the short run, we can specify the quarterly change in output as follows: Qt − Qt−1 = λ(Qt∗ − Qt−1 ) Substituting this into (6.1) gives us: Qt = (1 − λ)Qt−1 + λSte + λ(INVTt∗ − INVTt−1 )
(6.2)
We can now replace S e with the predicted level of sales from an appropriate forecasting model. The relationship between actual and predicted sales is given by St = Ste + et , where et is a random forecasting error. By substituting this definition into (6.2), we get: Qt = (1 − λ)Qt−1 + λSt + λ(INVTt∗ − INVTt−1 ) − λet
(6.3)
If λ < 1 due to adjustment costs, (6.3) shows that the current output level depends on the lagged output level, actual sales, the desired change in inventories, and a prediction error. When there are random shocks to sales, expectational errors in anticipating future sales result in unintended accumulations or depletions of the inventory buffer stock. One way to convert (6.3) into an estimable form is to assume that the desired inventory level is proportional to the expected level of sales (INVTt∗ = δSte ) and unplanned inventories follow a stationary AR process of the form φ(L)et = εt , where εt is a white noise error term. Plugging these into (6.3) produces a dynamic equation involving Qt , St and INVTt−1 . Unfortunately, we could not estimate this equation because of the difficulty encountered in constructing an inventory stock series from the available data on the change in stocks. Figure 6.1 shows that since about 1996, the change in inventories has swung in the negative direction in a big way. With any reasonable starting value, an inventory stock time series constructed from the data in Figure 6.1 will produce negative values towards the end of the sample period. While it is true that the Singapore economy has been in turbulence since the outbreak of the the Asian financial crisis in 1997, such a large inventory decumulation is unlikely to occur unless there are measurement errors. Due to the apparent data problem discussed above, we make no attempt to decompose the observed change in inventories into its planned and unplanned components i.e. INVTt = (INVTt∗ − INVTt−1 ) − et . As quarterly data on gross output and sales is also not available, we have to work with value-added (VA) and
2 Firms can adjust their inventory levels without changing production levels through imports. More than 85% of the change in inventories in Singapore is made up of imports.
Sectoral production 89
Figure 6.1 Change in inventories ($ million).
final demand (FD). Deducting intermediate products (including imported inputs) from both output and sales proceeds, (6.3) can be expressed as: VAt = (1 − λ)VAt−1 + λFDt
(6.4)
where FDt = Ct + It + Gt + TXt + INVTt . The formulation in (6.4) provides the basic structure for our exercise. Statistical data on final demand are compiled for the Singapore economy in terms of the standard categories of consumption, investment, government spending, net exports and the change in stocks. Following (6.4), we can in principle regress the valueadded of each sector on these expenditure components individually (after replacing net exports with our own series on total exports of goods and services, denoted TX above). However, this creates a severe multicollinearity problem in practice. To circumvent the problem, we work out extraneous estimates of the final demand for the ith sector in a way analogous to the computation of final import demand in Section 4.7: FDit =
Ci Ii Gi TXi INVTi Ct + It + Gt + TXt + INVTt C I G TX INVT
(6.5)
where Zi is the value-added contribution of the ith sector for a given type of final expenditure. Singapore’s input-output (IO) tables directly provide these contributions as a ratio to expenditures i.e. the Zi /Z shares. Based on them, we can construct the final demand series for the value-added output of each sector.
90
6.3
Sectoral production
Sectoral equations
Table 6.1 shows the sectoral value-added shares of different categories of final demand, taken from the Singapore IO Tables for the years of 1988, 1990 and 1995. The shares for the manufacturing sector are given for non-oil and oil exports separately while those for the construction industry are not shown since it is dealt with in a quite different manner below. The commerce sector subsumes two smaller sectors – wholesale and retail trade, and hotels and restaurants. It is seen that stock changes form a substantial part of the final demand for manufactured goods only, as inventory investment cannot be expected to play a significant role in services production. Except for the highlighted figures, the shares do not show much variation over the time span covered by the years. The substantial jumps seen in the bold numbers appear to reflect peculiar factors affecting the financial, business and other services sectors during the years in which the IO survey was conducted. As a remedy, we used simple growth rate regressions of value-added on expenditure components to calibrate the most suitable weights. Most of the calibrated weights that result are essentially the averages of the 1990 and 1995 figures in Table 6.1. The final demand equations are: FDMANt ≡ 0.0316Ct + 0.037It + 0.0125Gt + 0.0117ODXt + 0.1788NODXt + 0.1062INVTt
(6.6)
FDCOMt ≡ 0.1749Ct + 0.0413It + 0.022Gt + 0.0834TXt + 0.0079INVTt
(6.7)
Table 6.1 Sectoral value-added composition of final demand Year C Manufacturing
1988 1990 1995 Commerce 1988 1990 1995 Transport & 1988 communications 1990 1995 Financial & 1988 business 1990 services 1995 Others 1988 1990 1995
0.0429 0.0362 0.0271 0.1500 0.1652 0.1206 0.0621 0.0662 0.0704 0.1745 0.2612 0.2682 0.0997 0.0180 0.0150
NODX ODX
TX
INVT
I
G
0.0417 0.0363 0.0377 0.0413 0.0502 0.0462 0.0150 0.0108 0.0147 0.0571 0.0537 0.0836 0.0081 0.0039 0.0030
0.0198 0.1843 0.0124 0.1187 0.0132 0.1808 0.0146 0.0850 0.0118 0.1768 0.0087 0.1274 0.0220 0.0598 0.0079 0.0231 0.0718 0.0131 0.0187 0.0813 0.0104 0.0123 0.0835 0.0087 0.0207 0.0812 0.0047 0.0211 0.0709 0.0024 0.0520 0.0621 0.0117 0.7557 0.0739 0.0165 0.7317 0.0974 0.0114 0.5914 0.0161 0.0057 0.0073 0.0095 0.0062 0.0068 0.0076 −0.0067
Source: Singapore IO Table XVI, various years. Bold figures show substantial jumps in the coefficients.
Sectoral production 91 FDT &Ct ≡ 0.12Ct + 0.0128It + 0.0209Gt + 0.0335TXt + 0.0036INVTt
(6.8)
FDF&Bt ≡ 0.2647Ct + 0.0571It + 0.7317Gt + 0.0621TXt + 0.0117INVTt FDOTHERt ≡ 0.0165Ct + 0.0035It + 0.0071Gt + 0.0086TXt FDCONt ≡
p ICONt
g + ICONt
(6.9) (6.10) (6.11)
where MAN = manufacturing, COM = commerce, T&C = transport and communications, F&B = financial and business services, CON = construction, and OTHER = GDP − sum of the value-added of the above sectors. Clearly, final demand for the construction sector depends only on fixed capital formation in buildings and structures, so we set it equal to the sum of private construction investment (ICON p ), modelled in Chapter 3, and exogenously determined public construction investment (ICON g ). Figure 6.2 plots value-added and final demand for the major sectors in the Singapore economy over the period 1990Q1–2003Q4 (the graph for the other services sector has different vertical scales). Apart from transport and communications, all the sectors show a strong co-movement of the major turning points in VA and FD, even though their levels do not coincide. The graphs suggest that these two variables are co-integrated for the most part, a feature consistent with the partial adjustment equation we derived for sectoral value-added in (6.4). A serious drawback of the partial adjustment model, however, is that it imposes rather implausible restrictions on production dynamics. Instead of using this model, we employ more general log-linear ADL and ECM formulations for estimating the long-run and short-run relationships, respectively, between the value-added and final demand of each sector. The estimation results are presented in Table 6.2 and the actual and predicted quarterly growth rates are plotted in Figure 6.3. The error-correction coefficients in the equations for all the sectors are statistically significant, ranging in magnitude from −0.08 to −0.3. In terms of diagnostic statistics, all the models fit the data adequately. However, these mask substantial differences in predictive performance. Visual inspection of Figure 6.3 shows that the construction sector has the best fit. The models for the manufacturing, commerce, and other services sectors also provide reasonably good historical forecasts. The poorest performances occur in the transport and communications, and financial and business services sectors. One would have expected the former, which includes air and sea cargo haulage, to be more predictable; given the weak correlation between the sector’s VA and FD in Figure 6.2, however, this does not seem to be so. In any case, growth in this sector is the least variable. By contrast, the growth rates of financial and business value-added appear to be just white noise with mild negative first-order autocorrelation. This erratic behaviour
92
Sectoral production
Figure 6.2 Sectoral value-added (solid line) and final demand (dashed line).
Sectoral production 93
Table 6.2 Error correction models of sectoral value-added Manufacturing Construction Commerce Transport & Financial & Others communications business Constant ECt−1 ln FDt
0.953 (2.93) −0.287 (−2.94) 0.634 (8.05)
−0.204 (−2.99) −0.226 (−3.03) 0.865 (20.3)
ln FDt−1
0.424 (2.62) −0.236 (−2.64) 0.789 (8.83) 0.149 (1.64)
0.023 (6.22) −0.083 (−2.51) 0.198 (3.10)
0.284 (4.38) −0.299 (−4.12)
ln FDt−2 ln FDt−3 ln FDt−4 ln PSTOCK Long-run coefficient Sample period R2 SE DW AR(4) F ARCH(4) Normality Hetero RESET
0.66
1.00
1992–2003
1987–2003
0.61 0.025 2.05 0.64 (0.638) 0.81 (0.525) 0.23 (0.892) 0.62 (0.651) 2.58 (0.115)
0.88 0.016 2.18 2.01 (0.090) 0.61 (0.658) 3.31 (0.191) 0.37 (0.827) 0.90 0.345)
0.75
1.00
1992–2003 1990–2003 0.70 0.016 2.19 1.09 (0.372) 0.54 (0.707) 3.28 (0.194) 0.42 (0.858) 0.55 (0.462)
0.80 0.01 1.70 1.88 (0.130) 0.80 (0.529) 3.81 (0.149) 0.52 (0.791) 0.52 (0.476)
0.162 (4.61) 0.90
0.453 (2.92) −0.134 (−2.85) 0.279 (2.46) −0.140 (−1.25) 0.379 (3.25) −0.158 (−1.25) −0.250 (−2.04)
0.80
1990–2003 1990–2003 0.45 0.026 2.06 0.50 (0.739) 0.96 (0.442) 3.18 (0.204) 0.65 (0.631) 2.45 (0.123)
0.41 0.02 1.95 0.33 (0.858) 0.40 (0.810) 2.95 (0.229) 0.82 (0.631) 0.13 (0.719)
Notes: Figures in parentheses are t-statistics and p-values. The long-run coefficient refers to the slope coefficient of the ADL regression of ln VA on ln FD. Two impulse dummies were used in the T&C regression to remove the instability in recursive estimates in 2003Q2 and Q3 caused by SARS; without these dummies, the R2 is only 0.29.
94
Sectoral production
Figure 6.3 Actual (solid line) and predicted (dashed line) sectoral growth rates.
seems to emanate from stock market activities, prompting us to add the change in share prices ( ln PSTOCK) to the short-run model. This led to an improvement, though the equation’s forecasting performance is still not completely satisfactory. In Chapter 8, however, we shall see that the prediction errors for the output of the financial and business services sector, as well as those for the other major sectors, are relatively low when the overall model is simulated.
7
7.1
Ancillaries and identities
Introduction
This chapter is devoted to the ancillary equations and identities in the ESU01 model. The ancillary equations are so labelled because they provide essential support to the ESU01 model, though the variables involved sometimes are of interest in themselves. For example, the first group of stochastic equations that we describe concern the determination of key prices in the economy – consumer, producer, import and export. In other macroeconometric models, these could have been part of a “price block”. A second group of tax and CPF equations elucidate the fiscal revenues of the government and contributions to and withdrawals from the provident fund. We also list and discuss the identities in the ESU01 model without which it cannot be solved and simulated. These include the usual national income accounting definitions, trade identities, and an identity for inventories that serves to close the model.
7.2
Consumer prices and the Balassa-Samuelson effect
As far as the consumer price equations of macroeconometric models are concerned, a veritable hodgepodge has been used as explanatory variables, ranging from unit labour costs and import prices to money supply, asset prices and proxies for excess demand. No sensible linear combination of these forms a co-integrating relationship with the CPI in Singapore, despite the fact that they are all integrated of order one. We therefore had to break new ground by incorporating the BalassaSamuelson effect of the international trade literature into our equation for consumer prices. In a small open economy such as Singapore, the following relations are expected to hold approximately: 1−α α CPIt = PtT PtNT (7.1) PtT = Et PtF
(7.2)
Equation (7.1) states that the general price level as measured by the CPI depends on both tradable (P T ) and non-tradable (P NT ) prices, with α and 1−α being the shares
96
Ancillaries and identities
of the sectors producing traded and non-traded goods and services, respectively. The price of traded goods is in turn determined by the purchasing power parity (PPP) condition in (7.2), where E denotes the exchange rate and P F is a composite foreign price index. Taking logarithms of the variables in (7.1) transforms the CPI into: ln CPIt = α ln PtT + (1 − α) ln PtNT
(7.3)
The Balassa-Samuelson effect asserts that the price of non-traded goods relative to that of traded goods is explained by variations in marginal costs generated by the productivity differential between the two sectors. This can be shown easily. Under perfect competition, the real wage in each sector is tied to marginal productivity: (W /P)Tt = MPtT ,
(W /P)NT = MPtNT t
With labour being mobile, the nominal wage for the same job will be equalized across the sectors, implying that MPtT · PtT = MPtNT · PtNT
(7.4)
Substituting (7.4) into (7.3) yields: ln CPIt = ln PtT + (1 − α) ln MPtT − ln MPtNT
(7.5)
As productivity grows faster in the traded goods sector, wages in this as well as the non-traded sector rise correspondingly, which has the effect of increasing the marginal costs and relative price of non-tradables. Consequently, the overall price level will creep up gradually over time even as P T is kept stable by wage increases that match productivity gains in the traded goods sector.1 A satisfactory concordance between theoretical and empirical concepts is achieved if we let P T be the merchandise import price index and MP be replaced by the average product under the assumption of constant returns to scale. Beyond that, the modelling exercise is severely constrained by the difficulties involved in identifying the traded and non-traded sectors of the economy and calculating their productivity levels. In view of the highly aggregative nature of the available data, it is not easy to separate out the non-traded sector without risking a misclassification of the type of output produced. We had to engage in a lengthy search procedure to arrive at a satisfactory solution to this problem.
1 An analogous expression to (7.5) can be derived from the so-called Scandinavian model of inflation, which explains the domestic rate of inflation in an economy with an “exposed” sector and a “sheltered” sector through an exogenously given rate of increase in the foreign price level and the development of labour productivity in the two sectors.
Ancillaries and identities
97
Figure 7.1 Wages, productivity and the CPI.
The key to the formulation in (7.5) is that nominal wages should be the same in the traded and non-traded sectors. Figure 7.1a plots the nominal wage rates for the major economic sectors discussed in the last chapter, relative to manufacturing wages.2 These wage ratios invariably show a downward trend in the early 1990s. Subsequently, however, the wage ratios for the F&B, T&C and other services sectors settled down to constants, indicating a proportional relationship with manufacturing wages. The wages of the construction and commerce sectors, on the other hand, have continued to fall relative to manufacturing wages, reflecting low labour mobility, skill mismatches and the large presence of foreign workers. The equation in (7.5) can be modified in many ways to account for these wage differences. The simplest improvisation is to assume that WtNT = kWtT , where k may be constant or time-varying, thereby adding a relative wage variable to the model. After experimenting with numerous possibilities and also taking into consideration the overall structure of the ESU01 model, we decided to treat the manufacturing industry as the sole traded sector and the rest of the economy jointly as the non-traded sector. Figure 7.1b plots the traded and non-traded sectoral wage rates defined in this way (see Appendix A for further details on how the wage series are constructed). Interestingly, the divergent trends observed earlier in
2 Quarterly wage rates by sector are available only from 1994Q3. We interpolated the data for the period before this.
98
Ancillaries and identities
manufacturing and non-manufacturing wages are not so prominent now, resulting in the wage ratio W NT /W T being redundant for explaining the CPI level as well as its rates of change. There is one more recalcitrant data problem that needs to be fixed. The published statistics on nominal and real manufacturing value-added show that the implicit manufacturing price deflator is very similar to the CPI, whereas the model above requires value-added to be deflated by tradable goods prices, as proxied by the import price index (P m ). Even though the choice of deflator does not make much difference to a growth rate regression, it matters a lot for the levels regression. When import prices are not used, the latter always produces a negative coefficient for the P T variable, in contradiction to (7.5) which requires it to be unity. Hence, we compute the real value-added of the traded sector as VATt = (VAMANt · CPIt )/Ptm and that of the non-traded sector as VANT = GDPt − VATt . Correspondingly, the average product terms are calcut NT lated as PRODtT = VATt /NtT and PRODtNT = VANT t /Nt , where N stands for employment and it is assumed that working hours in both sectors are identical. As predicted by the Balassa-Samuelson theory, Figure 7.1c shows that the productivity gap between the two sectors has widened over the years due to more rapid productivity growth in the manufacturing sector. Johansen’s trace test shows that the logarithms of CPI, P m and the labour productivity ratio form a co-integrating relationship, albeit with a very small short-run adjustment coefficient. Given that CPI and P m have moved in opposite directions, as shown in Figure 7.2, Figure 7.1d demonstrates that the residual upward trend that remains in the former after removing the effect of the latter is very nicely picked up by the growing productivity differential between the traded and non-traded sectors. We also found that the coefficient of import prices can be constrained to unity in the Johansen estimation. A dynamic OLS regression estimated over the period 1979Q1–2003Q4 with this restriction imposed produces the following empirical
Figure 7.2 Consumer, producer and import prices.
Ancillaries and identities
99
analogue of (7.5): ln CPIt = −0.02 + ln Ptm + 0.55 ln(PRODtT /PRODtNT ) (26.4) (−2.23)
(7.6)
Robust estimation of this model over different sample periods indicates that the estimate of α falls in a narrow range of 0.4–0.5. According to Singapore’s IO tables, the import content of total consumption expenditures is also about 40%. Therefore, contradicting the common perception, we have to conclude that non-traded prices play a bigger role in setting the long-term direction of the CPI. Since nominal wages do not necessarily equalize across the sectors in the short run, we fall back on (7.3) in formulating a forecasting model for consumer prices.3 As P NT is not directly observed in this equation, we substitute it with ULCtNT = WtNT /PRODtNT and also let W NT be represented by the economy-wide nominal wage W (1+CPF emp ) instead of the non-tradable wage graphed in Figure 7.1 so as to have a longer time series for analysis (this should not create a problem because the bulk of employment is in the non-traded sector). Alternative calibrations of the tradable and non-tradable shares of output showed us that the estimate of α provided by (7.6) is in fact the best. Thus, the levels equation for the CPI is: ln CPIt = 0.45 ln Ptm + 0.55 ln ULCtNT
(7.7)
It should be reiterated that as the nominal wage levels of the traded and non-traded sectors converge in the very long run, (7.7) will be superseded by (7.6). In the short-run, incorporating (7.7) into an ECM estimated over 1987Q1–2003Q4 leads to the forecasting model: ln CPIt = 0.0025 + 0.46 ln CPIt−1 + 0.05 ln Ptm (4.44) (4.69) (2.41) −0.009 D_98Q1t −0.003 D_01Q1t −0.05 ECt−1 (−3.05) (−2.62) (−2.31)
(7.8)
R 2 = 0.53 SE = 0.003 DW = 2.32 LM(5) = 2.14 (0.07) ARCH(4) = 0.99 (0.42) Chow = 1.31 (0.26) Normality = 1.26 (0.53) Heteroscedasticity = 1.32 (0.25) RESET = 0.02 (0.90) where D_98Q1 is an impulse dummy for 1998Q1 and D_01Q1 is a step dummy for the period 2001Q1–2003Q4. These dummies are not essential for a good in-sample fit but they improve the long-term dynamic forecasts of consumer price inflation
3 Even though it was relatively easy to find good short-term forecasting models for the CPI inflation rate, it turned out to be quite a challenging task to find a specification that could provide good long-term forecasts for the level of the CPI. This explains the great lengths we went to in this exercise.
100
Ancillaries and identities
and the diagnostic performance of the model. The small magnitude of the adjustment coefficient highlights a typical feature of the data, namely, prices in general are stubbornly persistent. The short-run impact of import prices on the CPI is smaller than expected and it diminishes with time. In contradistinction, unit labour costs of the non-traded sector do not exert an immediate impact on the CPI but they have hump-shaped lagged effects that peak within four quarters and then decay geometrically.
7.3
Producer prices
The two indicators of producer prices available in Singapore are the domestic supply price index (DSPI) and the Singapore Manufactured Products Price Index (SMPI). The DSPI monitors the price changes of locally manufactured goods and imports which are retained for use in the domestic economy. It is compiled as a weighted composite of the import price index P m and the SMPI – a producer price index for manufacturing output. Thus, the DSPI can be thought of as a broad input-based wholesale price index. In this section, we present forecasting models for both DSPI and SMPI, even though the latter is essentially an exogenous variable in the ESU01 model. Despite some researchers reporting mixed results for the order of integration of prices, the ADF statistics computed for the indices discussed above suggest that it is reasonable to treat all of them as I (1) variables. Exploring the data further through simple correlation analysis, we also discovered that the link between the DSPI, SMPI and P m is very close, a fact evident in Figure 7.2. Not surprisingly then, the trace test finds co-integration between these variables during the overlapping sample period of 1978Q1–2003Q4. Nonetheless, the error-correction term in an ECM with the change in the DSPI as the dependent variable did not turn out to be statistically significant and a pure first differences specification suffices for predicting wholesale price inflation in Singapore: ln DSPIt = 0.87 ln Ptm + 0.26 ln SMPIt (16.0) (2.8)
(7.9)
R 2 = 0.952 SE = 0.006 DW = 1.68 LM(5) = 1.29 (0.28) ARCH(4) = 0.43 (0.79) Chow = 2.16 (0.02) Normality = 3.21 (0.20) Heteroscedasticity = 12.9 (0.00) RESET = 18.9 (0.00) Given the method of construction of the DSPI, it is not surprising that the coefficients on P m and SMPI add up to just over one. The model explains a very large proportion of the sample variation in domestic supply prices and closely tracks their changes. To construct a forecasting equation for SMPI which can be utilized in certain simulation exercises, we use P m and UBC as its predictor variables. The fitted equation over 1980Q1–2003Q4 is: ln SMPIt = −0.004 + 0.95 ln Ptm + 0.18 ln UBCt (2.96) (−3.53) (14.1)
(7.10)
Ancillaries and identities
101
R 2 = 0.70 SE = 0.013 DW = 1.96 LM(5) = 0.52 (0.76) ARCH(4) = 0.94 (0.44) Normality = 4.70 (0.10) Heteroscedasticity = 0.34 (0.85) RESET = 0.84 (0.36) Even though the predictive performance of this equation is not as good as that of DSPI, the model predicts the turning points reasonably well. The statistically significant constant term indicates that the two explanatory variables that are supposed to represent imported and domestic input costs do not fully capture the downward trend in the SMPI.
7.4
Import and export prices
In the ESU01 model, the plethora of import and export prices largely constitute what we have called “semi-endogenous” variables. These prices are basically exogenous in nature and for most practical purposes, are kept so. However, for the exchange rate-based policy simulations in Chapter 9, they need to be endogenized. Modelling import and export prices confers an additional benefit – it helps to shed light on the exchange rate pass-through in Singapore, a subject attracting a large literature which we do not intend to digress into.4 The essence of the pass-through argument can be stated by writing (7.2) in logarithms more generally as: ln Pti = α1 ln Et + α2 ln PtiF
(7.11)
where P i is the price of imports or exports in domestic currency units and P iF is the corresponding price in foreign currency. A complete pass-through requires that α1 = α2 = 1. Depending on how transaction invoicing is carried out (whether they are expressed in foreign currency or domestic currency) and a myriad of other factors such as demand elasticities, the exchange rate pass-through may be less than complete in the short run i.e. 0 ≤ α1 ≤ 1. However, a complete passthrough is expected in the long run in the case of price-takers. For exports in particular, this means a zero pass-through to P xF , as can be seen by writing (7.11) as ln P xF = ln P x − ln E and (∂P xF /∂E)(E/P F ) = (∂P x /∂E)(E/P x ) − 1, which is zero if the elasticity of P x with respect to E is unity. If this elasticity is less than unity in the long run, it is an indication that exporters have some price-setting power. Starting with import prices, we first calculated an import share-weighted P mF from producer price index data on Singapore’s trading partners (see Appendix A).5 For ease of interpretation, we also re-express the exchange rate in terms of foreign currency units per Singapore dollar and measure it using both the US dollar bilateral exchange rate (E US ) and the NEER. As a result, positive and negative movements coincide with the local dollar’s appreciation and depreciation, respectively, with the pass-through effect now bounded by −1 ≤ α1 ≤ 0. In addition, although
4 See Campa et al. (2005) for a recent study on the Euro area at a disaggregated level. 5 Our attempt to use export price indices failed because of data incompatibilities between countries.
102
Ancillaries and identities
P mF subsumes oil prices, we detected a robust and exogenous effect of the latter (in US dollars per barrel, denoted as P oil ) on Singapore’s import prices. Furthermore, P m (the merchandise import price index) and its sub-component P rm (the imported raw material price index) are found to have co-integrating relationships with NEER and P mF , but P k (the imported machinery and transport equipment price index) is not. This is probably because the aggregate P mF is quite different from the relevant sub-index for machinery and transport equipment. To find an empirical counterpart for P xF in the export price equations, we constructed a composite index based on selected categories from US and Japanese producer price indices. For notational clarity, we denote this variable as P USJap and the corresponding weighted exchange rate by E USJap . These contries being price-setters in the world market, the US-Japan index is more suitable for testing the pass-through effect and the price-taker assumption.6 After some experimentation, we arrived at the estimated equations for various import and export prices shown below. All the equations except those for P m and P rm are estimated over the period 1978Q1–2003Q4. We used shorter sample periods for P m (1986Q1–2003Q4) and P rm (1985Q1–2003Q4) because of the presence of a higher degree of co-integration over these sub-periods. Merchandise import price ln Ptm = 0.39 −0.31 ln NEERt + 0.92 ln PtmF (1.88) (−3.42) (4.91) + 0.07 ln Ptoil −0.09 ECt−1 (6.90) (−1.98) ECt = ln Ptm + ln NEERt − ln PtmF
(7.12)
R 2 = 0.71 SE = 0.011 DW = 1.42 LM(5) = 2.08 (0.08) ARCH(4) = 0.18 (0.95) Normality = 1.29 (0.53) Heteroscedasticity = 2.06 (0.06) RESET = 0.04 (0.85) Oil import price ln Ptom = 0.14 −0.53 ln EtUS + 0.68 ln Ptoil (3.37) (−2.27) (19.0) oil + 0.17 ln Pt−1 −0.10 ECt−1 (4.59) (−3.26)
ECt = ln Ptom + ln EtUS − ln Ptoil
(7.13)
6 Note that in Chapter 4, we had used only the US price index (expressed in Singapore dollars and denoted by P w ) in the preliminary analysis of different export models. As explained in Appendix A, we had to resort to using various measures of competing prices to circumvent data problems.
Ancillaries and identities
103
R 2 = 0.84 SE = 0.049 DW = 2.00 LM(5) = 1.45 (0.21) ARCH(4) = 3.65 (0.01) Normality = 2.85 (0.24) Heteroscedasticity = 2.90 (0.01) RESET = 6.59 (0.01) Raw material import price ln Ptrm = 0.25 −0.46 ln NEERt + 0.74 ln PtmF (2.09) (−5.34) (4.02) + 0.04 ln Ptoil −0.05 ECt−1 (4.54) (−2.12) ECt = ln Ptm + ln NEERt − 0.9 ln Ptm − 0.1 ln Ptoil
(7.14)
R 2 = 0.57 SE = 0.01 DW = 1.40 LM(5) = 1.86 (0.12) ARCH(4) = 0.06 (0.99) Normality = 6.91 (0.03) Heteroscedasticity = 0.87 (0.55) RESET = 0.23 (0.64) Machinery and transport equipment import price ln Ptk = −0.002 −0.97 ln NEERt + 0.87 ln PtmF (3.53) (−1.15) (−4.72)
(7.15)
R 2 = 0.41 SE = 0.012 DW = 1.89 LM(5) = 0.85 (0.52) ARCH(4) = 2.13 (0.09) Normality = 4.67 (0.10) Heteroscedasticity = 0.68 (0.84) RESET = 1.28 (0.26) Non-oil export price USJap
nodx ln Ptnodx = 0.19 + 0.19 ln Pt−4 −0.31 ln Et (2.36) (2.36) (−5.14) USJap
+ 0.84 ln Pt (4.34) USJap
ECt = ln Ptnodx − ln Pt
−0.02 D_01Q1t −0.04 ECt−1 (−3.07) (−2.39) USJap
+ ln Et
(7.16)
R 2 = 0.43 SE = 0.014 DW = 2.10 LM(5) = 1.50 (0.20) ARCH(4) = 0.64 (0.64) Normality = 0.12 (0.94) Heteroscedasticity = 0.56 (0.83) RESET = 1.51 (0.22)
104
Ancillaries and identities
Oil export price ln Ptox = 0.25 −0.38 ln EtUS + 0.58 ln Ptoil (4.25) (−1.80) (17.9) oil + 0.14 ln Pt−1 −0.19 ECt−1 (3.88) (−4.24)
ECt = ln Ptox + ln EtUS − ln Ptoil
(7.17)
R 2 = 0.83 SE = 0.044 DW = 2.08 LM(5) = 1.09 (0.37) ARCH(4) = 0.59 (0.68) Normality = 1.59 (0.45) Heteroscedasticity = 1.91 (0.07) RESET = 4.71 (0.03) Due to the lack of co-integration in the case of P k , what we have presented in (7.15) is the static solution to a model fitted with four lags of each explanatory variable, which captures the pass-through effect better (the full model is given in Appendix B). Even though this unrestricted dynamic model goes against the spirit of parsimony, we find that parsimony can only be achieved at the expense of forecast accuracy. In the case of P nodx , we had to use a step dummy for the period 2001Q1–2003Q4 to obtain a stable EC coefficient. This seems to be a result of the inadequacy of the P USJap variable as a proxy for the price of competing goods. The historical forecasts of P nodx follow the peaks and troughs in the actual data but they fall short of their amplitudes. Apart from this, the models above produce impressive in-sample fits and track turning points very well. At this juncture, it is useful to flesh out the key implications that follow from the models presented in (7.12)–(7.17). First, the exchange rate pass-through is complete in the long run. Even though the adjustment coefficients are relatively small in some cases – possibly reflecting the paucity of the data rather than the poverty of the theory – the estimates in general suggest that exchange rate fluctuations are transmitted fully to domestic prices within about five quarters. This clearly supports the presumption that Singapore is a price-taker in the world market. Second, the short-run pass-through taking place within a quarter for both merchandise imports and non-oil exports is about 30%. For oil products, the short-run pass-through is much higher – about 50% for oil imports and 40% for oil exports. Third, holding the exchange rate constant, an increase in foreign prices is transmitted onto domestic import and export prices fully and very quickly. We conclude this section by pointing out that the aggregative dynamics of prices in Singapore are such that import prices play a vital role in the price system by linking foreign and domestic price indices. As seen in Chapter 5 and Figure 7.2, import prices also drive a wedge between the consumer and producer price levels. In normal model simulation exercises, however, import as well as export prices can be taken to be given without affecting economic outcomes since they are a function of exogenous variables.
Ancillaries and identities
7.5
105
Tax and CPF equations
The amount of nominal tax revenues collected by the government is one of the determinants of disposable income in Singapore, the others being CPF contributions and withdrawals. Accordingly, we built an equation to predict tax revenues in toto, without distinguishing between direct and indirect taxes. Total tax receipts encapsulate contributions from income and corporate taxes, statutory boards, asset and vehicle taxes, excise duties and the Goods and Services Tax (GST). The only distinction we make is between tax revenues and government fees and charges for services rendered. Quite apart from the need to fully endogenize these variables, tax and CPF rates are key policy instruments that the Singapore government has tweaked from time to time and they therefore deserve to be examined in their own right. A number of sophisticated attempts have been made in journals to investigate the hypothesized “Laffer curve” effect of marginal tax rates on economic growth and budget deficits (see for example Ireland, 1994). Our model for tax revenues is a simple one that incorporates both average and marginal tax rate effects. The marginal effect claims that a tax rate above a certain threshold reduces work effort and therefore incomes and tax revenues. A useful accounting relationship in this regard states that Tt = τ t Yt where T is tax revenue, τ is the average tax rate and 2 Y is income. Assuming that ln Yt = α1 τtm + α2 τtm + ut , where τ m is the marginal tax rate (α1 > 0, α2 < 0), and letting u represent other effects on income growth, we can write the tax equation as: ln Tt = α0 + ln Yt−1 + α1 τt + α2 τt2 + β Xt + et
(7.18)
where X is a vector of other variables that capture the time-varying average tax rate and other effects present in u, and e is a standard error term. In view of the differential tax rates that apply to personal and corporate taxes, we would have liked to segregate these two major categories of tax receipts, if not for the fact that the data breakdown is not made public. The income variable we use is nominal GDP (GDPN ) net of CPF contributions (CPFCONN ), which are exempt from tax. Since direct taxes paid depend on both corporate and income tax rates, we created the average tax rate τ by assigning a weight of ²/³ to τ c (the corporate tax rate) and the remaining weight of ¹/³ to τ y (the average marginal income tax rate).7 Figure 7.3 is a scattergraph of Singapore’s total tax revenues (TAXN ) against our nominal income variable. Instead of an expected linear relationship between the two variables, tax collections seem to have increased at a decreasing rate with income growth. The figure also shows that the variability of the non-linear relationship has gone up over time (a plot in logarithmic scales does not alter
7 These weights correspond to the proportions of income tax collections coming from corporations and individuals (see the tables in the Public Finance section of the Yearbook of Statistics Singapore).
106
Ancillaries and identities
Figure 7.3 Relationship between taxes and income.
these peculiarities). Falling marginal tax rates and a reduction in the number of people falling into the tax net are likely to be the causes of these observations.8 As usual, we estimate the tax equation in a general co-integration framework using the functional form given in (7.18) as a guide. Even though quarterly data on government operating revenues are available from 1988, the estimation period has to be shortened by a year to 1989Q1–2003Q4 to remove the effect of some outliers. For this sample, the long-run tax equation is: ln TAXtN = −3.91 + 0.75 ln(GDPtN − CPFCONtN ) (−3.17) (8.11) + 0.40 τt −0.01 τt2 + 0.20 ln PPIt (3.52) (−3.28) (3.31)
(7.19)
where PPI is the property price index, included as a determinant of property taxes. We also attempted to incorporate the GST rate as a regressor but it turned out to be insignificant. The ECM test statistic with a value of −7.73 confirms the existence of a strong co-integrating relationship amongst the variables in the model. The estimated coefficients on the tax variables indicate that the “optimal” average tax rate from the point of view of maximizing direct tax revenues is about 25%. This should be contrasted with the Singapore government’s stated long-term target of a top rate of 20% for both corporate and income taxes. Needless to say, the maximization of tax revenues is probably not the government’s primary objective, as having a competitive tax regime is likely to be more critical for attracting foreign investment. Our finding nevertheless suggests that Singapore’s future tax
8 Currently, only about a third of Singaporeans pay income tax.
Ancillaries and identities
107
rates should not act as disincentives to investment and work effort since they are below the peak in the Laffer curve. Next, inserting the co-integrating vector in (7.19) into an ECM formulation delivers the simple model below: ln TAXtN = −3.06 −0.75 ECt−1 (−8.43) (−8.49)
(7.20)
R 2 = 0.55 SE = 0.058 DW = 1.91 LM(5) = 1.56 (0.20) ARCH(4) = 0.46 (0.77) Chow = 0.72 (0.72) Normality = 2.89 (0.24) Heteroscedasticity = 0.38 (0.68) RESET = 0.57 (0.45) Remarkably, (7.20) takes care of the non-linear and heteroscedastic behaviour apparent in Figure 7.3. The model’s diagnostics are excellent and recursive estimates of the parameters are very stable, in spite of the move to a greater reliance on indirect taxes with the introduction of the GST in April 1994. Moreover, the model’s fitted values capture most of the idiosyncratic factors that affect short-term tax collections. As mentioned previously, the other main source of government revenues is government fees and charges for services rendered (FEEN ), which includes the COE premiums on cars. Considering the complexity of the policy-influenced rates involved in this category, we simply construct a forecasting equation for FEEN using nominal GDP net of CPF contributions as the explanatory variable again. For policy analyses, nominal fees and charges will have to be treated as an exogenous variable. Disregarding the outliers, Figure 7.4 reveals an approximately linear relationship between the two variables, though volatility has also increased with rising incomes. The estimated log-linear relationship over the period 1989Q1– 2003Q4 is given by: ln FEEtN = −4.05 + 1.05 ln(GDPtN − CPFCONtN ) (−2.29) (6.11)
(7.21)
The elasticity of fees and charges with respect to nominal GDP excluding CPF contributions is insignificantly different from one. Here again, there is evidence of co-integration (the ECM t-statistic is −4.41). The model developed to explain short-run changes is: N N ln FEEtN = −1.15 −0.64 ln FEEt−1 −0.26 ln FEEt−2 (−2.45) (−4.40) (−2.05) N N + 2.26 ln(GDPt−1 − CPFCONt−1 ) (2.24) N N + 2.5 ln(GDPt−2 − CPFCONt−2 ) −0.27 ECt−1 (2.36) (−2.36)
(7.22)
108
Ancillaries and identities
Figure 7.4 Relationship between government fees and income.
R 2 = 0.510 SE = 0.190 DW = 2.15 LM(5) = 1.73 (0.16) ARCH(4) = 0.52 (0.72) Chow = 0.40 (0.96) Normality = 5.11 (0.08) Heteroscedasticity = 0.49 (0.88) RESET = 2.27 (0.14) The numerous lags and error-correction term stand in for the dynamic influences acting on government fees and charges at short time horizons. Notwithstanding the much slower pace of adjustment to the long-run equilibrium compared to tax revenues, both the in-sample and post-sample predictive performances of fees and charges are fine. For the CPF equations, we revert back to the time period 1980Q1–2003Q4. In the long run, gross nominal CPF contributions (CPFCONN ) are a function of wage income (the wage rate multiplied by employment) and the overall rate of contribution for employers and employees (CPF): ln CPFCONtN = −7.50 + 0.98 ln(W · N )t + 2.45 CPFt (14.8) (31.3) (73.7)
(7.23)
According to the ECM test, this is a co-integrating regression with a coefficient of effectively unity on wage income. In the short run, CPF contributions depend on changes in the same two variables and an error-correction term: ln CPFCONtN = −5.00 + 0.47 ln(W · N )t−1 + 1.18 CPFt (4.98) (−7.56) (2.19) −0.67 ECt−1 (−7.55)
(7.24)
Ancillaries and identities
109
R 2 = 0.598 SE = 0.045 DW = 1.92 LM(5) = 1.06 (0.39) ARCH(4) = 1.75 (0.15) Chow = 4.28 (0.00) Normality = 11.8 (0.00) Heteroscedasticity = 0.64 (0.69) RESET = 0.09 (0.76) The equation’s fit and diagnostics are satisfactory. However, both the recursive parameter estimates and the Chow test suggest that the equilibrium relationship between CPF contributions and their determinants has become less stable since the onset of the Asian financial crisis. We will still use the equation in simulations since it plays an auxiliary role in the scheme of things. Rather than focus on total withdrawals from the CPF, we zoom in on nominal housing withdrawals (CPFHOUSEN ) in view of their importance as an explanatory variable for private consumption expenditures in Singapore. As Chapter 2 argues, withdrawals to finance house purchases and mortgages depend predominantly on the residential property price level (PPIRES ) in the long run – with an estimated elasticity of one as it transpires. To account for a lag effect, we set the RES . In view of the difficulty long-run relationship to ln CPFHOUSEtN = ln PPIt−2 of accounting for a myriad of rules and regulations that govern CPF housing withdrawals, the only additional variable we bring into the model is the prime lending rate (PLR) to act as a proxy for the mortgage rate. This leads us to the following ECM estimated over 1990Q1–2003Q4: N ln CPFHOUSEtN = −0.13 −0.37 ln CPFHOUSEt−1 (−1.25) (−3.19)
+ 0.06 PLRt−2 −0.07 ECt−1 (5.31) (−3.23)
(7.25)
R 2 = 0.46 SE = 0.06 DW = 1.92 LM(5) = 1.34 (0.27) ARCH(4) = 0.81 (0.53) Chow = 0.62 (0.81) Normality = 1.82 (0.40) Heteroscedasticity = 0.98 (0.45) RESET = 2.01 (0.16) The negative autoregressive effect present in the above model captures the volatile fluctuations in housing withdrawals that are caused by the omitted variables. CPF withdrawals for other purposes such as education, investment and retirement (CPFOTHERN ) are even more volatile than withdrawals for housing and we treat them as an exogenous variable in the ESU01 model. Total withdrawals play a role in the evolution of households’ financial wealth over time, as explained in the next section.
7.6
Bridging equations
The last set of ancillaries in the ESU01 model are three bridging equations. These reconcile inconsistencies in data definitions and link flow variables back to asset stocks. The first bridging equation in the model relates the net exports of goods and services (NX ) in the national income accounts to the difference between total
110
Ancillaries and identities
exports (TX ) and imports (TM ) in the trade data (see I.9, I.15 and I.16 below).9 The estimated model over the period 1986Q1–2003Q4 is: NXt = −19.5 +0.62 (TX − TM )t (−0.16) (6.03)
(7.26)
R 2 = 0.441 SE = 813.6 DW = 2.13 LM(5) = 1.75 (0.15) ARCH(4) = 0.31 (0.87) Normality = 0.22 (0.89) Heteroscedasticity = 0.36 (0.70) RESET = 1.17 (0.29) This takes the form of a simple first differences equation since the objective is merely to reconcile the two definitions of net exports and also because the variables in (7.26) do not seem to be co-integrated. To link investment flows to capital accumulation, we use the standard identity stating that the change in the stock of physical capital equals gross investment minus depreciation: Kt ≡ It − δKt−1
(7.27)
where δ is the depreciation rate. In Appendix A, we computed the capital stock for the manufacturing sector using annual survey data. As published investment data relate to the whole economy, the stock-flow identity in (7.27) will not hold exactly for our constructed capital stock series. Hence, we treat (7.27) as a regression and estimate a bridging equation to connect private investment in machinery and transport equipment (IMT p ) to the manufacturing capital stock (K). The equation estimated from data for 1978Q1–2003Q4 is: p
Kt = 129.55 + 0.68 Kt−1 + 0.07 IMTt −0.009 Kt−1 (2.70) (8.87) (2.00) (−1.97)
(7.28)
R 2 = 0.590 SE = 183.3 DW = 1.85 LM(5) = 1.80 (0.12) ARCH(4) = 0.86 (0.49) Normality = 20.8 (0.00) Heteroscedasticity = 1.47 (0.21) RESET = 0.06 (0.81) Due to the data incompatibility, the estimated equation shows a large autoregressive effect. Converting it into a static solution yields Kt = 400.6 + 0.22It − 0.03Kt−1 , indicating that a $1 rise in investment spending on machinery and transport equipment increases the manufacturing net capital stock by about 22¢. The solution also implies that the annual depreciation rate for manufacturing capital is about 12%. The last bridging equation converts savings flows into changes in the stock of nominal financial wealth (FWN ), taking into account asset price appreciation.
9 The NX estimates are derived from the Balance of Payments, which is also the source of service exports and imports data, while merchandise trade statistics are reported by International Enterprise (IE) Singapore. For reasons that are not clear, these figures do not tally.
Ancillaries and identities
111
Instead of adopting the complicated formula described in Appendix A, a simpler option is resorted to in simulations of the ESU01 model: N + 547 PSTOCKt FWtN = −1727.9 + 0.12 FWt−1 (17.8) (−0.98) (2.35) + 1.08 (CPFCONtN − CPFWITHtN ) (5.20) + CPIt · (Ydt − Ct ) − ICONHOUSEtN
(7.29)
R 2 = 0.794 SE = 11478.8 DW = 1.6 LM(5) = 2.20 (0.06) ARCH(4) = 3.12 (0.02) Normality = 13.9 (0.00) Heteroscedasticity = 21.4 (0.00) RESET = 31.4 (0.00) where CPFWITHN are total withdrawals from the CPF and ICONHOUSEN is nominal investment in housing. The first independent variable in (7.29) is a stock price revaluation term while the second in square parentheses represents the sum of “forced” CPF savings and “voluntary” financial savings – both expressed in nominal terms – after deducting housing outlays.
7.7
Identities in the ESU01 model
We begin the description of identities in the ESU01 model by restating the definitions of the final demand variables used to estimate sectoral value-added and retained imports in Chapters 6 and 4: FDMANt ≡ 0.0316Ct +0.037It +0.0125Gt +0.0117ODXt +0.1788NODXt +0.1062INVTt
(I.1)
FDCOMt ≡ 0.1749Ct + 0.0413It + 0.022Gt + 0.0834TXt + 0.0079INVTt (I.2) FDT &Ct ≡ 0.12Ct + 0.0128It + 0.0209Gt + 0.0335TXt + 0.0036INVTt (I.3) FDF&Bt ≡ 0.2647Ct + 0.0571It + 0.7317Gt + 0.0621TXt + 0.0117INVTt
(I.4)
FDOTHERt ≡ 0.0165Ct + 0.0035It + 0.0071Gt + 0.0086TXt p
g
FDCONt ≡ ICONt + ICONt
(I.5) (I.6)
FDMt ≡ 0.4088Ct + 0.6644It + 0.2866Gt + 0.5788(TX − RX )t + 0.8492INVTt
(I.7)
112
Ancillaries and identities
Adding up the final demand for the outputs of the various sectors results in total final demand net of imports: FDt = FDMANt + FDCOMt + FDT &Ct + FDF&Bt + FDOTHERt + FDCONt = Ct + It + Gt + NXt + INVTt This familiar equality is at the heart of the expenditure method for calculating GDP. By the value-added approach, GDP in the ESU01 model is obtained as: GDPt = VAMANt + VACOMt + VAT &Ct + VAF&Bt + VAOTHERt + VACONt
(I.8)
Reconciliation of the expenditure and output approaches to the measurement of GDP gives us our model closure procedure: GDPt = FDt + STATt or GDPt − (Ct + It + Gt + NXt ) = INVTt + STATERRORt
(I.9)
STATERROR is the discrepancy that arises because of errors in the collection of statistics. Identity (I.9) makes inventory changes the mechanism in the ESU01 model through which short-term disequilibria between the economy’s aggregate demand (C + I + G + NX ) and aggregate supply (GDP) are eliminated. If demand picks up faster than anticipated, stocks will be run down; if demand slackens unexpectedly, stocks will pile up. This way of closing the model is consistent with the supply-side framework adopted in Chapter 6 and also sidesteps the need to model inventories – an arduous task given their recent behaviour (recall Figure 6.1). Finally, the income perspective is embodied by the concept of private disposable income, which is real GDP less taxes, fees and net CPF contributions, all deflated by the CPI: Ydt = GDPt − (TAXtN + FEEtN )/CPIt − (CPFCONtN − CPFWITHtN )/CPIt (I.10) Conversely, the nominal GDP measure used in the tax and fee equations is arrived at by taking the product of real GDP and the CPI: GDPtN = GDPt · CPIt
(I.11)
Total withdrawals from the CPF are obtained as the sum of housing withdrawals – constituting the lion’s share – and (non-modelled) withdrawals for all other
Ancillaries and identities
113
purposes: CPFWITHtN ≡ CPFHOUSEtN + CPFOTHERN t
(I.12)
Housing withdrawals together with bank advances to professional and private individuals constitute the loans proxy for the “price effect” in the aggregate consumption function: LOANtN ≡ BANKtN + CPFHOUSEtN
(I.13)
Using the superscripts g and p to index the public and private sectors, respectively, the investment and trade components of national income are retrieved from the following identities: p
g
g
p
It ≡ IMTt + IMTt + ICONt + ICONt
(I.14)
TXt ≡ NODXt + ODXt + RXt + SXt
(I.15)
TMt ≡ RMt + RXt + SMt
(I.16)
where IMT g is government investment in machinery and transport equipment and ICON g is public sector construction investment. The following two equations define the user and rental costs of capital utilized in the modelling of investment expenditures and employment, respectively: UCCt = (0.12 + PLRt −
4
ln PtK )
1 − τ c · 1 − e−3PLRt /3PLRt PtK 1 − τc CPIt (I.17)
rt = PLRt − 4 ln DSPIt
(I.18)
Moving to the labour market, the definitional equation for the unemployment rate found in Chapter 5 is repeated here: ut = [(Lt − Nt )/Lt ] · 100
(I.19)
Another definition worth commenting on is that of unit labour costs: emp
ULCt = [Wt (1 + CPFt
) · Nt ]/GDPt
(I.20)
This equation can be interpreted in two ways. The expression as it stands says that ULC is the quotient of the overall wage bill to the output produced by the economy. Alternatively, if employment N is brought down to the denominator, ULC can be viewed as average earnings divided by the productivity of employed workers.
114
Ancillaries and identities
Labour costs in turn affect international competitiveness through the following estimated relationships with absolute and relative unit business costs: UBCt = 0.49ULCt + 0.51NLCt + UBCERRORt
(I.21)
RUBCt = UBCt /(MULCt /EtRM )
(I.22)
NLC is non-labour costs, UBCERROR is an error series that emanates from the mismatch between the published UBC series and our measures of labour and non-labour costs, MULC is Malaysia’s ULC index and E RM is the Singapore dollar bilateral exchange rate against the Malaysian Ringgit. Both variables in the denominator of (I.22) are considered to be exogenous. Lastly, we have four identities for calculating value-added, employment, productivity and wages in the traded and non-traded sectors of the economy: VATt = (VAMANt · CPIt )/Ptm
(I.23)
NtNT = N NT SHAREt · Nt
(I.24)
PRODtNT = (GDPt − VATt )/NtNT
(I.25)
emp
ULCtNT = Wt (1 + CPFt
)/PRODtNT
(I.26)
where the employment share in the non-traded sector (N NT SHARE) is treated as an exogenous variable which can easily be predicted by a simple AR model. That concludes our discussion of the 26 non-stochastic identities in the ESU01 model. In addition, there are 36 behavioural equations containing estimated coefficients that are subject to uncertainty. Nine of these behavioural equations pertain to variables that can be exogenized, which we have labelled as semi-endogenous variables in Appendix B (we developed these nine regression models to enhance our understanding of certain aspects of the Singapore economy and to carry out specific policy simulations). Thus, the core of the model consists of 27 behavioural equations and 26 identities. The system described up to this point is complete in the sense that there are as many equations and identities as endogenous variables. This equality permits a reduced form solution of the structural model in terms of its exogenous variables (35 of them in the present form of the model), including variables whose values are wholly or largely determined abroad, the instruments of government policy, demographic indicators, and domestic variables which can be projected outside of the main model through the use of satellite equations (property and stock prices are prime examples). Appendix B lists the full set of equations, identities and variables in the ESU01 model.
8
8.1
Multiplier analysis
Introduction
Having fleshed out the different blocks, equations and identities of the ESU01 macroeconometric model in the last six chapters, it is now time to run the full model with its dynamic structure and feedback loops. There are several objectives associated with the exercise. In this chapter, the initial aim is to find a well-behaved solution to the estimated model. Once such a solution is found, the model can be formally evaluated on its ability to mimic the actual workings of the Singapore economy, with its unique structural features, through an examination of the system’s in-sample predictive performance. After validating the ESU01 model, we proceed with a multiplier analysis. This entails coming up with quantitative estimates of the impact that changes in foreign trade conditions and fiscal policies have on aggregate economic activity over time, as measured by real GDP and related variables. The dynamic and long-run multipliers that we calculate from perturbed solutions of the model will form the basis of the policy lessons to be learnt from this chapter. In the next chapter, we shall go a step further to conduct simulations under various policy scenarios.
8.2
Model validation
For a large non-linear system of difference equations such as the ESU01 model, an analytic solution cannot be found and numerical algorithms have to be resorted to. We employ the most frequently used method – the Gauss-Seidel procedure. This technique starts with a set of initial conditions, usually the observed values of variables at time zero, and solves for the values of the endogenous variables in the first period sequentially, taking one equation at a time and fixing other variables at either their starting or solved values. The first round of trial solution values are then used in the next iteration, and the process is repeated until convergence is achieved according to a pre-set level of tolerance. The algorithm then moves on to solve for the second period, using the previous period’s final solution if needed. The individual equations we estimated in the previous chapters all have a good statistical fit, but the ESU01 model as a whole may or may not. To assess how well
116
Multiplier analysis
the model reproduces historical reality, we compute the root mean square percentage error (RMSPE), mean absolute percentage error (MAPE) and Theil’s U -statistic for each of the endogenous variables: n yi − yˆ i 2 n RMSPE = 100 · yi
(8.1)
i=1
n yi − yˆ i MAPE = 100 · y n i=1
(8.2)
i
n i=1 yi − yˆ i 2 n U = n 2 n i=1 (yi − y˜ i )
(8.3)
Here, yi are the actual values of the endogenous variables, yˆ i are the solved values and n is the number of time periods covered. Obviously, the lower the RMSPE or MAPE measures of empirical accuracy, the better it speaks of the model. The U -statistic is simply the ratio of the root mean square error of yˆ i to that of y˜ i , which are the predictions from a random walk model (with drift if necessary). A Theil statistic with a value smaller than 1, signifies that the solution from the full model performs better on average than a naïve forecast. The error statistics for the static (i.e. one-step ahead) and dynamic (i.e. multiperiod) solutions of the model over the period 1994Q1–2003Q4 are presented in Table 8.1 for 51 of the endogenous macroeconomic aggregates.1 The RMSPEs from the static solution are very similar to the SE statistics reported for the individual equations in each chapter. This shows that running the model as a system does not lead to an undesirable accumulation of errors that might result from poorly fitted single equation models. In other words, the individual equations appear to be well specified despite the data constraints mentioned in each chapter and Appendix A. The worst performer is the FEEN equation which records an RMSPE of 20.5%. This figure again is not very different from the 19% standard error recorded in Equation (7.21). The high error simply reflects the volatility of the FEEN variable, which has grown at erratic rates such as 160% in 1992Q3 and −36% in 1995Q3. In general, many variables have RMSPE and MAPE measures of less than 5% in the static solution. For the dynamic solution, the statistics are understandably larger due to the snowballing of errors as the forecast horizon lengthens. Interestingly however, the Theil U -statistics are not only less than 1, but they also show a substantial drop in the dynamic solution. We take this as an indication that the
1 The variables listed as semi-endogenous variables in Appendix B were exogenized for these solution runs. Moreover, INVT and NX are left out since these take on negative values. We did not compute the dynamic error statistics separately for each set of ith-step ahead forecasts. Thus, the reported statistics are simply averaged agglomerations of solution errors across different horizons.
Multiplier analysis 117 Table 8.1 Tracking performance of the ESU01 model Static Variable
RMSPE
C 1.50 CPFCON N 5.67 CPFHOUSE N 5.67 CPFWITH N 3.17 CPI 0.28 DSPI 0.59 FDCOM 1.73 FDCON 3.24 FDF&B 1.38 FDMAN 2.79 FDOTHER 3.14 FDT &C 1.74 FDM 1.49 FEE N 20.46 FW N 1.41 GDP 1.15 GDP N 1.25 I 4.22 ICON p 4.90 IMT p 8.17 K 0.56 L 0.28 LOAN N 0.40 N 0.39 N NT 0.39 NODX 3.53 ODX 6.16 PRODNT 1.17 r 0.54 RM 3.97 RUBC 0.99 RX 3.12 SM 3.90 SX 3.19 TAX N 7.41 TM 2.47 TX 2.33 u 0.15 UBC 1.01 UCC 1.65 ULC 1.90 ULC NT 2.06 VACOM 2.79 VACON 1.90 VAF&B 2.91
Dynamic MAPE
U-statistic
RMSPE
MAPE
U-statistic
1.17 4.26 4.79 2.73 0.22 0.47 1.35 2.57 1.08 2.33 2.51 1.35 1.16 14.27 1.09 0.90 0.99 3.29 3.91 6.50 0.44 0.18 0.34 0.31 0.31 2.76 4.41 0.95 0.42 3.07 0.77 2.38 2.61 2.49 5.68 1.85 1.76 0.11 0.77 1.55 1.47 1.61 2.29 1.31 2.22
0.87 0.82 0.91 0.44 0.86 0.58 0.80 0.84 0.68 0.87 0.83 0.81 0.79 0.94 0.50 0.76 0.68 0.87 0.91 0.89 0.81 0.59 0.33 0.69 0.65 0.87 0.87 0.83 0.23 0.92 0.41 0.80 0.85 0.84 0.97 0.77 0.78 0.54 0.72 0.45 0.93 0.97 0.83 0.83 0.84
1.84 8.62 12.23 6.65 0.97 2.17 2.31 5.07 1.74 3.35 4.61 2.33 1.99 23.36 5.96 2.20 2.75 5.45 7.39 8.82 2.35 2.69 0.81 3.02 3.02 5.18 6.50 4.22 1.20 5.92 1.56 4.23 6.17 4.58 9.66 3.67 3.06 0.32 1.57 2.16 3.00 2.95 2.94 2.72 6.06
1.59 7.15 9.39 5.29 0.82 1.82 1.92 3.81 1.43 2.80 3.97 1.94 1.64 16.23 4.98 1.84 2.26 4.19 5.68 7.49 1.97 2.26 0.65 2.51 2.51 4.48 4.91 3.56 0.87 4.97 1.20 3.14 4.74 3.76 7.55 2.84 2.57 0.20 1.21 1.98 2.31 2.30 2.42 2.16 5.33
0.53 0.69 0.74 0.57 0.53 0.61 0.42 0.45 0.43 0.46 0.50 0.43 0.42 0.92 0.67 0.44 0.45 0.49 0.52 0.60 0.55 0.82 0.22 0.80 0.69 0.52 0.75 0.89 0.19 0.51 0.49 0.45 0.53 0.66 0.64 0.42 0.43 0.67 0.65 0.37 0.96 0.74 0.44 0.42 0.49 Continued
118
Multiplier analysis
Table 8.1—cont’d Static
Dynamic
Variable
RMSPE
MAPE
U-statistic
RMSPE
MAPE
U-statistic
VAMAN VAOTHER VAT &C VAT W Yd
2.65 2.72 1.78 0.97 1.76 2.07
2.29 2.22 1.47 0.74 1.24 1.64
0.92 0.83 0.89 0.64 0.90 0.75
6.76 3.00 3.10 1.67 4.24 2.07
5.45 2.44 2.54 1.19 3.29 1.61
0.68 0.56 0.95 0.52 0.77 0.48
Note: For r and u, the reported values are absolute errors (RMSE and MAE).
ESU01 model produces far superior forecasts in the long run compared to naïve random walk forecasts. The general impression conveyed by visual inspection of the simulated and observed trajectories of the key endogenous variables is one of statistical adequacy (Figures 8.1 and 8.2). That is to say, the model tracks actual time series well during the validation period with no egregious errors occurring. Most notably, the static and dynamic projections are able to replicate the turning points in the data closely, without at the same time exhibiting oscillatory or unstable behaviour. Bearing in mind that all economic models are simplified representations of the real world, we surmise that the performance of the ESU01 model is good enough for it to be deployed in our next exercise – multiplier analysis.
8.3
Foreign trade multipliers
A multiplier analysis for our model can in principle be performed by linearizing the reduced form solution obtained in the previous section. Given its unwieldiness, however, we eschewed this option and follow the usual practice of deriving multiplier estimates by means of appropriately designed simulation experiments. The model simulation methodology is explained in detail in the next chapter but the eager reader may skip it for the time being. Suffice it to say that the multiplier for a given endogenous variable Y is computed as the difference between the “shock” and “control” solutions for a given unit, change in the exogenous variable. In this section, the multipliers we are interested in are those with respect to increases in foreign income (Y w ) and global chip sales (CHIP): Y ∗ − Yt+k ∂Yt+k = t+k w , w ∂Yt Yt
Y ∗ − Yt+k ∂Yt+k = t+k ∂CHIP CHIPt
(8.4)
∗ and Y where Yt+k t+k are the perturbed and control dynamic solutions, respectively, of the endogenous variable at k quarters after the initial shock takes place.
Figure 8.1 Static simulation (solid line = actual, dashed line = simulated).
Figure 8.2 Dynamic simulation (solid line = actual, dashed line = simulated).
Multiplier analysis 121 Putting k = 0 yields the impact multiplier showing the immediate effect of a change in external trade conditions on the Singapore economy. Summing the multiplier in (8.4) for consecutive values of k gives us the dynamic multipliers: m ∂Yt+k k=0
∂Ytw
,
m ∂Yt+k , ∂CHIPt
m = 1, 4, 8, 12, 36
(8.5)
k=0
These represent the accumulated change in Y arising from a sustained increase in world income or electronics demand – they are thus the responses to permanent shocks. For ease of comparison with the dynamic elasticities reported in Chapter 4, we express the above multipliers also in elasticity form.2 These are calculated for the indicated values of m, with m = 36 quarters representing the long-run multiplier effect. Figure 8.3 illustrates the impact over time measured in quarters of a temporary 1% increase in external demand on Singapore’s trade and macroeconomic variables while Table 8.2 tabulates the dynamic multipliers. All categories of exports rise, albeit to varying degrees. The impact on NODX and RX is delayed by a quarter and it lingers on for much longer whereas the effect on SX is instantaneous and vanishes quickly. Non-oil export, oil export and re-export volumes grow by 3.5%, 0.9% and 1.9%, respectively, in the long term, making for a total rise in merchandise exports of 2.5%. As Singapore is essentially a “re-export economy”, additional exports lead to a commensurate 2.6% increase in retained imports, suggesting that foreign trade is balanced. The heightened trade activities generate a strong multiplier effect on domestic income that causes real GDP to increase by 0.28% three months after the rise in foreign income and by 1.3% after 36 quarters. As predicted by Okun’s Law, the unemployment rate falls by two-tenths of a percentage point and employment expands by 0.5%. Due to the rise in aggregate demand, the CPI starts to inch up in the medium term, rising by 0.7% in the long run. It is instructive to compare the full model multipliers for NODX and SX with those computed from the single equations estimated in Chapter 4. For NODX, the close correspondence between the long-run multiplier effects hides two important mechanisms that come into play when the model is solved simultaneously. The first is a rise in unit business costs as wages are bid up, which dampens exports subsequently. However, this is offset by an expansion of the manufacturing capital stock that is induced by higher output and investment. In the case of service exports, both model as well as equation elasticities decline straightaway, having overshot the long-term value. However, the long-run system multiplier is 1.1, below the analogous figure of 1.5 from the single equation, because of a greater erosion of service export competitiveness. Thus, when account is taken of the feedback
2 In this section, we use the words “multiplier” and “elasticity” interchangeably although the former typically presents values in dollar terms and the latter in percentage terms.
122
Multiplier analysis
Figure 8.3 Macroeconomic impact of an increase in foreign income (%). Note: For the unemployment rate, the percentage point difference between the shock and control solutions is graphed.
relationships in the complete model, a rise in foreign demand will benefit the economy less. The dynamic effects of the increase in world income on sectoral value-added are also shown in Figure 8.4 and Table 8.2. Not surprisingly, the responses of the manufacturing and commerce sectors are similar to those of NODX and RX as exports fuel the growth engines of these industries. The interim effects on the
Multiplier analysis 123 Table 8.2 Dynamic multipliers for an increase in foreign income Variable
0 qtr
1 qtr
4 qtrs
8 qtrs
12 qtrs
36 qtrs
GDP CPI u N NODX ODX RX SX RM NODX + ODX + RX VAMAN VACOM VAT &C VAF&B
0.04 0.00 0.00 0.01 0.00 0.00 0.00 2.33 0.40 0.00 −0.05 0.23 0.05 0.00
0.28 0.00 −0.04 0.04 1.17 0.44 0.43 1.95 1.15 0.78 0.69 0.54 0.12 0.02
0.70 −0.01 −0.23 0.32 2.87 0.57 1.21 1.65 2.02 1.95 1.60 1.05 0.35 0.22
0.93 0.03 −0.29 0.57 3.50 0.62 1.63 1.45 2.30 2.40 1.94 1.22 0.63 0.50
1.07 0.14 −0.27 0.64 3.57 0.67 1.78 1.30 2.41 2.49 2.05 1.26 0.86 0.68
1.31 0.71 −0.18 0.47 3.46 0.87 1.86 1.06 2.56 2.49 2.11 1.29 1.46 0.95
transport and increases of financial services sectors, by contrast, tend to lag the business cycle and are more persistent. The long-run sectoral impacts are also quite diverse, ranging from increases of 1% to 2.1%. We have witnessed in Chapter 4 the short-run dependence of Singapore’s trade activities on the demand for electronic products. To confirm this stylized fact in
Figure 8.4 Sectoral impact of an increase in foreign income (%).
124
Multiplier analysis
Figure 8.5 Macroeconomic impact of an increase in chip sales (%). Note: For the unemployment series, the percentage point difference between the shock and control solutions is graphed.
the full model setting, Figures 8.5 and 8.6 present the dynamic multipliers for a 1% increase in global chip sales, with the same information shown in Table 8.3. We chose to plot the cumulative multipliers defined in (8.5) rather than those in (8.4) as the former can be conveniently interpreted to be the responses of the economy to a permanently higher growth rate in chip sales. Resulting from the built-in features
Multiplier analysis 125
Figure 8.6 Sectoral impact of an increase in chip sales (%).
of the model, a faster growth of chips will have only a transitory effect on non-oil exports. Nonetheless, it has a permanent effect on the level of re-exports in the long run, as the charts in Figure 8.5 show. What is also striking about the dynamic multipliers plotted is their broad resemblance to those in Figure 8.3 for an increase in foreign income. The subtle difference lies in the contemporaneous effect that semiconductor sales have on foreign trade, estimated to be 0.3% for both NODX and RX (oil exports are unaffected).
Table 8.3 Dynamic multipliers for an increase in chip sales Variable
0 qtr
1 qtr
4 qtrs
8 qtrs
12 qtrs
36 qtrs
GDP CPI u N NODX RX SX RM VAMAN VACOM VAT &C VAF&B
0.06 0.00 −0.01 0.01 0.33 0.27 0.01 0.15 0.18 0.10 0.02 0.00
0.08 0.00 −0.02 0.02 0.35 0.30 0.01 0.25 0.21 0.14 0.03 0.01
0.05 0.00 −0.02 0.04 0.11 0.36 −0.01 0.06 0.07 0.09 0.04 0.04
0.04 0.01 −0.01 0.03 0.02 0.39 −0.02 0.04 0.03 0.07 0.06 0.05
0.05 0.01 −0.01 0.02 0.00 0.40 −0.02 0.04 0.02 0.07 0.06 0.05
0.06 0.01 −0.01 0.02 −0.01 0.41 −0.02 0.05 0.02 0.08 0.09 0.05
126
Multiplier analysis
Given that the long-run multiplier effect on NODX is zero while that for RX is only 0.4%, we do not observe much of a sustained increase in retained imports. On the other hand, service exports decline slightly due once again to higher labour costs. Still, there is a mini trade boom as exports and re-exports of electronic products and components are boosted. Real GDP and total employment increase for up to a year following the exogenous shock to chip sales and then plateau off to reflect small positive effects that can be ascribed to the higher volume of re-export trade. Mirroring this, the unemployment rate declines and then stabilizes with hardly any impact on CPI inflation. In terms of sectoral distribution, the gains in output and employment accrue mostly to the commerce and transport sectors, since manufacturing value-added effectively returns to its original level (Figure 8.6). Industries such as wholesale trading and cargo handling in these two sectors stand to benefit permanently from more buoyant global electronics demand as value-added is raised by nearly 0.1% in each case.
8.4
Fiscal multipliers
We now wish to derive fiscal policy multipliers for the Singapore economy, which are estimates that indicate by how much national income will change as a consequence of an initial autonomous increase in government spending. By way of introducing the issue, let us consider the analytical formula for the income multiplier in a Keynesian fix-price model where wages and prices are assumed to be rigid and there are unutilized resources in the economy (see Artis, 1984, Chapter 4): 1 − mg ∂Y = ∂G 1 − c(1 − CPF − τ y ) (1 − m)
(8.6)
In the formula, mg and m are the marginal propensities to import (MPI) out of government spending and total final expenditures, respectively, c is the marginal propensity to consume (MPC) out of disposable income, τ y is the marginal income tax rate and CPF is the overall (i.e. employer plus employee) contribution rate to the provident fund. Whilst the denominator in the multiplier expression allows for leakages in the process of income determination in Singapore, the term in the numerator is explained as follows: since a fraction mg of any change in government expenditures will fall immediately on the demand for imports, this leaves only the fraction 1 − mg to impact on domestic output right at the outset. From the Singapore IO Tables, we find that mg is 0.29 for government consumption expenditures. Converting the income elasticity of private consumption estimated in Chapter 2 into the MPC using the APC ratio at the end of 2003, we obtain a value of 0.43 for c. An estimate of the MPI out of the final demand for goods and services is provided by the average ratio of RM to
Multiplier analysis 127 (C + I + G + INVT + TDX ), which is 0.23.3 Substituting these numbers into (8.6) and holding the effective tax and CPF rates at their terminal values (6.4% and 33%, respectively), the multiplier is found to be equal to 0.9. That means to say that a $1 injection of government consumption spending raises real national income by only 90¢. When we set mg = 0.66 for the case of investment expenditures, the multiplier drops further to 43¢. These results are hardly surprising, considering the withdrawals from the circular flow of income taking place all the time through imports and compulsory CPF contributions, which are invested overseas by the Government of Singapore Investment Corporation (GIC) rather than re-injected back into the domestic economy. We shall proceed to compare these findings from what is but a simple static model to the results from dynamic simulations of the ESU01 model involving the two different types of government expenditures. In these experiments, the change in inventories is kept exogenous to ensure that the results are compatible with the demand-driven Keynesian model discussed above. The reason why we decided not to explain government consumption is now clear – our first fiscal impulse consists of a one-off increase in such expenditures amounting to 1% of real GDP. In Figure 8.7, we plot the unit multipliers for Y , C, IMT p , RM and SM i.e. the change in these endogenous variables in real dollar terms at different time periods k for a $1 rise in G. The impact on current national income is found to be just 5¢ although this increases to 10¢ in the next two quarters. Thereafter, the beneficial effects of an autonomous increase in public spending begin to fall off. Table 8.4 presents the corresponding dynamic multipliers showing the cumulative dollar sums after m periods have elapsed. The long-run fiscal policy multiplier is 79¢, lower than the calibrated value from (8.6) but similar to the magnitude reported by Lim and Associates (1988, Chapter 16). Turning to the other dynamic profiles in Figure 8.7 and Table 8.4, we see that private consumption expenditures rise very little in the short run and by 35¢ in the long run. Similarly, firms’ investment expenditures are stimulated by less than 3¢ in the short term. In contrast, the demand for retained imports of consumption goods shoots up by 15¢ in the immediate period on account of higher government purchases. Service imports are less affected initially but ultimately swell by 30¢ as household consumer demand is encouraged. Thus, the overall leakage through imports is 48¢. For the remaining variables, we report the percentage changes from the control solution: in the long run, employment is lifted by 0.32%, the unemployment rate falls by 0.13 percentage points, while the cumulative increase in the CPI is 0.12%. Quite a different picture emerges in Figure 8.8 from an increase in government construction spending (ICON g ) equivalent to 1% of national income. Instead of the small impetus we get from government consumption, real GDP is immediately boosted by 44¢ – almost exactly the multiplier value we found for
3 The demand for re-exports is excluded since including it will bias the MPI upwards, as pointed out in Chapter 4.
128
Multiplier analysis
Figure 8.7 Macroeconomic impact of government consumption ($). Notes: The impact on CPI and N is expressed in percentage terms. For the unemployment series, the percentage point difference between the shock and control solutions is graphed.
Multiplier analysis 129 Table 8.4 Dynamic multipliers for government consumption Variable
0 qtr
1 qtr
4 qtrs
8 qtrs
12 qtrs
36 qtrs
GDP C IMT p CPI u N RM SM TM
0.05 0.01 0.00 0.00 0.00 0.01 0.15 0.01 0.15
0.15 0.02 0.01 0.00 −0.02 0.03 0.19 0.03 0.22
0.38 0.09 0.06 −0.02 −0.12 0.17 0.22 0.09 0.31
0.56 0.16 0.05 −0.04 −0.17 0.32 0.22 0.14 0.37
0.66 0.22 0.04 −0.02 −0.16 0.38 0.21 0.19 0.40
0.79 0.35 −0.08 0.12 −0.13 0.32 0.18 0.30 0.48
investment expenditures using the analytical formula! The increase in production is attributable to an instantaneous jump of 36¢ in construction value-added, as seen in the last row of Table 8.5. In the ensuing periods after the shock, however, the extra stimulus to aggregate output drops drastically to below 5¢ and fades away after a year. As a result, the long-run government investment multiplier is 70¢, smaller than the corresponding fiscal multiplier for recurrent spending. A frontloading of private consumption expenditures is evident in Figure 8.8, with personal spending increasing by 7¢ in the current quarter. In the next quarter, investment in machinery and equipment also receives a one-off shot in the arm, rising by 14¢. The impact and long-run multipliers for retained imports in Table 8.5, at 35¢ and 39¢, respectively, are double the estimates for government consumption, something to be expected since the IO tables tell us that the import requirements of investment spending in general are two times that of public consumption expenditures. In contrast, service imports are not sensitive to the type of fiscal shock and they eventually rise by about 30¢ in both cases. All in all, total imports increase by 67¢ when ICON g goes up by a dollar. This explains why, compared to an increase in G, the multiplier effect on domestic output is more restrained. Finally, the dynamic responses of employment, the unemployment rate and consumer prices have the same shapes in Figures 8.8 and 8.7; their long-run effects in Table 8.5 are also of the same order as those in Table 8.4.
8.5
Policy lessons
The present chapter first demonstrated that the excellent statistical properties of our single behavioural equations are well retained by the simultaneous equations ESU01 model. In a sense, the multiplier analysis that followed can also be viewed as a model validation exercise since it showed that the economic properties of the system are reasonable. We began by calculating multipliers with respect to improvements in external demand conditions, as measured by variables such as foreign GDP and global chip sales. Our analysis suggests that a 1% increase in
130
Multiplier analysis
Figure 8.8 Macroeconomic impact of government investment ($). Notes: The impact on CPI and N is expressed in percentage terms. For the unemployment series, the percentage point difference between the shock and control solutions is graphed.
Multiplier analysis 131 Table 8.5 Dynamic multipliers for government investment Variable
0 qtr
1 qtr
4 qtrs
8 qtrs
12 qtrs
36 qtrs
GDP C IMT p CPI u N RM SM TM VACON
0.44 0.07 0.00 0.00 −0.05 0.05 0.35 0.08 0.43 0.36
0.48 0.08 0.14 −0.01 −0.13 0.14 0.50 0.10 0.59 0.37
0.55 0.15 0.08 −0.04 −0.21 0.35 0.46 0.14 0.60 0.39
0.61 0.20 0.05 −0.03 −0.17 0.40 0.43 0.18 0.60 0.41
0.64 0.24 0.03 0.01 −0.15 0.38 0.41 0.20 0.62 0.42
0.70 0.32 0.01 0.16 −0.12 0.29 0.39 0.27 0.67 0.42
foreign income ultimately generates 1.3% more local output as export-led growth spills over to domestic demand.4 This finding is simply a quantitative affirmation of the observation that a flourishing world economy is a definite boon for Singapore. One policy lesson to be learnt from the multiplier analysis is that a faster growth rate in the worldwide demand for electronics does not benefit the Singapore economy as much as an acceleration of foreign economic growth. This is to be expected since the demand for electronics cannot be sustained without higher growth in world income. The main beneficiaries of improved chip sales, it would appear, are re-exporters of goods and workers employed in the commerce and transport sectors of the economy. On the negative side, Chapter 4 has shown that the short product cycles of the global electronics industry often result in undesirable swings in Singapore’s output and employment. Looking into the future, our findings have yet another implication: if the long-term trend growth rate of global semiconductor sales slows down to a single-digit figure – as widely anticipated – then the electronics-related industries can be expected to provide a smaller fillip to the local economy than in the past. While the government’s anti-cyclical Keynesian philosophy implemented via construction spending (cf. Section 3.5) and off-budget packages is commendable, both analytical and simulation methods confirmed the oft-heard hypothesis that the fiscal policy multiplier in Singapore is small. Instead of magnifying the initial effect of an increase in public expenditures upon income, our calibrated multipliers serve to dilute the impact due to savings and import leakages abroad. Illustrating very starkly the dilemma of fiscal policy in a small open economy, our econometric
4 Our multipliers should not be confused with the traditional foreign trade multipliers quantifying the increase in national income that results from export earnings, a study of which has been made by Wilson (1995) based on an earlier ESU model. He found that a $1 change in total export earnings generated an additional 87¢ of real income in the long run.
132
Multiplier analysis
simulations verify the heavily damped nature of government spending. Indeed, we have rediscovered what Wong (1974, pp. 43–44) found 30 years ago: As government expenditure consists chiefly of wages and salaries to public servants, its direct import content is small enough to be ignored . . . one might be tempted to recommend the use of government consumption spending rather than government investment spending as a stimulant to the economy, since the former entails less imports and has a larger multiplier effect on income than the latter. Like Wong, we might be tempted on the face of our results to conclude that government consumption expenditures generate the biggest “bang-for-the-buck” in terms of stimulating national income in the short run, but this is too narrow a view to adopt. Every one dollar spent on public construction and infrastructure projects adds to the national capital stock and facilitates, or even stimulates, private sector investment, leading to capacity expansion in the long term. The fact that our model has not fully allowed for these positive externalities does not mean that such “crowding in” effects are not present. When all is said and done, the supplyside effects of fiscal policy are very likely to raise the government investment expenditure multiplier well above the consumption multiplier, and perhaps even above unity.
9
9.1
Policy simulations
Introduction
In the final chapter of this book, we turn to the arena of macroeconomic policy. The term “policy evaluation”, sui generis, refers to any attempt by social scientists – judgemental or statistical – to assess the pros and cons of a specific policy intervention. Here, we are only concerned with the use of model simulation and control techniques to examine the economic impact of, and feedback on, public policies. The simulations described in this chapter are of course not exhaustive; rather, they illustrate how a broad range of real-world policy options faced by the Singapore government can be illuminated by econometric analysis. Still, these exercises are prescriptive because they lead to conclusions on how policy could or should be conducted. Another key word in the lexicons of econometric policy evaluation (and cliometrics) is “counterfactuals”. McCloskey (1987) defines it thus: “Counterfactuals are what ifs, thought experiments . . . alternatives to actual history; they imagine what would have happened to an economy if, contrary to fact, some present conditions were changed”. He goes on to warn that counterfactual experiments are bedevilled by a problem of vagueness. The solution to this dilemma is a detailed economic model – such as the ESU01 model we have at our disposal. Our explorations into counterfactual worlds are restricted to aspects of labour market regulation, monetary policy and the long-term growth path, involving the following conditional assumptions and questions: • • •
What would have happened had Singapore adopted a radical closed-door policy on foreign workers? Assume that the government froze the count of migrant labourers at the level prevailing in early 1994. Will the economy be more stable if the MAS adheres to a rule-based exchange rate policy that compels it to respond less aggressively to deviations of inflation from an implicit target and more vigorously to deviations of output from trend? For real GDP to grow at a healthy rate during the next 10–15 years, what is required by way of policy instrument settings?
134
Policy simulations
The last of these scenarios is really an exercise in optimal control, but we shall first begin with the counterfactual analysis by describing our model simulation procedures.
9.2
The mechanics of model simulation
For a long time now, simulation techniques have been used for a variety of purposes in the economic and physical sciences, including model validation, historical simulation and conditional forecasting. In policy analysis, governmental actions as enshrined in each alternative option are combined with exogenous events to determine simulated outcomes for the economy. An estimated macroeconometric model can therefore be envisaged as a laboratory for trying out new economic measures. Indeed, simulation is a controlled statistical experiment designed and performed on a reasonably comprehensive model, with the intention of indicating to policymakers not only the immediate consequences of a policy stance, but also its longer-term macroeconomic ramifications. This avoids having to test out ill-advised policy on the actual economy, possibly with dire consequences. The normal mode of model simulation is the dynamic one where the solutions for the current endogenous variables are computed using the solved values of the same variables in previous periods, as compared to a static simulation that uses their actual values.1 In generating a dynamic “baseline” solution of the model from the estimated equations and identities, the first few observations are used for initialization and the simulation horizon is also rendered sufficiently long to bring the model solutions into steady-state equilibrium. This baseline simulation is then compared to an alternative “shock” solution of the model where one or more of the exogenous variables are perturbed. When the exogenous variable concerned is a policy instrument, this comparison yields estimates of policy multipliers, as in the previous chapter. In counterfactual exercises, we follow the usual practice of adopting as time paths for the exogenous variables those actually observed over the historical sample, or to be specific, sub-periods within 1989Q1–2003Q4. This obviates the need to specify their trajectories in an out-of-sample simulation, which is a non-trivial matter given that macroeconomic variables are typically subject to less predictable stochastic trends. It should also be noted that, unless the model is highly non-linear, the exact pattern of the baseline solution does not matter much since simulation results are conventionally reported as percentage deviations of the alternative solution values from the baseline scenario. The observed discrepancies in endogenous variables are naturally attributed to the effects of the economic policy under investigation. Given that no econometric model can fit the data perfectly, however, it would be cavalier to interpret the results literally.
1 Such dynamic multi-period “forecasts”, together with their static counterparts, form the basis of the model validation exercise carried out in Chapter 8.
Policy simulations
135
In a “deterministic” simulation, the stochastic components of endogenous variables are suppressed in the model solution step i.e. the errors in behavioural equations are set to their expected values of zero. In a “stochastic” simulation, by contrast, part of the uncertainty associated with any macroeconometric model is internalized by administering representative shocks to the equation residuals (in the application below, the uncertainty in coefficient estimates and exogenous variables is ignored to minimize the computational burden). At each replication of a stochastic simulation, independent random numbers are drawn from the multivariate normal distribution and scaled to match the estimated variance-covariance matrix of the error terms. In other words, a stochastic simulation adds in shocks that mimic the type of disturbances to have hit the economy in the past. The result of the simulation is a probability distribution of outcomes for the endogenous variables in every period, from which statistics of interest such as the mean and variance can be calculated directly.2 To reduce simulation variability in the stochastic mode, the same pseudo-random shocks are used to obtain the baseline and perturbed model solutions.
9.3
A closed-door foreign worker policy
The first simulation we discuss is the restrictive migrant worker policy described in the introductory section. Under this scenario, the values of the exogenous variable POP, representing the total working age population, are replaced by the counterfactual path that results from reducing the rate of foreign worker inflow to nil beginning in 1994Q2. This deterministic simulation gives us a flavour of the economy-wide impact of such a policy. The qualitative results of the simulation experiment can be explained well ex post, though the quantitative effects cannot be divined ex ante: equilibrium and actual nominal wages increase while employment declines. Unemployment also falls because the downward adjustment in the labour force outweighs the retrenchment in employment (but remember that the overall jobless rate continues to be underestimated due to the existing pool of foreign workers). The rise in wages is subsequently transmitted to costs and prices; thus, ULC, UBC and the CPI all increase steadily (but not monotonically) throughout the simulation period. This in turn reduces exports because of a loss of competitiveness, particularly in the services trade. The quantitative findings are shown in Table 9.1. As explained in the foregoing section, the numbers in the cells are the percent deviations of the endogenous variables in the counterfactual scenario from the historical baseline at various intervals after the policy intervention (we start with four quarters since the economic impact prior to that is negligible). In the labour market, the contraction in
2 Incidentally, it should be mentioned that the mean values from a stochastic simulation are equal to the expected values of the probability distributions of endogenous variables in a non-linear model while this is generally not true for the solutions computed from a deterministic simulation.
136
Policy simulations
Table 9.1 Foreign worker policy simulation (% deviation from baseline) Variable
4 qtrs
8 qtrs
12 qtrs
36 qtrs
Y I CPI N L W ULC NODX SX
−0.02 −0.02 0.06 −0.13 −1.65 1.94 1.82 −0.07 −0.49
−0.09 −0.10 0.48 −0.69 −2.80 4.80 4.17 −0.32 −1.48
−0.17 −0.16 1.13 −1.47 −4.14 6.96 5.57 −0.55 −2.14
−0.90 −0.96 6.20 −6.90 −10.2 18.7 11.5 −1.77 −5.49
the supply of workers from 1994 onwards would have exerted upward pressure on nominal wages, so much so that they would rise by nearly 19% over 9 years. By early 2003, real earnings would have gone up by 12.5% despite a 6% rise in the consumer price level. The wage increases would then have discouraged the demand for labour by business firms so that employment ends up 7% below its baseline level. On the supply side of the economy, the concomitant 12% increase in ULC depresses NODX by 1.8% and SX by 5.5%. By the end of the simulation period, real GDP would have been about 1% lower compared to the baseline scenario. Investment spending would also be cut back by the same quantum because of accelerator-like effects. The normative consequences of the closed-door foreign worker policy are fairly obvious. No doubt, the average real wage is higher and the jobless rate is lower. However, total employment will shrink mutatis mutandis, or by about 145,000 workers according to our econometric estimates. It is inconceivable that Singaporeans and permanent residents will not be retrenched. Local and foreign manpower may be competitive in some occupations, yet they are likely to be complementary in others where domestic talent and expertise are lacking. Without foreigners to fill such jobs, hiring of locals will be curtailed, although we cannot tell by how much since no distinction is made between domestic and imported workers. One thing is clear, however: if the government were to stop foreigners from coming here to work, the severe labour shortage that develops will raise domestic costs and prices substantially, with detrimental effects on exports and economic growth.
9.4
The conduct of monetary policy
Our next simulation experiment investigates whether an exchange rate policy rule could have been used by the central bank more effectively for demand management than the discretionary policies that were actually employed. To be sure, Singapore has been extremely successful in relying on the latter, as evidenced
Policy simulations
137
by the low inflation rate over the last 30 years, save for the two big inflationary episodes stemming from the world oil shocks during the 1970s. By running a stochastic simulation of the ESU01 model, we wish to see whether the Singapore economy could have performed even better had the MAS followed alternative monetary policy rules. For this stochastic simulation, we use our full system of equations which include the import and export price relationships developed in Chapter 7. Unfortunately, our attempts to model the service export price index (P sx ) and service import price index (P sm ) along the same lines did not succeed. Considering the problems involved in constructing proper price indices for services, however, we are forced to reject the data rather than the maintained hypothesis of full exchange rate passthrough and simply let ln Ptsx = ln Ptsm = − ln NEERt . Apart from the NEER, we include three other exchange rates in the model, namely E US , E RM and E USJap . In the stochastic simulation exercise, we set the appreciation rates of all these bilateral exchange rates equal to that of the NEER. Figure 9.1 portrays the chain of events unleashed by a contractionary monetary policy, executed through a temporary 1% appreciation of the NEER (care must be exercised in interpreting these graphs because their vertical scalings are not the same). The effects on macroeconomic variables are as expected, with the timing and scale of many responses displaying the familiar “V” shape. First of all, the implied appreciation of the real effective exchange rate precipitates a decline of Singapore’s merchandise exports. As noted in Chapter 4, the negative impact of a currency appreciation on service exports is more severe due to the high proportion of domestic inputs used in their production process. In contrast, service imports increase due to the substitution effects arising from a resurgent Singapore dollar. The overall reduction in the final demand for goods and services leads automatically to lower sectoral output, especially in the manufacturing and commerce industries. This shows up as a decline in aggregate output that lasts for about a year after the monetary policy shock. However, the quantitative impact of a mild appreciation appears to be rather negligible (the cumulative effects are of course larger). In characteristic fashion, the inevitable correction in employment trails behind the GDP cycle. The stronger exchange rate also has the direct effect of lowering import and consumer price levels. With weaker economic activity, a downward wage-pricecost spiral is initiated, aggravating the deflationary process but mitigating the drop in output by partially restoring export competitiveness. The dynamic repercussions take about 3 years to filter through the economic system in the case of real variables and considerably longer for nominal prices and wages, which do eventually return to their baseline levels. Now we know that the ESU01 model produces plausible responses to exchange rate policy changes, we can go ahead with the stochastic simulation. The baseline scenario is the model solution where the behaviour of the monetary policy instrument coincides with the way it was actually deployed over the period 1990Q1–2003Q4. In other words, the exchange rate is deemed to be exogenous.
138
Policy simulations
Figure 9.1 Effects of a contractionary monetary policy (% deviation from baseline).
The alternative is a contingent policy rule consistent with the central bank’s declared macroeconomic objective of non-inflationary economic growth: Et = α + β(π − π ∗ )t + γ (Y − Y ∗ )t
(9.1)
Policy simulations
139
This exchange rate policy rule (or policy reaction function) says that the MAS reacts to cyclical fluctuations in both inflation and output by either appreciating or depreciating the S$ NEER in order to compensate for the impact of exogenous shocks.3 The shocks could be of domestic or foreign origin, transitory or permanent, though asymmetry in the response pattern is ruled out by the linear specification here. The constant term allows for an autonomous drift in the exchange rate when π = π ∗ and Y = Y ∗ . By conveniently setting the annual inflation target π ∗ to 2%, extracting the potential output level Y ∗ from the underlying GDP series as its Hodrick-Prescott trend, expressing the output gap as a percentage of the trend output level and setting the constant term to 1.5% (the average annual appreciation rate observed over 1990Q1–2003Q4), we can experiment with different values of the β and γ coefficients in (9.1) to arrive at a “policy frontier”. After several trial runs, we fixed γ at 0.5 – very close to Parado’s (2004) estimates – and derived the policy frontier shown in Figure 9.2 from 1000 stochastic simulations of the ESU01 model for each set of parameter values. The frontier represents the set of feasible combinations of inflation and output growth variabilities (measured by average standard deviations) that can be attained by changing the weight on the inflation deviation term in the policy rule. The U-shaped frontier indicates, not surprisingly, that there is a trade-off between
Figure 9.2 The policy frontier. Note: Numbers inside the graph are inflation weights.
3 The contemporaneous rule in (9.1) assumes that the MAS has information on the current year-onyear inflation rate and detrended output, which is not too unrealistic given that the CPI is compiled monthly and preliminary estimates of real GDP are available. In his study of monetary policy rules in Singapore, Parrado (2004) used forward-looking inflation forecasts and also allowed for exchange rate smoothing.
140
Policy simulations
output growth variability and inflation volatility for very large values of β. Values of β between 2 and 5 lower inflation variability without having any adverse effect on output growth uncertainty. As β drops below 2, however, both inflation and output volatilities rise, suggesting that a meek policy response to deviations of inflation from target is self-defeating. The basic desideratum of macroeconomic policy is price stability and high economic growth. Our experiments with alternative policy rules confirm that focusing on controlling inflation is the right thing to do at the MAS. In practice, the managed floating system in Singapore is relatively flexible and has indeed placed a heavy emphasis on maintaining low and stable inflation. As we have taken pains to explain in Chapter 4, the exchange rate is not a potent tool for fine-tuning output growth because of its offsetting effects on export and import prices. Given the ultra openness of the Singapore economy, it makes much more sense to use the nominal exchange rate to fight imported inflation than to smooth business cycles. For all its limitations, fiscal policy is arguably better suited for macroeconomic stabilization and growth in view of its positive impact on aggregate demand in the short run and output-enhancing effects in the longer run.
9.5
The quest for growth: An exercise in optimal control
The method of optimal control involves solving a macroeconometric model for “control” variables in order to attain specific targets with respect to key endogenous variables. In the language of Tinbergen (1956), a generic term that describes the means for achieving targets is “policy instruments”. For example, if Singapore’s GDP were to grow by 5% over the next 10–15 years, what would be the required levels of FDI and foreign worker inflows into the country? It is not necessary that control theory techniques will actually be used to steer the economy in an optimal manner, as the projected paths of instruments themselves can be extremely helpful in the formulation of macroeconomic policies to achieve targets. Moreover, they often tell us something about the properties of estimated macroeconomic systems, and by extension, of the economy under study. The solution to the optimal control problem is relatively straightforward if the number of targets and instruments is the same. When there are more instruments than targets, however, a unique solution for all instruments is not possible without pre-setting the values of some exogenous variables so as to equalize the number of targets and instruments. A different set of problems arises if there are more targets than instruments. In this case, one way out is to specify an objective function, which may be a loss function that captures the costs both of not achieving the target and of instrument instability.4 Given the complexity of the computations required to carry out a proper exercise in optimal control, plus the fact that many of the instruments that we are
4 See Klein et al. (1999, Chapter 8) for a brief discussion of optimal control in the context of macroeconometric models.
Policy simulations
141
interested in (e.g. employment and net investment commitments) are only indirectly affected by public policies, we decided to proceed in a less formal way using simulation methodology. Our target variables are the real GDP growth rate and CPI inflation. Instead of fixing their optimal paths, however, we set the dynamic paths of the exogenous variables – most of which are beyond the control of policymakers – to plausible values and simulate the model into the future until the growth rates of GDP, CPI and other variables converge to constants. The time paths of the control variables can then be manipulated until the desired output growth and inflation rates are achieved. This exercise provides backward and internally consistent solutions for all the endogenous variables in the model. The full ESU01 model as set out in Appendix B has 62 endogenous variables and 35 exogenous variables. Based on historically valid relations, we endogenize five of the exogenous variables using the following identities: G = 0.1GDP, ICON g = 0.06GDP, IMT g = 0.09GDP, ICONHOUSE N = 0.04GDP N and BANK = 0.04FW N . For the remaining exogenous variables, we rely on historical records and subjective inputs to set their expected (growth) paths over the next 10–15 years. The long-run values that we chose are shown in Table 9.2. By far the most critical exogenous variable affecting Singapore’s economic growth prospects is Y w , our proxy for world income. In view of the rise of China and India, it is not unreasonable to assume that the average growth rate of 4% observed over the previous 15 years can continue for at least another decade. The variable CHIP, which measures world demand for Singapore’s electronics exports, is assumed to grow at the same rate as Y w because it is ultimately income growth that drives the electronics industry. The growth rate of V (tourist expenditures as proxied by visitor days), also set at 4%, is based on what has been observed in the past. Domestically, the rate of increase of POP (Singapore’s working age population) is simply taken to be its average growth rate over the last 15 years, or 2.8%. This rate is consistent with the government’s aim to increase Singapore’s total population to about 6 million people in the long run. By contrast, the growth rate
Table 9.2 Assumptions on exogenous variables Growth rate (%) Yw CHIP V NLC MULC POP CPFOTHERN PPI PPI RES PSTOCK
4 4 4 0.7 4.6 2.8 8 4 4 4.8
Growth rate (%) NEER E RM E US E USJap P USJap P mf P sm P sx P bm P oil
1.2 1.2 1.2 1.2 1.2 1.2 1.2 1.2 1.2 2.8
% PLR CPF CPF emp τ τc τy θ N NT SHARE STATERROR UBCERROR
5.5 33 13 16.85 20 6.4 73 80 0 0
142
Policy simulations
of property prices (PPI, PPI RES ) is a calibrated figure based on the observation that the long-term increase in property values since 1975 has been the same as that of the per capita GDP of the working age population. The rate of change of PSTOCK (the Singapore stock price index) is also the average registered during the last 15 years. With regard to the Singapore dollar, we set the trend appreciation of the NEER and other bilateral exchange rates to 1.2% per annum based on the postulated rise in P mF (a proxy for the world price level). Of course, this implicitly assumes that the MAS will continue to offset imported inflation through exchange rate manoeuvring. We then allow the other import and export price indices to appreciate at the same rate as the NEER except for the world oil price (P oil ), which rises at a faster pace in the light of recent developments in the global oil market. The figure for the prime lending rate (PLR) is a rounded average for the recent past. The CPF contribution rates and tax rates are direct policy variables but since we do not treat them as control variables, they are set equal to the values at the end of the estimation period. The same goes for θ (the share of value-added in manufacturing) and N NT SHARE (the share of employment in the non-traded sector). Three points must be made before we present the simulation results. First, just as Y w is the critical exogenous variable for Singapore’s GDP growth, so its coefficient in the co-integrating equation for NODX (Equation 4.11) turns out to be the most important. This income elasticity coefficient is estimated to be 3.55 during the historical period, but as we have argued in Chapter 1, such a high elasticity relative to that for imports means that Singapore will enjoy trade surpluses even if its GDP were to grow at the same rate as that of her trading partners. Even though such a scenario is not unduly optimistic if Singapore conquers new export markets, we decided to fix the income elasticity at 3 in order to err on the conservative side (a further reduction is not possible without violating the co-integrating condition). Second, it will be recalled that the labour force equation in the ESU01 model does not take into account wage pressures because of the missing data problem. Therefore, we ignored this equation in the optimal control application. Third, the simulation run had to be carried forward for 15 years in order for the GDP growth rate to converge to a constant. Table 9.3 presents the growth rates that can be expected over a 15-year time frame, conditional on the assumptions contained in Table 9.2. The table begins with the growth target and factor input requirements and ends with inflation and cost variables. As far as the growth objective is concerned, the results of the control exercise show that, in the absence of binding input constraints, Singapore may experience a robust GDP growth rate of 6.5% over the next 10–15 years – much higher than the official target of 3–5%. Despite this, we refrained from manipulating the exogenous variables further because our assumed growth rates, especially that for Y w , seem to be quite realistic. Nevertheless, the projected GDP growth can materialize only if investment levels are kept at 33% of aggregate output and total employment is allowed
Policy simulations
143
Table 9.3 Projected growth rates of endogenous variables Growth rate (%) GDP I (excluding inventory changes) I (machinery & transport equip, private) I (construction, private) Net investment commitments (nominal) Construction contracts awarded (nominal) I/GDP K (manufacturing) N (total employment) CPF contributions (nominal) CPF withdrawal for housing (nominal) CPF total withdrawals (nominal) Fees and charges (nominal) Tax revenue (nominal) Yd Financial wealth (nominal) C NODX ODX RX SX RM SM
Growth rate (%)
6.5 5.6
Total X Total M
9.0 9.1
6.2
VA manufacturing
6.3
1.8
VA construction
4.2
VA commerce
5.9 7.9
2.2 4.6 9.5
VA transport & communication VA finance & business VA other CPI DSPI
6.5 1.4 −0.6
7.0
SMPI
−1.5
7.2
Pm
9.8
Pnodx
6.7 6.2 4.9
Pom Pox Prm
13.7 2.7 0.33
6.2 12.8 2.6 9.5 7.6 6.2 9.4
Pk W ULC UBC UCC RUBC
6.7
0.0 −0.1 1.6 1.6 0.2 −1.0 4.8 2.9 1.8 −2.4 −3.9
Note: The investment share in GDP (I /GDP) is a ratio.
to rise at 4.6% per annum. With slower growth in private construction investment, maintaining a one-third share of investment in national income requires steady increases in manufacturing investment commitments of about 14% per year. At the same time, the high investment rate needs to be supplemented by increases in employment that far exceed the growth of the resident population in order for the unemployment rate to be kept at its natural level. Consequently, the reliance on
144
Policy simulations
foreign workers cannot be curtailed without a substantial increase in wage costs, as we discovered earlier.5 On the inflation front, Table 9.3 shows that, with a moderate increase in the world price level taking place, imported inflation can be kept close to zero with an exchange rate appreciation policy without having much of an adverse effect on total exports. Non-oil exports in particular are still projected to expand at a vigorous rate of 13% per annum. As a result, consumer price inflation can be delinked from import prices and is only determined by domestic costs. With the provision that employment grows at 4.6%, unit labour costs will rise by only 2.9% and keep CPI inflation below 2%. An obvious caveat here is the possible risk posed by a long-term surge in oil prices that may elicit a more aggressive exchange rate policy response. The ESU01 model also predicts a mild drop in producer prices resulting from the negative trend in the SMPI that we observed in Chapter 7. A similar downward movement in imported capital goods prices will feed into a falling UCC.6 In contrast, the drop in RUBC is the result of a very mild rise in Singapore’s non-labour costs (NLC) and a large increase in Malaysia’s ULC (Table 9.2). Overall, the results suggest that Singapore is unlikely to lose its cost competitiveness if the labour market is managed carefully. The upshot of our results is that, if world income grows by 4% and the current liberal policies to encourage FDI and foreign talent to the shores of Singapore are continued, the country can attain a real GDP growth rate that is higher than the global benchmark, notwithstanding its mature status. Remarkably too, this can be done without stoking inflation. In other words, sustained non-inflationary growth over the next lap is feasible, at least according to the ESU01 model.
5 During the boom years of the nineties, Singapore’s employment grew by more than 6% per year while the resident population (Singapore citizens and permanent residents) increased by less than 2%. 6 Note that the P k regression is not a co-integrating relationship and the coefficient of NEER is bigger in absolute value than that of P mF (see Equation 7.15).
Appendix A: Computational methods of variables
In this Appendix, we present the methodologies used to create the data series that are not directly available from local and international databases. The sources of the data that are publicly available are detailed in the individual chapters of the book. Disposable income We calculate real disposable income as: Yd = GDP − taxes − government fees & charges − net CPF contributions where taxes include both direct and indirect taxes, and CPF contributions are net of all withdrawals. Data on taxes and government fees and charges are only available on a quarterly basis starting from 1988, so observations going back to 1977 had to be interpolated from annual data using the Chow-Lin technique. The variables that enter into the construction of the real disposable income series are all expressed in real terms; where necessary, they are deflated by the CPI. Wealth The nominal wealth variable that we constructed for Singapore has two components: financial wealth (FW N ) and housing wealth (HW N ). The stock of housing wealth in each period is derived by revaluing the previous period’s stock in line with changes in house prices as follows: N HWtN = ICONHOUSEtN + HWt−1
PPItRES PPItRES
where ICONHOUSEN is nominal investment expenditures on residential construction (obtained by multiplying real expenditures by the housing investment deflator), and PPIRES is the residential property price index in Singapore. The housing stock above is essentially calculated using the perpetual inventory method,
146
Computational methods of variables
in conjunction with flow data on residential construction investment and assuming an initial stock of S$14,276 million in 1975 based on Rao and Lee (1995). Due to the paucity of data on the saving and investment patterns of households in Singapore, we consider only three types of financial assets: CPF savings, bank deposits and equities. The aggregate stock of financial wealth can be written as: depo
+ At FWtN = ACPF t
PSTOCKt PSTOCKt−1 depo = CPIt (Ydt − Ct ) − ICONHOUSEtN − St
equi
where At
equi
and St
equi
+ At
equi
= St
equi
+ At−1
Accumulated CPF savings are directly measured by net CPF balances (inclusive of interest payments) while data on the amount of bank deposits held by Singapore residents are also available. Additions to the stock of equities held by households come from savings invested in shares, which are in turn calculated as nominal disposable income less consumption expenditures, housing investment, and savings in bank deposits (the initial value of equities is taken to be two-fifths of the amount of bank deposits at the end of 1976). Like housing wealth, the FW N equation also captures changes in the valuation of equities that result from movements in PSTOCK, the Stock Exchange of Singapore share price index. Export and import price indices Price indexes for different merchandise export and import categories (in Singapore dollars) are lodged in the Department of Statistics’ (DOS) STS database. The price index for imported raw materials (P rm ) was computed as a weighted average of the import price indices for machinery and transport equipment, mineral fuel, and chemical and chemical products. The weights chosen are the same as those used by DOS for computing the overall import price index. As for the service import price index (P sm ), we first noted that the overall goods and services import price index (P gsm ) and the goods import price index (P gm ) move very closely with each other, reflecting the dominance of goods imports in the overall index (P gm is referred to as P m in the book). Therefore, we converted the annual P gsm observations to a quarterly time series using the spline interpolation method and then used the following two different methods to construct P sm . Method 1 Quarterly data on nominal service imports (SM N ) and nominal goods imports (GMN ) are available. We derive service imports in real terms using the relationship: GM N GM N + SM N − SM = P gsm P gm Using nominal and real service imports, we obtain P sm = SM N /SM .
Computational methods of variables 147 Method 2 Assuming a geometric average, we have: P gsm = (P gm )α (P sm )(1−α) where α is the share of merchandise imports in total imports (retained imports plus service imports). Note that α is not a constant. Taking log differences of the above equation, we get ln P sm = ln P gsm − α ln P gm (1 − α) By setting the starting value to 100, we can compute the P sm index using the above growth rates. Both methods produced very similar estimates. In this book, we use the index derived from the first method. World price indices and NEER The foreign price index (P mF ) entering the import price equations of Chapter 7 is a geometric average of either the producer or wholesale price indices of Singapore’s major trading partners in Asia and the OECD countries: Australia, France, Germany, Japan, South Korea, Malaysia, Netherlands, Taiwan, Thailand, UK and USA. If a country’s price index is denoted as Pj for j = 1, 2, 3, . . ., 11 and its bilateral exchange rate in terms of Singapore dollars as Ej , the geometric average can be written as: 11
Pj Ej
j=1
wj
=
11 i=1
(Pj )wj
11
(Ej )wj = P mF · NEER
i=1
where the weights wj are the 12-quarter moving averages of Singapore’s import shares with each of the eleven countries. We use the reciprocal of the exchange rate so that an increase indicates appreciation. The export equations estimated in Chapter 4 also required a world price index (P w ) to represent the prices of goods that compete with Singapore exports. Following a common practice in the literature, we initially computed P w as an export share-weighted average of the producer or wholesale price indices in Singapore dollars of the eleven major trading partners that we listed above. We found this to be a poor proxy for the prices of goods competing with Singapore’s exports, however. The indices of the trade partners include diverse products ranging from aircraft and heavy machinery to agricultural products, many of which are absent in Singapore’s exports. Ideally, we should select only the most relevant sub-categories of the price indices and then compute an average index to represent P w . Unfortunately, a detailed breakdown of producer prices is not easily available for the major trade partners, except for the USA and Japan. Due to the
148
Computational methods of variables
fact that these two countries are major export destinations for Singapore’s exports and also price-setters in the world market, we decided to use the relevant subcategories of their producer price indices to construct a price index for competing goods. From the US producer price index we selected 7 sub-categories. They are processed food (WPU02), textile products and apparel (WPU06), chemicals and allied products (WPU06), general purpose machinery (WPU114), electrical machinery and equipment (WPU117), electronic components and accessories (WPU1178), and medical, surgical and personal aid devices (WPU156), all of which were downloaded from the Bureau of Labor Statistics website. The above seven categories account for about 95% of Singapore’s non-oil exports to the USA, with electronics receiving the largest weight of 55%. We constructed P US as an export share-weighted geometric average of these sub-indices, with the shares based on Singapore’s nominal exports to the USA in the year 2000. We then used P US expressed in Singapore dollars as P w in Chapter 4 because it produced more meaningful estimates in the Johansen co-integration procedure. To construct the price index for Japan (P Jap ), we used 11 categories from Japan’s wholesale price index (now renamed as the corporate good price index) obtained from Bank of Japan publications. The label numbers of the selected sub-categories are 67, 68, 71–74, 77, 87, 51–56 & 59, 57–58, 64, 26 & 65 and 2. We combined some categories to match Singapore’s export classifications. We then constructed P Jap as the export-share weighted geometric average of the above 11 categories, using the shares of Singapore’s exports to Japan in 2000. By exploiting the property of the geometric average, we finally computed a combined USA-Japan price index expressed in local currency as P USJap · E USJap . Again, we use the 12-quarter moving averages of Singapore’s export shares with the USA and Japan as weights. The forecasting equation developed for P nodx in Chapter 7 uses this USA-Japan index. Foreign income Our foreign income variable Y w is an export share-weighted geometric average of the real GDPs of the ASEAN4 countries (Indonesia, Malaysia, Philippines and Thailand), NIE3 economies (Hong Kong, South Korea and Taiwan), China, Japan, USA, and the rest of the OECD as one group. The share of exports going to each foreign country is a 12-quarter moving average. By allowing the export shares to vary over time, our measure of world income reflects changes in the country composition of Singapore’s trade. Computational details for GDP and the export shares are given in Abeysinghe and Forbes (2005). Manufacturing net capital stock Like many countries, there is no data on the capital stock (K) in Singapore. Many researchers overcome this problem by resorting to the use of the perpetual inventory method to construct the capital stock. However, the unsatisfactory nature
Computational methods of variables 149 of the method led us to devise a new way of estimating the capital stock of the manufacturing sector in Singapore. The Report on the Census of Industrial Production published annually by the Economic Development Board of Singapore provides annual (year-end) data on the value of gross fixed assets and net fixed assets. The former is the accumulated cost of capital expenditures and the latter is net of depreciation. The ratio of the two series shows that the depreciation rate has increased over time, which is quite plausible, especially for the electronics sector where product obsolescence has been very rapid. Still, it implies that the use of fixed depreciation rates in the perpetual inventory method will produce misleading estimates. We can easily obtain an annual series on the net capital stock based on net fixed assets data. The annual change in this series represents net investment expenditures. We deflated this series by the deflator for gross fixed capital formation to obtain a constant dollar net investment series. Then using 1990 as the base year, we computed an annual series for the net capital stock. Due to changes in the survey coverage, however, this series shows level shifts in 1997 and 2001. To adjust for this, we ran a regression of the capital stock on a trend term and dummies and then obtained adjusted growth rates for the two years as the predicted values from the regression. Keeping the other growth rates intact, we worked the data backward to obtain a more refined series. The next problem is how to convert the annual series to quarterly figures. We decided to use the Chow-Lin technique which depends on the availability of quarterly data on related variables to predict the capital stock (see Abeysinghe and Gulasekaran, 2004). To search for such related series, we use the following approach. First-order conditions for profit maximization based on a Cobb-Douglas production function show that the logarithm of the capital stock (k ∗ ) is a linear function of the logarithm of output (q) and the real price of capital. Since there is no readily compiled data on the price of capital, we regress k ∗ on q, as represented by direct exports excluding re-exports (which account for about 60% of manufacturing output), the logarithm of manufacturing employment and a time trend. The latter two variables turn out to be insignificant, leaving us with a highly stable and significant relationship between and k ∗ , lagged k ∗ and q. The presence of the lagged capital stock makes the direct application of the Chow-Lin procedure difficult. Consequently, we took the approach in Abeysinghe (1998). Let the regression for the quarterly data be written as: ∗ + ut kt∗ = β0 + β1 qt + λkt−1
The transformation given in Abeysinghe (1998) enables us to estimate the quarterly model from annual data on k ∗ and quarterly data on q. The transformed model is kt∗ = β0 (1 + λ + λ2 + λ3 ) + β1 (qt + λqt−1 + λ2 qt−2 + λ3 qt−3 ) ∗ + λ4 kt−4 + vt
150
Computational methods of variables
This model can be estimated by non-linear OLS. The estimation results based on data over 1978–2002 produce the following results, where the numbers in parentheses are t-statistics: βˆ0 = 0.6411 (9.25) βˆ1 = 0.0469 (5.40) λˆ = 0.8944 (62.61) R2 = 0.99, Durbin h = 1.17 Plugging these estimates into the original capital stock equation and taking the year-end values of k ∗ as starting values, we are able to generate a quarterly k ∗ series. This step is the same as the Chow-Lin method because the errors are serially uncorrelated. However, a minor refinement was needed to remove some spikes that occurred in 4th quarter growth rates. This was easily achieved by making small adjustments to the third and fourth decimal places of the AR parameter estimate λˆ and keeping it constant at the adjusted value until the next adjustment was necessary. User cost of capital We use the following formula from Schaller (2006) to compute the user cost of capital for investment in machinery and transport equipment: K P 1 − τ cζ − u UCC = (i + δ + γ − π K ) 1 − τc P = UCC r · UCC tax · UCC price where i is the nominal interest rate (proxied by the annual prime lending rate), δ is the depreciation rate (set at an annual rate of 0.12 based on the equation for the capital stock above), γ is a risk premium (set to zero), π K is the rate of inflation for investment goods, τ c is the annual corporate tax rate, ζ is the present value of depreciation allowances, u is the investment tax credit rate, P K is the price of capital goods, and P is the price of output (represented by the CPI). We Tan follow and Thia (2004b) in setting u to zero and also use their formula ζ = 1 − e−3i 3i to compute the present value of depreciation allowances. Since quarterly price indices for P K are not available in Singapore and most capital goods are imported, we use the import price index for machinery and transport equipment as a proxy for P K . It follows that the annual inflation rate of this variable represents π K . Employment Quarterly data on employment levels are not available from public sources. The Ministry of Manpower (MOM), based on the Labour Force Survey (LFS)
Computational methods of variables 151 conducted every June, publishes annual employment levels, quarterly labour productivity growth and the quarterly change in employment. We first use the published data on the change in employment and the 1999 June employment level as the base to construct quarterly employment levels beginning from 1983. Then the quarterly growth rates of labour productivity and GDP growth are used to extend the employment series back to 1979. It should be noted that the mid-year survey excludes construction workers living at work sites and daily commuting workers from abroad. The changes in employment data, which are taken from administrative records, include these workers. As a result, our 2nd quarter employment estimates do not necessarily coincide with the annual figures published by the MOM. Labour force and unemployment rate Neither the MOM nor the DOS publish quarterly data on the labour force. However, the MOM has been announcing quarterly unemployment rates since 1986. To maintain consistency across data series, we first derive the quarterly labour force by dividing the employment level by one minus the seasonally adjusted unemployment rate. To extend the labour force data all the way back to 1979, we employ the following approach. First, we adjusted the annual unemployment rates from the June LFS for a seasonal effect – the second quarter unemployment rate tends to be higher – by running a regression of the available quarterly unemployment rates on its own lag and seasonal dummies. The second quarter seasonal coefficient was then subtracted from the annual unemployment rate figures for the period 1979–85. We then divided the seasonally adjusted second quarter employment levels by one minus the adjusted unemployment rates to obtain the annual series for the labour force. Lastly, we applied the Chow-Lin procedure using employment as the predictor variable to interpolate the quarterly labour force observations. Wages The DOS publishes an annual series on average monthly earnings based on the June LFS. It also publishes a quarterly series on economy-wide unit labour costs (ULC), a key cost variable that is monitored regularly in Singapore. To construct a wage series consistent with ULC and our derived employment (N ) series, we first use the growth rate relationship: ˙ = U LC ˙ − N˙ + G DP ˙ W to obtain the quarterly growth rates of wages. Using the end-1994 wage level as the starting value, we then recursively derived our average monthly earnings series. For the major sectors of the economy, quarterly data from 1994Q3 and annual data from 1989 on average monthly earnings (inclusive of employer’s
152
Computational methods of variables
CPF contributions) are available. To backdate the quarterly data, we constructed the annual wage share of the ith sector as Si = Wi Ni / Wi Ni , where Ni is the employment level in the sector, and then applied the spline method to convert Si to the quarterly frequency. Then using the quarterly data on W and N , we computed the sectoral wage bills as Wi Ni = Si W · N . For consistency with the official data, we converted these series into growth rates and worked backward from 1994Q3 to construct Wi Ni from 1989 onwards. The nominal wage series the traded (W T ) for and non-traded (W NT ) sectors can then be calculated as Wi Ni / Ni for appropriately defined i. In particular, manufacturing is the traded sector and the other industries constitute the non-traded sector. Working age population Population data are only available on an annual basis while its detailed composition by age group is restricted to the resident population. However, Singapore’s working age population (15–64 years old) includes a substantial number of nonresidents. To impute the working population on a quarterly basis, we had to use an indirect method. Since the DOS publishes annual labour force participation rates (LFPR) and these change very gradually, we could simply use a univariate interpolation to convert the annual figures to quarterly ones and then divide our quarterly labour force figures obtained above by the interpolated LFPR to obtain the working age population. These LFPR figures, however, contain both a seasonal effect and a census effect. The former is a result of the timing of the LFS (as mentioned earlier) and the latter comes from the census years of 1980, 1990 and 2000. To avoid these effects, we divided the DOS labour force figures by the LFPR statistics to obtain the annual working age population. We then divided our 2nd quarter labour force figures by these population figures to obtain a new annual series for the LFPR. As seen in Figure 5.3 of this book, the constructed series shows more pronounced cyclical effects because our data capture foreign workers at construction sites and daily commuters, who move out of the labour force when they are unemployed. Finally, we applied the linear spline method to interpolate the new quarterly LFPR series and then divided our labour force figures by this series to obtain the quarterly working age population. As by-products, we also compiled the following quarterly population series for inclusion in our database: (a) resident population; (b) working age resident population; (c) non-resident population; and (d) total population. By definition, the resident population is made up of Singapore citizens and permanent residents. Incidentally, it should be noted that the total resident population is not necessarily a good proxy for the working age resident population because of the so-called “dragon-year” effect (the jump in births during the year of the dragon in the Chinese zodiac). Furthermore, we had to make an adjustment to this population series to remove another spike in the growth rate in 1990 due to a change in the enumeration method. In the 2000 census, the DOS shifted to a de jure population counting method and adjusted the annual mid-year population figures from
Computational methods of variables 153 1990 onwards to correspond to this method. The figures before 1990 represent the de facto population. A plot of the annual statistics shows a slight upward shift of the trend line from 1990. To adjust for this shift, we fitted a linear trend to the data from 1990 to 1999 and predicted the figure for 1989, from which an adjusted growth rate for 1990 can be worked out. Using this adjusted figure and the growth numbers for the earlier years, we managed to backcast the population series to 1980. We performed a similar adjustment for the working age resident population. The refined resident population shows almost a linear increase with steadily falling growth rates. Hence, univariate spline interpolation should provide reasonably accurate quarterly figures. The total population also includes foreigners who live in Singapore for at least 1 year. This group is labelled the non-resident population and accounts for close to 20% of the current population. In contrast to the resident population, the nonresident population consists mostly of foreign workers and their dependents, and so its size at any one time depends heavily on the business cycle and policy responses to booms and busts. Even though we also applied the univariate spline method on the non-resident population to get quarterly figures, the resulting growth rates cannot be very accurate. Average personal income tax rate The usual measure of the average tax rate is obtained by taking the ratio of total tax revenues to some income variable, such as nominal GDP. The tax rate so calculated is endogenous, however, and not a direct policy variable since the government only fixes marginal tax rates. For example, between 1980 and 2003, the Singapore authorities have adjusted the marginal tax rates eight times. Creating a measure of the average tax rate from marginal tax rates requires a set of weights that are exogenous. We tried several weighting schemes to obtain the average marginal tax rate but the outcomes were not very promising. We finally settled down for the usual income-weighted measure. This can be computed as follows. Let τim be the marginal tax rate for the ith income bracket (i = 1, 2, . . ., k), Yi be the assessed income of the ith tax group, and wi be the income share of the ith group. Then the average marginal income tax rate τ y can be written as: τy =
k i=1
Yi τ m Total tax assessed wi τim = i = Yi Total assessed income
where wi = Yi / Yi . If the weights remain fixed over time, movements in τ y will directly reflect changes in marginal tax rates. Figures A.1 and A.2 show the weights (income shares) over selected years computed from the income distribution data published by the Inland Revenue Authority of Singapore (IRAS). In the 1980s, the income shares shifted a lot
154
Computational methods of variables
Figure A.1 Income shares for average marginal tax rates, 1980–90.
Figure A.2 Income shares for average marginal tax rates, 1994–2003.
compared to the 1990s. As the income distribution evolved over time, the income shares tended to stabilize. In recent years, it is interesting to note that the shares have exhibited an increasingly bi-modal pattern.
Appendix B: Listing of equations and variables
Equations Where applicable, the first equation is the long-run co-integrating relationship and the second represents the short-run error-correction model (ECM). Private consumption expenditures (1978Q1–2003Q4)
ln
(FWtN /CPIt ) (LOANtN /CPIt ) Vt Ct = 0.16 ln − 0.22 ln + 0.19 ln + εt Ydt (2.25) Ydt Ydt (6.50) Ydt (−5.71)
ln Ct = −0.03 − 0.26 ln Ct−1 + 0.25 ln Ydt + 0.08 ln (FWtN /CPIt ) (2.36) (−4.12) (−3.15) (4.05) N − 0.03 ln (CPFHOUSEt−1 /CPIt ) − 0.24 ECt−1 (−2.06) (−5.94)
(E.1)
Net investment commitments (1985Q1–2003Q4) ln NICtN = −9.75 − 0.92 ln RUBCt + 1.99 ln VAMANt + εt (17.60) (−10.8) (−3.20)
(E.2)
Machinery and transport equipment investment (1985Q1–2003Q4) p
p
ln IMTt = 2.56 − 0.23 ln IMTt−1 + 2.10 ln GDPt−1 (3.53) (−1.67) (3.03) − 0.28 ln UCCt − 0.20 ln (NIC N /P k )t−1 (−1.75) (−3.16) p − 0.35 ln IMTt−1 (NIC N /P k )t−1 (−3.63)
(E.3)
156
Listing of equations and variables
Private construction contracts awarded (1981Q1–2003Q4) N ln CAN t = −0.03 − 0.41 ln CAt−1 + 2.0 ln PPIt (2.84) (−0.79) (−4.45)
+ 0.23 ln PPIt−1 + 1.27 ln PPIt−2 − 0.24 PLRt−1 (0.29) (2.19) (−3.86) (E.4) Construction investment (1981Q1–2003Q4) p
ln ICONt = 0.513 + 0.045 ln(CAN /P bm )t − 0.038 ln(CAN /P bm )t−1 (4.82) (2.55) (−1.88) − 0.052 ln(CAN /P bm )t−1 (−3.97) p − 0.155 ln ICONt−1 (CAN /P bm )t−1 (−7.38)
(E.5)
Non-oil domestic exports (1981Q1–2003Q4) ln NODXt = −7.1 + 3.55 ln Ytw + 2.7 ln Kt (4.38) (21.5) (50.7) −1.0 θt (ln P rm − ln P nodx )t − (1 − θt )(ln UBC − ln P nodx )t (−6.11) (1989Q1–2003Q4) N ln NODXt = −2.3 + 0.31 ln CHIPtN + 0.15 ln CHIPt−1 (1.61) (−3.56) (3.43)
− 0.35 ECt−1 (−3.59)
(E.6)
Oil exports (1978Q1–2003Q4) ln ODXt = 3.36 + 0.57 ln Ytw + 0.32 ln Kt − 0.65 ln Ptom + 0.42 ln Ptox (6.75) (4.77) (5.4) (−3.23) (1.93) ln ODXt = 2.76 + 0.75 ln Kt − 0.45 ln Ptom − 0.82 ECt−1 (9.09) (2.76) (−7.55) (−9.11) (E.7)
Listing of equations and variables 157 Re-exports (1989Q1–2003Q4) ln RXt = −0.66 + 1.87 ln Ytw + 0.41 ln CHIPtN (5.04) (−0.75) (8.08) ln RXt = −0.15 + 0.27 CHIPtN + 0.13 D_94Q2t − 0.23 ECt−1 (3.85) (3.20) (4.19) (−3.46) (E.8) Service exports (1986Q1–2003Q4) ln SXt = − 0.08 + 1.50 ln Ytw + 0.50 ln Vt − 0.89 ln (UBCt /Ptsx ) (12.7) (−6.10) (−0.55) (29.2) ln SXt = −0.17 + 2.31 ln Ytw + 0.30 ln Vt − 0.30 ln (UBCt /Ptsx ) (7.08) (−4.22) (3.85) (−1.57) + 0.02 D_01Q1t − 0.51 ECt−1 (2.08) (−4.30)
(E.9)
Retained imports (1986Q1–2003Q4) ln RMt = −2.95 + 1.21 ln FDMt − 0.87 ln (P m /SMPI )t (−6.12) (25.4) (−4.41) ln RMt = −0.68 − 0.25 ln RMt−1 + 1.09 ln FDMt (13.0) (−2.52) (−2.29) + 0.52 ln FDMt−1 − 0.23 ECt−1 (3.50) (−2.50)
(E.10)
Service imports (1986Q1–2003Q4) ln SMt = −2.86 + 1.56 ln GDPt − 0.83 ln Ptsm (−1.15) (−3.01) (11.3) ln SMt = −0.44 + 0.81 ln GDPt − 0.83 ln Ptsm − 0.075 D_03Q2t (−2.24) (2.56) (−6.14) (−1.64) + 0.084 D_03Q3t − 0.13 ECt−1 (1.98) (−1.74) Employment (1981Q1–2003Q4) ln Nt = 1.19 − 0.41 ln W (1 + CPF emp )/DSPI t + 0.002 rt (4.69) (−6.76) (1.95) + 0.91 ln GDPt (13.8)
(E.11)
158
Listing of equations and variables ln Nt = 0.0001 + 0.57 ln Nt−1 + 0.11 ln GDPt (0.14) (8.71) (3.74) + 0.11 ln GDPt−1 + 0.07 ln GDPt−2 (3.69) (2.17) ∗ − 0.06 ln Nt−1 − ln Nt−1 (−3.71)
(E.12)
Labour force (1981Q1–2003Q4) ln Lt = 0.80 ln Lt−1 + 0.24 ln POPt + 0.58 2 ln POPt (14.7) (2.8) (7.52) − 0.003 ut−1 (−5.37)
(E.13)
Equilibrium and disequilibrium wages (1982Q1–2003Q4) ln
W CPI
∗ t
GDP = 5.27 + 0.001 rt + ln POP t (287.0) (0.33) emp CPIt (1 + CPFt ) − 0.20 ln y emp DSPIt (1 + CPFt − τt ) (−3.23) emp
ln
W CPI
− ln (1 + CPFt
t
y
− τt )
= 0.016 − 0.57 CPFt − 0.74 ln POPt−1 (7.0) (−4.92) (−2.83) W ∗ W − ln − 0.23 ln CPI CPI (−7.05) t−1
(E.14)
Manufacturing value-added (1992Q1–2003Q4) ln VAMANt = 0.66 ln FDMANt ln VAMANt = 0.95 + 0.63 ln FDMANt − 0.29 ECt−1 (2.63) (8.05) (−2.94) Commerce value-added (1992Q1–2003Q4) ln VACOMt = 0.75 ln FDCOMt
(E.15)
Listing of equations and variables 159 ln VACOMt = 0.42 + 0.79 ln FDCOMt + 0.15 ln FDCOMt−1 (2.62) (8.83) (1.64) −0.24 ECt−1 (−2.64)
(E.16)
T&C value-added (1990Q1–2003Q4) ln VAT &Ct = ln FDT &Ct ln VAT &Ct = 0.02 + 0.20 ln FDT &Ct − 0.11 D_03Q2 (6.22) (3.10) (−10.23) + 0.06 D_03Q3 − 0.08 ECt−1 (5.16) (−1.64)
(E.17)
F&B value-added (1990Q1–2003Q4) ln VAF&Bt = 0.9 ln FDF&Bt ln VAF&Bt = 0.24 + 0.16 ln PSTOCKt − 0.30 ECt−1 (4.38) (4.61) (−4.12)
(E.18)
Other value-added (1990Q1–2003Q4) ln VAOTHERt = 0.8 ln FDOTHERt ln VAOTHERt = 0.45 + 0.28 ln FDOTHERt (2.92) (2.46) − 0.14 ln FDOTHERt−1 (−1.25)
(E.19)
+ 0.38 ln FDOTHERt−2 − 0.16 ln FDOTHERt−3 (−1.25) (3.25) − 0.25 ln FDOTHERt−4 − 0.13 ECt−1 (−2.04) (−2.85)
Construction value-added (1987Q1–2003Q4) ln VACONt = ln FDCONt ln VACONt = − 0.20 + 0.87 ln FDCONt − 0.23 ECt−1 (E.20) (−2.99) (20.3) (−3.03)
160
Listing of equations and variables
Consumer price index (1987Q1–2003Q4) ln CPIt = 0.45ln Ptm + 0.55ln ULCtNT ln CPIt = 0.0025 + 0.46 ln CPIt−1 + 0.05 ln Ptm − 0.0088 D_98Q1t (4.44) (4.69) (2.41) (−3.05) − 0.0026 D_01Q1t − 0.05 ECt−1 (−2.62) (−2.31)
(E.21)
Domestic supply price index (1978Q1–2003Q4) ln DSPIt = 0.87 ln Ptm + 0.26 ln SMPIt (16.0) (2.8)
(E.22)
Singapore manufactured products price index (1981Q1–2003Q4) ln SMPIt = − 0.005 + 0.95 ln Ptm + 0.17 ln UBCt (2.53) (−3.51) (13.7)
(E.23)
P m (Import price index) (1986Q1–2003Q4) ln Ptm = ln PtmF − ln NEERt ln Ptm = 0.39 − 0.31 ln NEERt + 0.92 ln PtmF (1.88) (−3.42) (4.91) + 0.07 ln Ptoil − 0.09 ECt−1 (6.90) (−1.98)
(E.24)
P rm (Imported raw material price index) (1985Q1–2003Q4) ln Ptrm = − ln NEERt + 0.9 ln Ptm + 0.1 ln Ptoil ln Ptrm = 0.25 − 0.46 ln NEERt + 0.74 ln PtmF (2.09) (−5.34) (4.02) + 0.04 ln Ptoil − 0.05 ECt−1 (4.54) (−2.12)
(E.25)
Listing of equations and variables 161 P k (Imported capital goods price index) (1978Q1–2003Q4) k ln Ptk = −0.002 + 0.31 ln Pt−1 − 0.35 ln NEERt (3.03) (−1.14) (−3.73)
− 0.15 ln NEERt−1 − 0.05 ln NEERt−2 (−1.40) (−0.50) − 0.08 ln NEERt−3 − 0.05 ln NEERt−4 (−0.41) (−0.50) mF mF + 0.28 ln PtmF − 0.11 ln Pt−1 + 0.25 ln Pt−2 (1.84) (1.53) (−0.51) mF mF − 0.001 ln Pt−3 + 0.18 ln Pt−4 (1.15) (−0.004)
(E.26)
P om (Oil import price index) (1978Q1–2003Q4) ln Ptom = ln Ptoil − ln EtUS oil ln Ptom = 0.14 − 0.53 ln EtUS + 0.68 ln Ptoil + 0.17 ln Pt−1 (3.37) (−2.27) (19.0) (4.59)
− 0.10 ECt−1 (−3.26)
(E.27)
P ox (Oil export price index) (1978Q1–2003Q4) ln Ptox = ln Ptoil − ln EtUS oil ln Ptox = 0.14 − 0.53 ln EtUS + 0.68 ln Ptoil + 0.17 ln Pt−1 (3.37) (−2.27) (19.0) (4.59)
− 0.10 ECt−1 (−3.26)
(E.28)
P nodx (Non-oil domestic export price index) (1978Q1–2003Q4) USJap
ln Ptnodx = ln Pt
USJap
− ln Et
USJap
nodx ln Ptnodx = 0.19 + 0.19 ln Pt−4 − 0.31 ln Et (2.36) (2.36) (−5.14) USJap
+ 0.84 ln Pt (4.34)
− 0.02 D_01Q1t − 0.04 ECt−1 (−3.07) (−2.39)
(E.29)
162
Listing of equations and variables
Taxes (1989Q1–2003Q4) ln TAXtN = −3.91 + 0.75 ln GDPtN − CPFCONtN (−3.17) (8.11) + 0.40 τt − 0.01 τt2 + 0.20 ln PPIt (3.52) (−3.28) (3.31) ln TAXtN = −3.06 − 0.75 ECt−1 (−8.43) (−8.49)
(E.30)
Fees and charges (1989Q1–2003Q4) ln FEEtN = −4.05 + 1.05 ln GDPtN − CPFCONtN (−2.29) (6.11) N N ln FEEtN = −1.15 − 0.64 ln FEEt−1 − 0.26 ln FEEt−2 (−2.45) (−4.40) (−2.05) N N + 2.26 ln GDPt−1 − CPFCONt−1 (2.24) N N − 0.27 ECt−1 − CPFCONt−2 + 2.5 ln GDPt−2 (−2.36) (2.36)
(E.31) CPF contributions (1980Q1–2003Q4) ln CPFCONtN = −7.50 + 0.98 ln(W · N )t + 2.45 CPFt (14.8) (−31.3) (73.7) ln CPFCONtN = −5.00 + 0.47 ln(W · N )t−1 + 1.18 CPFt (4.98) (−7.56) (2.19) − 0.67 ECt−1 (−7.55)
(E.32)
CPF housing withdrawals (1980Q1–2003Q4) RES ln CPFHOUSEtN = ln PPIt−2 N ln CPFHOUSEtN = −0.13 − 0.37 ln CPFHOUSEt−1 (−1.25) (−3.19)
+ 0.06 PLRt−2 − 0.07 ECt−1 (5.31) (−3.23)
(E.33)
Listing of equations and variables 163 Net exports (1986Q1–2003Q4) NXt = −19.5 + 0.62 (TX − TM ) (−0.16) (6.03)
(E.34)
Manufacturing capital stock (1978Q1–2003Q4) p
Kt = 129.55 + 0.68 Kt−1 + 0.07 IMTt − 0.009 Kt−1 (2.70) (8.87) (2.00) (−1.97)
(E.35)
Nominal financial wealth (1988Q1–2003Q4) N FWtN = −1727.9 + 0.12 FWt−1 + 547 PSTOCKt (17.8) (−0.98) (2.35) + 1.08 CPFCONtN − CPFWITHtN (5.20) +CPIt · (Ydt − Ct ) − ICONHOUSEtN
(E.36)
Identities FDMANt ≡ 0.0316Ct + 0.037It + 0.0125Gt + 0.0117ODXt + 0.1788NODXt + 0.1062INVTt
(I.37)
FDCOMt ≡ 0.1749Ct + 0.0413It + 0.022Gt + 0.0834TXt + 0.0079INVTt
(I.38)
FDT &Ct ≡ 0.12Ct + 0.0128It + 0.0209Gt + 0.0335TXt + 0.0036INVTt
(I.39)
FDF&Bt ≡ 0.2647Ct + 0.0571It + 0.052Gt + 0.0621TXt + 0.0117INVTt FDOTHERt ≡ 0.0165Ct + 0.0035It + 0.0071Gt + 0.0086TXt p
g
FDCONt ≡ ICONt + ICONt
(I.40) (I.41) (I.42)
FDMt ≡ 0.4088Ct + 0.6644It + 0.2866Gt + 0.5788(TX − RX )t + 0.8492INVTt
(I.43)
GDPt = VAMANt + VACOMt + VAT &Ct + VAF&Bt + VAOTHERt + VACONt
(I.44)
164
Listing of equations and variables INVTt = GDPt − (Ct + It + Gt + NXt ) − STATERRORt Ydt = GDPt − (TAXtN + FEEtN ) CPIt − (CPFCONtN − CPFWITHtN ) CPIt
(I.45)
GDPtN = GDPt · CPIt
(I.47)
CPFWITHtN ≡ CPFHOUSEtN + CPFOTHERN t
(I.48)
LOANtN ≡ BANKtN + CPFHOUSEtN
(I.49)
p
g
p
g
(I.46)
It ≡ IMTt + IMTt + ICONt + ICONt
(I.50)
TXt ≡ NODXt + ODXt + RXt + SXt
(I.51)
TMt ≡ RMt + RXt + SMt
(I.52)
UCCt = (0.12 + PLRt − 4 ln PtK ) 1 − τ c · 1 − e−3PLRt 3PLRt PtK × 1 − τc CPIt
(I.53)
rt = PLRt − 4 ln DSPIt
(I.54)
ut = [(Lt − Nt )/Lt ] · 100 emp ULCt = Wt (1 + CPFt ) · Nt GDPt
(I.55)
UBCt = 0.494ULCt + 0.506NLCt + UBCERRORt RUBCt = UBCt MULCt /EtRM
(I.57)
(I.56)
(I.58)
VATt = (VAMANt · CPIt ) Ptm
(I.59)
NtNT = N NT SHAREt · Nt
(I.60)
PRODtNT = (GDPt − VATt ) NtNT emp ULCtNT = Wt (1 + CPFt ) PRODtNT
(I.61) (I.62)
Endogenous variables Variable
Description
Equation/identity
C
Private consumption expenditures (S$ million at 1995 constant prices)
Eq (1, 36), Id (37–41, 43, 45)
Listing of equations and variables 165 Variable
Description
Equation/identity
CPFCONN
CPF gross contributions (S$ million) CPF housing withdrawals (S$ million) CPF gross withdrawals (S$ million) Consumer price index (1995 = 100) Domestic supply price index (1995 = 100) Final demand for the commerce sector (S$ million at 1995 constant prices) Final demand for the construction sector (S$ million at 1995 constant prices) Final demand for the financial and business services sector (S$ million at 1995 constant prices) Final demand for retained imports (S$ million at 1995 constant prices) Final demand for the manufacturing sector (S$ million at 1995 constant prices) Final demand for the other services sector (S$ million at 1995 constant prices) Final demand for the transport and communications sector (S$ million at 1995 constant prices) Government fees and charges (S$ million) Nominal financial wealth (S$ million) Real gross domestic product (S$ million at 1995 constant prices) Nominal gross domestic product (S$ million)
Eq (30, 31, 32, 36), Id (46) Eq (1, 33), Id (48, 49)
CPFHOUSEN CPFWITHN CPI DSPI FDCOM
FDCON
FDF&B
FDM
FDMAN
FDOTHER
FDT&C
FEEN FWN GDP
GDPN
Eq (36), Id (46, 48) Eq (1, 14, 21, 36), Id (46, 47, 53, 59) Eq (12, 14, 22), Id (54) Eq (16), Id (38)
Eq (20), Id (42)
Eq (18), Id (40)
Eq (10), Id (43)
Eq (15), Id (37)
Eq (19), Id (41)
Eq (17), Id (39)
Eq (31), Id (46) Eq (1, 36) Eq (3, 11, 12, 14), Id (44–47, 56, 61) Eq (30, 31), Id (47)
166
Listing of equations and variables
Variable
Description
Equation/identity
INVT
Changes in stocks (S$ million at 1995 constant prices) Total investment (S$ million at 1995 constant prices) Private construction investment (S$ million at 1995 constant prices) Private machinery and transport equipment investment (S$ million at 1995 constant prices) Capital stock (S$ million at 1995 constant prices) Labour force (number of people) Total loans (S$ million) Total employment (thousands)
Id (37–40, 43, 45)
I ICON p IMT p
K L LOANN N N NT NODX NX ODX PRODNT r RM RUBC RX SM SX TAX N TM TX u
Employment in the non-tradable sector (thousands) Non-oil domestic exports (S$ million at 1995 constant prices) Net exports (S$ million at 1995 constant prices) Oil domestic exports (S$ million at 1995 constant prices) Productivity in the non-tradable sector (S$ thousands per worker) Real interest rate (% per annum) Retained imports (S$ million at 1995 constant prices) Relative unit business costs (1995 = 100) Re-exports (S$ million at 1995 constant prices) Service imports (S$ million at 1995 constant prices) Service exports (S$ million at 1995 constant prices) Taxes (S$ million) Total imports (S$ million at 1995 constant prices) Total exports (S$ million at 1995 constant prices) Unemployment rate (%)
Id (37–41, 43, 45, 50) Eq (5), Id (42, 50)
Eq (3, 35), Id (50)
Eq (6, 7, 35) Eq (13), Id (55) Eq (1), Id (49) Eq (12, 32), Id (55, 56, 60) Id (60, 61) Eq (6), Id (37, 51) Eq (34), Id (45) Eq (7), Id (37, 51) Id (61, 62) Eq (12), Id (54) Eq (10), Id (52) Eq (2), Id (58) Eq (8), Id (43, 52) Eq (11), Id (52) Eq (9), Id (52) Eq (30), Id (46) Eq (34), Id (52) Eq (34), Id (38–41, 43, 51) Eq (13), Id (55)
Listing of equations and variables 167 Variable
Description
Equation/identity
UBC UCC ULC ULCNT
Unit business costs (1995 = 100) User cost of capital (1995 = 100) Unit labour costs (1995 = 100) Unit labour costs in the non-tradable sector (1995 = 100) Value-added in the commerce sector (S$ million at 1995 constant prices) Value-added in the construction sector (S$ million at 1995 constant prices) Value-added in the financial and business services sector (S$ million at 1995 constant prices) Value-added in the manufacturing sector (S$ million at 1995 constant prices) Value-added in the other services sector (S$ million at 1995 constant prices) Value-added in the transport and communications sector (S$ million at 1995 constant prices) Value-added in the tradable sector (S$ million at 1995 constant prices) Nominal wages (average monthly earnings in S$) Private disposable income (S$ million at 1995 constant prices)
Eq (6, 9, 23), Id (57, 58) Eq (3), Id (53) Id (56, 57) Eq (21), Id (62)
VACOM
VACON
VAF&B
VAMAN
VAOTHER
VAT&C
VAT
W Yd
Eq (16), Id (44)
Eq (20), Id (44)
Eq (18), Id (44)
Eq (15), Id (44, 59)
Eq (19), Id (44)
Eq (17), Id (44)
Id (59, 61)
Eq (12, 14, 32), Id (56, 62) Eq (1, 36), Id (46)
Semi-endogenous variables Variable
Description
Equation/identity
CAN
Contracts awarded by the private sector (S$ million) Net investment commitments in the manufacturing sector (S$ million)
Eq (4, 5)
NICN
Eq (2, 3)
168
Listing of equations and variables
Variable
Description
Equation/identity
Pm Pk
Import price index (1995 = 100) Price of capital goods (1995 = 100) (Imported machinery and transport equipment) Oil import price index (1995 = 100) Oil export price index (1995 = 100) Imported raw material price index (1995 = 100) Non-oil domestic export price index (1995 = 100) Singapore manufactured products price index (1995 = 100)
Eq (10, 21–25), Id (59) Eq (3, 26), Id (53)
P om P ox P rm P nodx SMPI
Eq (7, 27) Eq (7, 28) Eq (6, 25) Eq (6, 29) Eq (10, 22, 23)
Exogenous variables Variable
Description
Equation/identity
BANK
Bank lending to professionals and private individuals (S$ million) Global semiconductor sales (US$ billion) Employers’ CPF contribution rate (%)
Id (49)
CHIPN CPFemp CPF CPFOTHERN E RM E US E USJap G ICONg ICONHOUSEN
Overall CPF contribution rate (%) Other withdrawals from CPF (S$ million) Malaysian Ringgit/S$ exchange rate (period average) US$/S$ exchange rate (period average) Average of US$/S$ + Yen/S$ exchange rate (1995 = 100) Government consumption expenditures (S$ million at 1995 constant prices) Government construction investment (S$ million at 1995 constant prices) Nominal housing investment (S$ million)
Eq (6, 8) Eq (12, 14), Id (56, 62) Eq (14, 32) Id (48) Id (58) Eq (27, 28) Eq (29) Id (37–41, 43, 45)
Id (42, 50) Eq (36)
Listing of equations and variables 169 Variable
Description
Equation/identity
IMTg
Government machinery and transport equipment investment (S$ million at 1995 constant prices) Malaysian unit labour costs (1995 = 100) Share of employment in the non-tradable sector Nominal effective exchange rate of the Singapore dollar (1995 = 100) Non-labour costs (1995 = 100) Building materials price index (1995 = 100) Foreign price for imports (1995 = 100) Brent spot crude oil price (US$ per barrel) Price of competing goods for exports (1995 = 100) Service import price index (1995 = 100) Service export price index (1995 = 100) (Implicit price deflator for service industries in Singapore) Property price index (1995 = 100) Residential property price index (1995 = 100) Stock exchange of Singapore share price index (1995 = 100) Prime lending rate (% per annum)
Id (50)
MULC N NT SHARE NEER NLC P bm P mF P oil P USJap P sm P sx
PPI PPIRES PSTOCK PLR POP STATERROR τc τy τ θ
Working age population (thousands) Statistical discrepancy (S$ million at 1995 constant prices) Corporate tax rate (%) Average marginal personal income tax rate (%) Average of corporate and personal income tax rate (%) Share of value-added in manufacturing output
Id (58) Id (60) Eq (24–26) Id (57) Eq (5) Eq (24–26) Eq (24, 25, 27, 28) Eq (29) Eq (11) Eq (9)
Eq (4, 30) Eq (33) Eq (18, 36) Eq (4, 33), Id (53, 54) Eq (13, 14) Id (45) Id (53) Eq (14) Eq (30) Eq (6)
170
Listing of equations and variables
Variable
Description
Equation/identity
UBCERROR V
Error term in the UBC equation Visitor expenditures as represented by total visitor days Foreign GDP (1995 = 100)
Id (57) Eq (1, 9)
Yw
Note: S$ = Singapore dollar.
Eq (6–9)
Bibliography
Abeysinghe, T. (1992) “A small-scale model to forecast GDP growth of Singapore”, in Y. Dodge and J. Whittaker (eds) Computational Statistics, Vol.1, Germany: Physica-Verlag, pp. 297–302. Abeysinghe, T. (1998) “Forecasting Singapore’s quarterly GDP with monthly external trade”, International Journal of Forecasting, 14: 505–13. Abeysinghe, T. (2001a) “Thai meltdown and transmission of recession within ASEAN4 and NIE4”, in S. Classens and K. Forbes (eds) International Financial Contagion. Boston: Kluwer Academic Publishers, pp. 225–40. Abeysinghe, T. (2001b) “Electronics and growth cycles in Singapore”, Applied Economics, 32: 1657–63. Abeysinghe, T. and Choy, K.M. (2004) “The aggregate consumption puzzle in Singapore”, Journal of Asian Economics, 15: 563–78. Abeysinghe, T. and Gulasekaran, R. (2004) “Quarterly GDP estimates for China and ASEAN4 with a forecast evaluation”, Journal of Forecasting, 23: 431–47. Abeysinghe, T. and Forbes, K. (2005) “Trade linkages and output-multipliers: A structural VAR approach with a focus on Asia”, Review of International Economics, 13: 356–75. (First appeared as NBER Working Paper, W8600, 2001). Abeysinghe, T. and Lee, C. “Sectoral forecasts of the Singapore economy: A new quarterly model”, paper presented at the Econometric Studies Unit Conference, Singapore, March 1997. Abeysinghe, T. and Tan, L.Y. (1998) “Exchange rate appreciation and export competitiveness”, Applied Economics, 30: 51–55. Abeysinghe, T. and Wilson, P. “Forecasts for the Singapore economy”, paper presented at the ISEAS Regional Outlook Conference, Singapore, February 2001. (Available at the ESU website: http://nt2.fas.nus.edu.sg/ecs/cent/ESU/esu011.pdf ). Alston, J.M., Carter, C.A., Green, R. and Pick, D. (1990) “Whither Armington trade models?”, American Journal of Agricultural Economics, 72: 455–67. Ando, A. and Modigliani, F. (1963) “The ‘life-cycle’ hypothesis of saving: Aggregate implications and tests”, American Economic Review, 53: 55–84. Armington, P.S. (1969) “A theory of demand for products distinguished by place of production”, IMF Staff Papers, 26: 159–78. Artis, M.J. (1984) Macroeconomics, Oxford: Oxford University Press. Athukorala, P. and Riedel, J. (1991) “The small country assumption: A reassessment with evidence from Korea”, Weltwirtschaftlichess Archiv, 127: 138–51.
172
Bibliography
Athukorala, P. and Riedel, J. (1994) “Demand and supply factors in the determination of NIE exports: A simultaneous error-correction model for Hong Kong: A comment”, Economic Journal, 104: 1411–14. Athukorala, P. and Riedel, J. (1996) “Modelling NIE exports: Aggregation, quantitative restrictions, and choice of econometric methodology”, Journal of Development Studies, 33: 81–98. Ball, L. and Mankiw, N.G. (2002) “The NAIRU in theory and practice”, Journal of Economic Perspectives, 16: 115–36. Banerjee, A., Dolado, J.J. and Mestre, R. (1998) “Error-correction mechanism tests for cointegration in a single-equation framework”, Journal of Time Series Analysis, 19: 267–84. Blinder, A.S and Deaton, A. (1985) “The time series consumption function revisited”, Brookings Papers on Economic Activity, 2: 465–511. Boltho, A. (1996) “The assessment: International competitiveness”, Oxford Review of Economic Policy, 12: 1–16. Browne, F. (1982) “Modelling export prices and quantities in a small open economy”, Review of Economics and Statistics, 64: 346–47. Campa, J.M., Goldberg, L.S. and Minguez, J.M.G. (2005) “Exchange rate pass-through to import prices in the Euro area”, Staff Paper #219, Federal Reserve Bank of New York. Chew, R. (1988) The national wages council: An incomes policy experience, unpublished thesis, National University of Singapore. Chia, S.Y. (1986) “Direct foreign investment and the industrialization process in Singapore”, in C.Y. Lim and P.J. Lloyd (eds) Singapore: Resources and Growth, Singapore: Oxford University Press, pp. 79–117. Chirinko, R.S. (1993) “Business fixed investment spending: Modeling strategies, empirical results, and policy implications”, Journal of Economic Literature, 31: 1875–1911. Chow, H.K. and Choy, K.M. (1995) “Economic leading indicators for tracking Singapore’s growth cycle”, Asian Economic Journal, 9: 177–91. Deaton, A.S. (1977) “Involuntary saving through unanticipated inflation”, American Economic Review, 67: 898–910. Disney, R. and Ho, S.K. (1990) “Do real wages matter in an open economy? The case of Singapore 1966–1987”, Oxford Economic Papers, 42: 635–57. Eckstein, O. (1983) The DRI Model of the U.S. Economy, New York: McGraw-Hill Book Company. Enzler, J.J., Murphy, C.W., Ng, H.T., Phang, A.S.J. and Robinson, E. (2005) “Two decades of macromodelling at the MAS”, MAS Staff Paper No. 39. (Available at http://www.mas.gov.sg). Ermisch, J.F. and Huff, W.G. (1999) “Hypergrowth in an East Asian NIC: Public policy and capital accumulation in Singapore”, World Development, 27: 21–38. Ericsson, N.R. and Irons, J.S. (1995) “The Lucas critique in practice: Theory without measurement”, in K.D. Hoover (ed.) Macroeconometrics: Developments, Tensions and Prospects, Dordrecht: Kluwer Academic Press, pp. 263–312. Faini, R., Clavijo, F. and Senhadji, A. (1992) “The fallacy of composition argument: is it relevant for LDC manufactured exports?”, European Economic Review, 36: 865–82. Fair, R.C. (1984) Specification, Estimation and Analysis of Macroeconometric Models, Cambridge, MA: Harvard University Press. Goldstein, M. and Khan, M.S., (1978) “The supply and demand for exports: A simultaneous approach”, Review of Economics and Statistics, 60: 275–86.
Bibliography 173 Gwartney, J., Skipton, C.D., and Lawson, R.A. (2000) “Trade, openness, and economic growth”, Working Paper, Southern Economic Association Meeting. Hadjimatheou, G. (1987) Consumer Economics After Keynes: Theory and Evidence of the Consumption Function, Brighton: Wheatsheaf Books. Hall, R.E. and Jorgenson, D.W. (1967) “Tax policy and investment behaviour”, American Economic Review, 53: 247–59. Harberger, A.C. (1953) “A structural approach to the problem of import demand”, American Economic Review, 43: 148–59. Harris, D. (1997) “Principal components analysis of cointegrated time series”, Econometric Theory, 13: 529–57. Hayashi, F. (2000) Econometrics, Princeton: Princeton University Press. Hendry, D.F. (1995) Dynamic Econometrics, Oxford: Oxford University Press. Hendry, D.F. and Doornik, J.A. (2001) Empirical Econometric Modelling Using PcGive 10: Volume I, London: Timberlake Consultants. Holt, C.C., Modigliani, F., Muth, J.F. and Simon, H.A. (1960) Planning Production, Inventories and Work Force, Englewood-Cliffs, NJ: Prentice Hall. Hong, P. (1999) “Import elasticities revisited”, DESA Discussion Paper No. 10, United Nations. Hooper, P., Johnson, K. and Marquez, J. (2000) “Trade elasticities for the G-7 countries”, Princeton Studies in International Economics No. 87, Princeton University. Horsnell, P. (1997) Oil in Asia, Oxford: Oxford University Press. Houthakker, H.S. and Magee, S.P. (1969) “Income and price elasticities in world trade”, The Review of Economics and Statistics, 51: 111–25. Huff, W.G. (1994) The Economic Growth of Singapore: Trade and Development in the Twentieth Century, Cambridge: Cambridge University Press. Husain, M. (1995) “Determinants of private saving in Singapore”, in K. Bercuson (ed.) Singapore: A Case Study in Rapid Development, IMF Occasional Paper No. 119, Washington, D.C.: International Monetary Fund. Ireland, P.N. (1994) “Supply-side economics and endogenous growth”, Journal of Monetary Economics, 33: 559–71. Johansen, S. (1995) Likelihood-based Inference in Cointegrated Vector Autoregressive Models, Oxford: Oxford University Press. Jorgenson, D.W. (1963) “Capital theory and investment behavior”, American Economic Review, 53: 247–59. Kapur, B. (1983) “A short term analytical model of the Singapore economy”, Journal of Development Economics, 12: 355–76. Khor, H.E., Robinson, E. and Lee, J. (2004) “Managed floating and intermediate exchange rate systems: The Singapore experience”, MAS Staff Paper No. 37. (Available at http://www.mas.gov.sg). Klein, L.R., Welfe, A. and Welfe, W. (1999) Principles of macroeconometric modeling, Amsterdam: North-Holland, pp. 285–318. Kopcke, R.W. (1993) “The determinants of business investment: Has capital spending been surprisingly low?”, New England Economic Review, Federal Reserve Bank of Boston, Jan/Feb, pp. 3–31. Krause, L.B., Koh, A.T. and Lee, Y. (1987) The Singapore Economy Reconsidered, Singapore: Institute of Southeast Asian Studies. Kuznets, S. (1946) National Product Since 1869, New York: National Bureau of Economic Research, Princeton University Press.
174
Bibliography
Leamer, E.E. (1981) “Is it a demand curve, or is it a supply curve–partial identification through inequality constraints”, Review of Economics and Statistics, 63: 319–27. Lim, C.Y. (1997) “NWC: Targets and goals”, Accounting and Business Review, 4: 165–85. Lim, C.Y. and Associates (1988) Policy Options for the Singapore Economy, Singapore: McGraw-Hill. Lim, K.G., Chow, K.B. and Tsui, K.C. (1996) “Estimating Singapore’s import function using demand systems theory”, Singapore Economic Review, 41: 1–12. Lloyd, P.J. and Sandilands, R.J. (1986) “The trade sector in a very open re-export economy”, in C.Y. Lim and P.J. Lloyd (eds) Singapore: Resources and Growth, Singapore: Oxford University Press, pp. 183–219. Low, V. (1994) “The MAS model: Structure and some policy simulations”, in A. Chin and K.J. Ngiam (eds) Outlook for the Singapore Economy, Singapore: Trans Global Publishing, pp. 19–32. Lucas, R.E. (1976) “Econometric policy evaluation: A critique”, Carnegie-Rochester Conference Series on Public Policy, 1: 19–46. Ludwig, A. and Sløk, T. (2002) “The impact of changes in stock prices and house prices on consumption in the OECD countries”, IMF Working Paper WP/02/1. MacKinnon, J.G. (1991) “Critical values for cointegration tests”, in R.F. Engle and C.W.J. Granger (eds) Long-Run Economic Relationships, Oxford: Oxford University Press, pp. 267–76. Maddala, G.S. and Kim, I.M. (1998) Unit Roots, Cointegration, and Structural Change, Cambridge: Cambridge University Press. Mathews, R.C.O. (ed.) (1982) Slower Growth in the Western World, London: Heinemann. McCloskey, D.N. (1987) “Counterfactuals”, in J. Eatwell, M. Milgate and P. Newman (eds) The New Palgrave: A Dictionary of Economics, New York: W.W. Norton, pp. 701–03. Monetary Authority of Singapore (1993) Annual Report 1993/1994. Monetary Authority of Singapore (2000) An Introduction to the Monetary Model of Singapore, Economics Department. Monetary Authority of Singapore (2004) “Singapore’s natural rate of unemployment”, Macroeconomic Review, 3: 46–51. Muscatelli, V.A. (1995a) “NIE export performance revisited: The estimation of export demand elasticities and the role of product differentiation and growth”, in D. Vines and D. Currie (eds) North-South Linkages and International Macroeconomic Policy, New York: Cambridge University Press. Muscatelli, V.A. (1995b) “Demand and supply factors in the determination of NIE exports: A reply”, Economic Journal, 104: 1415–17. Muscatelli, V.A., Srinivasan, T.G. and Vines, D. (1992) “Demand and supply factors in the determination of NIE exports: A simultaneous error-correction model for Hong Kong”, Economic Journal, 102: 1467–77. Muscatelli, V.A., Srinivasan, T.G. and Vines, D. (1994) “The empirical modeling of NIE exports: An evaluation of different approaches”, Journal of Development Studies, 30: 279–302. Muscatelli, V.A., Stevenson, A.A., and Montagna, C. (1995a) “Modelling aggregate manufactured exports for some Asian Newly Industrialized Economies”, Review of Economics and Statistics, 77: 147–55. Muscatelli, V.A., Stevenson, A.A., and Montagna, C. (1995b) “An analysis of disaggregated manufacturing exports of the Asian NIEs to EEC, USA, and Japan”, Applied Economics, 27: 17–24.
Bibliography 175 Ng, Y.P., Tu, S.P., Robinson, E. and Choy, K.M. (2004) “Using leading indicators to forecast the Singapore electronics industry”, MAS Staff Paper No. 30. (Available at http://www.mas.gov.sg). Nguyen, D.T. (1989) “The demand for LDC exports of manufactures: Estimates from Hong Kong: A comment”, Economic Journal, 99: 461–66. Nickell, S.J. (1988) “The supply side and macroeconomic modelling”, in R.C. Bryant et al. (eds) Empirical Macroeconomics for Interdependent Economies, Washington, DC: Brookings Institution, pp. 202–21. Orcutt, G. (1950) “Measurement of price elasticities in international trade”, Review of Economics and Statistics, 32: 117–32. Parrado, E. (2004) “Singapore’s unique monetary policy: How does it work?” MAS Staff Paper No. 31. (Available at http://www.mas.gov.sg). Peebles, G. and Wilson, P. (1996) The Singapore Economy, Cheltenham and Brookfield: Edward Elgar. Peebles, G. and Wilson, P. (2002) Economic Growth and Development in Singapore: Past and Future, Cheltenham, UK: Edward Elgar. Pesaran, M.H. and Shin, Y. (1998) “An autoregressive distributed-lag modelling approach to cointegration analysis”, in S. Steinar (ed.) Econometrics and Economic Theory in the 20th Century: The Ragnar Frisch Centennial Symposium, Cambridge: Cambridge University Press. Phang, S.Y. (2004) “House prices and aggregate consumption: Do they move together? Evidence from Singapore”, Journal of Housing Economics, 13: 101–19. Phillips, P.C.B. (1994) “Some exact distribution theory for maximum likelihood estimators of cointegrating coefficients in error correction models”, Econometrica, 62: 73–93. Rao, V.V.B. and Lee, C. (1995) “Sources of growth in the Singapore economy and its manufacturing and service sectors”, Singapore Economic Review, 40: 83–115. Riedel, J. (1988) “The demand for LDC exports of manufactures: Estimates from Hong Kong”, Economic Journal, 98: 138–48. Sachs, J.D. and Warner, A. (1995) “Economic reform and the process of global integration”, Brookings Papers on Economic Activity, 1: 1–95. Schaller, H. (2006) “Estimating the long-run user cost elasticity”, Journal of Monetary Economics, 53: 725–36. Scott, M.F.G. (1989) A New View of Economic Growth, Oxford: Clarendon Press. Scott, M.F.G. (1992) “Policy implications of ‘A New View of Economic Growth”’, Economic Journal, 102: 622–32. Sims, C.A., Stock, J.H. and Watson, M.W. (1990) “Inference in linear time series models with some unit roots”, Econometrica, 58: 113–44. Singapore Economic Development Board website http://www.edb.gov.sg/edb/sg/en_uk/ index.html, 2005. Stock, J. (1987) “Asymptotic properties of least squares estimators of cointegrating vectors”, Econometrica, 55: 1035–66. Stock, J. and Watson, M. (1993) “A simple estimator of cointegrating vectors in higher order integrated systems”, Econometrica, 61: 783–820. Tan, K.E. and Thia, J.P. (2004a) “Private consumption in Singapore: Trends and development”, Economic Survey of Singapore, First Quarter, Ministry of Trade and Industry, Republic of Singapore, pp. 45–52. Tan, K.E. and Thia, J.P. (2004b) “What explains private non-residential gross fixed capital investment in Singapore?”, Economic Survey of Singapore, Third Quarter, Ministry of Trade and Industry, Republic of Singapore, pp. 56–74.
176
Bibliography
Tan K.Y., Wu, F., Toh, M.H., Thia, J.P. and Yeo, H.S. (2002) “Employment growth will trail economic recovery in 2002: Evidence from econometric estimates”, Economic Survey of Singapore, First Quarter, Ministry of Trade and Industry, Republic of Singapore, pp. 45–52. Teh, K.P. and Shanmugaratnam, T. (1992) “Exchange rate policy: Philosophy and conduct over the past decade”, in L. Low and M.H. Toh (eds) Public Policies in Singapore: Changes in the 1980s and Future Signposts, Singapore: Times Academic Press, pp. 285–314. Teo, E., Thangavelu, S.M. and Quah, E. (2004) “Singapore’s Beveridge curve: A comparative study of the unemployment and vacancy relationship for selected East Asian countries”, in Economic Survey of Singapore, Second Quarter, Ministry of Trade and Industry, Republic of Singapore, pp. 54–64. Tinbergen, J. (1956) Economic Policy: Principles and Design, Amsterdam: North-Holland. Tobin, J. (1962) “A general equilibrium approach to monetary theory”, Journal of Money, Credit and Banking, 1: 115–29. Toh, M.H. and Low, L. (1990) “An econometric model of the Singaporean economy: Modelling trade sector”, in J.T. Yap and Y. Nakamura (eds) ASEAN Link: An Econometric Study, Singapore: Longman Singapore Publishers, pp. 138–77. Toh, M.H. and Ramstetter, E.D. (1994) “A structural link model of Singapore for Asian Link”, in S. Ichimura and Y. Matsumoto (eds) Econometric Models of Asian-Pacific Countries, Tokyo: Springer-Verlag, pp. 359–90. Urata, S. and Kiyota, K. (2003) “Services trade in East Asia”, in T. Ito and A.O. Krueger (ed.) Trade in Services in the Asia-Pacific Region, Chicago: University of Chicago Press, pp. 379–428. Whitley, J. (1994) A Course in Macroeconomic Modelling and Forecasting, Hemel Hempstead: Harvester Wheatsheaf. Wilson, P. (1995) “Export instability in Singapore 1972 to 1986: A time series simulation approach”, Asian Economies, 24: 5–25. Wong, K.P. (1974) “A macroeconomic model of Singapore, 1960–69”, Singapore Economic Review, 19: 35–45.
Index
ADF test see Dickey-Fuller residual-based test ADL see autoregressive distributed lag model aggregate consumption function 16 APC and key ratios 26, 27 decline in APC 16–30, 126 dynamic elasticities 28 house prices 20, 22–4 policy recommendations 28–30 regression results 24–6 short-run fluctuation 26, 27 traditional consumption functions 19–22 Akaike Information Criterion (AIC) 50 APC see average propensity to consume ARIMA model 7 Armington trade model 46–7 Asian financial crisis 2–4, 20, 35, 41, 56, 83, 88, 109 autoregressive distributed lag (ADL) model 8, 10 average propensity to consume (APC) 12 decline 16–30, 126 Balassa-Samuelson effect 6, 15, 95, 96, 98 behavioural equations 12, 114 Bowman-Shenton test 10 Breusch-Godfrey LM (Lagrange multiplier) test 10 bridging equations 109–11 business costs 34, 36, 44, 67 capital accumulation 110 capital goods: import price index 161 capital inflows 2
capital stock 54, 58, 60, 69, 87, 110, 148–50, 163 car prices 22 Central Provident Fund (CPF) contributions 4, 18, 74, 85 equations 105–9, 162 withdrawals 24, 30, 109, 111, 112 chip sales see electronic products Chow test 10, 78, 109 Cobb-Douglas production function 71 co-integration methodology 8, 38, 43, 50, 51, 54, 59–60, 61, 66, 98 labour demand 75 commerce value-added 90–4, 158–9 competitiveness see international competitiveness computational methods of variables average personal income tax rate 153–4 disposable income 145 employment 150–1 export and import price indices 146–7 foreign income 148 manufacturing net capital stock 148–50 unemployment 151 user cost of capital 150 wages 151–2 wealth 145–6 working age population 84, 152–3 world price indices 147–8 construction contracts 42–4, 156 construction investment 40–4, 127, 132, 156 construction value-added 90–4, 159 consumer price index (CPI) 6, 20, 22, 33, 66, 72, 95, 98 consumer prices 95–100, 160
178
Index
consumption private consumption expenditures 155 see also aggregate consumption function; average propensity to consume counterfactual analysis 15, 29, 133, 134 CPF see Central Provident Fund CPI see consumer price index data constraints 7, 36 demand variables 111–12 demographic change 18, 76–7 working age population 84, 152–3 depreciation 110 Dickey-Fuller residual-based test 8, 21 disposable income 18–20, 25–6, 28, 30, 64–5, 105, 112, 126, 145 domestic demand 12, 16, 64 domestic supply price index 75, 100–1, 160 Durbin-Watson statistic 10, 21, 36 dynamic adjustment processes 8, 10, 73 ECM see error correction model economic openness 1, 3, 46 electronic products 2, 56–8, 60–2, 118, 121, 123–6, 131 employment levels 150–1, 157–8 see also labour market endogenous variables 12, 114, 141, 143, 164–7 semi-endogenous variables 101, 167–8 Engle’s test 10 equilibrium relationships 8, 9 error 112, 114, 116 error correction model (ECM) 8–11 ESU01 model 4–5 equation format 7–11 history of ESU models 5–6 identities 111–14 Lucas critique 11–12 modelling philosophy 6–7 multiplier analysis see multiplier analysis validation 115–18 exchange rates 3, 32–3, 55–6, 68, 101, 137, 139, 140 exogenous variables 5, 12, 26, 51, 101, 104, 114, 141, 142, 168–70 export equation 6, 14, 46 export functions 48–54 demand 46–8 non-oil domestic exports (NODX) 50, 54–9, 121–2, 156
oil exports 59–60, 156 policy implications 67–9 re-exports 46, 56, 60–2, 157 service exports 62–3, 157 supply 46–8 export prices 101 indices 146–7, 161 non-oil exports 103, 161 oil exports 104, 161 exports 1–2 bridging equations 109–10 net exports 89, 163 re-export economy 46, 56, 121 external demand 14, 52, 129 FDI see foreign direct investment fees and charges 18, 107, 116, 162 female workers 76–7 financial policies 2 fiscal multipliers 126–9 fiscal policy 3, 45, 131–2, 140 see also tax equations fixed capital formation 3, 31 forecasting 5–7, 11, 12, 140–4 foreign direct investment (FDI) 2, 31, 34, 69 foreign income 7, 54, 58, 67, 118, 130–1, 141, 148 foreign trade multipliers 118–26 foreign workers 70, 73, 77, 81, 84, 85, 135–6 Gauss-Seidel procedure 115 gross domestic product (GDP) 1, 18–20, 29, 31, 36, 39, 40, 66, 75–6, 83, 112, 140–2 gross national income (GNI) 1 growth forecasts 5–7, 140–4 house prices 20, 22–4, 30, 41 household wealth 20, 23, 145–6 housing withdrawals 24, 30, 109, 112–13, 162 identities 12, 111–14, 163–4 import functions 14, 63 retained imports 64–5, 157 service imports 65–7, 157 import prices 61, 64, 96, 98, 100–2, 104 capital goods 161 indices 146–7, 160–1 machinery and transport equipment imports 103
Index merchandise imports 102, 160 oil imports 102–3, 161 raw material imports 103, 160 imports 2 re-export economy 46, 56, 121 income tax rate 18, 38, 74, 79, 105–7, 153–4 incomes policy 4 industrial relations 4 see also labour market inflation 3, 6, 21–2, 100, 139, 140, 144 interest rates 3, 21–2, 38, 43, 75, 142 international competitiveness 32–4, 44, 58, 63, 67–9, 85, 114, 121, 144 inventory levels 87–9, 112 investment expenditures 31–2 construction investment 40–4 machinery and transport equipment 36–40, 110, 155 modelling investment commitments 34–6 net investment commitments 155 and international competitiveness 32–4 policy options 44–5 investment flows 110 investor confidence 45 Johansen’s likelihood ratio (LR) test 50, 51, 54, 60, 61, 62 Johansen’s maximum likelihood (ML) method 8, 25, 50, 55 Johansen’s trace test 8, 38, 43, 50, 51, 54, 59–61, 66, 98, 100 labour demand 75 Jorgenson’s capital-stock adjustment model 37 knowledge-based industries 44 labour market 4, 7, 14, 33, 58–9, 70–3, 135–6 disequilibrium dynamics 74–5 labour demand 75–6 labour supply 76–8 wage determination 78–81 employment levels 135–6, 150–1, 157–8 female participation rate 76–7 foreign workers 70, 77, 84, 85, 135–6 government intervention 4, 70 labour force 71–2, 158
179
National Wages Council (NWC) 4, 70, 80, 84, 85 policy conclusions 84–6 unemployment 72, 73, 78, 81–4, 85–6, 113, 135–6 computation 151 unit labour costs 14, 33–4, 113–14 wage-price mechanism 71 working age population 18, 78, 84, 152–3 Laffer curve 105, 107 life-cycle theory 18 Lucas critique 11–12 machinery 36–40 import prices 103 investment 155 macroeconometric models ESU01 see ESU01 model simulation see policy simulations manufacturing 2, 56 manufacturing capital stock 58, 69, 87, 110, 148–50, 163 manufacturing output 32, 35, 36 manufacturing value-added 90–1, 158 marginal propensity to consume (MPC) 21, 126 MAS see Monetary Authority of Singapore merchandise imports 14, 64, 102 modelling ESU01 see ESU01 model simulation see policy simulations Monetary Authority of Singapore (MAS) 3, 34, 68–9, 71 monetary policy 3, 136–40 MPC see marginal propensity to consume multinational corporations (MNCs) 2, 52 multiplier analysis 115 fiscal multipliers 126–9 foreign trade multipliers 118–26 model validation 115–18 policy lessons 129–32 National Wages Council (NWC) 4, 70, 80, 84, 85 nominal effective exchange rate (NEER) 3, 59, 62, 67, 101, 102, 137, 139 nominal financial wealth 20, 110–11, 145, 163 non-oil domestic exports (NODX) 50, 54–9, 121, 156 export prices 103, 161 normalization paradox 47 NWC see National Wages Council
180
Index
oil exports 59–60, 156 export prices 104, 161 oil import prices 59–60, 102–3, 161 optimal control 140–4 output levels 53–4, 87–9 parameter stability 10–12 partial adjustment model 36, 48, 88, 91 policy analysis 5–7, 11, 12, 134 policy simulations closed-door foreign worker policy 135–6 counterfactual analysis 133, 134 deterministic simulation 135 dynamic baseline simulation 134 growth rates 140–4 monetary policy 136–40 optimal control 140–4 real-world policy options 133 stochastic simulation 135 population data 152–3 see also demographic change 152–3 prediction see forecasting price equations consumer prices 95–100 export prices 101 non-oil exports 103 oil exports 104 import prices 101–2 machinery and transport equipment imports 103 merchandise imports 102 oil imports 102–3 raw material imports 103 producer prices 71, 100–1, 147–8 price indices 146–8, 160–1 price-taking behaviour 1, 47–9, 52, 60, 104 private consumption expenditures 16–17, 19–20, 29, 155 producer prices 71, 100–1, 147–8 productivity 14, 32, 35, 36, 73, 96, 98, 113, 114 property prices 20, 22–4, 30, 41, 42, 106, 114 Raffles, Sir Stamford 1 Ramsey’s RESET test 10 raw material imports 34, 48, 55, 103 raw material prices 58, 103, 146, 160 re-exports 46, 56, 60–2, 64, 121, 157 retained imports 64–5, 157
sales levels 87–9 savings behaviour 18, 21 savings flows 110–11, 146 Schwarz Bayesian Criterion (SBC) 50 sectoral production 14, 87 sectoral equations 90–4 supply-side modelling 87–9 semiconductor trade see electronic products service exports 2–3, 62–3, 68–9, 157 service imports 65–7, 157 share prices 94, 111, 146 simulation techniques see policy simulations Singapore manufactured products price index 64, 100, 160 Singstat Time Series (STS) 7, 146 statistical authorities 7, 18, 77 statistical error 112, 116 stock flows 110, 145–6 stock market 94 supply-side modelling 87–9 tax cuts 3, 18, 45, 106 tax equations 105–9, 162 average personal income tax rate 74, 105–7, 153–4 time series 7–8 Tobin’s “q model” 37 tourism 3, 24, 25, 28–9, 62, 63, 69 trade figures 1–2, 110 trade sector 14, 46 export functions 48–54 non-oil domestic exports (NODX) 50, 54–9 oil exports 59–60 policy implications 76–9 re-exports 46, 56, 60–2 service exports 62–3 import functions 63 retained imports 64–5 service imports 65–7 literature review 46–8 trade surpluses 7 transport equipment 36–40 import prices 103 investment 155 UCC see user cost of capital unemployment 72, 73, 78, 81–4, 85–6, 113 computation 151
Index unit labour costs 14, 33–4, 113–14 unit root tests 8, 10, 50, 100 user cost of capital (UCC) 37, 38, 40, 113, 150 value-added contribution 35, 56, 89–93, 111–12, 158–9 vector autoregressive (VAR) methodology 5, 25, 50 visitor expenditures 24, 28–9, 63, 69
181
wage policy 4 National Wages Council (NWC) 4, 70, 84, 85 wage rates 7, 33, 36, 58–9, 67–8, 70, 74, 78–81, 85, 97–8, 108, 135–6 computation 151–2 equilibrium and disequilibrium wages 158 unit labour costs 14, 33–4, 113–14 wealth variables 19, 20–1, 110–11, 145–6 White test 10 working age population 18, 78, 84, 152–3 world price indices 147–8