Dynamic Modeling and Applications for Global Economic Analysis

DYNAMIC MODELING AND APPLICATIONS FOR GLOBAL ECONOMIC ANALYSIS A sequel to Global Trade Analysis: Modeling and Applicat...

Author: Elena Ianchovichina; Terrie Louise Walmsley

98 downloads 1066 Views 6MB 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

DOWNLOAD PDF

DYNAMIC MODELING AND APPLICATIONS FOR GLOBAL ECONOMIC ANALYSIS

A sequel to Global Trade Analysis: Modeling and Applications (Cambridge University Press, 1997, edited by Thomas W. Hertel), this book presents the technical aspects of the Global Trade Analysis Program’s global dynamic framework (GDyn) and its applications within important global policy issues. The book covers a diverse set of topics including trade reform, growth, investment, technology, demographic change, and the environment. Environmental issues are particularly well suited for analysis with GDyn, and this book covers its uses with climate change, resource use, and technological progress in agriculture. Other applications presented in the book focus on integration issues such as rules governing foreign investment, e-commerce regulations, trade in services, harmonization of technical standards, sanitary and photo-sanitary regulations, streamlining of customs procedures, and demographic change and migration. Elena I. Ianchovichina is Lead Economist for the Middle East and North Africa region of The World Bank. Since joining The World Bank in 2000, she has served in its Research Department, East Asia and Pacific Region, and managed the Economic Policy and Debt Department’s program on inclusive growth. Her work has focused on country-specific analyses of economic growth, emerging Asia, and fiscal and trade reform. Dr. Ianchovichina has published more than twenty articles in a variety of journals, including the Canadian Journal of Economics, Contemporary Economic Policy, Review of International Economics, World Bank Economic Review, and Ecological Economics. She received Purdue University’s 2008 Apex award for outstanding contributions to quantitative trade analysis. Terrie L. Walmsley is an Associate Professor at Purdue University and a Principal Fellow and Associate Professor at the University of Melbourne, Australia. Dr. Walmsley is also the Director of the Center for Global Trade Analysis, the Purdue home of the Global Trade Analysis Project, a global network of 8,500 researchers from 150 countries (www.gtap.org). Dr. Walmsley leads the construction of the GTAP Data Base, a global database used worldwide to examine the impact of international trade and environmental policies. Her research has focused on international trade in goods and services and the movement of capital and labor across national boundaries; it has been used extensively by The World Bank.

http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139059923


Dynamic Modeling and Applications for Global Economic Analysis Edited by ELENA I. IANCHOVICHINA The World Bank

TERRIE L. WALMSLEY Purdue University and University of Melbourne, Australia


cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, S˜ao Paulo, Delhi, Mexico City Cambridge University Press 32 Avenue of the Americas, New York, NY 10013-2473, USA www.cambridge.org Information on this title: www.cambridge.org/9781107002432 C Cambridge University Press 2012

This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2012 Printed in the United States of America A catalog record for this publication is available from the British Library. Library of Congress Cataloging in Publication data Dynamic modeling and applications for global economic analysis / [edited by] Elena Ianchovichina, Terrie Walmsley. p. cm. Includes bibliographical references and index. ISBN 978-1-107-01169-4 (hardback) – ISBN 978-1-107-00243-2 (paperback) 1. International trade – Mathematical models. 2. International economic relations – Mathematical models. 3. International trade. 4. International economic relations. I. Ianchovichina, Elena. II. Walmsley, Terrie Louise. III. Title. HF1379.D957 2011 2011025080 337.01 5195–dc23 ISBN 978-1-107-01169-4 Hardback ISBN 978-1-107-00243-2 Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party Internet Web sites referred to in this publication and does not guarantee that any content on such Web sites is, or will remain, accurate or appropriate.


Contents

Contributors

page vii

Acknowledgments

ix PART I INTRODUCTION AND OVERVIEW

1 Introduction Elena I. Ianchovichina

3

PART II STRUCTURE OF THE DYNAMIC GTAP FRAMEWORK

2 Theoretical Structure of Dynamic GTAP Elena I. Ianchovichina and Robert A. McDougall 3 Behavioral and Entropy Parameters in the Dynamic GTAP Model Alla Golub and Robert A. McDougall 4 An Overview of the Dynamic GTAP Data Base: The Data Base Construction and Aggregation Programs Robert A. McDougall, Terrie L. Walmsley, Alla Golub, Elena I. Ianchovichina, and Ken Itakura

13

71

120

5 A Baseline Scenario for the Dynamic GTAP Model Terrie L. Walmsley, Betina V. Dimaranan, and Robert A. McDougall

136

6 Welfare Analysis in the Dynamic GTAP Model Terrie L. Walmsley, Robert A. McDougall, and Elena I. Ianchovichina

158

7 Implementing the Dynamic GTAP Model in the RunDynam Software Ken Itakura, Elena I. Ianchovichina, Csilla Lakatos, and Terrie L. Walmsley v


173

vi

Contents PART III APPLICATIONS OF DYNAMIC GTAP

8 Assessing the Impact of China’s WTO Accession on Investment Terrie L. Walmsley, Thomas W. Hertel, and Elena I. Ianchovichina 9 Dynamic Effects of the “New-Age” Free Trade Agreement between Japan and Singapore Thomas W. Hertel, Terrie L. Walmsley, and Ken Itakura

205

235

10 Resource Use and Technological Progress in Agriculture Elena I. Ianchovichina, Roy Darwin, and Robin Shoemaker

269

11 Global Economic Integration and Land-Use Change Alla Golub and Thomas W. Hertel

290

12 The Contribution of Productivity Linkages to the General Equilibrium Analysis of Free Trade Agreements Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer

312

13 Global Demographic Change, Labor Force Growth, and Economic Performance Rod Tyers and Qun Shi

342

PART IV EVALUATION OF THE DYNAMIC GTAP FRAMEWORK

14 Household Saving Behavior in the Dynamic GTAP Model: Evaluation and Revision Alla Golub and Robert A. McDougall 15 Implications for Global Economic Analysis Elena I. Ianchovichina and Terrie L. Walmsley Appendix: Negative Investment: Incorporating a Complementarity into the Dynamic GTAP Model Terrie L. Walmsley and Robert A. McDougall Glossary of GDyn Notation Terrie L. Walmsley Index

379 406

413

421

429


Contributors

Elena I. Ianchovichina (The World Bank, USA) Robert A. McDougall (Purdue University, USA) Terrie L. Walmsley (Purdue University, USA, and University of Melbourne, Australia, Australia) Roy Darwin (U.S. Department of Agriculture, USA) Betina V. Dimaranan (IFPRI, USA) Alla Golub (Purdue University, USA) Thomas W. Hertel (Purdue University, USA) Ken Itakura (Purdue University, USA) Csilla Lakatos (Purdue University, USA) Jeffrey J. Reimer (University of Wisconsin, USA) Qun Shi (Australian National University, Australia) Robin Shoemaker (U.S. Department of Agriculture, USA) Rod Tyers (University of Western Australia, Australia)

vii



Acknowledgments

We would like to thank a number of individuals for their help in making this book a reality. The dynamic GTAP work builds on the intellectual accomplishments of the GTAP project, and we are indebted to the many individuals who have contributed to GTAP over the years. We also recognize that the dynamic GTAP project would not have become a reality without the unwavering support of Professor Thomas W. Hertel – the founder of the GTAP project – who encouraged and steered this work in the right direction from the very beginning. We are grateful to several individuals for their advice and guidance in the early stages of the dynamic GTAP project. We were fortunate to have Robert A. McDougall on the dynamic GTAP team. His leadership and vast technical expertise were invaluable. Philippa Dee encouraged us to embark on this work because she foresaw the appeal of a dynamic GTAP version to analysts of a wide range of global economic policy issues. As in the case of GTAP, Alan Powell was generous with his time and support, whereas Ken Pearson and Mark Horridge helped us operationalize the dynamic model and address GEMPACK and other software issues. Over the years a number of researchers contributed to this project in different ways. Ken Itakura worked on updating the tab file to match the GTAP v6.2 tab file, whereas Csilla Lakatos developed postsimulation processing programs. We are grateful to the instructors of and participants in the dynamic GTAP short courses held at Purdue University in 2000, 2006, 2008, and 2010. These courses provided an opportunity for fruitful exchanges on important research questions concerning the theory behind the model and many of the applications discussed in this book. We would like to recognize in particular Robert A. McDougall, Anna Strutt, Thomas W. Hertel, Alla Golub, Csilla Lakatos, Ken Itakura, Amer Ahmed, Angel Aguiar, Peter Minor, Peter Dixon, and Kevin Hanslow. Finally, we are grateful to the three anonymous reviewers for their useful insights and comments. ix



PART I

INTRODUCTION AND OVERVIEW



ONE

Introduction Elena I. Ianchovichina

The objective of the Global Trade Analysis Project (GTAP), launched in 1992, was to lower the cost of entry into the world of applied computable general equilibrium (CGE or AGE) modeling using a global, economy-wide framework. The birth of the GTAP project and the subsequent publication of the GTAP book (Hertel 1997), which documented the model structure, data, and software, were timely because there was an increasing demand for quantitative analyses of trade policy issues on a global basis. Most notably, the Uruguay Round negotiations under the auspices of the General Agreement on Tariffs and Trade (GATT) were a catalyst in moving forward the GTAP database and model, as were the heated debates over the North American Free Trade Agreement (NAFTA) and subsequently the World Trade Organization’s (WTO) Doha Development Agenda. In response to this demand, the GTAP project grew from a few people in a handful of countries in 1992 to more than 8,500 people from 140 countries in 2010; the GTAP book has been widely cited; and the GTAP model and data have been actively used by a large number of public institutions around the world and in analyses on various topics published in numerous refereed journals, books, and reports. The dynamic GTAP model (GDyn) is a follow-up to the GTAP model (Hertel 1997). It is a recursively dynamic applied computable general equilibrium framework of the world economy that extends standard GTAP to include features that improve the treatment of the long-run in the model, but retains all its other features. Part II of this book documents these extensions to the GTAP model structure and data, the construction of a baseline, and the software. Part III consists of six applications of the model that highlight the versatility of the modeling framework.

3

4

Elena I. Ianchovichina

1. Motivation for the GDyn The main objectives of GDyn are to provide a better treatment of the longrun within the GTAP framework and a way of tracing the evolution of the global economy through time. A good treatment of the long-run and issues of timing is essential when analyzing the economics of some of today’s most prominent global issues, such as climate change, natural resource management, globalization, and demographic change. For a good long-run treatment, we need international capital mobility that will allow us to capture how policy shocks and other developments diversely affect incentives to invest in different regions. We also need to determine regional capital stocks, which is most satisfactorily done in a dynamic model. GDyn aims to facilitate analysis of the economic implications of climate change, economic growth, and other issues affecting the global economy in a dynamic context.

2. Data for GDyn With capital mobile between regions, the database for GDyn needs to extend beyond the standard GTAP Data Base. It needs to allow for foreign and domestic ownership of regional capital stocks, as well as international income payments and receipts. This is necessary because the assets owned by a region need no longer be the assets located in that region and the income generated by the assets in a region need no longer accrue to that region’s residents. To limit the burden of data construction for GDyn, and because data on foreign assets and liabilities are limited and inconsistent globally, we prefer a treatment of foreign assets that is parsimonious in its data requirements. This treatment is discussed in Chapters 2 and 4. New pieces of data are also needed to accommodate the new lagged adjustment, adaptive expectation theory of investment. In GDyn investors act so as to eliminate disparities in expected rates of return, not instantaneously, but progressively over time. The parameters determining the speed of convergence in rates of return are presented in Chapter 3. These parameters have either been estimated econometrically or have been informed by econometric or empirical evidence. Finally, macroeconomic and policy projections data are needed for the construction of a baseline scenario. On the macroeconomic side these data include projections of gross domestic product, gross domestic investment, population, and skilled and unskilled labor. On the policy side, these data include policies that are important elements of the baseline scenario, and their inclusion will depend primarily on the issue being examined. For

Introduction

5

example, if one is interested in free trade agreements among the Association of South-East Asian Nations (ASEAN) countries, it would be important to incorporate those agreements that have already been ratified. However, if one is interested in agreements between the EU and South Africa, then agreements between ASEAN countries may be of limited concern. Chapter 5 discusses the construction of the macroeconomic and policy projections. The data for GDyn adhere to the same principles as the GTAP Data Base – public availability at cost, upgrades coordinated with the release of the standard GTAP Data Base, and broad participation. The network of GDyn users includes those who would identify areas for improvement or extension of the database and who are encouraged to work with GTAP staff to incorporate their ideas into future database releases. The operational concept that has worked for years for the GTAP Data Base continues to apply in the case of GDyn: “If you do not like it, help fix it!”

3. Model and Software The investment theory and the treatment of financial assets and associated income flows in GDyn are discussed in Chapter 2. The main features include the treatment of time; the distinctions between physical and financial assets, and between domestic and foreign financial assets; and the treatment of capital and asset accumulation, assets and liabilities of firms and households, income from financial assets, and the investment theory of adaptive expectations. The discussion focuses on those areas in which the new treatment of the long-run required us to make changes to the standard GTAP model. All distinguishing features of GTAP, apart from those discussed in Chapter 2, remain unchanged. These include the treatment of private household behavior, international trade, and transport activity. Auxiliary variables in GTAP that facilitate the construction of alternative closures, including partial equilibrium specification, are preserved in GDyn. The GDyn model is implemented using the GEMPACK software suite (Harrison and Pearson 1998) developed at the Centre of Policy Studies and IMPACT Project, Monash University, under the direction of Kenneth Pearson. Other general equilibrium models solved using the GEMPACK software suite include the standard GTAP model and the Monash model of Australia. The software that makes it easy to run GDyn is RunDynam, which is a program created by Ken Pearson and specially tailored to the needs of GDyn. It allows users to examine the data, construct and modify experiments, produce solutions, and examine results. Users who wish to alter the underlying theory of the model will need to acquire GEMPACK

6


and RunDynam from the Centre of Policy Studies at Monash University, Australia. Those who wish to make their own data aggregations will need to purchase the GTAP Data Base and GDyn extensions from the GTAP Center, Purdue University, United States. At the time of publication of this book, a number of applications of GDyn have been published in refereed journals, professional books, and magazines, and a half dozen are currently underway worldwide. These applications address a variety of issues, including trade policy reform, regional integration, equilibrium real exchange rate analysis, technical change, natural resource management, global climate change, and demographic change. Six of these applications were selected for inclusion in this book. They are representative of the work being undertaken currently with GDyn.

4. Short Course in Dynamic Global Economic Analysis Although its dynamic nature makes GDyn somewhat more complex than standard GTAP, the use of GDyn has spread around the world. Since 2000, when the Center for Global Trade Analysis held the first short course in Dynamic Global Economic Analysis, the model has been used by economists in universities and public research institutions in more than 20 countries on five continents. The dynamic GTAP course was offered again in 2006, 2008, and 2010, and there are plans to have this course offered at regular intervals in different parts of the world.

5. Overview of the Book This book is divided into four parts, of which this chapter is the first. Part II presents the technical aspects of the GDyn framework. In six chapters it covers data and new theoretical extensions that enable improved treatment of the long-run in GDyn, the construction of a baseline, welfare analysis in a dynamic model, and the software used to run the model. Chapter 2 presents an in-depth exposition of the investment theory and the treatment of financial assets and associated income flows. Chapter 3 discusses the techniques used to determine the magnitude of behavioral and entropy parameters used in the theory presented in Chapter 2. Chapter 4 discusses the data construction and aggregation programs. Chapter 5 documents the steps involved in building a baseline for GDyn. Chapter 6 develops a method for decomposing welfare in GDyn. Chapter 7 presents the software for running the model and analyzing model solutions.

Introduction

7

Part III of the book is a collection of six applications of GDyn. These are grouped by topic and examine a diverse set of issues: trade reform, growth and investment, climate change, natural resources, technology, and demographic change. The first of these applications, Chapter 8, is authored by Walmsley, Hertel, and Ianchovichina. This application formally explores the linkage between China’s WTO accession and investment in China in the period between 1995 and 2020. The application is similar to the one presented in Walmsley and Hertel (2001), who also used GDyn, but makes a number of enhancements, including the depiction of the duty drawback regime in China and the liberalization of trade and investment in services. Walmsley, Hertel, and Ianchovichina find that investment in China has increased substantially as a result of China’s accession. Accession doubles the extent of foreign ownership of Chinese assets relative to the no-accession baseline by 2020. Central to this increase in foreign ownership is the expected catch-up in the productivity of the services sectors driven by the opening of these sectors to foreign investment. The resulting impact on GDP is also large – 22.5% higher than the baseline by 2020. China’s welfare gain (15% by 2020) is dampened because a substantial share of additional investment comes from overseas. These estimates are far larger than those predicted by earlier studies, which ignored the reforms affecting the services sectors of China and which also abstracted from capital accumulation and international capital mobility. In Chapter 9 Hertel, Itakura, and Walmsley turn their attention to issues that are the focus of “new-age” free trade agreements (FTAs), such as rules governing foreign investment, e-commerce regulations, trade in services, harmonization of technical standards, sanitary and phyto-sanitary regulations, and the streamlining of customs procedures. They use GDyn, which is well suited to capturing the long-run effects of new-age FTAs, to assess the dynamic gains from the FTA between Japan and Singapore. They find that the impacts of this new-age FTA on bilateral trade and investment flows are significant – with customs automatization playing the most important role in driving increases in merchandise trade. The FTA also boosts rates of return in the two economies, thereby increasing both foreign and domestic investment as well as GDP. Chapters 10 and 11 address issues related to climate change and the natural resource implications of technical change in agriculture. In Chapter 10, Ianchovichina, Darwin, and Shoemaker incorporate different types of land in GDyn to analyze the impact of a slowdown in agricultural total factor productivity (TFP) growth on agriculture and forest resources. They find that a slowdown in agricultural TFP growth might lead to higher crop

8


prices in all regions, with South East Asia facing the steepest increases. A slowdown in agricultural TFP growth might also be accompanied by increases in conversion rates of forestland to farmland as well as by a worsening of environmental or ecological damages on the remaining forestland. In Chapter 11, Golub and Hertel investigate the role of globalization and growth in determining long-run patterns of land-use change. They are able to isolate the impact on land markets of the following elements of globalization: population growth, real income growth, access to forestland, and international trade. Of the two demand-side factors, real income growth is more important than population growth. The potential for accessing new forestland plays a small role in dampening the growth in global land rent, whereas international trade plays a very substantial role in mediating between the land-abundant, slower growing economies of the Americas and Australia/New Zealand, and the land-scarce, rapidly growing economies of Asia. When combined, the forces of globalization are expected to play a large role in determining the pattern of land-use change. Chapter 12, authored by Itakura, Hertel, and Reimer, incorporates into GDyn existing econometric evidence of the strong correlations that exist between firm productivity, on the one hand, and investment and trade on the other, to study the economic effects of a recently proposed East Asian FTA. They find that although conventional applied computable general equilibrium (AGE) modeling effects predominate and are reinforced by the productivity effects, in some cases, the latter actually reverse the changes predicted by the conventional effects. In the final application, Chapter 13, Tyers and Shi introduce a global demographic submodel into a version of GDyn in which regional households differ by gender and age group. Their goal is to study the global implications of demographic change. They find that increased longevity of the global population slows down growth in real per capita incomes, lowers saving rates, and alters the distributions of global economic activity in favor of those regions with high aged labor force participation. Part IV of the book offers an evaluation of the GDyn framework. In Chapter 14, Golub and McDougall evaluate the evolution of foreign assets and liabilities in GDyn by comparing the model-determined and actual trends in foreign assets, foreign liabilities, and net foreign assets – all as a share of wealth. Their choice of the first two indicators is driven by their interest in regional wealth allocation and the composition of capital in GDyn. The third indicator is of more general interest and is often used in discussions in the financial press and among policy makers in reference

Introduction

9

to the sustainability of net foreign liabilities. How large can gross and net foreign assets and liabilities get in GDyn simulations? Have such changes been observed in past data? These are the types of questions Golub and McDougall ask in Chapter 14. For this comparison they use a country portfolio database constructed by Kraay et al. (2000). The database covers 68 countries, including all industrial countries and many developing countries, for the years ranging from 1966 to 1997. Golub and McDougall find that, unlike in real data, net foreign positions in GDyn grow without bounds in the very long-run. As economies with high saving rates in the initial year (e.g., China) grow, there is a glut of savings in GDyn, and rates of return to capital fall without bound in the very long-run. The reason for this is the assumption of fixed propensity to save in each region in GDyn. To ensure that, as in reality, gross foreign assets and liabilities do not diverge without bound in the very long-run, Golub and McDougall endogenize the saving rate in GDyn and make it a function of the share of wealth in income. The new theoretical structure supports balanced growth scenarios, stabilizes global rates of return to capital, and prevents net external assets or liabilities from growing implausibly large. Chapter 15 by Ianchovichina and Walmsley provides an overall evaluation of the effort to date, as well as some observations about the future course of global economic analysis with GDyn.

6. Reader’s Guide For those who wish to master the material presented in the book, the most efficient approach is to read the book chronologically – beginning with the model structure, reading about the parameters and data, accessing the software from the Web site, mastering the construction of the baseline, and only then turning their attention to the applications and the evaluation of the model. However, many readers may be interested in a particular topic. Those interested in climate change, natural resources, and demographics might want to start with Chapters 10, 11, and 13, respectively. Having read these chapters and understanding the essence of the changes introduced to address these topics, you might want to study the features of the main dynamic model. For this you need to backtrack to Chapter 2 and then go to Chapters 3–7. Those interested in trade policy and investment might want to look at Chapters 8, 9, and 12 first, before going back to the details of the dynamic model presented in Chapters 2–7.

10


References Harrison, W. J. and K. R. Pearson. 1998. An Introduction to GEMPACK. GEMPACK Document No. GPD-1 (4th ed.). Melbourne, Australia: Centre of Policy Studies and Impact Project, Monash University. Hertel, T. W. (ed.). 1997. Global Trade Analysis Modeling and Applications. Cambridge: Cambridge University Press. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. National Bureau of Economic Research Working Paper Series No. 7795. Washington, DC: NBER. Walmsley, T. L. and T. W. Hertel. 2001. “China’s Accession to the WTO: Timing is Everything.” World Economy 24(8), 1019–49.

PART II

STRUCTURE OF THE DYNAMIC GTAP FRAMEWORK



TWO

Theoretical Structure of Dynamic GTAP Elena I. Ianchovichina and Robert A. McDougall

1. Introduction GDyn is a recursively dynamic AGE model of the world economy. It extends the standard GTAP model (Hertel 1997) to include international capital mobility, capital accumulation, and an adaptive expectations theory of investment. This chapter discusses the rationale behind the design decisions affecting GDyn and presents its technical features in detail. The main objective of GDyn is to provide a better treatment of the long-run within the GTAP framework. In standard GTAP, capital can move between industries within a region, but not between regions. This inability of capital to move between regions impedes the analysis of policy shocks and other developments that affect incentives to invest in different regions. For a good long-run treatment, then, we need international capital mobility. With capital mobile between regions, we need to expand the national accounts to allow for international income payments. Policies that attract capital to a region may have a strong impact on the gross domestic product, but if the investment is funded from abroad, the impact on the gross national product and national income may be much weaker. Therefore, to avoid creating spurious links between investment and welfare, we need to distinguish between asset ownership and asset location: The assets owned by a region need no longer be the assets located in that region, and the income generated by the assets in a region need no longer accrue to that region’s residents. To distinguish between asset location and ownership, we introduce a rudimentary representation of financial assets. Regions now accumulate not only physical capital stocks but also claims to the ownership of physical capital. These ownership claims are financial assets of some kind. Thus 13

14

Elena I. Ianchovichina and Robert A. McDougall

international income receipts and payments emerge as part of the system of accounting for financial assets. With capital internationally mobile, we need to determine regional capital stocks. This is most satisfactorily done in a dynamic model. First, tracing out the investment and capital stock time paths is the best way to assure ourselves that the end-of-simulation capital stocks are reasonable. Second, the immediate impact of the earlier period investments required to achieve the end-of-simulation stocks in regional economies is itself of some interest. Accordingly, we make the model dynamic and incorporate the stock flow or intrinsic dynamics of investment and capital accumulation. Likewise, we incorporate the intrinsic dynamics of saving and wealth accumulation. The key features of this extension are endogenous regional capital stocks, international financial assets and liabilities, international investment and income flows, and intrinsic dynamics of physical and financial asset stocks. While introducing these new features we seek to preserve the strengths of the standard GTAP model, including the abilities to work with empirical rather than highly stylized databases and to solve the model in a reasonable time on reasonable computing platforms, while preserving a detailed regional and sectoral disaggregation, a money metric of utility, and an associated decomposition. The GDyn model is suitable for medium- and long-run policy analysis, in which the comparative statics of the end-of-simulation solution is supplemented with time paths leading to the solutions. It has enough dynamics and a sufficient treatment of financial assets to support this analysis, but not enough to support short-run macroeconomic dynamics or financial or monetary economics. A salient technical feature of GDyn is the treatment of time. Many dynamic models treat time as an index, so that each variable in the model has a time index. In GDyn, time itself is a variable, subject to exogenous change along with the usual policy, technology, and demographic variables. Section 2 elucidates the mechanics and motivation of this treatment, and Section 3 applies it to the capital accumulation equation. This lays the groundwork for the discussion in Section 4 of wealth accumulation, financial asset determination, and foreign income flows. Section 5 describes the investment theory, incorporating lagged adjustment of capital stocks and adaptive expectations for the rate of return. Section 6 discusses the properties of the complete model and the existence of and convergence toward a long-run equilibrium. The chapter concludes in Section 7 with a summary of the strengths and limitations of the model.

Theoretical Structure of Dynamic GTAP

15

We encourage the reader to consult the glossary for a list of variables and coefficients and the TABLO code for the model equations.1 Doing so will help the reader follow the notation and relate the chapter to the solution program source code. We give each equation appearing in the model in two or three forms: the levels equation (if appropriate) in mathematical notation, the differential (change) equation in mathematical notation, and the differential equation as coded in the model. The coded equations are close but not literal transcriptions from the source code.

2. Time As noted in Section 1, a key technical feature of GDyn is the treatment of time not as a discrete index but as a continuous variable. However, because the continuous-time treatment may be less familiar to many readers, we first provide an overview of the more familiar discrete-time approach and then contrast the two. Within the very large class of dynamic economic models, we confine our discussion to recursively solvable AGE models. In discussing solution methods, we assume the use of the GEMPACK suite of economic-modeling software. We use a simplified wealth accumulation equation to illustrate and contrast the discrete- and continuous-time approaches. In a closed economy with a single capital good, which constitutes the sole economic asset and hence the sole vehicle for saving, real wealth may be defined as the size K of the capital stock, and the evolution of the capital stock through time is given by an integral equation,2 T K = K0 + I (τ)dτ, (2.1) T0

where K 0 denotes the capital stock at some base time T0 , and I, net investment.

2.1 The Discrete-Time Approach Within a recursively solvable discrete-time framework, there is typically a concept of a time period. A given database refers to a given time period; 1 2

The model code is available on the Web site at https://www.gtap.agecon.purdue.edu/ models/Dynamic/applications.asp. Note that equation (2.1) combines features that might be separately represented by the capital and wealth accumulation equations in a more complex setting.

16


a simulation takes the database to the next time period, with simulation results representing changes between the initial period and the next. Within such a framework, the database might include a representation of the economy in the current period, together with some extra data pertaining to the next period. The representation of the economy might contain values as of the start of the period, or as of the midpoint of the period, or average values over the period. The extra data might be only the period length or might include, for example, values of stocks at the start of the next period. Suppose that the database contains a representation of the economy at the start of the period, together with the period length. We have from equation (2.1), by the mean value theorem and assuming a continuous time path for investment I, K = K 0 + (T − T0 )I (Tm ),

(2.2)

for some Tm between T0 and T , where we now interpret the base time T0 as the start of the period represented by the initial database. For small T − T0 , we have I(Tm ) ≈ I(T0 ), so K ≈ K 0 + I 0L ,

(2.3)

where L denotes the interval length T − T0 . By differentiating equation (2.3), we obtain the percentage change in the capital stock k within the simulation: k ≈ 100

I 0L . K0

(2.4)

We may calculate the right-hand side as a formula outside the model and apply it as a shock to k, or to avoid performing a separate calculation before the simulation, we may include a capital accumulation equation within the solution program, writing k ≈ 100

I 0L h, K0

(2.5)

where h is an artificial variable, sometimes called a homotopy variable, which is always exogenous and always receives a shock of 1 in a dynamic simulation. Note that the coefficients I0 and K0 refer to the start-of-simulation database and are not updated within the simulation. Note that equation (2.1) is true only approximately, not exactly. This is not because of a linearization error arising as a result of the passage from the levels to the change equation: Indeed, there is no such error, because the levels equation (2.3) is itself linear. The change equation instead inherits an


17

error from the levels equation because the levels equation is itself inexact. The error cannot be reduced by refinements in the solution procedure, such as using smaller step sizes, because it is inherent in the levels equation. The only way to reduce the error is by revising the simulation strategy and using more simulations with shorter time intervals. Once the time interval is set, we have an irreducible inaccuracy in the accumulation equation. Readers familiar with the discrete-time approach may object that their own favorite discrete-time model does not suffer from this particular inaccuracy. In general, it appears that it is possible to change the form of the inaccuracy, but not to eliminate it. Suppose, for example, that the database represents the average state of the economy through the period, together with start-of-period and end-of-period stocks. Then we can derive exact equations for the start-of-period and end-of-period stocks for the next period, given initial-period and next-period investment. To calculate the next-period average capital stock value, however, we need to know how investment is distributed in time through the next period, but we cannot possibly know this. Therefore the determination of the through-period average capital stock is necessarily approximate. For sufficiently small time steps, this inaccuracy does not matter much, but for larger time steps, we must replace equation (2.5) by some other, more complex equation that offers a better approximation over longer periods. For example, in our closed economy we may equate investment with saving. Then, we have I = S/, where S denotes nominal net saving, and is the price of investment goods. If S AP is the average propensity to save, we then have S = S AP Y and I = S AP Y/, where Y denotes nominal income. Writing Y as the product of real income YR and some price index PY , investment is given by I =

S AP P Y YR .

Substituting equation (2.6) into equation (2.1), we have T S AP (τ)P Y (τ)YR (τ) dτ. K = K0 + (τ) T0

(2.6)

(2.7)

Now it is possible to solve equation (2.7) in terms of initial and final values of the variables under the integral, only with the aid of various supplementary assumptions. For example, one might assume that real income YR maintains some constant growth rate between one period and the next; the average propensity to save, S AP , maintains some constant time rate of change; and the prices PY and jump immediately to their final values. With prices

18


being liable to overshooting, we might prefer this assumption to a steadygrowth assumption. The resulting equation will obviously be quite different from and far more complex than equation (2.5). Less obviously, it will, like that equation, include period-length-dependent parameters. Thus by making assumptions about time paths of variables between adjacent periods, we might derive a longer-run wealth accumulation equation. The details of the assumptions are not important. The point is that, to implement the discrete-time approach for longer time intervals, we would need to make strong assumptions about the time paths of various economic variables between time periods (e.g., that the variables involved are typically endogenous to the system and that the assumptions must be applied not at run time but in developing the accumulation equation). The method we have outlined is just one of many ways to implement a discrete-time treatment of capital accumulation, but it serves to illustrate some common features of this approach. r The database represents the economy in some period of time, possibly

but not necessarily at a single time point within the period.

r The capital accumulation equation includes coefficients derived not

r

r r

r

from the current but from the start-of-simulation database. It may also include some current coefficients, although in our illustrative example it does not. The capital accumulation equation includes parameters that depend on the size of the time step for the simulation. In our illustration, the time step size itself is L. Given the size of the time step, there is some inaccuracy built into each experiment that cannot be removed by refining the solution procedure. Major changes in the step size are liable to require revision not only of the parameters but also of the form of the capital accumulation equation. For longer time intervals, the accumulation equations are liable to embody strong assumptions about time paths of endogenous variables.

In conclusion, the discrete-time treatment of capital accumulation is viable, but it is apt to suffer from some problems, including inaccuracy, special assumptions about investment paths, and inflexibility in the size of the time step. Fortunately, there is an alternative. Capital accumulation lends itself naturally to a continuous-time approach, described next.


19

2.2 The Continuous-Time Approach Returning to equation (2.1), we now reinterpret the database as representing the economy at some point in time. Both stock data and flow data refer to the same time point. In addition we treat T not as a discrete index but as a variable within the model. Totally differentiating equation (2.1), we then obtain the equation k = 100

I t, K

(2.8)

where k represents percentage change in the capital stock, and t represents change in time. This is very similar in form to the discrete-time equation (2.5). There are, however, two differences: The time variable t replaces the homotopy variable h, and the equation uses the current rather than initial values of investment I and the capital stock K. These differences have major consequences. First, the new equation, being the linearized form of equation (2.1), involves a linearization error, but not an irreducible error. Thus the error in the calculation of the capital stock may be made as small as desired by refining the solution procedure; for example, by increasing the number of subintervals. Second, because there is no irreducible error, the equation is equally valid for any time interval. Third, because the length of the time interval is given by a variable (t) rather than by a parameter (L), the time-interval length is determined at run time rather than in the database. In contrast to the discrete-time approach, our approach does the following: r uses the database to represent the economy at a point in time r in a multistep solution, uses no coefficients derived from the start-ofr r r r

simulation database, but only current values involves no parameters that depend on the length of the time interval involves no irreducible inaccuracies in dynamic relations uses the same accumulation equation for any time interval relies on no prior assumptions about the time paths of endogenous variables

The notion of time as a variable can be explained in terms of the sources of change in an economy. An economy may change not only in response to changes in external circumstances such as technology, policy, or endowments but also through the intrinsic dynamics of its stock-flow relationships. In the presence of nonzero net investment or saving, the passage of time

20


leads to change in the stock of capital goods or of wealth. Furthermore, with adaptive expectations or lagged adjustment, the passage of time leads to the revision of expectations or the adjustment of target variables toward equilibrium. Such changes, arising not from changes in external circumstances but autonomously through the passage of time, can be captured in time terms using the time variable t in the equation system. The shock to t defines the change in time through the simulation. Shocks to other exogenous variables represent accompanying changes in external circumstances.

3. Capital Accumulation We now begin applying the time treatment described in Section 2 to the GDyn equation system. We start with the capital accumulation equation, deriving the capital stock variable used both in the financial assets theory presented in Section 4 and the investment theory discussed in Section 5. We begin with the integral equation for the capital stock, QK = QK 0 +

TIME

QCGDSNET × dτ,

(2.9)

TIME 0

where QK(r) represents the capital stock in region r, QK_0(r) is the capital stock at some base time TIME_0, TIME is current time, and QCGDSNET(r) is net investment. Totally differentiating equation (2.9), we obtain QK(r) ×

qk(r) = QCGDSNET(r) × time, 100

(2.10)

where qk(r) represents percentage change in the capital stock in region r, and time is change in time. Multiplying both sides by 100 times the price of capital goods, we obtain VK(r) × qk(r) = 100 × NETINV(r) × time,

(2.11)

where VK(r) denotes the money value of the capital stock in region r, and NETINV(r) represents the money value of net investment. In a static simulation, with time equal to zero, we see from equation (2.11) that the percentage change in the capital stock qk is also zero. Sometimes, however, we wish to impose some nonzero change in capital stocks. To that end we introduce into the accumulation equation a region-generic shift factor SQKWORLD and a region-specific factor SQK(r). Incorporating


21

those factors, we obtain the final version of the levels equation, QK(r) = SQKWORLD × SQK(r) TIME QCGDSNET(r)dτ . × QK 0(r) +

(2.12)

TIME 0

The differential equation is then given by the following equation, VK(r) × qk(r) = VK(r)[sqkworld + sqk(r)] + 100 × NETINV(r) × time,

(2.13)

represented as follows in the model code: Equation KBEGINNING #capital accumulation # (all,r,REG) VK(r)∗ qk(r) = VK(r)∗ [sqkworld + sqk(r)] + 100∗ NETINV(r)∗ time.

4. Financial Assets and Associated Income Flows As discussed in the introduction, to model international capital mobility we need to distinguish between asset location and ownership. For this purpose, we introduce financial assets. In GDyn, regional households do not own physical capital; only firms do. Households own not physical capital but financial assets, which represent indirect claims on physical capital. In this section, we show how the model determines agents’ financial assets and liabilities, as well as the associated income receipts and payments. We begin with a discussion of the treatment’s general features in subsection 4.1 and follow with a note on notation in subsection 4.2. Stock-flow accumulation relations determine two key financial asset variables presented in subsection 4.3. With those as constraints, we use an atheoretic mechanism to determine the composition of firms’ liabilities and regional households’ assets in subsection 4.4. We complete the module with equations for the assets and liabilities of the global financial intermediary in subsection 4.5 and for income flows associated with financial assets in subsection 4.6.

4.1 General Features In addition to the prime motivation to take account of international capital mobility, several other requirements have shaped the treatment of financial assets in GDyn. For reasons discussed in subsection 5.1, we do not enforce rate-of-return equilibration over the short-run. This means that we need to represent gross ownership positions. It is not enough, for example, to know

22


a region’s net foreign assets. We must know both its gross foreign assets and its gross foreign liabilities, because their rates of return may differ. To limit the burden of data construction for the extended model, and because data on foreign assets and liabilities are limited and inconsistent, we prefer a treatment of foreign assets that is parsimonious in its data requirements. We also want the treatment to accommodate the salient empirical regularity of local specialization – that countries do not hold globally balanced asset portfolios, but specialize strongly in holding local assets. With the new treatment, we do not aim to give a full or accurate representation of financial variables. The financial assets in GDyn are there not to provide a good representation of financial assets in the real world, but to let us represent international capital mobility without creating leaks in the foreign accounts. Our treatment of financial assets accordingly is minimalist and highly stylized. Influenced by these considerations, we determine some broad features of the financial assets module. First, we elect not to adopt a full financetheoretic treatment of financial assets, but to take an ad hoc or heuristic approach. The attraction of a finance-theoretic approach is that it would let us account in a principled way for investors’ holding assets with different rates of return, rather than only the highest yielding asset. It would recognize that investors are concerned not only with return but also with risk. It would relate their decisions on risk-return tradeoffs and their consumption and saving behavior to the same set of underlying preferences, preserving thereby the rigor of the welfare analysis. Yet introducing a finance-theoretic treatment would add greatly to the complexity of the model and create perhaps as many difficulties as it would solve. There are a number of paradoxes in international financial behavior, empirical regularities that are difficult to account for theoretically. Most relevant here, it is difficult to account for observed disparities between countries’ rates of return, which far exceed those predicted with simple financetheoretic models, plausible behavioral parameter settings, and observed risk levels. This does not rule out the finance-theoretic approach, but it does make the cost-benefit balance less attractive. On balance then, we elect not to implement such a treatment in this version of GDyn, while acknowledging its attractiveness as an area for future research. After this basic decision, there are several other design decisions to make. First, we must decide which physical assets should back financial assets – in other words, to which assets should financial assets represent indirect claims. To allow for international capital mobility, we must include physical capital in this set. We may also include primary factors, also referred to as “endowment commodities” in GTAP jargon, other than labor. In the GTAP


23

version 4 Data Base (McDougall, Elbehri, and Truong, 1998), there are two of these endowment commodities: (1) agricultural land and (2) other natural resources such as mineral deposits, fisheries, and forests. Although it would be more logical to let all these commodities back financial assets, it is easier to let only physical capital back financial assets. In this version of the model, we take the easier approach. Accordingly, in GDyn, firms own physical capital, but rent land and natural resources. Conversely, regional households own land and natural resources, which they lease to firms, and financial assets, which may be construed as indirect claims on physical capital. The next question is which classes of financial assets to represent in the model. In the real world there are three broad classes of financial assets – money, debt, and equity – which are divided in turn into many subclasses. Recognizing more asset classes would potentially improve the realism of the model. However, for reasons discussed earlier, realism in the representation of financial assets is not a priority for this model. In light of this, and consistent with our stance that the role of the financial asset module is to support international capital mobility rather than to depict the financial sector realistically, we include in the model just one asset class, equity. Accordingly, in GDyn, firms have no liabilities and only one asset, physical capital. By the fundamental balance sheet identity (assets = liabilities + proprietorship), shareholder equity in the firm is equal in value to the physical capital that the firm owns. Next we ask which agents can hold equity in firms. The simplest design would be to let all regional households hold equity in firms in all regions. This, however, would require bilateral data on foreign assets and liabilities. Unfortunately, the available data, especially the foreign direct investment data, are insufficient and internally inconsistent. To minimize the data requirements, we adopt instead the fiction of a global trust that serves as a financial intermediary for all foreign investment. In GDyn, regional households do not hold equity directly in foreign firms, but only in local firms and the global trust. In turn the global trust holds equity in firms in all regions. The trust has no liabilities and no assets other than its equity in regional firms. Therefore, by the balance sheet identity, total equity in the trust is equal in value to total equity held by the trust. A minor defect of this treatment is that it leads the model to misreport foreign asset holdings. We identify each region’s equity in the global trust with its foreign assets, when in fact some portion of it represents indirect ownership of local assets. This misreporting is trivial for small regions, but more considerable for large regions such as the United States. In terms of analysis this defect has minimal effect on the results of a simulation because regional investment is driven by rates of return and not affected

24

Elena I. Ianchovichina and Robert A. McDougall Region r

WQHTRUST(r)

WQHHLD(r) Global Trust WQHFIRM(r)

冱r WQHTRUST(r) = 冱r WQTFIRM(r)

WQ_FIRM(r)

WQTFIRM(r)

Figure 2.1. Wealth linkages.

by the treatment of ownership. However, the treatment will result in rental income from capital ownership being biased toward the rental obtained in the home country; the extent of this bias will depend on the share of the home region in the global trust. Figure 2.1 summarizes the financial asset framework. Firms in each region r have a value WQ _FIRM(r), of which the local regional household owns WQHFIRM(r) and the global trust WQTFIRM(r). In turn the global trust is owned by the regional households, each region r owning equity WQHTRUST(r). The total financial wealth of the regional household comprises equity WQHFIRM(r) in local firms and equity WQHTRUST(r) in the global trust. We discuss these relations further in subsections 4.3, 4.4, and 4.5. The concept of income from and investment in physical and financial assets remains to be discussed. We count as income the earnings of the asset, but not capital gains or losses arising from asset price changes. For physical capital, we also exclude physical depreciation from the definition of income, just as in the standard GTAP model. For equity in firms or in the global trust, we count as investment the money value of net change in the quantity of the entity’s assets, but exclude capital gains.


25

This treatment has two merits. First, it imposes consistency between income and investment in financial assets: Both exclude capital gains, so saving, calculated as total investment in financial assets, is consistent with income. Second, it supports a simple decomposition of change in proprietorship. Consider an entity that has no liabilities, but owns several assets. Let WAi denote the value of assets of type i, and W = i WAi is the total asset value. Then percentage change w in total asset value is given by the equation W = i WAi (p Ai + q Ai ), where p Ai denotes percentage change in the price of asset i, and q Ai denotes percentage change in the quantity. We can use this equation to decompose this change in total asset value into two components: the money value of net change in the quantity of the entity’s assets, (1/100) i WAi q Ai , and the money value of change in the prices of the quantity’s assets, (1/100) i WAi p Ai . Now by the balance sheet identity, total proprietorship in the firm is equal to total asset value W, so w = p Q + q Q , where pQ and qQ denote percentage change in the price and volume, respectively, of the firm’s stock. We can decompose this into an investment component, (1/100)Wq Q , and a our conventional definition capital gain component, (1/100)WpQ . Then, by of investment, Wq Q = i WAi q Ai , so Wp Q = i WAi p Ai (i.e., the price of equity in the firm is proportional to an index of prices of the firm’s assets). Thus, the price and quantity components of change in total proprietorship equate to the corresponding components of change in total assets. Another way to look at this is to imagine that firms and the trust fully distribute their net earnings as dividends to shareholders, and they fund their net asset purchases entirely through new stock issues. Under this supposition, the value of dividends coincides with the GDyn definition of income, and the value of stock issues coincides with the GDyn definition of financial investment.

4.2 Notation We use a systematic notational convention to present this accounting framework. Percentage change variables are written in lowercase, whereas uppercase variables are data coefficients, parameters, levels variables, or ordinary change variables. In general, the first character of a variable or a coefficient shows its type: W (wealth) for asset values, and Y for income flow. The second character identifies the asset type: In the current version of the model, this is always Q for eQuity. The third character indicates the sector that owns the asset or receives the income it generates, whereas the fourth character identifies the sector that owes the asset or pays the associated income. For

26


example, F designates investment in regional firms, T denotes investment in the global trust, and H stands for investment by the regional household. Thus, a name beginning with WQHF refers to the wealth in equity owned by the regional household and invested in domestic firms, whereas a name beginning with YQHF refers to the income from equity paid to the regional household by the domestic firms. An underscore is used in cases where the distinction pertaining to a particular character is not in point. The underscore is left out if it is located at the end of the name.

4.3 Asset Accumulation The financial assets module revolves around two key variables: the ownership value of firms in region r and the equity holdings of the household in region r. Both these equations are given, directly or indirectly, by accumulation relations. In GDyn, firms buy intermediate inputs, hire labor, and rent land, but own fixed capital. They have no debt. In accounting terms, they have no liabilities and no assets, except fixed capital. Conversely, only firms own fixed capital. Therefore the ownership value WQ_FIRM(r) of firms in region r is equal to the value of their fixed capital, which is the value of all local fixed capital, which in turn is equal to the product of the corresponding price and quantity: WQ FIRM(r) = VK (r) = PCGDS(r) × Q K (r),

(2.14)

where PCGDS(r) denotes the price of capital goods in region r. Differentiating, we obtain wq f(r) = pcgds(r) + qk(r),

(2.15)

where wq_f(r) denotes percentage change in WQ_FIRM(r), and pcgds(r), the percentage change in PCGDS(r). In the model, we write Equation REGEQYLCL #change in VK(r) [qk]# (all,r,REG) wq_f(r) = pcgds(r) + qk(r) + swq_f(r); where swq_f(r) is a region-specific shift factor.3 Thus the total equity value of each region’s firms is given indirectly by the capital accumulation equation (2.13). 3

The variable swq_f (r) is a shift factor used for modeling purposes. It is exogenous and equal to zero in the model.


27

For future use we note that, by the conventions discussed in subsection 4.1, the price PQ_FIRM(r) of equity in firms in region r is proportional to the price of capital goods in region r, pq f(r) = pcgds(r),

(2.16)

where pq_f denotes the percentage change in PQ_FIRM. As with capital stocks and investment, we use the variable time to capture the intrinsic dynamics of regional wealth and savings. For the regional household’s ownership of domestic assets, we have the accumulation equation: TIME QQHFIRM(r)dτ, (2.17) WQHFIRM(r) = PQ FIRM(r) TIME 0

where PQ_FIRM(r) is the price of stocks in local firms in region r, and QQHFIRM(r) is the number of stocks purchased by the regional household. Similarly, for the regional household’s equity in the global trust, we have TIME QQHTRUST(r)dτ, (2.18) WQHTRUST(r) = PQTRUST TIME 0

where PQTRUST is the price of equity in the global trust, and QQHTRUST(r) is the volume of equity purchases by the regional household. Then the total wealth of the regional household is given by equation (2.19): TIME QQHFIRM(r)dτ WQHHLD(r) = PCGDS(r) TIME 0

+ PQTRUST

TIME

QQHTRUST(r)dτ. (2.19) TIME 0

Differentiating, and substituting for pq_f from equation (2.16), we obtain WQHHLD(r) × wqh(r) = WQHFIRM(r) × pcgds(r) + WQHTRUST(r) × pqtrust + 100 × (VQHFIRM(r) + VQHTRUST(r)) × time,

(2.20)

where pqtrust denotes the percentage change in PQTRUST, and VQHFIRM(r) denotes the value of new investment by the regional household in domestic firms in region r, VQHFIRM(r) = PCGDS(r) × QQHFIRM(r).

(2.21)

28


VQHTRUST(r), the value of new investment by the regional household in the global trust, is as follows: VQHTRUST(r) = PQTRUST(r) × QQHTRUST(r).

(2.22)

Total investment by the regional household in domestic and foreign equity is equal to saving by the regional household; that is, VQHFIRM(r) + VQHTRUST(r) = SAVE(r), where SAVE(r) denotes savings in region r. So equation (2.20) simplifies to WQHHLD(r) × wqh(r) = WQHFIRM(r) × pcgds(r) + WQHTRUST(r) × pqtrust + 100 × SAVE(r) × time.

(2.23)

In the model code, we write as follows: Equation REGWLTH#change in wealth of the household [wqh(r)]# (all,r,REG) WQHHLD(r)∗ wqh(r) = WQHFIRM(r)∗ pcgds(r) + WQHTRUST(r) ∗ pqtrust + 100.0∗ SAVE(r)∗ time + WQHHLD(r)∗ swqh(r); where swqh(r) is a region-specific shift factor in the wealth of the region.

4.4 Assets and Liabilities of Firms and Households In subsection 4.3, we determined the value WQ_FIRM of equity in firms in each region. As shown in Figure 2.1, this equity has two components: equity belonging to the local regional household, WQHFIRM(r), and that belonging to the global trust, WQTFIRM(r): WQ FIRM(r) = WQHFIRM(r) + WQTFIRM(r).

(2.24)

Differentiating, we obtain WQ FIRM(r) × wq f(r) = WQHFIRM(r) × wghf(r) +WQTFIRM(r) × wqtf(r),

(2.25)

where wqhf(r) and wqtf(r) denote percentage changes in WQHFIRM(r) and WQTFIRM(r), respectively.


29

Equation (2.25) appears in the model as Equation EQYHOLDFNDLCL #total value of firms in region r# (all,r,REG) WQ_FIRM(r)∗ wq_f(r) = WQHFIRM(r)∗ wqhf(r) + WQTFIRM(r) ∗ wqtf(r). Also in subsection 4.3, we determined the wealth in equity of the regional households, WQHHLD. As shown in Figure 2.1, this also has two components – equity in domestic regional firms, WQHFIRM, and in the global trust, WQHTRUST: WQHHLD(r) = WQHFIRM(r) + WQHTRUST(r).

(2.26)

Differentiating, we obtain WQHHLD(r) · wqh(r) = WQHFIRM(r) · wqhf(r) +WQHTRUST(r) × wqht(r),

(2.27)

where wqhf(r), and wqht(r) denote percentage changes in WQHFIRM(r) and WQHTRUST(r), respectively. This equation appears in the model as Equation EQYHOLDWLTH #total wealth of the household# (all,r,REG) WQHHLD(r)∗ wqh(r) = WQHFIRM(r)∗ wqhf(r) + WQHTRUST(r) ∗ wqht(r). Thus far, for each region r we have two accounting identities, equations (2.24) and (2.26), and three variables to determine: WQHFIRM, WQTFIRM, and WQHTRUST. Equivalently, for each region the identities suffice to determine the net value of foreign assets, WQHTRUST(r) − WQTFIRM(r) = WQHHLD(r) − WQ FIRM(r) (2.28) but not gross foreign assets and liabilities: WQHTRUST(r) and WQTFIRM(r). Obviously many different gross foreign asset positions are consistent with the net position. In this model, we do not make use of portfolio allocation theory, so we have no theory explaining the gross ownership position. Over the long-run, rates of return on capital are equalized across regions. With no portfolio allocation theory, investors care only about returns, so with returns equalized the allocation of assets is arbitrary. Over the short-run, we allow interregional differences in rates of return (subsection 5.1). We need investors to hold several assets (because net foreign ownership positions must be nonzero), but we have no theory explaining why investors would hold any assets other

30


than the highest yielding one. Accordingly, we can determine portfolio allocation over the short- or long-run only by applying a heuristic rule. In selecting a portfolio rule, we have some constraints to guide us. First and most obviously, the three variables WQHFIRM(r), WQHTRUST(r), and WQTFIRM(r) must satisfy the two identities (2.24) and (2.26). Furthermore, we want to obtain positive values for those three variables. This is possible, provided that WQHHLD(r) and WQ_FIRM(r) are positive. Although it is possible in the real world to short-sell stocks, we do not observe large, long-lasting negative equity holdings. If we nevertheless allowed negative holdings in the model, they would be liable to generate strange welfare results. If, for example, we allowed the global trust to hold negative equity in Taiwan, then the income of the trust and, consequently, the foreign asset income of each region would vary not directly, but inversely with Taiwanese capital rentals. Given the real-world absence of stable negative equity holdings, this inverse relationship would be unrealistic. Finally, we want the allocation rule to preserve as nearly as possible the initial allocation of each region’s wealth between domestic and foreign assets. One of the objectives of the asset treatment is to allow the model to respect the empirical regularity that regions tend to specialize in their own domestic assets. If the initial database respects this, we want updated databases to respect it too. One possible approach is to assume that each region allocated its wealth between domestic and foreign assets in fixed proportions. This assumption is simple and in some ways appealing, but it has one defect: It makes it too easy for foreign liabilities to become negative. For example, a negative shock to productivity in Taiwan might cause the value of capital located in Taiwan to fall more rapidly than the value of equity owned by the Taiwanese. With the fixed shares approach, the value of domestic equity owned by the Taiwanese might easily come to exceed the value of the Taiwanese capital stock, so that the value of foreign ownership of Taiwanese industry would become negative. As discussed earlier, we wish to avoid such outcomes. If, conversely, we assumed that the composition of the source of funds were fixed in each region, so that foreign and domestic equity in local capital varied in fixed proportion, we would be assured that foreign ownership of local capital would not turn negative. However, growth in the local capital stock might easily lead to negative local ownership of foreign assets. To avoid negative values in both gross foreign assets and gross foreign liabilities, we need a more sophisticated approach. We find this in entropy theory. In particular, cross-entropy minimization gives us a way of dividing a strictly positive total into strictly positive components, subject to various constraints, while staying as close as possible to the initial shares. A full


31

exposition of the relevant concepts would take us too far afield here. For example, Kapur and Kesavan (1992) present a modern treatment emphasizing aspects of interest to economists. Cross entropy is an indicator of the degree of divergence between two partitions Si , i = 1, . . . , n of a total value. Writing Si (0) for the initial shares and Si (1) for the final shares, the cross entropy is

S i (1) log

i

S i (1) . S i (0)

(2.29)

This takes a minimum when, for all i, Si (1) = Si (0); that is, when the final shares are equal to the initial shares (Kapur and Kesavan 1992). The advantages of the cross-entropy approach become apparent when we impose constraints on the final shares. For a wide variety of constraints, the constrained optimization problem leads to a simple and transparent set of first-order conditions. In addition, with strictly positive initial shares, we are guaranteed, constraints permitting, strictly positive final shares. We are concerned with two sets of shares: the shares of domestic and foreign equity in domestic wealth and the shares of domestic and foreign funds in ownership of local capital. With each of these we associate a crossentropy measure. For shares in domestic wealth in region r, the cross entropy is CEHHLD(r) = WQHFIRMSH(r) · log

WQHFIRMSH(r) WQHFIRMSH 0(r)

+WQHTRUSTSH(r) · log

WQHTRUSTSH(r) , WQHTRUSTSH 0(r)

(2.30)

where WQHFIRMSH(r) denotes the current share of local firms; WQHTRUSTSH(r), the current share of the global trust in the equity portfolio of the household in region r; and WQHFIRMSH_0(r) and WQHTRUSTSH_0(r) denote the initial levels of those shares. By definition, we have WQHFIRM(r) =

WQHFIRM(r) , WQHHLD(r)

WQHFIRMSH 0(r) =

WQHFIRM 0(r) , WQHHLD 0(r)

WQHTRUSTSH(r) =

WQHTRUST(r) , WQHHLD(r)

WQHTRUSTSH 0(r) =

WQHTRUST 0(r) . WQHHLD 0(r)

(2.31)

32


Substituting these into equation (2.30), we obtain WQHFIRM(r) WQHFIRM 0(r) WQHTRUST(r) + WQHTRUST(r) × log WQHTRUST 0(r) WQHHLD(r) −WQHHLD(r) × log . WQHHLD 0(r) (2.32)

WQHHLD(r) × CEHHLD(r) = WQHFIRM(r) × log

Because WQHHLD(r) and WQHHLD_0(r) are given, maximizing CEHHLD(r) is equivalent to maximizing FHHLD(r) = CEHHLD(r) + WQHHLD(r) × log

WQHHLD(r) . WQHHLD 0(r) (2.33)

Then, WQHHLD(r) × FHHLD(r) = WQHFIRM(r) WQHFIRM(r) + WQHTRUST(r) × log WQHFIRM 0(r) WQHTRUST(r) × log . (2.34) WQHTRUST 0(r) Similarly, maximizing the cross entropy associated with the local capital ownership shares is equivalent to maximizing FFIRM(r), where WQ FIRM(r) × FFIRM(r) = WQHFIRM(r) WQHFIRM(r) + WQTFIRM(r) × log WQHFIRM 0(r) × log

WQTFIRM(r) . WQTFIRM 0(r)

(2.35)

We seek to minimize a weighted sum of the two cross entropies: WSCE(r) = RIGWQH(r) × WQHHLD(r) × CEHHLD(r) + RIGWQ F(r) × WQ FIRM(r) × CEFIRM(r).

(2.36)

The two cross entropies are weighted by the corresponding total values, WQHHLD(r) and WQ_FIRM(r), and explicitly by the rigidity parameters RIGWQH(r) and RIGWQ_f(r). If RIGWQH(r) is assigned a high value and RIGWQ_f(r) a low one, then the solution will, if possible, keep the allocation of household wealth nearly fixed and put most of the onus of adjustment


33

on the source shares for equity in local firms. If RIGWQ_F(r) is assigned a high value and RIGWQH(r) a low one, the equity source shares will tend to remain near their initial values, and the household wealth allocation shares do most of the adjusting. From the foregoing, minimizing WSCE is equivalent to minimizing the somewhat simpler equation: F = RIGWQH(r) × WQHHLD(r) × FHHLD(r) + RIGWQ F(r) × WQ FIRM(r) × FFIRM(r) ⎛ ⎞ WQHFIRM(r) ⎜WQHFIRM(r) × log WQHFIRM 0(r) ⎟ ⎜ ⎟ ⎜ = RIGWQH(r) × ⎜ WQHTRUST(r) ⎟ ⎟ ⎝+WQHTRUST(r) × log WQHTRUST 0(r) ⎠ ⎞ ⎛ WQHFIRM(r) ⎜WQHFIRM(r) × log WQHFIRM 0(r) ⎟ ⎟ ⎜ ⎟ +RIGWQ F(r) × ⎜ ⎜+ WQTFIRM(r) × log WQTFIRM(r) ⎟ . (2.37) ⎝ WQTFIRM 0(r) ⎠ To determine the three wealth variables, we minimize this objective function subject to the constraints (2.26) and (2.24). The Lagrangean contains corresponding multipliers: XWQHHLD(r) for the household wealth constraint (2.26) and XWQ_FIRM(r) for the firm value constraint (2.24). The first-order conditions include the two constraints and three equations corresponding to the three net wealth variables. Thus, differentiating the Lagrangean with respect to foreign equity in domestic capital, WQTFIRM(r), we obtain the first-order condition WQTFIRM(r) + 1 . (2.38) XWQ FIRM(r) = RIGWQ F(r) × log WQTFIRM 0(r) Differentiating again, we obtain xwq f(r) = RIGWQ F(r) × wqtf(r),

(2.39)

where xwq_f(r) denotes change in the Lagrange multiplier XWQ_FIRM(r). In TABLO code, we have Equation EQYHOLDFNDHHD #equity holdings of trust in the firms [wqtf(r)]# (all,r,REG) xwq_f(r) = RIGWQ_F(r)∗ wqtf(r).

34


Likewise, for domestic ownership of foreign equity, we have the levels form of the first-order condition, WQHTRUST(r) +1 (2.40) XWQHHLD(r) = RIGWQH(r) × log WQHTRUST 0(r) and the differential form of the first-order condition, xwqh(r) = RIGWQH(r) × wqht(r) + swqht(r),

(2.41)

where xwqh(r) denotes change in the Lagrange multiplier XWQHHLD(r), and swqht(r) is a region-specific shift variable. In TABLO code we have Equation EQYHOLDHHDFND #equity holdings of households in trust[xwqh(r)]#(all,r,REG) xwqh(r) = RIGWQH(r)∗ wqht(r) + swqht(r). Finally, for domestic ownership of domestic equity, we have the levels form of the first-order condition, XWQHHLD(r) + XWQ FIRM(r) = (RIGWQH(r) + RIGWQ F(r)) WQHFRIM(r) +1 , × log WQHFIRM 0(r) (2.42) the differential form of the first-order condition, xwqh(r) + xwq f(r) = (RIGWQH(r) + RIGWQ F(r)) × wqhf(r), (2.43) and the TABLO code specification, Equation EQYHOLDHHDLCL #shift variable wealth of firms [xwq_f(r)]# (all,r,REG) [RIGWQH(r) + RIGWQ_F(r)]∗ wqhf(r) = xwqh(r) + xwq_f(r) + swqhf(r). where swqhf(r) is a region-specific shift variable. Note that, substituting for wqtf from equation (2.39), and for wqht from equation (2.41) into equation (2.43), we obtain (RIGWQH(r) + RIGWQ F(r)) × wqhf(r) = RIGWQH(r) × wqht(r) + RIGWQ F(r) × wqtf(r). (2.44)


35

This equation shows that the adjustment in WQHFIRM(r) is an average of the adjustments in WQTFIRM(r) and WQHTRUST(r). Note also that if, for example, we assign a high value to RIGWQH(r) and a low value to RIGWQ_F(r), then xwqh(r) will assume a relatively large value and xwq_f(r) a relatively small value, so that xwqh(r) ≈ RIGWQH(r) · wqhf(r), and wqhf(r) ≈ wqht(r) = RIGWQH(r) − 1 xwqh(r); that is, the household wealth allocation shares are nearly fixed, as previously asserted and discussed further in Chapter 3.

4.5 Assets and Liabilities of the Global Trust There are three accounting identities associated with the global trust. First, the value of assets owned by the global trust, WQTRUST, is equal to the sum across regions of foreign ownership of firms: WQTFIRM(r). (2.45) WQTRUST = r

In percentage change form, we have WQTFIRM(r) × wqtf(r), WQTRUST × wqt =

(2.46)

r

where wqt is the percentage change in WQTRUST, and in the TABLO code, Equation TOTGFNDASSETS #change in the value of assets owned by global trust# WQTRUST∗ wqt = sum{s, REG, WQTFIRM(s)∗ wqtf(s)}. The second identity is that the value of the trust, WQ_TRUST, is equal to the sum of the regions’ equity in the trust; that is, to the sum across regions of ownership of foreign assets: WQHTRUST(r). (2.47) WQ TRUST = r

In percentage change form, equation (2.47) transforms into WQ TRUST × wq t = WQHTRUST(r) × wqht(r),

(2.48)

r

where wq_t is the percentage change in WQ_TRUST and in the TABLO code is given as

36


Equation TOTGFNDPROP #value of trust as total ownership of trust# wq_t = sum{s, REG, [WQHTRUST(r)/WQ_TRUST]∗ wqht(s)}. Finally, the total value of the trust is equal to the total value of its assets: WQ TRUST = WQTRUST.

(2.49)

This equation as written would be redundant in the model, because it is implicit in other relations. The accumulation equations, together with the equivalence of global investment and global saving, ensure that the total value of physical capital is always equal to the total value of financial asset ownership by regions: so WQ FIRM(r) = WQHHLD(r). (2.50) r

r

Then WQ TRUST =

WQHTRUST(r)

r

=

WQHHLD(r) − WQHFIRM(r)

r

=

WQHHLD(r) −

r

=

WQHFIRM(r)

r

WQ FIRM(r) −

r

=

WQHFIRM(r)

r

WQ FIRM(r) − WQHFIRM(r)

r

=

WQTFIRM(r)

r

= WQTRUST.

(2.51)

To verify that simulation results satisfy the identity, we include the following equation: WQTRUST = WTRUSTSLACK × WQ TRUST,

(2.52)

where WTRUSTSLACK denotes an endogenous slack variable. In percentage change form, wqt = wq t + wtrustslack,

(2.53)

where wtrustslack denotes percentage change in WTRUSTSLACK. In the TABLO code, we have


37

Equation GLOB_BLNC_SHEET # ownership by the trust equals ownership for the trust # wqt = wq_t + wtrustslack. Provided that the model database respects the asset accounting identities, and assuming no errors in the equations, the variable wtrustslack is endogenously equal to zero in any simulation. Thus the result for the slack variable provides a check on the validity of the model. Figure 2.1 illustrates these accounting relations. Corresponding to equation (2.46) for asset values, we have a price equation. As discussed in subsection 4.1, we can divide growth in assets and in proprietorship into matching investment and capital gain components. For the global trust, equating the capital gain components of assets and proprietorship yields the equation WQTFIRM(r) pcgds(r). (2.54) pqtrust = WQTRUST r In the code, this becomes Equation PKWRLD #change in the price of equity in the global fund# WQTRUST∗ pqtrust = sum{r, REG, WQTFIRM(r)∗ pcgds(r)}.

4.6 Income from Financial Assets Having determined stocks of financial assets in the previous subsections, we now determine the associated income flows. We do so in three stages. First, we determine payments from firms to households and to the global trust. Second, we calculate the total income of the global trust and determine payments from the trust to regional households. Third, we calculate the equity income of regional households as the sum of receipts from local firms and from the global trust. For an overview of the equity income flows, we refer to Figure 2.2. Firms in region r distribute to shareholders equity income payments YQ_FIRM(r), of which YQHFIRM(r) goes to the local regional household and YQTFIRM(r) to the global trust. Summing these receipts YQTFIRM(r) across regions, we obtain the total income YQTRUST of the global trust. The trust distributes this total income among the regional households, with region r receiving an amount YQHTRUST(r). Thus the total equity income of region r, YQHHLD(r), is the sum of receipts YQHFIRM(r) from local

38

Elena I. Ianchovichina and Robert A. McDougall INCOME(“r1”)

INCOME(“r2”)

YQHHLD(“r1”)

YQHHLD(“r2”)

YQHTRUST(“r1”)

YQHTRUST(“r2”)

YQTRUST

YQHFIRM(“r1”)

YQTFIRM(“r1”)

YQHFIRM(“r2”)

YQTFIRM(“r2”)

YQ_FIRM(“r2”)

YQ_FIRM(“r1”)

Figure 2.2. Income linkages.

firms and receipts YQHTRUST(r) from the global trust. This is summed with nonequity factor income and indirect taxes, which yields total regional income INCOME(r). We begin the detailed discussion with payments by firms. Firms buy intermediate inputs, hire labor, and rent land, but own fixed capital. By the zero pure profits condition, their profits are equal to the cost of capital services, excluding any factor usage or income taxes, less depreciation. These profits accrue to shareholders. Thus total income payments by firms in region r to shareholders, YQ_FIRM(r), are equal to net after-tax capital earnings: YQ FIRM(r) = VOA(“capital”, r) − VDEP(r),

(2.55)

where VOA(“capital”, r) is the value of capital earnings, and VDEP(r) is the value of capital depreciation. Differentiating, we obtain YQ FIRM(r) × yq f(r) = VOA(“capital”, r) × (rental(r) + qk(r)) −VDEP(r) × (pcgds(r) + qk(r)),

(2.56)

where yq_f(r) denotes the percentage change in income payments by firms in region r, and rental(r) denotes the percentage change in the rental price


39

of capital. In the code, this equation is written as Equation REGINCEQY #income from capital in firms in region r# (all,r,REG) YQ_FIRM(r)∗ yq_f(r) = sum{h, ENDWC_COMM, VOA(h,r)∗ [ps(h,r) + qo(h,r)]} − VDEP(r)∗ [pcgds(r) + qk(r)]. To relate the equation in the TABLO code to its mathematical form, note that ENDWC_COMM is a set with just one element: “capital”, with ps(“capital”, r) = rental(r) and qo(“capital”, r) = qk(r). Firms distribute payments among shareholders in proportion to their shareholdings. The local regional household owns WQHFIRM, and the global trust owns WQTFIRM with a total equity value WQ_FIRM (see subsection 4.4). So for payments YQHFIRM(r) to the local regional household, we have WQHFIRM(r) YQHFIRM(r) = × YQ FIRM(r). (2.57) WQ FIRM(r) Differentiating, we obtain yqhf(r) = yq f(r) + wqhf(r) − wq f(r),

(2.58)

where yqhf(r) denotes the percentage change in YQHFIRM(r). In the TABLO code this equation is given as Equation INCHHDLCLEQY #income of the household from local firms]# (all,r,REG) yqhf(r) = yq_f(r) + wqhf(r) − wq_f(r). Similarly, payments to the global trust, YQTFIRM(r), are given by YQTFIRM(r) =

WQTFIRM(r) × YQ FIRM(r). WQ FIRM(r)

(2.59)

Differentiating, we obtain yqtf(r) = yq f(r) + wqtf(r) − wq f (r),

(2.60)

where yqtf(r) is the percentage change in YQTFIRM(r). In the TABLO code, we write Equation INCFNDLCLEQY #income of trust from equity in region r# (all,r,REG) yqtf(r) = yq_f(r) + wqtf(r) − wq_f(r).

40


In the second stage, we compute total income receipts and the various income payments of the global trust. The total income of the trust, YQTRUST, is equal to the sum of equity receipts from firms in each region. In levels, we express this as YQTFIRM(r), (2.61) YQTRUST = r

in percentage changes as yqt =

YQTFIRM(r) r

YQTRUST

× yqtf(r),

(2.62)

where yqt denotes the percentage change in YQTRUST, and in the TABLO code as Equation INCFNDEQY #change in the income of the trust# yqt = sum{r, REG, [YQTFIRM(r)/YQTRUST]∗ yqtf(r)}. The trust distributes its income among its shareholders, so that each region r receives income YQHTRUST(r) in proportion to its ownership share. This is expressed in the levels equation, YQHTRUST(r) =

WQHTRUST(r) × YQTRUST, WQ TRUST

(2.63)

the differential equation, yqht(r) = yqt + wqht(r) − wq t,

(2.64)

where yqht(r) denotes the percentage change in YQHTRUST, and in the TABLO code as Equation REGGLBANK #income of household r from its shares in the trust# (all,r,REG) yqht(r) = yqt + wqht(r) − wq_t. In the third and final stage we compute the financial asset income of regional households. Total equity income YQHHLD(r) of regional household r equals the sum of equity income received from domestic firms and from the global trust: YQHHLD(r) = YQHFIRM(r) + YQHTRUST(r).

(2.65)


41

In percentage changes, this equation transforms into yqh(r) =

YQHFIRM(r) YQHTRUST(r) × yqhf(r) + × yqht(r), YQHHLD(r) YQHHLD(r)

(2.66)

where yqh(r) denotes percentage change in YQHHLD. In the TABLO code, it is given as Equation TOTINCEQY #total income from equity of households in r#(all,r,REG) yqh(r) = [YQHFIRM(r)/YQHHLD(r)]∗ yqhf(r) + YQHTRUST(r)/ YQHHLD(r)]∗ yqht(r).

5. Investment Theory In this section we describe a lagged adjustment, adaptive expectations theory of investment. Investors act so as to eliminate disparities in expected rates of return not instantaneously, but progressively through time. Moreover, their expectations of rates of return may be in error, and these errors are also corrected progressively through time. Finally, in estimating future rates of return, they allow for some normal rate of growth in the capital stock; this normal rate too is an estimated rate that investors adjust through time.

5.1 The Required Rate of Growth in the Rate of Return In a simple perfect adjustment model of investment, profit-maximizing investors would keep rates of return uniform across regions, because any differences in rates of return would be immediately eliminated by a reallocation of capital from regions with lower rates of return to regions with higher rates. This equalization would apply to net rates of return, so that we might write, for each region r, RORNET(r) = RORCOMM, where RORNET(r) denotes the net rate of return on capital in region r, and RORCOMM the common world rate of return. If we allow for region-specific risk premia RRISK(r), then we postulate equalization not of the actual net rates of return RORNET(r) but of the risk-adjusted rates RORNET(r) − RRISK(r), so that, for all regions r, RORNET(r) = RORCOMM + RRISK(r). Furthermore, as we find later, it is convenient to express the investment theory in terms of gross rather than net rates of return; anticipating this, we write RDEP(r) for the depreciation

42


rate in region r and obtain, for the gross rate of return the equilibrium condition, RORGROSS(r) − RORCOMM − RRISK(r) − RDEP(r) = 0.

(2.67)

In principle, the gross rate of return RORGROSS(r) includes both an earnings component and a capital gains component: RORGROSS(r) =

RENTAL(r) + RG PCGDS(r), PCGDS(r)

(2.68)

where RENTAL(r) denotes the rental price of capital in region r, and RG_PCGDS(r) denotes the rate of growth in the purchase price of capital. In practice, with a period-by-period solution method, we do not know the rate of growth in the purchase price of capital,4 so we neglect it and define the gross rate of return as the earnings rate only: RORGROSS(r) =

RENTAL(r) . PCGDS(r)

(2.69)

Differentiating equation (2.69), we obtain the percentage change equation: rorga(r) = rental(r) − pcgds(r),

(2.70)

where rorga(r) denotes percentage change in RORGROSS. In the TABLO code, this equation is represented as follows: Equation RATERETURNP #identity for rate of return# (all,r,REG) rorga(r) = rental(r) − pcgds(r). We now consider the investment response to sudden (i.e., instantaneous) price changes. For example, sudden price changes may occur as the result of sudden tax rate changes. Sudden changes in output or input prices affect the capital rental price, RENTAL(r), and thereby the rate of return. In a perfect adjustment model with capital gains, these changes in the rate of return must be offset by some sudden change in PCGDS or RG_PCGDS, or by some sudden offsetting influence on RENTAL, so as to maintain international equality in rates of return as defined in equation (2.68). 4

In fact, we can estimate the backward-looking growth rate, limH->0- (PCGDS(r,T + H)PCGDS(r,T ))/H, where PCGDS(r,T ) denotes the value of PCGDS(r) at time T. This, however, is liable to differ from the forward-looking growth rate, limH->0 + (PCGDS(r,T + H)PCGDS(r,T ))/H, which is the one needed in the rate-of-return formula.


43

Suppose initially that the supply of capital goods is perfectly elastic. Then a first-round improvement in profitability (i.e., a first-round positive effect on RENTAL) leads to an increase in the capital stock – increasing output supply and possibly increasing demand for noncapital inputs – and thereby negating the first-round effect on RENTAL. If the initial shock is sudden, then so also must be the increase in the capital stock, implying an infinite rate of investment over an infinitesimal time period. Of course, in the real world capital stocks do not adjust in this manner. Instantaneous adjustment of capital stocks is precluded by gestation lags, adjustment costs, imperfect elasticity of the supply of capital, and other factors. In addition, in a CGE model, even if other realistic features are lacking, the supply of capital is typically imperfectly elastic. If we rule out infinite rates of investment, how can rate-of-return equalization be maintained in the face of sudden shocks affecting profitability? The answer is through sudden changes in the price of capital goods. A sudden improvement in earnings leads to a sudden increase in demand for capital goods, and that in turn leads to a sudden increase in the price of capital goods. This price increase helps stabilize the rate of return in two ways. First, it reduces the earnings rate RENTAL(r)/PCGDS(r). Second, it leads to a decrease in the rate of capital gain, RG_PCGDS. As demand for capital goods eases through time after the initial spike, or the supply of capital goods gradually rises, the price of capital goods tends to fall through time after its initial increase. In our model, we cannot capture the capital gains effect of an increase in demand for capital goods, but we can capture the earnings rate effect. Thus the way appears open in principle to use a perfect adjustment mechanism for investment. However, because we do not capture all the effects of the increase in demand for capital, it is likely that the model will require unrealistically large increases in the price of capital goods and in the level of investment. Indeed, there are several reasons why the model would tend to exaggerate investment volatility, some of which have already been mentioned and some not: r The model does not capture the capital gain effect of capital goods

price changes.

r As we typically use it in dynamic simulations, the model assumes

perfect capital mobility within regions. Accordingly, it overstates the elasticity of the supply of capital goods. r The model does not incorporate other real-world effects such as gestation lags or adjustment costs.

44


For all these reasons, the perfect adjustment approach is unrealistic in the context of this model. We pursue accordingly a lagged adjustment approach. Recalling equation (2.67), we rewrite it as RORGROSS(r) − RORGTARG(r) = 0,

(2.71)

where RORGTARG(r) denotes the target rate of return in region r. To move to a lagged adjustment approach, we replace this in turn by RRG RORG(r) = LAMBRORG(r)∗ log

RORGTARG(r) , RORGROSS(r)

(2.72)

where RRG_RORG(r) denotes the required rate of growth in the rate of return, and LAMBRORG(r) denotes a coefficient of adjustment. Differentiating, we obtain rrg rorg(r) = LAMBRORG(r)∗ [rorgt(r) − rorga(r)],

(2.73)

where rrg_rorg(r) denotes (absolute) change in the required rate of growth in the rate of return in region r, and rorgt(r) denotes the percentage change in the target rate of return. Note that this is not the final form of the equation. We present it in subsection 5.3, following further theoretical development. Referring back to equation (2.67), we note that the target rate of return includes both region-specific components RRISK(r) and RDEP(r) and a region-generic component RORCOMM. In the present context there is another possible region-generic component, a worldwide drift in rates of return such as to accommodate the global level of investment. We do not represent all these components explicitly in the model, but instead write simply RORGTARG(r) = SDRORTWORLD + SDRORTARG(r),

(2.74)

where SDRORTWORLD denotes a region-generic component in the target rate of return, and SDRORTARG(r) a component specific to region r. Differentiating, we obtain DRORT(r) = SDRORTW + SDRORT(r),

(2.75)

where DRORT(r) denotes the absolute change in the target rate of return; SDRORTW, a region-generic shift; and SDRORT(r), a region-specific shift. We use here the absolute rather than the percentage change form for the target rate, to ensure that any worldwide shift SDRORTW leads to equal percentage-point changes in rates of return in different regions; equivalently, to ensure that any cross-region differentials are maintained in percentagepoint rather than percentage terms (so, for example, we might maintain a


45

risk premium of two percentage points, but not a risk premium equivalent to 20% of the rate of return). We have then DRORT(r) = SDRORTW + SDRORT(r),

(2.76)

or in TABLO code, Equation NET_ROR #equilibrium condition for rate of return# (all,r,REG) DRORT(r) = SDRORTW + SDRORT(r). We relate the absolute-change variable DRORT to the percentage-change variable rorgt with the equation RORGTARG(r) × rorgt(r) = DRORT(r),

(2.77)

and in the code, Equation GROSS_ROR #identity for target gross rate of return# (all,r,REG) RORGTARG(r)∗ rorgt(r) = DRORT(r).

5.2 The Expected Rate of Growth in the Rate of Return Having determined in subsection 5.1 the required rate of growth in the rate of return, we now relate it to the level of investment through an equation linking the expected rate of growth in the rate of return to investment, with a condition that the expected rate should be equal to the required rate. This brings us to one of the central elements of the investment theory in GDyn, the expected rate-of-return schedule. Investors understand that, the higher the level of the capital stock at any given time, the lower the rate of return at that time. Accordingly, the rate of return expected to prevail at any future time depends on the level of the capital stock at that time. Consequently, the expected rate of growth in the rate of return depends on the rate of growth in the capital stock or, equivalently, on the level of investment. We describe investors’ understanding of the investment environment through a rate-of-return schedule, relating the expected rate of return to the size of the capital stock: QK(r) −RORGFLEX(r) RORGEXP(r) = , (2.78) RORGREF(r) QKF(r) where RORGEXP(r) denotes the expected gross rate of return, and RORGFLEX(r) denotes a positive parameter, representing the absolute magnitude of the elasticity of the expected rate of return with respect to the size

46


of the capital stock. RORGREF(r) is a reference rate of return in region r, and QKF(r), a reference capital stock. Investors expect that if the actual capital stock QK is equal to the reference stock QKF, then the rate of return will be equal to the reference rate RORGREF. If the capital stock exceeds the reference stock, the expected rate of return is less than the reference rate. If the capital stock is less than the reference stock, the expected rate is greater than the reference rate. In dealing with expectations, as in equation (2.78), there are two relevant times: the time at which the expectations are held and the time to which they refer. We call these, respectively, the expectation time and the realization time. So for example, in describing an investor in 2000 holding an expectation about the rate of return in 2005, the expectation time is 2000, and the realization time is 2005. In the theory underlying the investment module, expectation time is always just the current time, TIME, for the model. For example, if the model represents the state of the world economy in the year 2000, then the expectation time is 2000. However, realization time, TREAL, may be either the current or some future time. In the model itself, as opposed to the underlying theory, expectation time and realization time are always equal to the current time. So in the model equations TREAL would be redundant, and we use only the current time, TIME. To complete our description of investor expectations in equation (2.78), we need to specify how the reference rate of return and the reference capital stock depend on realization time. We postulate that the reference rate of return is independent of realization time, whereas the reference capital stock grows at some normal rate KHAT(r): QKF(r) = QKO(r)e KHAT(r) TREAL ,

(2.79)

where QKO(r) denotes the reference capital stock at some base time TREAL = 0. Under this treatment, the normal rate of growth KHAT(r) is the rate at which the capital stock can grow without (as investors expect) affecting the rate of return. If the capital stock grows at a rate greater than KHAT(r), investors expect rates of return to decline through time. If the capital stock grows at less than KHAT(r), investors expect rates of return to increase. The specification of expectations in equations (2.78) and (2.79), although simple, is intended to approximate the actual investment schedule. In particular, it allows a range between zero and infinity for the gross rate of return, RORGROSS(r). This allows, realistically, the net rate of return to be negative sometimes. Whether the specification is locally model consistent depends on the setting of the normal growth rate KHAT(r) and the elasticity


47

RORGFLEX(r). As discussed in subsection 5.4, we allow model-consistent adjustment of KHAT(r), but RORGFLEX(r) is fixed. We can set it initially at a locally model-consistent value, but through a simulation or series of projections, it typically becomes more or less inconsistent. This is undesirable, but also unavoidable without a considerable increase in the complexity of the theory. To find the expected rate of growth in the rate of return, we differentiate equation (2.78) with respect to realization time. Substituting for QKF(r) from equation (2.79), we obtain ERG RORG(r) = −RORGFLEX(r) × (RG Q K (r) − KHAT(r)),

(2.80)

where ERG_RORG(r) denotes the expected rate of growth in the rate of return in region r, and RG_QK(r), the rate of growth in the capital stock. The rate of growth in the capital stock is RG Q K (r) =

QCGDS(r) − RDEP(r)Q K (r) dQK(r)/dTIME = Q K (r) Q K (r) QCGDS(r) − RDEP(r), (2.81) = Q K (r)

where QCGDS(r) denotes the level of investment in region r. Substituting the expression for RG_QK(r) into equation (2.80), we obtain ERG RORG(r) = −RORGFLEX(r) QCGDS(r) − RDEP(r) − KHAT(r) . × Q K (r)

(2.82)

Totally differentiating equation (2.82), we obtain erg rorg(r) = −RORGFLEX(r) × [IKRATIO(r) ×(qcgds(r) − qk(r)) − DKHAT(r)],

(2.83)

where erg_rorg(r) denotes the absolute change in the expected rate of growth in the rate of return in region r; IKRATIO(r), the ratio QCGDS(r)/QK(r) of gross investment to the capital stock; qcgds(r), the percentage change in gross investment; and DKHAT(r), the absolute change in the normal rate of growth in the capital stock. Then, in the TABLO code we have Equation EGROWTH_ROR #rule for expected rate of growth in rate of return# (all,r,REG) erg_rorg(r) = − RORGFLEX(r)∗ IKRATIO(r)∗ [qcgds(r) − qk(r)] + RORGFLEX(r)∗ DKHAT(R).

48

Elena I. Ianchovichina and Robert A. McDougall R

RORGROSS RORGTARG RORGREF

(A)

QK

QKT QKF

QK

Figure 2.3. Actual investment schedule.

As equation (2.83) shows, the expected rate of growth in the rate of return varies inversely with the level of investment. The level of investment is given implicitly by the condition that the expected rate of growth is equal to the required rate: ERG RORG(r) = RRG RORG(r).

(2.84)

We depict some aspects of the investment theory in Figure 2.3. Each point in the figure represents a (capital stock, rate-of-return) pair (QK, R). The curve (A) represents the expected rate-of-return schedule for realization time equal to expectation time. It is downward sloping, with the slope related to the elasticity RORGFLEX, a vertical asymptote at QK = 0, and a horizontal asymptote at R = 0. It passes through the point (QK, RORGROSS) representing the current capital stock and rate of return, and also through the reference point (QKF, RORGREF). For a realization time greater than the expectation time, the curve would be similar to (A), but dilated about the vertical axis. Assuming a positive normal growth rate, the curve would dilate rightward as realization time advanced. As drawn in Figure 2.3, the actual rate of return RORGROSS exceeds the target rate RORGTARG. From equation (2.72), this implies that the required rate of growth in the rate of return is negative. On the diagram in Figure 2.3, this implies some required downward vertical speed. By inverting equation (2.82) and equating the required rate of growth in the rate of return to the expected rate, we find the required level of investment: QCGDS(r) = Q K (r) × [RDEP(r) + KHAT(r) − (RORGFLEX(r))−1 × RRG RORG(r)].

(2.85)

Here QK(r)×RDEP(r) is the investment level required to maintain the capital stock QK(r) at its current level, QK(r)×KHAT(r) is the further


49

investment required to keep pace with the rightward dilation of the rate of return curve (A), and QK(r) × (RORGFLEX(r)) − 1 × RRG_RORG(r) is the further investment required to maintain the required vertical speed down the curve.

5.3 Adaptive Expectations In practice, the investment theory as presented to this point in equations (2.73), (2.83), and (2.84) has a significant disadvantage. Using information in the benchmark data, we can calculate the actual rate of return, RORGROSS(r), in the initial year. The rate of return and the equations of the model allow us to determine the level of investment QCGDS(r). However, the benchmark data also specify the level of investment. In general this level will be inconsistent with the level calculated with the theory. Consider, for example, the region with the highest rate of return in the database. In this region the actual rate of return exceeds the target rate, so the required rate of growth in the rate of return RRG_RORG(r) is negative. This, in turn, implies that the normal rate of growth in the capital stock KHAT(r) and investment QCGDS(r) should be high. However, it may be that the level of investment recorded in the database is not particularly high. In this case, theory and data are inconsistent. We can resolve this inconsistency by modifying either the data or the theory. One approach to modifying the data involves equalizing rates of return across countries in the database. This conflicts with one of our objectives for the dynamic model, specifically that it should work with databases that conform closely to observed statistics, rather than requiring a heavily recalibrated or stylized database. Another approach would be to account for investment-level anomalies through risk premia. We can do this readily in the database, without touching the flows data, by adjusting the target rates. This option is sometimes appealing. We do not wish, however, to force it on users. Rather we recognize that reality is under no obligation to respect our or any other investment theory and that, for a multitude of reasons, observed investment levels will surely differ from any theoretical prediction. We therefore extend the theory so that it does not prescribe investment levels, but accommodates the observed investment levels over the short-run, while still maintaining the old theory’s long-run properties. We achieve the desired relaxation by letting investors react to expected rather than actual rates of return. With this approach we can account for any observed level of investment by setting the expected rate of return so as to warrant that investment level. At the same time, by incorporating an adjustment mechanism that draws the expected rate of return gradually

50


toward the actual rate, we retain the long-run properties of the simpler theory, including long-run equalization of rates of return. Furthermore, this way of accounting for observed investment levels has some theoretical appeal. Investment is undertaken with the expectation of deriving returns over some period of time. Thus, investors are concerned with the rate of return not only at the moment of purchasing an asset but also throughout its life. Investors’ expectations are also “sticky” or “sluggish.” When the observed rate of return changes, investors are unsure whether this change is transient or permanent. They adjust their expectations of future rates of return only with a lag. At first investors make a small adjustment; then if the change in the actual rate persists, they make further changes in expectations, until eventually the expected rate conforms to the observed rate. Earlier, in subsection 5.1, we represented investors’ reactions to current returns through equation (2.73). To let investors react to the expected rate of return rather than the actual rate, we replace the actual rate-of-return variable rorga in equation (2.73) with the expected rate variable rorge. At the same time, we enforce the condition that the expected rate of growth in the rate of return will be equal to the required rate by replacing the required rate rrg_rorg with the expected rate erg_rorg. This changes equation (2.73) into the following equation: erg rorg(r) = LAMBORG(r) × [rorgt(r) − rorge(r)].

(2.86)

In the code, we implement this equation as Equation INVESTMENT # rule for investment # (all,r,REG) erg_rorg(r) = LAMBRORG(r)∗ [rorgt(r) − rorge(r)]. We now need to specify an error-correction mechanism that brings the expected rate, rorge, closer through time to the actual rate, rorga. We recall equation (2.78) for the expected rate-of-return schedule and note a few points: r Even before we introduced the theoretical extension in this subsection,

we already had a concept of an expected rate of return.

r Previously, however, the expected rate-of-return schedule was such

that, at the current capital stock and the current time, the expected rate of return was equal to the actual rate of return. Now we allow the expected rate of return to differ from the actual rate.


51

r As specified by the expected rate-of-return schedule, the expected rate

of return is conditional on the capital stock and also on realization time. This rules out a simple adjustment rule for the expected rate of return, such as rorga(r) × time. rorge(r) = 100 × LAMBRORGE(r) × log rorge(r)

(2.87)

This adjustment rule would represent investors as perversely ignoring the effects of investment and economic growth on the rate of return. Rather than an adjustment rule for the rate of return itself, we need an adjustment rule for the rate-of-return schedule, shifting so that through time the expected current rate of return draws closer to the actual current rate. From equations (2.78) and (2.79), we note that the position of the rateof-return schedule is given by the reference rate of return, RORGREF(r), and the base time value of the reference capital stock, QKO(r). To specify an error-correction mechanism for the rate-of-return schedule, we define the warranted reference rate of return, RORGFWARR(r), as value for the reference rate that equates the expected rate RORGEXP(r) to the actual rate RORGROSS(r) in equation (2.78). Then the warranted reference rate of return is given implicitly by the equation: Q K (r) −RORGFLEX(r) RORGROSS(r) = . (2.88) RORGFWARR(r) Q K F (r) From equations (2.78) and (2.88), we have RORGROSS(r) RORGFWARR(r) = . RORGREF(r) RORGEXP(r)

(2.89)

We postulate an error-correction process through which the reference rate of return draws closer through time to the warranted rate: RORGFWARR(r) × time, rorgf(r) = 100 × LAMBRORGE(r) × log RORGREF(r) (2.90) where rorgf(r) denotes the percentage change in the reference rate of return, and LAMBRORGE(r) is an adjustment coefficient. Substituting from equation (2.89), we obtain RORGEXP(r) time rorgf(r) = −100 × LAMBRORGE(r) × log RORGROSS(r) = −100 × LAMBRORGE(r) × ERRRORG(r) × time,

(2.91)

52


where ERRRORG(r) is a measure of error in the expected rate of return, ERRRORG(r) = log(RORGEXP(r)/RORGROSS(r)). Having specified this error-correction mechanism for the expected rateof-return schedule, we can now derive the error-correcting equation for the expected rate of return itself. By substituting equation (2.79) into equation (2.78), we obtain −RORGFLEX(r) Q K (r) RORGEXP(r) = . (2.92) RORGREF(r) QKO(r)e KHAT(r)TIME At this point, we add one final feature. For various reasons, users may sometimes wish to intervene in the expectation-setting process. They may wish to add some additional shock to the expected rate of return or to deactivate the expectations rule, so as, for example, to set the investment level directly. To allow this, we add a shift factor SRORGEXP(r) to the expected rate equation. This variable is normally exogenous and zero, but it may be given a nonzero value to add exogenous shocks to the expectationsetting process, or it may be endogenized to disable the expectations rule so that, for example, the investment level may be set directly. This gives us the final form for the levels equation: RORGEXP(r) = SRORGEXP(r) RORGREF(r) −RORGFLEX(r) Q K (r) × . QKO(r)e KHAT(r)TIME

(2.93)

Differentiating equation (2.93), we obtain rorge(r) = rorgf(r) − RORGFLEX(r) × (qk(r) − 100 × KHAT(r) × time) + srorge(r),

(2.94)

where srorge(r) denotes the percentage change in the expected rate shift factor. Substituting for rorgf from equation (2.91), we obtain rorge(r) = −RORGFLEX(r) × (qk(r) − 100 × KHAT(r) × time) −100 × LAMBRORGE(r) × ERRRORG(r) × time + srorge(r). (2.95) This equation shows three sources of change in the expected rate of return: (1) divergence between the actual rate of growth in the capital stock, qk(r)/[100 · time], and the normal growth rate KHAT(r); (2) a correction for the observed error in the expected rate; and (3) an exogenous shift


53

factor. We implement this in the model as follows: Equation EXPECTED_ROR #rule for expected gross rate of return# (all,r,REG) rorge(r) = −RORGFLEX(r)∗ [qk(r) − 100.0∗ KHAT(r)∗ time] −100.0∗ LAMBRORGE(r)∗ ERRRORG(r)∗ time + srorge(r).

5.4 The Normal Rate of Growth in the Capital Stock As noted in subsection 5.2, whether the expected growth rate ERG_RORG is model-consistent depends in part on the normal growth rate KHAT. In some early versions of the model, we treated KHAT as a fixed parameter, calibrating it before each base-case projection to ensure that it was consistent with the long-run behavior of the model. This had two disadvantages. It forced us to calibrate the parameter anew for each base-case projection, which was somewhat onerous. It also held KHAT constant within each projection, which was not always appropriate. For example, holding that variable constant would not be appropriate to a projection involving slow technological progress through the 1980s, but faster progress through the 1990s. We could avoid this problem by setting several different KHAT values for different periods within the projection, but that would involve yet more calibration simulations. To avoid these problems, we now treat the normal growth rate KHAT as an updatable coefficient within the model and provide an adjustment mechanism to bring it toward a model-consistent value through the course of a simulation. We postulate an adjustment mechanism DKHAT(r) = 100 × LAMBKHAT(r) × (KHAPP(r) − KHAT(r)) × time, (2.96) where LAMBKHAT(r) denotes a coefficient of adjustment, and KHAPP(r) the apparent current normal growth rate in region r. By the apparent current normal growth rate, we mean the normal growth rate implied by current changes in the capital stock and the rate of return and by the assumed elasticity RORGFLEX. If the rate of return is currently constant, then it appears that the capital stock is growing at the normal rate, so the apparent normal rate is equal to the actual rate. If the rate of return is rising, then the apparent normal rate is greater than the actual rate. If the rate of return is falling, the apparent normal rate is lower than the actual rate.

54


To calculate the apparent normal rate, we return to the expected investment schedule equation (2.78), assume that it agrees with the actual schedule, and solve for the apparent value KHAPP of the normal growth rate KHAT. We thus obtain KHAPP(r) = RORGFLEX(r)−1 ARG RORG(r) +

QCGDS(r) − RDEP(r). Q K (r) (2.97)

This shows that the apparent normal growth rate KHAPP(r) is equal to the actual growth rate QCGDS(r)/QK(r) − RDEP(r), plus an adjustment RORGFLEX(r) − 1 ARG_RORG(r) for current growth in the rate of return. Substituting into equation (2.96), we obtain DKHAT(r) = 100 × LAMBKHAT(r) × (RORGFLEX(r)−1 × ARG RORG(r) + QCGDS(r)/Q K (r) − RDEP(r) − KHAT(r)) × time. (2.98) Now adapting equation (2.10), we have QCGDS(r) − RDEP(r) × time. qk(r) = 100 × Q K (r)

(2.99)

Also, by definition of ARG_RORG(r) we have rorga(r) = 100 × ARG RORG(r) × time.

(2.100)

Substituting into equation (2.98), we obtain DKHAT(r) = LAMBKHAT(r) × (qk(r) + RORGFLEX(r)−1 rorga(r) + 100 × KHAT(r) × time).

(2.101)

Translating into TABLO code, we have in the model Equation KHATGROWTH # behavioral equation for estimated normal rate growth rate # (all,r,REG) DKHAT(r) = LAMBKHAT(r)∗ [qk(r) + (1/RORGFLEX(r))∗ rorga(r) −100.0∗ KHAT(r)∗ time] + SDKHAT(r). The exogenous shift variable SDKHAT(r) is included for modeling purposes. Figure 2.4 shows two rate-of-return curves: the expected rate-of-return curve (E), passing through the current capital stock and the expected current rate-of-return (QK, RORGEXP), and the warranted curve (A), passing


55

R

RORGROSS

RORGTARG RORGREF RORGEXP RORGEREF

(A) (E) QK QKEF QKT QKF

QKW

QK

Figure 2.4. Actual and expected investment schedules.

through the current capital stock and actual current rate of return (QK, RORGROSS). As before, the expected investment curve dilates rightward through time at a rate given by the normal rate of growth in the capital stock KHAT, or leftward, if KHAT is negative. Yet now it also dilates vertically, so as to draw closer to the warranted curve (A). The shape of the curve is such that any vertical dilation is equivalent to a horizontal dilation, and vice versa. Specifically, a vertical dilation by a factor V is equivalent to a horizontal dilation by a factor V. Therefore we may say simply that the curve dilates inward or outward, at a rate depending on the normal growth rate KHAT, but adjusted so as to draw closer to the warranted curve (A). As the expected rate curve dilates outward, so too does the warranted rate curve, at a rate described by the apparent normal growth rate KHAPP. If the error in expectations is zero (RORGEXP(r) = RORGROSS(r)) and the expected normal growth rate KHAT agrees with the apparent rate KHAPP, then the expected rate and warranted rate curves (E) and (A) coincide and also dilate outward together at the same rate so as to remain coincident. If the error in expectations is zero (RORGEXP(r) = RORGROSS(r)), but the apparent normal growth rate KHAPP exceeds the expected normal growth rate KHAT, then the expected rate curve (E) and the warranted rate curve (A) are initially coincident, but the warranted rate curve dilates outward faster than the expected rate curve. Through the normal rateadjustment process, the normal rate accelerates toward the apparent rate, pushing the velocity of the expected rate curve closer to that of the warranted rate curve, whereas the rate-of-return adjustment process pushes the position of the expected rate curve closer to that of the warranted rate curve.

56

Elena I. Ianchovichina and Robert A. McDougall Table 2.1. Investment module DRORT (r) = SDRORTW + SDRORT (r) RORGTARG(r) · rorgt(r) = DRORT (r) rorge(r) = − RORGFLEX (r)[qk(r) − 100 · KHAT (r) · time] − 100 · LAMBRORGE(r) · ERRRORG(r) · time + srorge(r) erg rorg(r) = LAMBRORG(r)[rorgt(r) − rorge(r)] DKHAT (r) = LAMBKHAT (r) [RORGFLEX (r) − 1 rorga(r) + qk(r) + 100 · KHAT (r) · time] erg_rorg(r) = − RORGFLEX (r){IKRATIO(r)[qcgds(r) − qk(r)] − DKHAT (r)}

(2.76) (2.77) (2.95) (2.86) (2.101) (2.83)

If the expected normal growth rate KHAT agrees with the apparent growth rate KHAPP, but the expected rate of return RORGEXP exceeds the actual rate RORGROSS, then the expected rate curve (E) lies outside the warranted rate curve (A). Then the expected rate curve dilates outward at less than the normal rate, allowing the warranted rate curve to catch up with it.

5.5 Summary Equations (2.86), (2.83), (2.101), and (2.95), shown in Table 2.1, comprise the investment theory of adaptive expectations and jointly determine the forward-sloping regional supply of investment funds. With this set of equations, there is perfect capital mobility only over the long-run as regional rates of return gradually adjust toward a common target rate. Equation (2.95) both determines the expected rate of return rorge and (in Fig. 2.4) governs the position of the expected rate-of-return curve (E). It lets the expected rate curve (E) dilate outward at a rate governed partly by equation (2.101) and partly by a catch-up component drawing toward the warranted rate curve (A). Equation (2.101) coordinates the movements of curves (A) and (E) so that, abstracting from the catch-up effect, their velocities draw together. Equation (2.86) specifies the required rate of growth in the expected rate of return – the required vertical velocity of the point (QK, RORGEXP) in Figure 2.4. Equation (2.83) translates this into a required level of investment, or horizontal velocity within the figure, given the vertical velocity and the requirement that the point lie on the expected rate curve (E). Thus equations (2.86), (2.83), (2.101), and (2.95) determine regional investment and, via the accumulation equation (2.13), regional capital stocks in GDyn. To illustrate the disequilibrium nature of the adjustment mechanism in this model, let us assume initial equality between the actual, expected, and


57

target rates of return. This equality implies that the actual and expected rate-of-return schedules overlap and move together in response to changes in the normal rate of growth in the capital stock KHAT(r). If there is a positive shock to productivity, the actual rate of return increases, and the warranted rate-of-return schedule (A) moves to the right of the expected rate-of-return schedule (E). The model detects the acceleration in economic growth via equation (2.101) in the initial period of the shock, which leads to an increase in regional investment via equation (2.83). In the next period, a further increase in the expected normal growth rate KHAT(r) leads to a further rise in regional investment. Graphically, this rise is represented by an outward dilation of the expected rate-of-return curve. In addition, via the second term of equation (2.95), investors realize that the expected rate of return is lower than the actual rate of return. This leads both to a further outward dilation of the expected rate-of-return curve toward the warranted rate-of-return curve and, through equation (2.86), a decrease in the required rate of growth in the rate of return, RRG_RORG. Equation (2.83) translates this into a rightward movement along the expected rate curve. These four equations, together with the equations (2.76) and (2.77) for rorgt, form an equation subsystem. With a normal closure, the subsystem takes as given variables qk, SDRORT, srorge, and time, and one degree of freedom of qcgds, because qcgds is constrained by the requirement that the money value of world investment must equal world saving. It determines the variables qcgds and SDRORTW. The variables DKHAT, DRORT, erg_rorg, rorge, and rorgt are internal to the subsystem. In its relations with the rest of the equation system, this subsystem has some notable features. It may be surprising that the capital stock qk(r) helps determine the investment level. Referring back to the derivation of equation (2.83), we see that achieving a given expected rate of growth in the rate of return entails achieving a certain rate of growth in the capital stock. The level of investment required to achieve that rate of growth depends on the size of the capital stock. The capital stock thus serves as a scaling factor for investment. Equally notable is the absence of certain links from the equation system. We expect the actual rate of return to affect the expected rate, yet in the expected rate-of-return equation (2.95), the variable rorga(r) does not affect rorge(r). Equally, we expect investment to affect the capital stock, yet in the capital accumulation equation (2.13), the variable qcgds(r) does not affect qk(r). The explanation is that these links do exist in the theoretical structure, but through data coefficients rather than variables. The level of

58


the actual rate of return affects the coefficient ERRRORG(r), which appears in the expected rate-of-return equation and affects the variable rorge(r). Similarly, the level of gross investment affects the coefficient NETINV(r), which appears in the capital accumulation equation and affects the variable qk(r).

5.6 Alternative Investment Determination In some simulations, the user may wish to disable the investment theory described in the preceding subsections and instead impose specific investment targets. For example, he or she may wish to use investment forecasts from macroeconomic models or to model sudden (perhaps dramatic) fluctuations in investment, such as those observed during financial crises. Imposing investment targets on all regions is harder than it may seem at first. Through the identity that world saving is equal to world investment, it would implicitly impose a target on world saving. To accommodate that, the user would need to change the treatment of saving in the closure. In this section we consider a more limited objective – imposing targets on regional shares in world investment – while allowing the usual saving theory to determine its level.5 To enable this, we use an equation QCGDS(r) = SQCGDSREG(r) × SQCGDSWORLD,

(2.102)

representing investment in region r as the product of a region-specific factor SQCGDSREG(r) and a region-generic factor SQCGDSWORLD. Differentiating, we obtain qcgds(r) = sqcgdsreg(r) + sqcgdsworld,

(2.103)

where sqcgdsreg(r) and sqcgdsworld denote percentage changes in SQCGDSREG(r) and SQCGDSWORLD, respectively. In the TABLO code, this equation is given as Equation GDI # region specific determination of investment # (all,r,REG) qcgds(r) = sqcgdsreg(r) + sqcgdsworld. When we wish to target the regional allocation of investment, we exogenize sqcgdsreg and endogenize either srorge or SDRORT. At the same time, 5

GDyn inherits from GTAP the fixed propensity to save treatment, which is adequate in the medium to long-run (5 to 15 years), but is inadequate in terms of depicting saving behavior in the very long-run. Saving behavior in GDyn is discussed in detail in Chapter 14.


59

we exogenize SDRORTW and endogenize sqcgdsworld, letting sqcgdsworld adjust so that global investment remains equal to global saving. If we wish to target the investment allocation in all periods, it does not matter whether we endogenize srorge or SDRORT. If, however, we wish to target it only in earlier periods, but let the investment theory drive it in later periods, then the choice of variable does matter. If in the earlier periods we endogenize srorge, the model achieves the investment targets by adjusting expected rates of return. In the later periods, with srorge exogenous, the expected rates converge toward the actual rates according to the usual GDyn theory. So under this treatment, the imposed investment allocation is transient. If, however, in the earlier periods we endogenize SDRORT, the model achieves the investment targets by adjusting target rates of return. In the later periods, with SDRORT exogenous, the differentials in the target rates remain in place unless and until we shock them back toward equality. So under this treatment, the imposed investment allocation is persistent.

6. Properties and Problems Having completed the presentation of the GDyn theoretical structure, we now discuss some properties of the system and issues arising in using it: r r r r r

existence and stability of long-run equilibrium cumulative and comparative dynamic results path dependence one-way relations capital account volatility and the propensity to save

6.1 Long-Run Equilibrium In the GDyn investment theory (Section 5), expected, target, and actual rates of return may all differ over the short-run. In the long-run equilibrium, these three rates are all equal and constant over time, as is also the normal growth rate for the capital stock: RORGEXP(r) = RORGTARG(r) = RORGROSS)(r), ∀r (2.104) ˙ ˙ ˙ RORGEXP(r) = RORGTARG(r) = RORGROSS(r) = 0, ∀r KHAT(r) = 0, DKHAT(r) = 0, ∀r

(2.105) (2.106)

60


These conditions imply in turn a constant investment-capital ratio. They are the same conditions characterizing the equilibrium solution of a multiregion q-investment model with convex adjustment costs. Ianchovichina (1998) demonstrates the existence and stability of the long-run equilibrium. Here we provide a numerical illustration. We use a three-region aggregation of the GTAP 3 Data Base (McDougall 1997) featuring the United States, the European Union (EU), and all other regions aggregated into a Rest of World region (ROW). The initial data (1992) reveal regional differences in rates of return, RORGROSS(r) (Fig. 2.5); normal rates of growth in capital KHAT(r) (Fig. 2.7); investment-capital ratios (Fig. 2.8), as well as sizable errors in expectations ERRRORG(r) (Fig. 2.6). In short, the benchmark data depict a world in disequilibrium. We test the long-run properties of the model over a 100-year period. The simulation represents the changes in the three economies occurring solely due to the passage of time. It depicts the movement from the initial disequilibrium state toward a long-run equilibrium. For simplicity, we assume zero regional risk premia. Figure 2.6 suggests that in the GTAP 3 Data Base, investors underestimated returns to capital in the United States and the rest of the world and overestimated returns to capital in the European Union. As investors realize their errors in predicting these returns, they adjust their expectations in an upward direction in the case of the United States and the ROW region, and in a downward direction in the case of the EU (via equation [2.100]). As a result, investment in the United States and ROW increases, whereas investment in the European Union declines (via equations [2.86] and [2.83]). It takes approximately 12 years for the model to eliminate errors in expectations (Fig. 2.6) and interregional differences in rates of return (equilibrium condition [2.104]; Fig. 2.5). However, because KHAT(r) is neither zero nor a constant in 2004, this is only a temporary equilibrium. Positive and nonconstant KHAT(r) (Fig. 2.7) implies that the expected investment schedule (2.78) will overshoot the warranted one (2.88) and over time will start moving back. We observe this type of oscillating behavior in Figures 2.5–2.8 around 2004. Only after further reduction in KHAT via equation (2.101), leading to a reduction in the investment-capital ratio via equation (2.83), will the model permanently eliminate errors in expectations and differences in interregional rates of return. Figure 2.5 shows the convergence of the regional rates of return RORGROSS(r) toward the target rate RORGTARG(r), and Figure 2.6 shows the elimination of errors in expectations ERRRORG(r) over time. Figure 2.7 displays the normal rate of growth in the capital stock KHAT(r) in its


61

14

12

10

6

4

2

USA

EU

ROW

2087

2082

2077

2072

2067

2062

2057

2052

2047

2042

2037

2032

2027

2022

2017

2012

2007

2002

1997

0 1992

Percent

8

TARGET

Figure 2.5. Actual and target rates of return. Source: Ianchovichina (1998).

movement toward 0 over the long-run, whereas Figure 2.8 demonstrates the process of adjustment toward constant investment-capital ratios. The four figures suggest that the stability conditions of the model are satisfied over time.

6.2 Cumulative and Comparative Dynamic Results GDyn is designed as a recursive dynamic model. To obtain projections through time, you run a sequence of simulations, one for each time period in the projection. To obtain comparative dynamic results, you run two sequences of simulations, one representing a base-case projection and the other representing a variant projection. From the period-by-period results, you then calculate cumulative results for each projection. Finally you find the difference between the two series of cumulative results to obtain comparative dynamic results.

62


0.2

0.15

0.1

Level

0.05

2087

2082

2077

2072

2067

2062

2057

2052

2047

2037

2042

2032

2027

2022

2017

2012

2007

2002

1992

1997

0

-0.05

-0.1

-0.15 USA

EU

ROW

Figure 2.6. Errors in expectations. Source: Ianchovichina (1998).

The formulas used for calculating cumulative results and differences in results are different for the different kinds of variables distinguished in GEMPACK: change and percentage change. For a change variable dV, the cumulative change over two periods 1 and 2, dV02 = dV01 + dV12 ,

(2.107)

where the subscript 01 denotes changes between the start and end of period 1, and 12, denotes changes between the end of period 1 and the end of period 2. For a percentage change variable v, we have a more complex formula, v 02 = 100

v 12 v 01 1+ −1 . 1+ 100 100

(2.108)

This procedure works for most of the variables in the model, but not for all. In particular, it does not work for the equivalent variation, EV(r), and


63

0.04

0.035

0.03

0.025

Level

0.02

0.015

0.01

0.005

-0.005 USA

EU

ROW

Figure 2.7. Normal rate of growth in capital stock, KHAT(r). Source: Ianchovichina (1998).

associated variables. The problem is that the variable is defined so that, in say the first period, the equivalent variation variable is EV01 = E (U1 , P 0 ) − E (U0 , P 0 ),

(2.109)

where E is the expenditure function, U is utility, P denotes prices, and the subscripts 0 and 1 refer to values at the beginning and end of the first period. In the second period, we have EV12 = E (U2 , P 1 ) − E (U1 , P 1 ).

(2.110)

Then the cumulative equivalent variation for the first and second periods is derived from EV02 = E (U2 , P 0 ) − E (U0 , P 0 ),

(2.111)

but we cannot calculate this from EV01 and EV12 . Thus we cannot calculate valid cumulative results for the equivalent variation nor, consequently, can we obtain valid comparative dynamic results. Similarly, we cannot calculate

2088

2084

2076

2080

2072

2068

2060

2064

2052

2056

2044

2048

2040

2032

2036

2028

2024

2016

2020

2012

2004

2008

2000

1992

1996

0

64


0. 0.1 0.09 0.08 0.07

0.05 0.04 0.03 0.02 0.01

USA

2088

2080

2084

2076

2068

2072

2064

2056

2060

2052

2044

EU

2048

2036

2040

2032

2028

2020

2024

2016

2008

2012

2004

1996

2000

0 1992

Ratio

0.06

ROW

Figure 2.8. Investment–capital ratio. Source: Ianchovichina (1998).

valid comparative dynamic results for the equivalent variation decomposition (Huff and Hertel 1996). This does not mean that we cannot obtain comparative dynamic results for equivalent variation. To obtain them, however, we need some computational machinery beyond the cumulating and differencing procedures used for other variables. The calculation of welfare in GDyn is discussed in Chapter 6 of this book.

6.3 Path Dependence GDyn is inherently a path-dependent model. That is, in GDyn, the effects of changes in exogenous variables depend not only on the overall changes in exogenous variables but also on the paths followed by them. When GDyn is used in a dynamic mode by shocking the time variable, the effects of


65

economic shocks depend not only on the size but also on the timing of the shocks. Path dependence is built into the theory in three places: wealth accumulation (subsection 4.3), the partial adjustment treatment of the capital stock (subsection 5.1), and the adaptive expectations treatment of the expected rate of return and the normal growth rate (subsections 5.3 and 5.4). In GDyn a region’s wealth depends on its past history; it cannot be determined from other current variables, such as income. The final level of regional wealth in any simulation depends on the original level and on the time paths of the exogenous variables within the simulation. For example, technological progress in a given region normally leads to an increase in its wealth, but the increase in wealth is greater if the technological progress occurs mostly near the beginning of the period than if it occurs mostly toward the end. Other path dependencies arise from the lagged adjustment treatment of the capital stock and from the adaptive expectations treatment of investment. Similarly, regional capital stocks cannot be inferred from other current variables for two reasons. Globally, the money value of net physical investment is equal to saving, so the money value of the global capital stock is determined by wealth accumulation (and capital gain), not by an equilibrium condition. In addition, the distribution of capital across regions is given not by an equilibrium condition, but by a partial adjustment process, as described in subsection 5.1. Investors do redistribute capital to equalize rates of return, but only gradually. Past shocks therefore affect the current international distribution of capital more if they occur in the more distant past and less if they occur in the more recent past. Finally, the level of investment depends not on the actual rate of return but on the expected rate. The expected rate of return cannot be inferred from other current variables, but adjusts toward the actual rate with a lag, as described in subsection 5.3. Here then is yet another adjustment process whose results depend not only on the size of the changes in its inputs but also on their timing. Given GDyn’s objectives, this path dependence must be construed not as a bug but a feature. Indeed, if we should extend GDyn to provide a better treatment of short-run dynamics, bringing in more macroeconomic content such as that found in such models as G-CUBED or FAIR, path dependencies will become more pervasive rather than less. In short, path dependence in GDyn is here to stay. Nevertheless, and this is why one might be tempted to construe it as a bug, path dependence imposes some practical inconveniences. It places on the

66


user an onus to represent accurately the time paths of exogenous variables, in circumstances where doing so would otherwise be unnecessary. Users need to take it into account in several places in their computational strategy. First, you need to set periods, within the overall projection time interval, to capture sufficient detail about the time profile of the shocks. With the continuous-time approach used in GDyn, you can run, say, a tariffreduction scenario over a single 10-year interval and get sensible and meaningful results. If, however, you want the tariff cuts to be back-loaded and not implemented at an even pace, then you need to use several shorter intervals, so that you may specify lower rates of tariff reduction in the earlier intervals and higher rates in the later intervals. Second, even if you wish to apply shocks evenly through time, you may wish to avoid long time intervals, if you do not like the rule TABLO uses to distribute shocks between steps. TABLO-generated programs distribute shocks so that the change in the levels variable is the same for all steps (Harrison and Pearson 1998, see chapter 4, “GEMSIM and TABLOGenerated Programs”). To take an extreme example, if you shock a variable by 300%, using a two-step solution procedure, your TABLO-generated program shocks the variable by an amount equivalent to 150% of the initial level in each step; that is, by 150% in the first step (going from 1 to 2.5 times the initial level), and 60% in the second step (going from 2.5 to 4 times the initial level). For most percentage change variables, in most applications, a more appealing default assumption is that the percentage change in the variable is constant across steps. For example, it is more natural to assume that the population grows at a constant rate through time (for example, by 1% per year) than that it changes at a constant rate (for example, by 200,000 persons per year). Likewise, in the extreme example given previously, we would typically prefer by default to shock the variable by 100% in each step, rather than by 150% in the first step and 60% in the second. GEMPACK wizards may perhaps know some way to coerce TABLO to use equal percentage changes. No such way, however, is apparent from the published documentation. The shock-splitting rule does not much matter when the shocks are small, but it does matter when they are large. More specifically, it matters when the total shock in a simulation is large, even if the shock is broken up into small pieces in individual steps. One way to work around the problem is to avoid long intervals always, even if all shocks are evenly distributed through time. Finally, path dependence rules out some common closure-swapping strategies. In GEMPACK, a common expedient is to let the model determine


67

the change in some instrument variable required to achieve a given change in a target variable, by making the naturally exogenous instrument variable endogenous, and the naturally endogenous target variable exogenous. For example, we may determine the rate of technological progress required to achieve a given improvement in welfare by endogenizing the technological change variable and exogenizing the welfare variable. If we then run a second simulation, using the natural closure and shocking the technological change variable according to the results from the first simulation, we get – with a path-independent model – the same results as in the first simulation. We can then investigate the effects of changes in other elements of the scenario on welfare as on other variables, using the natural closure and the calibrated technological change shock. With a path-dependent variable this approach does not work. The trouble is that the path of the technological change variable is different in the two simulations. In the second simulation, technology changes evenly throughout the simulation interval; in the first, it changes so as to keep GDP moving evenly through the interval. This difference is liable to affect the simulation results. In GDyn, for example, a front-loaded improvement in technology has more effect on end-of-interval wealth than the same total improvement distributed evenly through the interval. What we need (but, at the time of writing, do not have) for this problem is an automated algorithm for finding the constant rate of growth in an instrument variable (or the constant rates of growth in a set of instrument variables) that achieves given total growth in a target variable (or a set of target variables). Such a tool would be useful not only for single-simulation but also for multi-simulation projections. For example, in a projection made up of five 2-year simulations, each involving different tariff shocks, we would like to be able to find the constant rate of technological progress, through the complete 10-year projection interval, required to achieve a given welfare improvement over the interval. Such a multi-simulation facility would be useful even with path-independent models.

6.4 One-Way Relations A novel emergent feature of GDyn, relative to standard GTAP, is the appearance of what we here describe as one-way relations. In standard GTAP, as perhaps in most GEMPACK-implemented models, if an exogenous variable A affects an endogenous variable X, we can swap A and X in the closure and determine in a simulation the change in A required to bring about a given change in X. Of course, this may not work if X is not monotonic in A, but it works most of the time, and of course, it always works for sufficiently small

68


changes in X if X is locally nonstationary in A. In GDyn, however, it may easily fail. That is, there are relations between variables A and X such that the solution program can determine the change in X arising from a given change in A, but not the change in A required to bring about a given change in X – no matter how well behaved mathematically the relation is between A and X. These one-way relations appear when one variable affects another not through the equation system but through data updates. For example, investment, of course, affects the capital stock, yet the investment variable (qcgds) does not appear in the relevant equation (2.13). Instead, the equation contains the investment coefficient NETINV. In a single-step simulation, qcgds has no effect on qk. In a multistep simulation, it affects NETINV at each update of the data files and thereby affects qk. Now consider the effect of a shock to a normally exogenous, investmentrelated variable, for example, the target rate shift variable SDRORT. In each period, this leads through the investment module to some change in the investment variable qcgds, but that change in qcgds has no effect in the current period on qk. If we try to change the closure to find the SDRORT value consistent with a given change in qk, we find it impossible to do so. If we exogenize qk(r) for some region r, and endogenize SDRORT(r), we make the model singular, because we thereby make exogenous all variables in equation (2.13). The only natural way to exogenize qk(r) is to swap it with sqk(r), and that does not achieve the larger purpose, because it does not allow qk(r) to determine qcgds(r) or SDRORT(r). Another one-way relation is that between the actual rate of return RORGROSS and the expected rate RORGEXP. According to the GDyn theory, changes in RORGEXP cause changes in RORGROSS, yet rorga does not appear in the equation for rorge (2.95). In a single-step simulation, indeed, rorga has no effect on rorge. In a multistep simulation, however, it affects ERRRORG at each data file update and thereby affects rorge. If you shock some exogenous variable so as to increase the actual rate of return – if, for example, you apply a positive shock to labor supply qo(“labor”,r) – this has no effect on the expected rate rorge(r) in a singlestep simulation, but does affect it in a multistep simulation. However, if you want to find the labor supply change needed to achieve a given change in the expected rate of return, you find that you cannot exogenize rorge(r) and endogenize qo(“labor”,r). To do so would create a singular system: The twoequation subsystem comprising the capital accumulation equation (2.13) and the expected rate-of-return equation (2.95) for region r would contain only one endogenous variable, qk(r). The only natural way to exogenize


69

rorge(r) is to endogenize srorge(r), and this does not achieve the purpose of determining labor supply endogenously. Given that closure swaps do not work at all across these one-way relations, we evidently need some new computational machinery to let us target the naturally endogenous variables in them.

6.5 Capital Account Volatility and the Propensity to Save GDyn inherits from standard GTAP its specification of the regional household demand system and, in particular, the treatment of saving. As in standard GTAP, there is a fixed average propensity to save. In other words, saving is a fixed proportion of income in each region. One unwelcome implication is that the capital account and net foreign liabilities are highly volatile and can grow without bound in GDyn simulations. In the real world, for reasons that are poorly understood, saving and investment are highly correlated across countries, and international capital flows are much smaller and more stable than simple theory would suggest (Feldstein and Horioka 1980). In GDyn, we do not impose any such correlation, so relatively modest economic shocks can lead to unrealistically large international capital flows, as well as unrealistically large changes in regions’ net foreign liabilities. The second problem is that as economies with high savings rates, like China, grow, there is a glut of global savings and, as a result, of investment and capital in the world. Because of excessive investment, rates of return to capital fall without bound. This problem prevents us from running simulations with GDyn over the very long-run. Chapter 14 evaluates the behavior of foreign assets and liabilities in GDyn, comparing it to historical data on these indicators, and proposes a technique to address this shortcoming of the model.

7. Concluding Remarks This chapter presented a set of new equations added to the GTAP model to construct GDyn, a dynamic AGE model of the world. The new theory offers a disequilibrium approach to modeling endogenously international capital mobility in a dynamic applied general equilibrium setting, and it takes into account stock-flow dynamics and foreign asset income flows. The method can be especially attractive to policy modelers as it permits a recursive solution procedure, a feature that allows easy implementation of dynamics into any static AGE model without imposing limitations on the model’s size.

70


Key to the proposed approach is investors’ adaptive expectations about potential returns to capital. This type of expectation emphasizes errors in investors’ assessment of potential returns to capital – such as those observed during financial crises. It can also be shown that it ensures the convergence of the model toward a stable equilibrium and offers the flexibility of tailoring the model to observed data. Despite some limitations of GDyn, such as the lack of equity-for-debt substitution, the absence of bilateral detail, and the lack of forward-looking behavior, the model offers a unique and simple treatment of international capital mobility in a dynamic AGE context. It captures endogenously the economy-wide effects of capital and wealth accumulation and the income effects of foreign property ownership. References Feldstein, M. and C. Horioka. 1980. “Domestic Saving and International Capital Flows.” Economic Journal 90(358), 314–29. Harrison, W. J. and K. R. Pearson. 1998. An Introduction to GEMPACK. GEMPACK Document No. GPD-1 (4th ed.). Melbourne, Australia: Centre of Policy Studies and Impact Project, Monash University. Hertel, T. W. (ed.). 1997. Global Trade Analysis Modeling and Applications. Cambridge: Cambridge University Press. Hertel, T. W. and M. E. Tsigas. 1997. “Structure of GTAP.” In T. W. Hertel (ed.), Global Trade Analysis Modeling and Applications (pp. 13–73). Cambridge: Cambridge University Press. Huff, K. and T. W. Hertel. 1996. Decomposing Welfare Changes in GTAP. Technical Paper No. 5, Center for Global Trade Analysis. West Lafayette, IN: Purdue University. Ianchovichina, E. I. 1998. International Capital Linkages: Theory and Application in a Dynamic Computable General Equilibrium Model. Ph.D. thesis, Department of Agricultural Economics, Purdue University. Kapur, J. N. and H. K. Kesavan. 1992. Entropy Optimization Principles with Applications. New York: Academic Press. McDougall, R. A. (ed.). 1997. Global Trade, Assistance, and Protection: The GTAP 3 Data Base. Center for Global Trade Analysis. West Lafayette, IN: Purdue University. McDougall, R. A., A. Elbehri, and T. P. Truong (eds.). 1998. Global Trade, Assistance, and Protection: The GTAP 4 Data Base. Center for Global Trade Analysis. West Lafayette, IN: Purdue University. McDougall, R., M. E. Tsigas, and R. Wigle. 1997. “Overview of the GTAP Data Base.” In T. W. Hertel (ed.), Global Trade Analysis Modeling and Applications (pp. 74–124). Cambridge: Cambridge University Press.

THREE

Behavioral and Entropy Parameters in the Dynamic GTAP Model Alla Golub and Robert A. McDougall

1. Introduction The dynamic theory in Chapter 2 describes various new parameters governing international capital mobility. This chapter examines what we can learn from country-panel data about the magnitude of these additional parameters, corresponding calibration procedures, and the manipulation of the parameters with an aggregation program. The new parameter file containing parameters used in the dynamic theory is a GEMPACK header array file; its contents are listed in Table 3.1. The first new parameter listed in Table 3.1 is INC. This parameter is the initial income level across simulations and is used to calculate welfare measures in multiperiod simulations. It is represented in the unit US$ millions. The rest of the parameters in Table 3.1 can be grouped according to their role in the model: lagged adjustment parameters, elasticity of rate of return to capital with respect to capital stock, and parameters determining the allocation of regional wealth and composition of regional capital. These three parameter types are discussed in turn in this chapter.

2. Parameters Determining Lagged Adjustments The investment theory presented in Chapter 2 is expressed in terms of gross rather than net rates of return, and it allows only zero or positive gross rates of return. However, net of depreciation, rates of return may be negative, and they may decline to the negative of the depreciation rate. The long-run equilibrium in the GDyn model is defined as the convergence of the net rates of return to capital stock across regions. If region-specific risk premia are We are thankful to Thomas Hertel, Terrie Walmsley, Ken Foster, Elena Ianchovichina, and Paul Preckel for their valuable comments and suggestions.

71

72

Alla Golub and Robert A. McDougall Table 3.1. Contents of the dynamic parameters file

Coefficient name

Dimensions

Description

INC LAMBKHAT

REGa REG

LAMBRORGE LAMBRORG RORGFLEX

REG REG REG

RIGWQH

REG

RIGWQ_F

REG

initial income coefficient of adjustment in estimated normal growth rate coefficient of adjustment in expected rate of return coefficient of adjustment in rate of return elasticity of rate of return to capital with respect to capital stock rigidity of allocation of wealth by regional household rigidity of source of funding of enterprises

a

REG denotes number of regions.

allowed, then the long-run equilibrium in the model is defined as the convergence of risk-adjusted net rates of return to capital stock across regions. In the absence of risk premia, and if the depreciation rates are the same across regions, the convergence of net rates of return guarantees the convergence of gross rates of return. In this section, we construct cross-country time-series data on net rates of return to capital to test the convergence hypothesis and determine the speed of convergence in rates of return across countries. We then use the results to set the lagged adjustment parameters in the model in accordance with the observed behavior.

2.1 Data As in the standard GTAP model, the GDyn model is a real assets model; that is, there is no financial market. The gross rate of return to capital for each country is defined as the ratio of gross operating surplus to the capital stock, and the net rate of return to capital is the ratio of net (of depreciation) operating surplus to capital stock.1 To determine parameters that will quantify the degree of capital mobility in the GDyn model, the rates of return to capital are constructed in accordance with these definitions. Net rates of return to capital are often used to compare companies’ profitability across countries. Walton (2000a, 2000b) uses the net rate of return to capital to compare profitability of the corporate nonfinancial sector in the United Kingdom with profitability of the corporate nonfinancial sector in other countries. In these studies the rates of return were constructed using 1

In both definitions, the capital stock in the model is net capital stock.

Behavioral and Entropy Parameters in the Dynamic GTAP Model

73

data available in the national accounts of 19 countries. These rates of return are rather sparse for our purposes (the time period covered is too short for econometric investigation and the panel is unbalanced) and represent returns for nonfinancial corporations only, whereas rates of return to capital in the GDyn model represent the overall profitability of the economy. Therefore these studies are considered here for illustrative purposes only. There are two important features of the data highlighted in Walton (2000a, 2000b) that we should keep in mind when choosing data for our analysis. First, annual rates of return are calculated as the ratio of the operating surplus to capital employed. Profits, the main source of the operating surplus, are defined fairly precisely and measured reasonably consistently. However, capital employed is not defined as precisely; the definitions and methods used to estimate capital stock vary from country to country. Most of the national statistics data on capital are compiled using the perpetual inventory method (PIM), which is discussed in detail later. This method involves adding gross fixed capital formation to, and deducting consumption of fixed capital from, an initial estimate of capital stock. The variations come from the estimates of useful service lives by capital type and country and are influenced by the business cycle and technological change. As a result, the differences between countries’ rates of return, constructed using national account data, can reflect both differences in profitability and differences in calculation methods. Because countries have estimated profitability consistently over time, relative changes in net rates of return should reflect real changes in their economies and hence could be used in the cross-country profitability comparisons undertaken in Walton (2000a, 2000b). However, to test for convergence of rates of return to capital, we need a capital stock series that is constructed using uniform (across countries) assumptions. In addition, not many countries have data on profitability and/or capital stock, and time coverage varies from country to country (Walton 2000b). Our first step in constructing rates of return is to define profits associated with the use of capital stock. Using the income approach to gross domestic product, GDP can be represented as a sum of value added at factor costs plus indirect taxes. The value added at factor costs consists of labor earnings, capital earnings, and land earnings. Although time-series data on value added at factor costs with good country coverage are available from many sources (for example, the World Development Indicators (WDI) database supported by the World Bank), the labor earnings data are problematic. For this reason, the analysis begins with a set of countries for which these data are most readily available – the Organization for Economic Cooperation and Development (OECD) countries.

74

Alla Golub and Robert A. McDougall

Time-series data on the gross operating surplus at current prices are obtained from the SourceOECD database, Annual National Accounts Volume II – Detailed Tables – Main Aggregates Volume 2004 release 01 (SourceOECD). This database covers detailed national account data for most OECD countries, including components of value added, from 1970 to the present. It lists four components of GDP: (1) compensation of employees; (2) taxes less subsidies on production and imports; (3) gross operating surplus and gross mixed income; and (4) a statistical discrepancy, which is small or zero for most of the countries considered. The sum of the gross operating surplus and gross mixed income is used as a proxy for capital earnings. Note that this measure overestimates capital earnings because it includes land earnings, returns to natural resources, and that portion of self-employed labor earnings that is not accounted for by imputed wages. Land earnings should not be a big problem because they are expected to be small relative to capital earnings in developed countries. However, the potential inclusion of self-employed labor may result in a larger error in the capital earnings measure, but this error is expected to be much smaller in the OECD countries than in developing countries. The gross operating surplus measure also includes depreciation of capital stocks. As noted earlier, the convergence in GDyn is modeled as convergence in net rates of return. To test for convergence of net rates of return, we will construct net operating surplus measures. The second step in constructing rates of return to capital is to define capital stocks. Several alternative sources for capital stock data for the OECD countries could be used. In all these sources the capital stock estimates are derived using the perpetual inventory method (PIM). The first data source for capital stock data for the OECD countries is the OECD itself. Until 1997, OECD published annual data in a report titled Flows and Stocks of Fixed Capital (Statistics Directorate OECD, various years).2 However, in 1997, production of these data ceased because of the move to the new system of national accounts. Some countries are now starting to produce these data again, but not in a sufficient amount for the OECD to resume publishing them. The data are available for only a few of the OECD countries and cover different time periods for different countries. The data come from national account statistics, and as noted earlier, the assumptions made to construct these series differ from country to country.

2

The PIM and the estimation procedures used by the OECD countries are described in the manual issued by OECD (OECD 1993).


75

A second data source is Larson et al. (2000), who constructed capital stock time-series data for 62 industrial and developing countries for the period 1967–92. The main objective of this database was to provide sectoral and economy-wide capital stock data for countries both within and outside of the OECD. The same method was used in the calculations to facilitate comparisons across countries. Larson et al. (2000) constructed fixed-capital series based on national account investment data, using a modified version of the PIM. A third data source for capital stock data is Nehru and Dhareshwar (1993), who constructed capital stock time-series data for 92 developing and industrial countries from 1960 to 1990. Because the OECD capital stock data are sparse and constructed using assumptions that differ from country to country, we eliminated this source from consideration and made a choice between the Larson et al. (2000) and Nehru and Dhareshwar (1993) databases. The PIM used in the construction of these databases can be generalized in the following equation: K t = s t I t + s t−1 I t−1 + · · · + s t−L I t−L ,

(3.1)

where Kt is capital stock at the end of year t, It is investment made during year t, L is the lifetime of the capital good, t − L is the vintage of the oldest surviving capital asset, and sj is the productivity of investment of age j, 0 < s j < 1 for 0 < j < L; s0 = 1 and sj = 0 for j ≥ L. The main difference between the Larson et al. (2000) and Nehru and Dhareshwar (1993) methods for constructing the databases is in the assumptions made about the path of sj . Nehru and Dhareshwar (1993) assume that sj follows a geometric decay pattern with the rate of decay fixed at 4%, which is equivalent to the assumption of an infinite lifetime L of capital assets and a 4% decline in productivity every year. The method used to define productivity in Larson et al. (2000) is more general and closer to the one used in the OECD data. To restrict productivity to be non-negative, this method assumes a finite lifetime L of capital assets and a curvature parameter β bounded from above by 1, as shown in the following equation (Larson et al. 2000): s j = (L − j )/(L − βj ),

0 ≥ j ≥ L.

(3.2)

To generate economy-wide capital stocks from investment data, Larson et al. (2000) set β = 0.7 and defined service life L as a stochastic variable with a mean of 20 and standard deviation of 8 years. Analysis of equation (3.2) shows that productivity falls with the age of assets, and when β is positive but less than unity, the depreciation accelerates with the time of asset use. Figure 3.1 illustrates these points and the differences between productivity paths assumed in Nehru and Dhareshwar (1993) and Larson et al. (2000).

76


1 0.9

Relative productivity

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

2

4

6

8

10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 Time in use

Nehru and Dhareshwar (1993)

Larson et al. (2000)

Figure 3.1. Comparison of the relative productivity paths in Nehru and Dhareshwar (1993) and Larson et al. (2000). The geometric decay path in Nehru and Dhareshwar (1993) is based on a 4% decay rate. Larson et al. (2000) built the path of productivity of total economy-wide fixed capital assuming 20 years of service life and curvature parameter β = 0.7. Source: Authors’ calculations.

To construct a capital stock series, an assumption about the initial value of capital is needed. Many techniques to seed the initial values are discussed in Nehru and Dhareshwar (1993) and Larson et al. (2000). However, if the investment series are sufficiently long, and given that the productivity of old capital is low, the contribution of old capital to the current capital stock should be small. This view is supported by the analysis of sensitivity of constructed capital stock series with respect to the choice of initial values in Larson et al. (2000). Because the Larson et al. (2000) assumptions to measure economy-wide capital stocks are more realistic, and capital stock series constructed using their method are less sensitive to the choice of initial value, we chose this database to construct rates of return to capital. The choice of countries and years to be included in the analysis is dictated by the availability of data in both the OECD records on gross operating surplus and the capital stock data in the Larson et al. (2000) database. To construct a net operating surplus, depreciation should be subtracted from the gross operating surplus. Data


77

on depreciation are available from SourceOECD; however, these data are not consistent with the depreciation assumed in the calculations of net capital stocks in Larson et al. (2000). To calculate a net operating surplus that is consistent with the net capital stock data, the depreciation Dt is recovered from the Larson et al. (2000) time series using the following formula: D t = K t−1 − K t + I t .

(3.3) 3

It is then subtracted from the gross operating surplus. To calculate the values of depreciation (Dt ), current capital (Kt ), investment (It ), and capital in the previous year, (Kt −1 ) should be measured in constant prices. The capital stock data in Larson et al. (2000) are given in 1990 US$, whereas investments are given in 1990 local currency units (LCU). A real exchange rate et = 1/Et (P tUS /p t ) is constructed, where Et is the nominal exchange rate in US$ to local currency, P tUS is the dollar deflator, and pt is the domestic deflator.4 Using the real exchange rate, capital stocks are converted from 1990 US$ to 1990 LCU, and depreciation in 1990 LCU is calculated. Using the dollar deflator, capital stock and depreciation are converted into current US$. The gross operating surplus is also converted from current local currency to current dollars using SourceOECD exchange rates.5 The net operating surplus in current dollars is calculated by subtracting depreciation from the gross operating surplus, and then the net rates of return are calculated as a ratio of net operating surplus and net capital stock for 20 OECD countries from 1970 to 1992.6 The net rates-of-return series are shown in Figure 3.2. To simplify the figure, the rates of return are shown for only nine OECD countries. The time series of the other 11 countries are in the range between the rates of return in Portugal and Finland. The rates of return are very high in the beginning of the time period in Turkey, Greece, South Korea, and Portugal, but then decrease. This feature concurs with our expectations that the rates of return 3

4

5

6

For some reason, the calculated depreciation in 1980 is negative for all countries in the Larson et al. (2000) database. For each country, we use the arithmetic average of 1979 and 1981 depreciations for the 1980 depreciation. Investments series and all determinants of the real exchange rate are given in Larson et al. (2000), including the exchange rate, which is the market exchange rate from the International Monetary Fund. For European countries adopting the Euro, the SourceOECD database lists gross operating surplus in Euros. For this reason, SourceOECD exchange rates of US$ to local currency are different from ones obtained from the International Monetary Fund. These countries are Australia, Austria, Belgium-Luxemburg, Canada, Denmark, Finland, France, Greece, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, South Korea, Sweden, Turkey, United Kingdom, and the United States.

78


60%

Net rate of return to capital

50%

Finland Greece Japan Portugal South Korea Sweden Turkey United Kingdom United States

40%

30%

20%

10%

0% 1970 1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

-10% Year

Figure 3.2. Net rates of return to capital in OECD countries. Source: Authors’ calculations.

to capital are higher in the least developed countries because capital is a scarce resource. As these countries’ economies grow, capital expands and its marginal product falls, and the rates of return to capital decline. Note that in Finland in 1991, the return to capital is negative, although small in absolute magnitude. This may be because the Larson et al. (2000) method overestimates the depreciation, or possibly it is negative simply because of a decline in capital earnings.

2.2 Convergence of Rates of Return to Capital Convergence of different productivity measures is a popular topic (Bernard and Jones 1996a, 1996b; Nin et al. 2004). This section draws on the econometric techniques used in these studies to focus on the question of convergence in rates of return. Assuming the absence of risk premia, the long-run equilibrium in the GDyn model is defined as the convergence of the ratios of capital earnings to capital stock across regions. Thus an important research question is whether these measures actually converge. An initial look at the dispersion of the rates of return to capital across countries in Figure 3.2 shows that it does appear to decline over time. Time-series evidence can also be used to examine the convergence of the rates of return by applying


79

the test for unit roots in panel data. With such a short time series, unit root testing for pairs of countries would appear to be out of the question; however, the technique of testing unit roots in panel data would be appropriate (Bernard and Jones 1996b; Levin and Lin 1992). In conducting the unit root test, the United Kingdom is chosen as a benchmark country, and deviations from the United Kingdom’s rate of return for 19 OECD countries are constructed. Consider the following general model: rr it = μi + ρ · rr it−1 + εit ,

(3.4)

where rrit is the difference between country i and the benchmark country rates of return, with error term εit ∼ iid (0, σε2 ) and drift μi ∼ iid(μ, σμ2 ). Let ρˆ and t be the OLS parameter estimate and t-statistic from the regression with equation (3.4) above, respectively. Bernard and Jones (1996b) show that, under the null hypothesis of a unit root and nonzero drift, t approaches the standard normal distribution. We are testing the null hypothesis of no convergence, which means that the deviation of the rate of return to capital from a benchmark country is a nonstationary process with nonzero drift. The alternative hypothesis is that rates of return to capital are converging in the sense that deviations of rates of return to capital from the benchmark country across countries are stationary processes. Table 3.2 reports estimates of country-specific drifts μi , together with results of the test ρˆ = 1. The results show that ρˆ is significantly less than 1, providing evidence against the null hypothesis of no convergence. Because all considered countries are developed countries, this result is expected and similar to what was found in the literature on convergence of productivity for OECD countries (Bernard and Jones 1996b). The estimate of ρˆ reported in Table 3.2 implies a convergence rate of 9% per year for the net rates of return to capital in the OECD countries. It is also important to mention that model (3.4) allows for countryspecific intercepts. For 16 of the 19 (without benchmark) countries, the intercepts are not statistically different from zero. This result has implications for the target rate and closure in the dynamic model: If depreciation rates are assumed to be equal, a common target rate can be set in the initial database for 17 (16 plus the United Kingdom) countries. For three other countries – Turkey, Ireland, and Greece – the intercept is different from zero. One possible interpretation is that, even if the deviations from the benchmark are stationary, there is a nonzero long-run value of the deviation, which may, in turn, suggest country-specific target rates for these three countries in the GDyn. Another possible explanation is that we have too few

80

Alla Golub and Robert A. McDougall Table 3.2. Time-series test for convergence of rates of return to capital

for 20 OECD countries∗

Parameter Australia (μ1 ) Austria (μ2 ) Belgium-Luxemburg (μ3 ) Canada (μ4 ) Denmark (μ5 ) Finland (μ6 ) France (μ7 ) Greece (μ8 ) Ireland (μ9 ) Italy (μ10 ) Japan (μ11 ) Netherlands (μ12 ) New Zealand (μ13 ) Norway (μ14 ) Portugal (μ15 ) South Korea (μ16 ) Sweden (μ17 ) Turkey (μ18 ) United States (μ20 ) ρ R-square Test ρ = 1 ∗

Estimate

Std. Error

t-value

Pr > |t|

0.001 −0.002 0.003 0.002 −0.001 −0.004 −0.001 0.010 0.007 0.002 −0.003 0.000 0.003 0.000 0.003 0.005 −0.002 0.017 0.003 0.907 0.984 5.410

0.003 0.003 0.003 0.003 0.003 0.003 0.003 0.004 0.003 0.003 0.003 0.003 0.003 0.003 0.003 0.004 0.003 0.006 0.003 0.017

0.320 −0.880 0.920 0.600 −0.450 −1.300 −0.240 2.280 2.480 0.850 −1.110 0.160 0.940 0.130 0.860 1.280 −0.610 2.880 1.180 52.900

0.749 0.382 0.361 0.546 0.651 0.195 0.810 0.023 0.014 0.395 0.269 0.870 0.350 0.899 0.388 0.200 0.543 0.004 0.240 qˆ D ), the average level of technology in the industry rises (i.e., the efficiency variable is positive). Note that this formulation can also induce efficiency losses when exported output declines relative to output for domestic consumption, because technological gains are reversible (as was found by Bernard and Jensen). As with the procompetitive effect, this efficiency effect is incorporated into the GDyn model as a reduced-form representation of the more complex underlying process by which exporting affects firm-level productivity. Furthermore, this export-productivity linkage only applies to the export and output deviations from the baseline, not to the baseline itself.

2.3 The FDI-Productivity Effect Increased levels of foreign direct investment (FDI) have the potential to transfer technology and managerial skills to a host country, thereby enhancing productivity (Blalock 2001). Some authors, such as Rodrik (1999), point out that there is little hard evidence for the more extravagant claims linking

The Contribution of Productivity Linkages

323

FDI and productivity. However, an increasing number of studies confirm that there are indeed significant, positive technological spillovers, even if we cannot always identify the precise mechanism through which these spillovers occur. For example, in a study of FDI, research and development (R&D), and spillover efficiency in Taiwan (China), Chuang and Lin (1999) use firm-level data to confirm the existence of beneficial spillovers from FDI. Specifically they find that a 1.0% increase in an industry’s FDI ratio produces a 1.40% to 1.88% increase in domestic firm productivity. As indicated in Chapter 2, in the GDyn model regional capital is owned by domestic and foreign households via a global trust. This relationship is V = VH + VF , where V is the equity value of firms in a given country, and VH and VF are the domestic- and foreign-held components of V, respectively. Thus we can write the foreign equity share as θF ≡ VF /V. We use this as a proxy for the share of FDI in total capital stock. Because we want to relate productivity changes to changes in θF , we totally differentiate this to get ˆ Using Chuang and Lin’s (1999) lower bound estimate, we can θˆF = Vˆ F − V. write the percentage change in productivity associated with a capital inflow from abroad as ˆ efficiency = 0.014(Vˆ F − V). As with Chuang and Lin’s study, we implement this reduced-form relationship only for manufacturing sectors and incorporate it into the GDyn model as an additional equation determining the change in efficiency endogenously as a function of changes in the share of foreign ownership, owing to the FTA. We do not incorporate this productivity effect into the dynamic baseline. It only plays a role in the FTA counterfactual.

3. Data and Procedures 3.1 Data and Aggregation In this analysis we employ the GTAP 5 Data Base, which has a base year of 1997 and distinguishes 57 sectors and 66 regions (Dimaranan and McDougall 2002). Among its notable features are disaggregation of service sectors and the explicit treatment of international transport margins. We aggregate the GTAP data up to 23 sectors and 19 regions (see Appendices 12.2 and 12.3, respectively).7 Our regional aggregation emphasizes the individual countries involved in the Japan-ASEAN FTA. The GTAP data 7

In certain tables in this chapter we use higher sector aggregates (e.g., food & agriculture) to save space. Definitions are in Appendix 12.2. The analysis is otherwise done in terms of the 23 sectors.

324

Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer

distinguish six ASEAN nations (Singapore, Indonesia, Malaysia, Philippines, Thailand, and Vietnam), and when we refer to “ASEAN” later, we refer to these six only. Although the GTAP data do not disaggregate Brunei, Cambodia, Laos, and Myanmar, these four nations comprise only 3.25% of ASEAN GDP (ASEAN 2002).

3.2 Trade Flows and Tariffs In this section we use the aggregated GTAP data to provide an overview of the current trade and tariff relationships between Japan and ASEAN nations. The data indicate that, although ASEAN depends on Japan for about 19% of overall imports, Japan gets only about 11% of its imports from ASEAN.8 So despite their proximity, and dissimilarities in terms of endowments and technology (which may be a source of comparative advantage), these economies are not highly integrated, especially when compared to other regions such as Europe or North America. A deeper view of the current Japan-ASEAN trade relationship can be gained from Table 12.2, which breaks down the relative importance of merchandise trade with Japan for individual ASEAN members. The top half refers to the percent of ASEAN merchandise imports that originate in Japan. Clearly, Japan is not an important supplier of agriculture, resources, or light manufactures for ASEAN, because it supplies less than 10% of total imports in nearly all cases (Table 12.2). However, Japan is quite important as a source of high-technology manufactures. In the automotive sector Japan plays a particularly dominant role, with an import share of 60.9% for ASEAN overall. The bottom half of Table 12.2 depicts the relative importance of Japan as an export destination for the different ASEAN countries. Although this varies to a great degree across ASEAN countries, Japan generally plays a fairly large role as an export destination for ASEAN food and natural resource sectors. To the extent that comparative advantage is a driver of trade, it would appear that Japan and ASEAN are natural trading partners. Japan can play a role as a high-tech supplier, whereas the ASEAN countries as a group are presently well suited to meet Japan’s need for resources, agriculture, and light manufactures. Average tariff rates for all sectors are reported in Table 12.3. Japan is quite notable for its protection of food and agriculture (52.7% average tariff), which is driven to a large extent by protection of its rice market (rice has 8

Japan’s imports of goods and services from ASEAN nations totaled $52.3 billion in 1997, whereas Japan’s imports from the Rest of World (ROW) were $395.1 billion. ASEAN imported $81.2 billion from Japan and $276.2 billion from the ROW and had $79.4 billion worth of trade within itself.


325

Table 12.4. Average trade-weighted bilateral tariffs, 1997 (%) Japan Singapore Indonesia Malaysia Philippines Thailand Vietnam Japan Singapore Indonesia Malaysia Philippines Thailand Vietnam

– 1.3 5.4 1.9 5.5 13.4 11.4

0.0 – 0.2 0.2 0.2 0.2 0.6

9.6 4.5 – 7.9 3.6 8.3 3.5

8.4 5.1 11.0 – 2.3 7.4 24.8

6.2 4.5 7.8 5.5 – 3.9 19.0

16.8 11.2 15.4 11.4 8.3 – 10.9

17.5 15.3 9.4 18.6 4.4 23.6 –

a tariff equivalent of 409%); it protects its service sector to a relatively lesser extent (22.4% tariff equivalents). In contrast, Japan is fairly open with regard to light manufacturing (7.8% tariff), and its average tariff on high-tech manufactures is only 0.8%. ASEAN is more open in food and agriculture and more protective with regard to manufacturing compared to Japan. This is particularly the case for the automotive sector, where tariff equivalents range from 38% to 48% for Vietnam, Indonesia, Malaysia, and Thailand. So there appears to be a large degree of complementarity between the two regions in terms of the benefits that can accrue from reducing tariffs. An alternative view of the level of protection is provided in Table 12.4, which reports a matrix of trade-weighted, bilateral tariffs across all commodities traded between country pairs. Looking first at Japan’s column, its 1997 tariffs on goods and services from the ASEAN nations ranged from 1.3% for Singapore to 13.4% for Thailand. The top row reports ASEAN tariffs on Japanese exports. Whereas Thailand’s and Vietnam’s tariffs on Japanese exports were relatively high, other ASEAN nations appear to be fairly open, at least as far as the trade-weighted average tariff goes. Note that in our FTA simulation all Japan-ASEAN tariffs are eliminated.9 Clearly there will be a fair amount of Japan-Thailand and Japan-Vietnam trade response on the basis of the relatively large tariffs in place on both sides.

3.3 Baseline Simulation Our policy simulation results are obtained by comparing the counterfactual FTA policy scenario to our baseline. To produce meaningful results, the baseline should reflect as closely as possible the changes in the world 9

Table 12.4 also displays intra-ASEAN tariffs for 1997. As shown in Appendix 12.4, these are reduced in our baseline scenario in the manner prescribed by ASEAN’s Common Effective Preferential Tariff (CEPT) reduction program.

326


economy expected to occur over the period under study: 1997 to 2020. The baseline used in this chapter is built on the baseline developed in Chapter 5. It contains information on macroeconomic variables as well as expected policy changes. The macroeconomic variables in the baseline include projections for real GDP, gross investment, capital stocks, population, skilled and unskilled labor, and total labor. These projected macroeconomic variables were obtained for 211 countries over the period from 1997 to 2020. These projections for population, investment, skilled labor, and unskilled labor were aggregated, and growth rates were calculated to obtain the macro shocks describing the baseline. Changes in capital stocks were not imposed exogenously, but rather were determined endogenously as the accumulation of projected investment. Any changes in real GDP not explained by the changes in endowments are attributed to technological change. In addition, policy projections are also introduced into the baseline (these are summarized in Appendix 12.4). The policies included in the baseline are those that are already agreed on and are legally binding (e.g., Uruguay Round commitments and China’s WTO accession). Uruguay Round tariff commitments are assumed to be honored by all countries. Taiwan (China) and China’s accession to the WTO is phased in over two periods: a period of pre-WTO tariff reduction for 1997–2001 and the period from 2002–20. This accession also gives China and Taiwan (China) quota-free access to the North American and European textile and apparel markets by 2007. However, the liberalization of these quotas is assumed to be heavily back-loaded, with most of the liberalization occurring after 2002. The CEPT preferential tariff reduction program among ASEAN members and the Japan-Singapore FTA have also been incorporated into the baseline.

3.4 Simulation Design Once the baseline has been established, we are able to explore the impact of counterfactual policy simulations. Our simulations of the Japan-ASEAN free trade agreement involve four different model specifications, aimed at identifying the most important potential sources of productivity gain. Simulation (a) involves the complete elimination of tariffs among all countries involved in the Japan-ASEAN FTA (as well as removal of service trade barriers), but does not allow for the three new linkages described earlier. As such, it represents the standard types of effects that a conventional, dynamic AGE model would capture, including allocative efficiency, investment reallocation, and accumulation of capital stocks, as well as terms of trade effects. The remaining three simulations extend the first simulation (a) by adding


327

the three additional modeling effects one at a time. Simulation (b) adds export-productivity effects, simulation (c) adds procompetitive effects, and simulation (d) adds FDI-productivity effects. These additions are cumulative in nature, and therefore simulation (d) includes all three additional effects.

4. Results We begin this section by introducing some shorthand notation regarding the productivity linkages we incorporate into our analysis. In the tables discussed in this section, columns labeled “STD” are meant to represent the difference between the baseline simulation and simulation (a). As such, “STD” refers to the effects normally captured by standard dynamic AGE models, including allocative, investment, and terms of trade effects. Next, “EXP” is the difference between simulations (a) and (b), and it captures productivity effects related to the potential expansion of export-oriented firms under an FTA. “IMP” is the difference between simulations (b) and (c), and it captures procompetitive effects related to the exposure of local, imperfectly competitive firms to foreign competition. Finally, “FDI” is the difference between simulations (c) and (d), and it refers to productivity effects related to foreign investment in local firms.

4.1 Welfare and GDP Impacts of the FTA Table 12.5 reports regional welfare changes in the year 2020 resulting from the Japan-ASEAN free trade agreement. “Welfare” is defined as the percentage change in utility of a representative regional household in 2020 owing to the FTA. Consider first the change in welfare with all effects in place (i.e., the results of simulation [d]). These are reported in the “Total” column of Table 12.5. All of the member nations experience an increase in welfare relative to the baseline. In relative terms, Thailand has the most to gain from a Japan-ASEAN FTA, with a welfare level that is 3.32 percentage points above the baseline scenario. For ASEAN nations as a whole, the welfare gain is 1.04 percentage points over the baseline, with Japan having a lower figure of 0.23 percentage points. The nations that face relatively low barriers in Japan before the FTA, including Indonesia, the Philippines, and particularly Singapore (Table 12.4), tend to experience smaller improvements in welfare (0.26, 0.24, and 0.46 percentage points, respectively). We can also examine how these results would differ had we not incorporated the additional productivity-linkage effects. Refer to the following

328

Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer Table 12.5. Overall welfare effects, 2020

Region

Total

STD (% of total)

EXP

IMP

FDI

ASEAN Indonesia Malaysia Philippines Thailand Vietnam Singapore Japan

1.04 0.26 0.47 0.24 3.32 0.62 0.46 0.23

0.26 (25) 0.11 (42) −0.06 (−12) 0.13 (54) 0.82 (25) 0.22 (36) 0.15 (33) 0.20 (86)

0.05 0.04 0.06 0.02 0.10 0.01 0.01 0.01

0.35 0.07 0.32 0.07 1.06 0.25 0.14 0.01

0.38 0.05 0.14 0.03 1.34 0.14 0.16 0.02

Note: These values represent percentage point differences from the baseline scenario in 2020. Figures in parentheses are percentage contribution of standard AGE model effects to overall welfare change. Abbreviations are as follows: standard AGE modeling effects (STD), export-productivity effect (EXP), procompetitive effect (IMP), FDI-productivity effect (FDI).

columns in Table 12.5: “STD” (standard AGE modeling effects resulting from the tariff cuts), “EXP” (the export-productivity effect), “IMP” (the procompetitive effect), and “FDI” (the FDI-productivity effect). In terms of utility, the contributions of these extra effects are generally significant. In fact, we see that only 25% of the welfare change in ASEAN is related to standard AGE modeling effects (see the values within parentheses in the STD column). This figure is higher for Japan (86%), because of the absence of procompetitive and FDI effects for that country. Thailand shows the largest overall relative gains. In that country, the most important channel for welfare change is the FDI-productivity effect (1.34 out of 3.32 percentage points), in which higher levels of foreign ownership following the FTA led to improvements in domestic firm productivity. We now move on to other macroeconomic results presented in Table 12.6. As in Table 12.5, these changes are given as percentage point differences from the baseline, allowing us to gauge differences in relative terms. We first focus on the total change in GDP (the other variables in Table 12.6 are discussed in later sections). Although Japan’s 2020 GDP is only 0.14 percentage points higher than in the baseline scenario, ASEAN’s overall change is significantly higher (3.66 percentage points), with Thailand having the largest change by far (12.41 percentage points).10 In Thailand, most of the change is due to conventional AGE modeling effects (STD), followed by the FDI-productivity effect and the procompetitive effect. This is because 10

The small change in Singapore’s GDP (0.18 percentage points) reflects the fact that it has already formed an FTA with Japan and thus does not benefit to the extent that other ASEAN nations do.


329

Table 12.6. Effect of Japan-ASEAN FTA on selected macro variables, 2020 Effect

GDP

Imports

Exports

Capital

GDP

Japan Total STD EXP IMP FDI

0.14 0.15 0.01 −0.01 0.00

2.83 2.73 0.02 0.02 0.06

0.53 0.33 0.06 0.08 0.06

1.27 1.10 0.07 0.04 0.05

1.54 1.54 0.02 −0.03 0.02

0.27 0.29 0.01 −0.02 −0.01

3.66 2.06 0.08 0.77 0.75

1.56 0.73 0.12 0.47 0.24

2.46 1.91 0.07 0.20 0.29

3.28 2.92 0.12 0.13 0.11

1.23 0.93 0.10 0.11 0.08

0.99 0.83 0.03 0.10 0.04

3.36 2.59 0.09 0.34 0.35

0.18 −0.10 −0.02 0.18 0.12

0.99 0.38 0.01 0.30 0.30

0.58 0.19 0.00 0.22 0.17

4.03 2.79 0.07 0.43 0.73

6.24 4.46 0.07 0.98 0.73

5.71 3.61 0.08 1.08 0.93

2.35 2.21 0.04 0.06 0.05

2.84 2.62 0.04 0.14 0.04

2.23 1.93 0.03 0.19 0.07

Thailand 2.26 1.26 0.14 0.55 0.30

12.41 6.96 0.15 2.56 2.75

Singapore Total STD EXP IMP FDI

Capital

Philippines

Malaysia Total STD EXP IMP FDI

Exports

ASEAN

Indonesia Total STD EXP IMP FDI

Imports

15.44 10.15 0.21 1.77 3.31

23.96 16.44 0.09 4.11 3.33

16.22 10.19 0.12 3.03 2.88

Vietnam 0.30 −0.23 −0.01 0.32 0.21

2.72 2.04 0.06 0.50 0.12

4.93 4.98 −0.04 −0.45 0.44

11.11 8.27 0.43 3.22 −0.81

4.62 3.10 0.01 1.28 0.23

Note: These values represent percentage point differences from the baseline scenario in 2020.

Japan has notably high overall tariffs with respect to Thailand, and Thailand also displays relatively high tariffs with respect to Japan (Table 12.4). Figure 12.1 offers a temporal perspective regarding the changes in Thailand’s GDP. Here, deviations from the baseline attributed to each effect are provided separately. Begin by looking at the year 2006, the first year of the prospective FTA. There we see that without the procompetitive effect (IMP), we would have underestimated Thailand’s GDP change from the baseline by 0.8 percentage points. Although initially the procompetitive effect is the most important driver of Thailand’s GDP difference, by 2007 conventional AGE effects (STD) in the form of added investment take over as the most important contributor. Also observe that the FDI-productivity link (FDI) is

330


8.0

Percentage points difference from baseline

7.0

6.0 5.0

4.0

3.0 2.0

1.0 0.0 2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

2019

2020

-1.0 STD

EXP

IMP

FDI

Figure 12.1. Effect of Japan-ASEAN FTA on GDP, Thailand, relative to baseline.

unimportant in the first several years after implementation of the FTA, but continuously grows in importance along with the increased foreign investment until it is the second largest contributor to the growth in GDP by 2020. The sum contribution of all three additional productivity effects to GDP is 5.45 percentage points over Thailand’s baseline level in 2020, compared to 6.96 percentage points from the standard effects alone. On this basis it would appear that the productivity effects that are normally ignored in AGE analysis may indeed be important in the analysis of the Japan-ASEAN FTA, although one must bear in mind that these effects were somewhat more pronounced in Thailand than in the other nations (recall Table 12.6).

4.2 Effects on Trade and Foreign Capital Ownership Looking back at Table 12.6 we see that a Japan-ASEAN FTA results in higher overall imports and exports for all ASEAN nations, as well as Japan. Thailand has the largest increases: 15.44 and 23.96 percentage points over the baseline for imports and exports, respectively. In ASEAN most of the changes in trade volumes are due to conventional AGE modeling effects (STD). Thus relatively little is missed by ignoring effects on productivity arising from increased exports, imports, and foreign ownership of firms. Table 12.7 presents a sectoral decomposition of the changes in JapanASEAN trade resulting from the FTA. The upper half reports the differences over the baseline regarding exports from ASEAN to Japan.11 Not 11

Of course, changes involving ASEAN exports to Japan coincide exactly with changes involving Japanese imports from ASEAN.


331

Table 12.7. Change in trade volume between ASEAN and Japan due to

Japan-ASEAN FTA, 2020 Total Food and ag. Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products Automotive Machinery Electrical equipment Services

Food and ag. Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products

STD

EXP

IMP

FDI

5,064 (77.5) −613 (−4.0) 1,458 (73.3) 709 (159.0) 1,149 (16.2) 359 (14.3) 635 (18.8) 261 (45.3) 2,789 (24.7) 4,537 (15.7) 1,206 (11.6)

Exports from ASEAN to Japan 5,250 −13 −73 (80.3) (−0.2) (−1.1) −512 −9 −51 (−3.3) (−0.1) (−0.3) 1,362 13 50 (68.5) (0.7) (2.5) 687 3 19 (153.9) (0.6) (4.2) 1,087 19 19 (15.4) (0.3) (0.3) 221 3 114 (8.8) (0.1) (4.5) 476 16 109 (14.1) (0.5) (3.2) 157 1 90 (27.3) (0.2) (15.6) 2,234 51 238 (19.8) (0.5) (2.1) 3,327 25 539 (11.5) (0.1) (1.9) 1,151 −5 114 (11.1) (−0.0) (1.1)

−99 (−1.5) −41 (−0.3) 34 (1.7) 1 (0.3) 25 (0.3) 21 (0.8) 33 (1.0) 12 (2.1) 266 (2.4) 645 (2.2) −54 (−0.5)

532 (52.9) 170 (21.3) 1,264 (187.5) 32 (118.0) 550 (62.7) 12,018 (33.0) 7,524 (47.8)

Exports from Japan to ASEAN 491 1 18 (48.9) (0.1) (1.8) 132 1 19 (16.5) (0.1) (2.4) 1,258 4 12 (186.5) (0.6) (1.7) 32 0 0 (116.3) (0.2) (1.2) 541 −1 5 (61.7) (−0.1) (0.6) 11,926 32 −161 (32.7) (0.1) (−0.4) 6,972 24 184 (44.3) (0.2) (1.2)

22 (2.2) 17 (2.2) −10 (−1.4) 0 (0.3) 4 (0.5) 221 (0.6) 343 (2.2) (continued)

332

Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer Table 12.7 (continued)

Automotive Machinery Electrical equipment Services

Total

STD

EXP

IMP

FDI

8,850 (70.1) 11,474 (36.5) 3,446 (18.1) 312 (8.2)

8,993 (71.3) 10,690 (34.0) 3,245 (17.1) 258 (6.8)

15 (0.1) 56 (0.2) 2 (0.0) 2 (0.1)

−348 (−2.8) 243 (0.8) 106 (0.6) 12 (0.3)

191 (1.5) 485 (1.5) 93 (0.5) 40 (1.0)

Notes: Values represent differences in 1997 US$ millions from baseline scenario in 2020. Values in parentheses represent the corresponding percentage point differences.

surprisingly, the FTA leads to a great deal more trade in nearly every category. In relative terms, the biggest boost comes from increased exports of leather products (159 percentage points over the baseline, or $709 million). In absolute terms, the biggest trade increase is in food and agricultural products, which increase by $5,064 million (77.5 percentage points) over the baseline, followed by electrical equipment ($4,537 million, 15.7 percentage points) and machinery ($2,789 million, 24.7 percentage points). In all of these cases the increases are related mainly to standard AGE modeling effects (STD). Exports from Japan to ASEAN are given a big boost in general (lower half of Table 12.7). In absolute terms, the largest change derives from an increase in chemical exports by $12,018 million (33.0 percentage points). In relative terms, exports of textile and apparel products have the largest increase, at 187.5 percentage points ($1,264 million). In both cases, the majority of the change results from standard effects relating to the tariff cuts. This is related to the fact that Thailand and Vietnam had particularly large initial tariffs in these particular sectors (Table 12.3). (For a disaggregation of Table 12.7, see also Appendixes 12.2 and 12.3.) Recall from Table 12.6 that implementation of the Japan-ASEAN FTA results in higher capital stocks for all the countries involved in the FTA. For these increased capital stocks, it is of interest to focus on the change in foreign ownership because it is hypothesized to drive the efficiency gain. Table 12.8 reports the change in share of foreign capital ownership by 2020, compared to the baseline. A free trade agreement between Japan and ASEAN attracts investment from abroad to all the countries involved in the FTA, resulting in a higher share of foreign capital ownership. For Thailand the share of capital


333

Table 12.8. Change in share of foreign capital ownership, 2020 (percent

of total capital ownership)

Indonesia Malaysia Philippines Thailand Vietnam Singapore Japan

Total

STD

EXP

IMP

FDI

0.003 0.05 0.82 5.29 1.40 0.03 0.07

0.002 0.03 0.71 3.15 0.98 −0.08 0.08

0.001 0.00 0.01 0.05 0.00 0.00 0.00

0.00 0.01 0.08 0.96 0.36 0.06 −0.01

0.00 0.01 0.03 1.14 0.07 0.04 0.00

owned by foreign investors increases by 5.29 percentage points, relative to the baseline. The figure for Vietnam is 1.4 percentage points, whereas the remaining changes in foreign ownership share are all less than 1%. The conventional AGE modeling effects (STD) account for the majority of the increase, but the procompetitive effects (IMP) and FDI-productivity effects (FDI) also contribute a considerable amount, particularly in Thailand and Vietnam.

4.3 Effects on Efficiency Table 12.9 reports the sectoral efficiency gains in the manufacturing sectors for ASEAN countries by 2020. These values reflect the combined impact of the export-productivity, procompetitive, and the FDI-productivity effects (they are decomposed in Appendix 12.3). Here we see that the automotive industries of Thailand and Vietnam have the largest gains, at 4.50 and 3.44 percentage points over the baseline, respectively. Recall in Table 12.1 that the self-sufficiency ratios for these sectors were well below 95%, so the procompetitive effects were active and indeed account for the largest source of efficiency gain in these cases (Appendix 12.3). Although most ASEAN manufacturing industries attain higher efficiency levels due to the Japan-ASEAN FTA, some countries have sectors for which the agreement has no impact. Interestingly, there are even slight reversals in efficiency in a few cases. This happens, for example, in Vietnam’s electrical equipment sector (Table 12.9). In this case, it is the import-productivity (IMP) linkage coupled with the FDI-productivity linkage that gives rise to the technology reversal. Following the FTA, both a drop in foreign investment in the Vietnamese electrical equipment sector and a reorientation of the existing firms toward the domestic market contribute to a slight loss in overall sectoral productivity.

334

Ken Itakura, Thomas W. Hertel, and Jeffrey J. Reimer Table 12.9. ASEAN sectoral efficiency gains, 2020 (%)

Sector

Thailand Indonesia Malaysia Philippines Vietnam Singapore

Textile/apparel Leather Paper/wood prod. Chemical products Metal products Automotive Machinery Electrical equipment

0.86 0.93 0.83 2.40 1.33 4.50 0.73 0.94

0.08 0.08 0.15 0.32 0.43 0.73 0.22 0.21

0.15 0.08 0.19 0.78 0.27 2.02 0.11 0.10

0.05 0.08 0.28 0.20 0.16 0.78 0.12 0.02

0.06 −0.02 0.04 0.50 0.00 3.44 0.14 −0.06

−0.01 0.02 0.13 −0.05 0.04 −0.02 0.04 0.00

Note: These changes are due solely to the EXP, IMP, and FDI effects.

4.4 Effects on Sectoral Output Finally, we move on to consider changes in sectoral output relating to the Japan-ASEAN FTA. Table 12.10 provides this information for ASEAN in both absolute and relative terms, for the final year of the simulation, 2020. In general, there are output increases in every sector in ASEAN. Electrical equipment shows the largest increase in output over the baseline: $44,451 million, or 10.2 percentage points under the total column. Moving across the columns of Table 12.10, we see that more than half of this increase is related to conventional AGE modeling effects (almost $30 billion, 6.85 percentage points), with the procompetitive effect contributing a difference of $6.4 billion from the baseline (1.46 percentage points), and the FDIproductivity effect contributing $7.9 billion (1.8 percentage points). Chemical products and the automotive sector offer two interesting cases. Here, we observe negative impacts under the standard AGE closure (STD): − $568 million or − 0.32 percentage points, and − $1,805 million or − 2.95 percentage points, respectively). In both of the sectors, the procompetitive effects (IMP) together with the FDI-productivity effects (FDI) are positive and large enough to reverse the overall output changes. Although these additional effects are important particularly for chemical products and the automotive sector, it is nevertheless the conventional AGE effects related to Japan’s tariffs that are the most important reason for the increase in output for most of the sectors. Outside of the manufacturing sectors, output by ASEAN’s food and agricultural sector grows by $3.6 billion (0.96 percentage points) over the baseline, and the corresponding value for the service sector is $32.7 billion (2.04 percentage points) over the baseline. ASEAN’s natural resources sectors have much smaller changes. As with the manufacturing sectors, it is


335

Table 12.10. Sectoral output changes in ASEAN, 2020

Food and ag. Nat. resources Textiles/apparel Leather Paper/wood Chemical products Metal products Automotive Machinery Electrical equipment Services

Total

STD

EXP

IMP

FDI

3,595 (0.96) 311 (0.22) 6,436 (9.14) 1,515 (12.50) 4,051 (3.92) 4,226 (2.38) 4,673 (5.90) 2,228 (3.64) 24,300 (10.45) 44,451 (10.20) 32,702 (2.04)

3,539 (0.95) 228 (0.16) 4,505 (6.40) 1,199 (9.89) 2,635 (2.55) −568 (−0.32) 1,839 (2.32) −1,805 (−2.95) 15,632 (6.72) 29,865 (6.85) 19,083 (1.19)

2 (0.00) 14 (0.01) 84 (0.12) 20 (0.17) 145 (0.14) 185 (0.10) 195 (0.25) 163 (0.27) 668 (0.29) 347 (0.08) 714 (0.04)

143 (0.04) 46 (0.03) 824 (1.17) 185 (1.53) 570 (0.55) 3,127 (1.76) 1,622 (2.05) 2,445 (4.00) 3,495 (1.50) 6,380 (1.46) 6,540 (0.41)

−89 (−0.02) 23 (0.02) 1,024 (1.45) 111 (0.91) 701 (0.68) 1,483 (0.83) 1,016 (1.28) 1,425 (2.33) 4,505 (1.94) 7,860 (1.80) 6,364 (0.40)

Notes: Values represent differences in 1997 US$ millions from baseline scenario in 2020. Values in parentheses represent the corresponding percentage point differences.

generally the conventional AGE effects (STD) that drive most of the changes in output related to implementation of the FTA. (For a disaggregation of Table 12.10, see also Appendix 12.4.)

5. Conclusions AGE models are extensively used in the evaluation of FTAs, but they have often underpredicted the increases in trade and economic growth that followed FTA implementation (Kehoe 2002). Meanwhile, there have been a surge of new empirical trade studies demonstrating strong correlations between firm productivity, on the one hand, and exporting, importing, and investment, on the other. Because increasing these flows is a key objective of most FTAs, this raises the question: Might these additional productivity linkages have a significant impact on AGE-based analyses of FTAs? To test this hypothesis, we generalize the GDyn model to allow for the

336


productivity-enhancing effects of import competition, increased exports, and FDI-productivity linkages. We then incorporate the best econometric evidence currently available and proceed to examine the impact of the Japan-ASEAN FTA, signed in 2008. In general, we find that this FTA will result in increases in trade for most sectors of the countries involved and that the welfare of all participating countries will improve. By far the largest proportional gains accrue to Thailand, which currently has rather high bilateral tariffs on its trade with Japan. Importantly, we find that the effects normally captured by standard AGE models still play a key role in driving the results. Our conventional, dynamic AGE model captures more than half of the ensuing GDP and trade changes. Overall, we find that the procompetitive and FDI-productivity linkages were the most important, with the export-productivity linkage playing a minor role. These added effects generally serve to reinforce the direction predicted by the standard AGE model. However, addition of the procompetitive effects does lead to aggregate output increasing instead of falling in the case of the two most imperfectly competitive sectors in the ASEAN region: chemicals and automobiles. Therefore, further refinements of the associated econometric estimates would be very worthwhile. We can think of several ways that our results could be changed by future research. For example, the elasticity of productivity response to FDI employed was only 1.4%, and the estimate concerning the higher productivity of exporting firms was only 8%. It seems likely that these figures may be higher for the specific countries examined in this study, particularly those within ASEAN. Future econometric research concerning these parameters for the specific countries examined would facilitate the analysis of FTAs using the framework developed in this chapter. Additionally, sensitivity analysis concerning these parameters (perhaps based on econometric standard errors) would also aid in the progression of this literature.

APPENDIX 12.1: DERIVATION OF THE EXPORTS-PRODUCTIVITY EXPRESSION We begin with the identity: A O Q O ≡ A D Q D + A X Q X , where AO is an index of an industry’s technology, AD represents technology used by local market firms, and AX is technology used by export firms. QO, QD , and QX indicate the total output, the output for the domestic market, and output for export. Normalize such that A D ≡ 1, and let δ ≡ A X /A D . We then rewrite the above identity as A O Q O = Q D + δQ X . We totally differentiate to get


337

A O dQ O + Q O dA O = dQ D + δdQ X . Divide through by A O Q O , let s D ≡ Q D /Q O and s X ≡ Q X /Q O , and multiply both sides by 100 to get qˆ D s D qˆ X s X ˆq O + aˆ O = +δ , (12.A1) AO AO where the lowercase symbols with hats refer to percentage changes (e.g., qˆ O = (dQ O /Q O ) × 100%). Based on the earlier identity we can also derive A O = s D + δs X . Using this we can restate (12.A1): aˆ O =

s D qˆ D + δs X qˆ X − qˆ O . s D + δs X

(12.A2)

Using the identity Q O ≡ Q D + Q X we can totally differentiate and show that qˆ O = s D qˆ D + s X qˆ X , which can be plugged into (12.A2). With algebraic manipulation and the fact that s D + s X = 1, we obtain the equation for the rate of change in overall productivity as a function of the productivity differential between exporters and domestic firms and the differential growth in these two markets for output: efficiency = aˆ O =

(δ − 1)s X s D (qˆ X − qˆ D ) . s D + δs X

(12.A3)

APPENDIX 12.2: AGGREGATION OF GTAP 5 DATA BASE SECTORS No.

Sectors in this study

57 GTAP sectors

1 2 3

Rice Grains Othcrops

4

Meat

5

Othfood

Paddy rice, Processed rice Wheat, Cereal grains nec Veg., fruit, nuts; Oil seeds; Sugar; Fibers; Crops nec; Wool, silk-worm cocoons Cattle, sheep, goats; Animal products nec; Meat products nec Raw milk; Veg. oils and fats; Dairy; Sugar; Food products nec; Bev & tobacco Forestry Fishing Coal, Oil, Gas, Minerals nec Textiles, wearing apparel Leather products Wood products; Paper products, publishing Petroleum, coal products; Chemical, rubber, plastic prods; Mineral products nec

6 7 8 9 10 11 12

Forestry Fish Extract Texwap Leather Paperwood Chemicals

(continued)

338


APPENDIX 12.2 (continued) No.

Sectors in this study

13 14 15

Metals Autos Machinery

16 17

Electrequip Othservice

18 19 20 21 22 23

Construction Trade Transport Comm Insfinance Pubservice

57 GTAP sectors Ferrous metals; Metals nec; Metal products Motor vehicles and parts Transport equipment nec, Machinery and equipment nec, Manufactures nec Electronic equip. Electricity; Gas; Water; Business services nec; Recr. and oth. services; Dwellings Construction Wholesale/retail trade Transport nec, Sea transport, Air transport Communication Financial services nec, Insurance PubAdmin/Defence/Health/Educat

Notes: “Food & agriculture” is 1–5, “Natural resources” is 6–8, “Light manufactures” is 9–11, and “High tech” is 12–16.

APPENDIX 12.3: AGGREGATION OF GTAP 5 DATA BASE REGIONS No.

Regions

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Japan Korea Malaysia Philippines Indonesia Vietnam Thailand Singapore Taiwan HongKong China USA Canada Mexico AusNzl CSAmerica

66 GTAP regions Japan Korea Malaysia Philippines Indonesia Vietnam Thailand Singapore Taiwan (China) Hong Kong China USA Canada Mexico Australia, New Zealand Central Am., Carib, Colombia, Peru, Venezuela, Argentina, Brazil, Chile, Uruguay, Rest of South America


No.

Regions

66 GTAP regions

17

WEuro

18

SAsia

19

ROW

Austria, Belgium, Denmark, Finland, France, Germany, United Kingdom, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, Switzerland, Rest of EFTA Bangladesh, India, Sri Lanka, Rest of South Asia Hungary, Poland, Rest of Cent. Eur., Former S.U., Turkey, Rest of Mid-East, Moroc., Rest of N. Africa, Bots., Rest of SACU, Malawi, Moz., Tanz., Zam., Zimb., Other S. Africa, Uganda, Rest of Sub-Saha. Afr., Rest of World

339

APPENDIX 12.4: BASELINE POLICY SHOCKS Period 1997–2000

1997–2005

2002–2007

Import tariff adjustments 1. UR tariff reductions for all regions except China and Taiwan (China) (no shocks to agriculture) 2. Pre-WTO tariff reductions undertaken by China before 2002 ASEAN’s Common Effective Preferential Tariff (CEPT) reduction program (1997–2003) and Japan-Singapore Free Trade Agreements (2002) UR tariff reductions for all regions; Taiwan (China) and China’s WTO agreement included (no shocks to agriculture, except for Taiwan (China) and China

Export tax adjustments

USA and EU quotas increased on exports of textiles and wearing apparel for all regions except Taiwan (China) and China

USA and EU quotas increased on exports of textiles and wearing apparel for all regions including Taiwan (China) and China

Notes: Japan-Singapore FTA and CEPT are added to the baseline originally developed in Chapter 5. Their study is otherwise the source that should be consulted concerning the baseline.

340


References ASEAN. 2002. Unemployment. Available from www.aseansec.org. Aw, B. A., S. Chung, and M. J. Roberts. 2000. “Productivity and Turnover in the Export Market: Micro-Level Evidence from the Republic of Korea and Taiwan (China).” World Bank Economic Review 14(1), 65–90. Bernard, A. B. and J. B. Jensen. 2001. “Exporting and Productivity: The Importance of Reallocation.” Unpublished manuscript. Blalock, G. 2001. “Technology from Foreign Direct Investment: Strategic Transfer through Supply Chains.” Unpublished manuscript, University of CaliforniaBerkeley. Brooke, J. 2002. “Rumor’s of Japan’s Recovery Are, It Seems, Exaggerated.”New YorkTimes, Dec. 16. Chuang, Y. C. and C. M. Lin. 1999. “Foreign Direct Investment, R&D and Spillover Efficiency: Evidence from Taiwan’s Manufacturing Firms.” Journal of Development Studies 35(April), 117–37. Devarajan, S. and D. Rodrik. 1991. “Procompetitive Effects of Trade Reform: Results from a CGE Model of Cameroon.” European Economic Review 35, 1157–84. Dimaranan, B. V. and R. A. McDougall. 2002. Global Trade, Assistance, and Production: The GTAP 5 Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Dixit, A. K. and J. E. Stiglitz. 1977. “Monopolistic Competition and Optimum Product Diversity.”American Economic Review 67, 297–308. The Economist. 2002. “Unemployment.” Available from www.economist.com. Francois, J. F. 1998. Scale Economies and Imperfect Competition in the GTAP Model. GTAP Technical Paper No. 14. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Francois, J. F. and D. W. Roland-Holst. 1997. “Scale Economies and Imperfect Competition.” In J. Francois and K. Reinert (eds.), Applied Methods for Trade Policy Analysis: A Handbook (pp. 331–63). New York: Cambridge University Press. Francois, J. F. and C. R. Shiells. 1994. “AGE Models of North American Free Trade.” In J. F. Francois and C. R. Shiells (eds.), Modeling Trade Policy: Applied General Equilibrium Assessments of North American Free Trade (pp. 3–44). New York: Cambridge University Press. Hallward-Driemeier, M., G. Iarossi, and K. Sokoloff. 2002. Exports and Manufacturing Productivity in East Asia: A Comparative Analysis of Firm-Level Data. NBER Working Paper No. 8894. Washington, DC: NBER. Harris, R. G. 1984. “Applied General Equilibrium Analysis of Small Open Economics with Scale Economics and Imperfect Competition.” American Economic Review 74, 1016–32. Hertel, T. W. 1994. “The ‘Procompetitive’ Effects of Trade Policy Reform in a Small, Open Economy.” Journal of International Economics 36, 391–411. Hertel, T. W. (ed.). 1997. Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press. Hertel, T.W., T. L. Walmsley, and K. Itakura 2001. “Dynamic Effects of the New Age Free Trade Agreement between Japan and Singapore.” Journal of Economic Integration 16(4), 446–84.


341

Ianchovichina, E. I. 1994. The Procompetitive Effects of Foreign Competition on Optimal Markups in the Presence of Imperfect Competition. M.S. thesis, Purdue University. Ianchovichina, E., J. Binkley, and T. W. Hertel. 2000. “Procompetitive Effects of Foreign Competition on Domestic Markups.” Review of International Economics 8(1), 134–48. Kehoe, T. J. 2002. An Evaluation of the Performance of Applied General Equilibrium Models of the Impact of NAFTA. Research Department Staff Report. Minneapolis: Federal Reserve Bank of Minneapolis. Levinsohn, J. 1993. “Testing the Imports-as-Market-Discipline Hypothesis.” Journal of International Economics 35(1/2), 1–22. Markusen, J. R. 1981. “Trade and the Gains for Trade with Imperfect Competition. Journal of International Economics 11, 531–51. Markusen, J. R. and A. J. Venables. 1988. “Trade Policy with Increasing Returns and Imperfect Competition: Contradictory Results from Competing Assumptions.” Journal of International Economics 24, 299–316. Rodrik, D. 1999. The New Global Economy and Developing Countries: Making Openness Work. Overseas Development Council Policy Essay No. 24. Washington, DC: Johns Hopkins Press. Sj¨oholm, F. 1999. “Exports, Imports, and Productivity: Results from Indonesian Establishment Data.”World Development 27(4), 705–15. World Trade Organization (WTO). 2002. “Regionalism.” Available from http://www. wto.org.

THIRTEEN

Global Demographic Change, Labor Force Growth, and Economic Performance Rod Tyers and Qun Shi

1. Introduction Recent changes in global demographic behavior, including to fertility, mortality, migration, and the sex ratio at birth, have been considerable and many were not widely anticipated in recent decades. In most countries, consistent with the central phase of the global demographic transition, infant mortality fell through the course of the last century and adult life expectancy increased, causing a surge in population growth. The declines in birth rates as part of the final phase of this transition have been particularly sharp, first in developed countries and recently in many developing countries.1 Before this century is half over, populations in Japan and some European countries are likely to be smaller than they were in 1990, with these declines in total populations being preceded by declines in the number and proportion of people of working age.2 The economic implications of these demographic trends and uncertainties are the subject of an already substantial global literature.3 Recent 1 2 3

IMF (2004: Chapter 3), Lee (2003). Bryant and McKibbin (1998), United Nations (2003). At minimum, this literature spans demography (Booth et al. 2002; McDonald and Kippen 2001), population economics (Lee 2003; Mason 2003), public economics (OECD 1996, 1998), economic history (Bloom and Williamson 1997; Williamson 1998), and growth economics (Barro and Becker 1989).

Funding for the research described in this chapter is from Australian Rural Industries Research and Development Corporation Research Contract No. ANU-51A and from Australian Research Council Discovery Grant No. DP0557889. Thanks are due to Heather Booth, Siew Ean Khoo, and Ming Ming Chan for helpful discussions about the demography; to Jeff Davis, Brett Graham, Ron Duncan, Robert McDougall, and Hom Pant for their constructive comments on the economic analysis; and to Terrie Walmsley for technical assistance with the GTAP Data Base, as well as useful discussions on the subject of baseline simulations. Useful comments were also received at the Conference on Global Economic Analysis in June 2005 from Dominique van der Mensbrugghe and Nico van Leeuwen. Jahnvi Vedi and Jyoti Pant provided research assistance in the later stages of the study.

342

Global Demographic Change

343

macroeconomic studies of demographic change have been global in scope, emphasizing the effects of aging on average saving rates and financial flows (Bryant and McKibbin 1998, 2001; Bryant et al. 2003; Faruqee and Muhleisen 2002).4 This work has clearly demonstrated the substantial implications of demographic change in some regions for economic performance in others. It has, however, fallen short of the complete demographic modeling needed to capture the three principal avenues through which demographic change influences economic performance: labor force growth,5 average saving rates, and age-specific consumption variation. This chapter examines the economic implications of population change using a complete demographic model on 14 regions, which is constructed as integral with a dynamic model of the global economy. The latter model is a development of GDyn in which regional households are disaggregated by age group and gender. Our explicit incorporation of the demographic submodel allows age-gender distributions, migration flows, and, therefore, labor force participation rates, migration rates, average saving rates, and the age-gender effects on consumption to be endogenized. It is constructed around a baseline projection through 2030 in which populations and labor forces are projected to decline in Europe and Japan and to begin declining before the end of this period in China and elsewhere in East Asia. Notably, as age distributions change, the trends in labor forces are shown to diverge substantially from those in total populations. The behavior of the model is then illustrated by considering the effects of one alternative demographic scenario. In this case, we imagine that continued improvements in public health and medical science cause life expectancy beyond 60 to grow faster than anticipated in all regions of the world. This scenario causes significant departures from the baseline and has important implications for overall economic performance. In models of the Solow-Swan type, where endogenous growth takes the form of physical capital accumulation, the decelerating population growth that is prominent in our baseline projection reduces the GDP growth rate but raises that of real per capita income. Multiple trading regions 4

5

Much of this research was organized under a project coordinated by the Brookings Institution in the United States and involving staff from the International Monetary Fund. Financing is from the Economic and Social Research Institute of the Japanese Cabinet Office. Capturing the demographic influences on the labor force change requires consideration of age- and gender-specific participation rates and rates of part-time employment, as well as age- and gender-specific migration rates.

344

Rod Tyers and Qun Shi

notwithstanding, this tendency is prominent in our model, too. There are two complicating effects of population deceleration, however. First, age distributions change so that the average age rises, which alters the pattern of international financial flows and the distribution of global investment. In addition, because consumption preferences vary with age, the pattern of consumption also changes. Second, on the supply side, changes in the size and composition of the population correspondingly change the size and composition of the labor force, although as we show, these changes are most often far from proportionate. Increased longevity complicates this picture by both accelerating the aging process and raising population and labor force growth. Although increased longevity raises aged dependency everywhere, it tends to attract new investment to those regions whose elderly have high labor force participation rates. In Section 2, we introduce the demographic submodel and discuss its population and labor force projections. We describe the extension of the GDyn model to incorporate populations disaggregated by age and gender and then the full demographic submodel. Section 3 discusses the construction of the baseline scenario. The “accelerated aging” scenario is described in Section 4, and the implications for the performance of the global economy are quantified through a comparison of its simulation results with the baseline. Section 5 offers brief concluding remarks.

2. Modeling Global Demographic Change The approach adopted follows Tyers et al. (2005), in that it applies a complete demographic submodel that is integrated within a dynamic numerical model of the global economy.6 The economic model is a development of GDyn, the standard version of which has single households in each region and therefore no demographic structure.7 The version used has regional households that are disaggregated by age group, gender, and skill level.

2.1 Demography The demographic submodel tracks populations in four age groups and two genders, yielding a total of eight population groups in each of 14 regions.8 6 7 8

See also Shi and Tyers (2004) and Tyers et al. (2005). Earlier applications of the GDyn model to the issues raised in this paper include those by Shi and Tyers (2004) and Duncan, Shi and Tyers (2005). The demographic submodel has been used in a stand-alone mode for the analysis of trends in dependency ratios. For a more complete documentation of the submodel, see Chan and Tyers (2006).


345

The four age groups are the dependent young, adults of fertile and working age, older working adults, and the mostly retired over 60s. The resulting age-gender structure is displayed in Figure 13.1. The population is further divided between households that provide production labor and those providing professional labor.9 Each age-gender-skill group is a homogeneous subpopulation with group-specific birth and death rates and rates of both immigration and emigration.10 If the group spans T years, the survival rate to the next age group is the fraction 1/T of its population, after groupspecific deaths have been removed and its population has been adjusted for net migration. The final age group (60+) has duration equal to measured life expectancy at 60, which varies across genders and regions. The key demographic parameters, then, are birth rates, sex ratios at birth, age- and gender-specific death, immigration and emigration rates, and life expectancies at 60.11 Another key parameter is the rate at which each region’s education and social development structure transforms production worker families into professional worker families. Each year a particular proportion of the population in each production worker age-gender group is transferred to professional status. These proportions depend on the regions’ levels of development, the associated capacities of their education systems, and the relative sizes of the production and professional labor groups. In any year, for each age group a, gender group g, skill group s, region of origin r, and region of destination d, the volume of migration flow is t t R t Ma,g ,s,r,d = δd M a,g ,s,r,d N a,g ,s,d ,

∀a, g , r, d,

(13.1)

where δtd is a destination-specific factor reflecting immigration policy in R region d, set to unity in all but counterfactual simulations, and Ma,g ,s,r,d is the migration rate between r and d expressed as a proportion of the group population in region d, Na,g ,s,d . Given the migration matrix, Ma,g ,s,r,d , the population in each age, gender, and skill group and region can be constructed. We begin with the population 9

10

11

The subdivision between production and professional labor accords with the ILO’s occupation-based classification and is consistent with the labor division adopted in the GTAP Data Base. See Liu et al. (1998). Mothers in families providing production labor are assumed to produce children who will grow up to also provide production labor, whereas the children of mothers in professional families are correspondingly assumed to become professional workers. Immigration and emigration are also age and gender specific. The model represents a full matrix of global migration flows for each age and gender group. Each of these flows is currently set at a constant proportion of the population of its destination group. See Chan and Tyers (2006) and Vedi (2005) for further details.

346

Rod Tyers and Qun Shi Female population

Male Population D

D

Mo

Mo

Aged >60

Mi

Aged >60

Mi

S S D Working 40-60

Mo

D Working 40-60

Mi

S Mo

D

Mi

Working fertile 15-40

B

S D Mi

Deaths Survival Births Immigration Emigration Sex ratio at birth Figure 13.1. The demographic submodel.

Mo

D

Young 0-15

Glossary:

Mo

Mi

Working fertile 15-40

S

Mo

SRB

D S B Mi Mo SRB

Mi

S D

Young 0-15

Mo

Mi


347

of males aged 0–14 from professional families in region d (a = 014, g = m, s = sk, r = d). S dt B t N t−1 1 + S dt sk,d 1539,m,sk,d t−1 t − D t014,m,sk,d N014,m,sk,d + M014,m,sk,r,d r t−1 t − M014,m,sk,d,r + ρd N014,m,unsk,d

t−1 t N014,m,sk,d = N014,m,sk,d +

r

1 t−1 t−1 − N014,m,sk,d − D t014,m,sk,d N014,m,sk,d , 15

∀d (13.2)

where S dt is the sex ratio at birth (the ratio of male to female births) in region d, B tsk,d is the birth rate, D t014,m,sk,d the death rate, and ρd is the rate at which region d’s educational institutions and general development transform production into professional worker families. The final term is survival to the corresponding 15–39 age group. In the corresponding equation for young males from production worker families the penultimate term is negative. For females in professional families in this age group the corresponding equation is 1 B t N t−1 1 + S dt sk,d 1539,f ,sk,d t−1 t − D t014,f ,sk,d N014,f M014,f ,sk,r,d ,sk,d + r t−1 t − M014,f ,sk,d,r + ρd N014,f ,unsk,d r 1 t−1 t−1 N014,f ,sk,d − D t014,f ,sk,d N014,f − ,sk,d , 15

t−1 t N014,f ,sk,d = N 014,f ,sk,d +

∀d. (13.3)

For adults of gender g from professional families in the age group 15–39, the equation includes a survival term from the younger age group: 1 t−1 −D t N t−1 N 15 014,g ,sk,d 014,g ,sk,d 014,g ,sk,d t−1 t M1539,g − D t1539,g ,sk,d N1539,g ,sk,r,d ,sk,d + r t−1 t − M1539,g ,sk,d,r + ρd N1539,g ,unsk,d r 1 t−1 t−1 t ∀g , d N1539,g − ,sk,d − D 1539,g ,sk,d N 1539,g ,sk,d , 25

t−1 t N1539,g ,sk,d = N 1539,g ,sk,d +

(13.4)

348


where the second term is the surviving inflow from the 0–14 age group and the final term is the surviving outflow to the 40–59 age group. Again, the skill transformation term is negative in the case of the corresponding equation for production worker families. The population of professional adults of gender g in age group 40–59 follows: 1 t−1 t−1 N1539,g ,sk,d − D t1539,g ,sk,d N1539,g ,sk,d 25 t−1 t M4059,g − D t4059,g ,sk,d N4059,g ,sk,r,d ,sk,d + r t−1 t − M4059,g ,sk,d,r + ρd N 4059,g ,unsk,d r 1 t−1 t−1 ∀g , d. (13.5) − N4059,g ,sk,d − D t4059,g ,sk,d N4059,g ,sk,d , 20

t−1 t N4059,g ,sk,d = N 4059,g ,sk,d +

and for adults in the 60+ age group, the corresponding relationship is 1 t−1 t−1 t−1 t N60+,g N4059,g ,sk,d − D t4059,g ,sk,d N4059,g ,sk,d = N 60+,g ,sk,d + ,sk,d 20 t−1 t t M60+,g M60+,g + ,sk,r,d − ,sk,d,r + ρd N 60+,g ,unsk,d r

−

r

1 L t60+,g ,sk,d

t−1 N60+,g ,sk,d ,

∀g , d,

(13.6)

where the final term indicates that deaths from this group each year depend on its life expectancy at 60, L t60+,g ,sk,d . Again, the equation for aged production worker family members is the same except that the skill transformation term is negative.

Sources and Structure R Key parameters in the model are the migration rates Ma,g ,s,r,d , birth rates t t t B s,r , sex ratios at birth S r , death rates D a,g ,s,r , life expectancies at 60 L t60+,g ,s,r and the skill transformation rates ρd . The migration rates are based on recent migration records and are held constant through time.12 The skill transformation rates are based on changes during the decade prior to the base year, 1997, in the composition of aggregate regional labor forces as between production and professional workers. These are also held constant through time.13 12 13

The migration rates and the corresponding birth rates are listed in detail in Chan and Tyers (2006: Tables 2–5) and Vedi (2005). Note that, as regions become more advanced and populations in the production worker families become comparatively small, the skill transformation rate has a diminishing effect on the professional population.


349

Asymptotic Trends in Other Parameters The birth rates, life expectancy at 60, and the age-specific mortality rates all trend through time asymptotically. For each age group a, gender group g, and region r, a target rate is identified.14 The parameters then approach these target rates with initial growth rates determined by historical observation. In year t the birth rate of region r is (13.7) B tr = B 0r + B 0Tg t − B 0r 1 − e β t , where the rate of approach β is calibrated from the historical growth rate: 0 B Tg t − B 0r 1 − e β B 1r − B 0r 0 B x¨ r = = , (13.8) P r0 B 0r so that

B 0r Pˆ r0 β = ln 1 − 0 . B Tg t − B 0r

(13.9)

The birth rates and death rates, thus calculated, are summarized in Tables 13.1 and 13.2. The corresponding life expectancies at 60 are listed in Table 13.3.

Labor Force Projections To evaluate the number of full-time equivalent (FTE) workers we first construct labor force participation rates, Pa,g,r by gender and age group for each region from ILO statistics on the “economically active population.” We then investigate the proportion of workers who are part time and the hours they work relative to each regional standard for full-time work. The result is the number of FTEs per worker, Fa,g,r . The labor force in region r is then L¯ tr =

f unsk 60+

L ta,g ,s,r

t t where L ta,g ,s,r = μta,r P a,g ,r F a,g ,r N a,g ,s,r .

a=1539 g =m s=sk

(13.10) Here is a shift parameter reflecting the influence of policy on partict ipation rates. The time superscript on P a,g ,r refers to the extrapolation of 15 observed trends in these parameters. μta,r

14 15

In this discussion the skill index, s, is omitted because birth and death rates and life expectancies at 60 do not vary by skill category in the version of the model used. Although part-time hours may well also be trending through time, we hold F constant in the current version of the model.

350

Rod Tyers and Qun Shi Table 13.1. Birth rates and the sex ratio at birth Fertility Total Total Birth: births/ completed Birth: births/ completed Sex ratio 1000 women fertility: 1000 women fertility: at birth aged 15–39a children per aged 15–39a children per male/female (1997) woman (1997) (2030) woman (2030)

Australia North America Western Europe Central Europe, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of world

1.05 1.05 1.06 1.06 1.06 1.10 1.05 1.06 1.05 1.05 1.05 1.05 1.03 1.05

72 88 61 51 57 76 104 99 139 160 105 137 180 128

1.8 2.2 1.5 1.3 1.4 1.9 2.6 2.5 3.5 4.0 2.6 3.4 4.5 3.2

70 85 53 51 57 58 91 90 115 144 101 130 178 102

1.7 2.1 1.3 1.3 1.4 1.4 2.3 2.3 2.9 3.6 2.5 3.3 4.5 2.6

a

Birth rates are based on UN estimates and projections as represented by the U.S. Bureau of the Census. The latter representation has annual changes in rates, whereas the UN model has them stepped every 5 years. Aggregation is by population-weighted average. Initial birth rates are obtained from the UN model by dividing the number of births per year by the number of females aged 15–39. These rates change through time according to annualized projections by the U.S. Bureau of the Census. Source: Aggregated from United Nations (2003), U.S. Department of Commerce, U.S. Bureau of the Census “International Data Base.”

Asymptotic Trends in Labor Force Participation For each age group a, gender group g, and region r, a target country is identified whose participation rate is approached asymptotically. The rate of this approach is determined by the initial rate of change. Thus, the participation rate takes the form 0 t 0 0 βt (13.11) P a,g ,r = P a,g ,r + P Tg t − P a,g ,r (1 − e ), where the rate of approach, β, is calibrated from the initial participation growth rate: 0 0 t 0 P (1 − e β ) − P − P P a,g ,r Tg t a,g ,r a,g ,r 0 = , (13.12) Pˆ a,g ,r = 0 0 P a,g P a,g ,r ,r


351

Table 13.2. Age- and gender-specific death rates a (deaths per year per thousand in each age

and gender group) 0–14 Males

15–39

Females

Males

40–59

Females

Males

Females

1997 2030 1997 2030 1997 2030 1997 2030 1997 2030 1997 2030 Australia 1.6 0.6 1.3 0.5 1.3 North America 2.2 0.6 1.8 0.6 1.7 Western Europe 1.8 0.6 1.4 0.6 1.2 Central Europe, FSU 2.0 0.8 1.8 0.6 2.1 Japan 1.2 0.7 1.0 0.7 0.7 China 1.1 0.5 0.9 0.5 0.8 Indonesia 1.4 1.1 1.1 0.9 2.3 Other East Asia 1.4 0.7 1.1 0.6 2.3 India 8.2 3.8 9.4 4.5 1.3 Other South Asia 8.2 3.6 9.4 4.2 1.3 South America 1.8 1.4 1.4 1.0 1.3 Mid East Nth Africa 6.7 1.5 6.5 1.9 1.3 Sub-Saharan Africa 10.1 14.8 7.7 11.4 1.3 Rest of World 1.7 0.7 1.4 0.6 1.3

1.1 0.8 1.1 1.6 0.6 0.6 2.0 2.0 1.1 1.4 1.1 0.9 1.7 0.8

0.5 0.7 1.1 0.9 0.4 0.3 2.0 1.0 2.4 3.1 0.7 1.5 1.3 0.7

0.4 3.2 2.8 2 1.5 0.5 4.5 3.6 2.7 2.2 1.0 4.4 3.2 2.7 2.0 0.7 8.0 6.1 3.3 2.5 0.3 3.5 2.6 2.0 1.4 0.2 3.9 2.8 2.0 1.8 1.6 7.9 6.3 3.9 2.7 1.0 7.6 3.9 3.4 2.3 2.1 12.3 7.6 8.5 5.7 3.1 10.8 9.9 10.3 9.5 0.4 4.3 3.0 2.3 2.0 1.5 8.3 8.0 5.0 3.6 1.3 30.0 29.6 30.0 28.4 0.5 5.4 3.6 2.8 1.9

a

Aggregation is by population-weighted average. Projections of these parameters to 2020 assume convergence on target rates observed in comparatively “advanced” countries, as explained in the text. Only the endpoint values are shown here, but the model uses values that change with time along the path to convergence. Source: Aggregated from United Nations (2000) and WHO (2003).

so that

β = ln 1 −

0 ˆ0 P a,g ,r P a,g ,r 0 0 P Tg t − P a,g ,r

.

(13.13)

Target rates are chosen from countries considered “advanced” in terms of trends in participation rates. Where female participation rates are rising, therefore, Norway provides a commonly chosen target because its female labor force participation rates are higher than for other countries.16

Accounting for Part-Time Work For each age group a, gender g, and region r, full-time equivalency depends on the fraction of participants working full time, fa,g,r , and, for those working part time, the ratio of average part-time hours to full-time hours for that 16

The resulting participation rates are listed by Chan and Tyers (2006: Table 10).

352

Rod Tyers and Qun Shi Table 13.3. Baseline life expectancy at 60 (years)a Male

Australia North America Western Europe Central Europe, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World

Female

Initial

2030

Initial

2030

17 19 19 15 22 16 14 17 15 15 17 16 10 18

21 21 21 15 27 17 14 18 16 16 19 17 10 22

21 23 23 20 26 18 15 19 18 15 20 18 13 23

26 25 25 21 33 21 15 19 19 17 21 19 13 28

a

Aggregation is by population-weighted average. Projections of these parameters to 2020 assume convergence on target rates observed in comparatively “advanced” countries, as explained in the text. Only the endpoint values are shown here, but the model uses values that change with time along the path to convergence. Source: Aggregated from United Nations (2000b) as presented in Chan and Tyers (2006).

gender group and region, rg,r . For each group, the ratio of FTE workers to total labor force participants is then F a,g ,r = f a,g ,r + (1 − f a,g ,r )r g ,r .

(13.14)

Preliminary estimates of fa,g,r and rg,r are approximated from OECD (1999: Table 1.A.4) and OECD (2002: Statistical Annex, Table F).17

The Aged Dependency Ratio We define and calculate four dependency ratios: (1) a youth dependency ratio is the number of children per FTE worker, (2) an aged dependency ratio is the number of persons over 60 per FTE worker, (3) a nonworking aged dependency ratio is the number of nonworking persons over 60 per 17

No data have yet been sought on part-time work in non-OECD member countries. In these cases the diversity of OECD estimates is used to draw parallels between countries and regions and thus to make educated guesses. The results are listed by Chan and Tyers (2006: Tables 11 and 12).


353

FTE worker, and (4) a more general dependency ratio is defined that takes as its numerator the total nonworking population, including children.18 The one of most widespread policy interest is the nonworking aged dependency ratio: f unsk t N60+,g ,sk,r − L t60+,g ,sk,r

= R ANW r,t

g =m s=sk

t

Lr

.

(13.15)

2.2 The Global Economic Model The chapter extends GDyn, presented in Chapter 2. In the version used, the world is subdivided into 14 regions, one of which is China. Industries are aggregated into only three sectors: food (including processed foods), industry (mining and manufacturing), and services. To capture the full effects of demographic change, including those of aging, the standard model has been modified to include multiple age, gender, and skill groups in line with the structure of the demographic submodel. In the adapted model, these 16 groups differ in their consumption preferences, saving rates, and labor supply behavior. Unlike the standard GTAP models, in which regional incomes are split among private consumption, government consumption, and total saving via an upper level Cobb-Douglas utility function that implies fixed regional saving rates, this adaptation first divides regional incomes between government consumption and total private disposable income. The implicit assumption is that governments balance their budgets while private groups save or borrow. Private disposable income is then split among the eight age-gender groups in a manner informed by empirical studies of age- and genderspecific consumption behavior. For each age-gender group we then use a Keynesian consumption equation to split disposable income between saving and consumption expenditure. Group private saving rates then become endogenous, depending on real disposable income and the real interest rate, thereby relaxing the fixed average saving rate assumption in the standard model. Once group consumption expenditures are known, the standard GTAP CDE19 consumption preferences are applied to each, with preference parameters varying to reflect age-gender differences in tastes. Finally, consumption volumes are totaled across groups to obtain the final demand for each product, and consumption expenditures are subtracted from group 18 19

All these dependency ratios are defined in detail by Chan and Tyers (2006). This refers to the “constant difference of elasticities of substitution” demand system. See Hertel (1997) and, in particular, Huff et al. (1997).

354


disposable incomes to obtain group saving levels, which are then totaled across groups to obtain regional saving.

Income Splitting The first step is to split government from private disposable income. For this we retain the original Cobb-Douglas system, this time in a two-way split, and the governments’ income shares from the original database.20 Total regional disposable income is then split among the eight age-gender groups. For this we draw from other studies the distribution of disposable income between age-gender groups for “typical” advanced and developing countries. To ensure that changes in the age-gender distribution of each region’s population alter the corresponding age-gender distribution of income, we define a set of weights, Wa,g ,r , that represent the ratio of the per capita disposable income of group (a, g), to that of the (15–39, m) group, chosen as an arbitrary standard.21 The share of the disposable income of region r enjoyed by people of gender g and age group a is thus D Ya,g ,r

YrD

=

60+

Wa,g ,r Na,g ,r . f Wa,g ,r Na,g ,r

(13.16)

a=0−15 g =m

The adopted values of Wa,g ,r are listed in Table 13.4. Their selection is guided by the empirical studies of the age distribution of income and consumption noted in the table.

Splitting Savings and Consumption Expenditure from Group Disposable Income Our reduced form approach to the intertemporal optimization problem faced by each individual centers on an exponential consumption equation that links group per capita consumption expenditure to per capita disposable income and the real interest rate, r: εc D Ya,g C a,g ,r ,r =A r rεr , (13.17) c a,g ,r = C C Na,g ,r P a,g N P a,g ,r a,g ,r ,r 20 21

This implies the assumption that all governments balance their budgets and that all saving in the original database is private. To date we have not realized the opportunity to have the age-gender distribution of income depend on the income’s factor origin. Despite intuition suggesting a link – the aged in advanced countries receive retirement income stemming from capital ownership – consistent empirical work on this distribution is unavailable.


355

Table 13.4. Income weights, Wa,g ,r , by age-gender group 0–14


15–39

40–59

60+

Male

Female

Male

Female

Male

Female

Male

Female

0.60 0.40 0.50 0.50 0.60 0.60 0.50 0.60 0.50 0.50 0.40 0.50 0.50 0.60

0.60 0.40 0.50 0.50 0.60 0.60 0.50 0.60 0.50 0.50 0.40 0.50 0.50 0.60

1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

0.86 1.10 1.00 1.00 1.60 1.60 1.40 1.60 1.40 1.40 1.05 1.40 1.40 0.86

0.86 1.10 1.00 1.00 1.60 1.60 1.40 1.60 1.40 1.40 1.05 1.40 1.40 0.86

0.67 0.60 0.70 0.70 0.94 0.94 0.90 0.94 0.90 0.90 1.10 0.90 0.90 0.67

0.67 0.60 0.70 0.70 0.94 0.94 0.90 0.94 0.90 0.90 1.10 0.90 0.90 0.67

Source: Compiled from studies of consumption behavior on particular countries, including U.S. and UK: Attanasio and Banks (1998), Attanasio et al. (1999); Japan: Kitamura et al. (2001: Table 1); Mexico (standard for Latin America and an indicator for some other developing regions): Attanasio and Szekely (1998: Figure 1); New Zealand (standard for Australia and Western Europe): Gibson and Scobie (2001: Figure 1). C where P a,g ,r is a group consumption price index, group consumption expenditure is C a,g ,r and parameters εc and εr are income and interest elasticities. This equation is calibrated for each group and region based on the set of initial (1997) age-specific saving rates from per capita real disposable income listed in Table 13.5. These estimates are drawn from the same empirical studies of the age distribution of income and consumption as the income weights of Table 13.4. They are recalibrated for consistency with the overall private saving rate in each region indicated in the GTAP Data Base.22 22

The elasticities of consumption expenditure to disposable income suggested by the empirical literature seem to be poor choices as reduced forms for saving behavior in the long term because they imply high marginal saving rates. We calibrate these elasticities according to the following scenario: (a) North American per capita disposable income grows at 3%/yr for 100 years; (b) growth in all other regions is sufficient to attain North America’s per capita disposable income levels within that period; (c) when the other regions catch up, all regions attain identical group-specific saving rates; and (d) the income, consumption, and saving transitions are smooth and exponential. Our reduced-form consumption equation (13.17) can be simplified for a single age-gender group to c = A y εc r εr , where c is per capita consumption expenditure, y is per capita disposable income, and r is the real interest rate. To focus on the key elasticity, εc , imagine that the real interest rate is constant through time, so that the interest term can be embedded in the constant. Then, if per capita disposable

356


Table 13.5. Initial saving rates from personal disposable income by age-gender group

(percent) 0–14


15–39

40–59

60+

Male

Female

Male

Female

Male

Female

Male

Female

0 0 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0

7 14 10 4 24 35 23 36 19 7 7 8 2 5

7 14 10 4 24 35 23 36 19 7 7 8 2 5

31 19 39 18 28 40 34 40 28 10 17 19 6 23

31 19 39 18 28 40 34 40 28 10 17 19 6 23

−5 −30 −20 −6 22 31 23 32 19 7 6 7 2 −5

−5 −30 −20 −6 22 31 23 32 19 7 6 7 2 −5

Source: Compiled from studies of consumption behavior on particular countries, including Mexico: Attanasio and Szekely (1998); Japan: Kitamura et al. (2001); New Zealand: Gibson and Scobie (2001); US: Attanasio et al. (1999).

Consumption Preferences The construction of the CDE demand system requires the calibration of two sets of parameters by the method detailed by Huff et al. (1997). Its advantage over the CES, or constant elasticity of substitution, system is that it is nonhomothetic and therefore allows income elasticities of demand to vary between commodities. Elasticities of demand then depend on CDE “expansion” and “substitution” parameters, which are calibrated for each region’s aggregate household in the GTAP Data Base. We retain the calibrated values of these parameters for the eight age-gender groups. To complete the demand system we then need expenditure shares for each of the eight different age-gender groups in each region. For these shares we turn, once again, to the consumption analysis literature. Studies of consumption preferences by age group are available for a few of the identified countries, and we use those as a guide in the income grows at rate g y , the rate of growth of consumption expenditure is εc g y . If per capita consumption expenditure is initially c 0 and per capita disposable income is initially y 0 , we can calculate the group’s average saving rate in period t and invert the resulting expression to find the elasticity that is consistent with the target saving rate after T years: T εc = 1 + g y1T ln( 1−s ). For further details, see Tyers et al. (2005). 1−s 0


357

construction of the complete matrix of expenditure shares listed in Table 13.6. The study by Weber et al. (2002) is the most detailed, and it shows only very modest variation in expenditure shares by age group when commodities are highly aggregated. Although there is considerable variation when comparisons are at a high level of detail, such as between fresh food and restaurant meals or between health and other services, the broad shares are remarkably similar.23 For presentational economy, we focus in this chapter on the three-product version.24 Age-gender group expenditure shares are drawn initially from the literature (indicated in Table 13.6) and then rendered consistent with group expenditures on the one hand and GTAP Data Base values for aggregate expenditure shares on the other by using RAS techniques to concord the shares with row and column sums in the matrix of expenditures.

Elasticities of Substitution It is well known that general equilibrium simulation results are particularly sensitive to the assumed degree of differentiation between home and foreign goods and services. In models such as this one, in which products are highly aggregated, some of this differentiation reflects regional differences in sectoral product composition. Both the complementarity of product compositions and true regional product differentiation are therefore represented in the model via the choice of the elasticities of substitution between home and foreign products. Controversy has raged over the merits of various estimates, and the view is commonly expressed that the “standard” GTAP estimates, which range between 1.0 and 4.0, are too small. We agree, because when these elasticities are used, our baseline simulation yields substantial divergence between the paths of home and trading prices in different regions. In the absence of any new trade distortions, global markets appear far more segmented in 2030 than they are at present. Newer estimates by Harrigan (1995), Trefler and Lai (1999), and Obstfeld and Rogoff (2000) all support much higher values. We therefore use 7.0 for food products, 4.0 for manufactures, and 2.2 for the less tradable services, and we retain the traditional “rule of two” for substitutability of imports by region of origin. 23

24

It is of concern that some expenditure shares for detailed products and services appear to be changing very rapidly through time. Weber et al. (2002) show that the health share is rising rapidly for the aged and that this is associated with very rapid growth in the share of expenditure on drugs by all groups but particularly the aged. The GTAP commodity classification is production oriented, based on the International Standard Industrial Classification (ISIC), and so it differs from the classification used in expenditure surveys. We use the GTAP commodities throughout, weakening the sensitivity of our analysis to differences in preferences.

358

Rod Tyers and Qun Shi Table 13.6. Private expenditure shares by age-gender group 0–14

15–39

40–59

60+

Male

Female

Male

Female

Male

Female

Male

Female

Australia Food Manufactures Services

0.18 0.07 0.74

0.18 0.07 0.74

0.10 0.19 0.71

0.10 0.19 0.71

0.10 0.19 0.71

0.10 0.19 0.71

0.18 0.05 0.77

0.18 0.05 0.77

North America Food Manufactures Services

0.12 0.12 0.76

0.12 0.12 0.76

0.05 0.16 0.79

0.05 0.16 0.79

0.06 0.16 0.78

0.06 0.16 0.78

0.11 0.09 0.80

0.11 0.09 0.80

Western Europe Food Manufactures Services

0.18 0.12 0.70

0.18 0.12 0.70

0.09 0.30 0.61

0.09 0.30 0.61

0.09 0.30 0.61

0.09 0.30 0.61

0.18 0.09 0.73

0.18 0.09 0.73

Central Europe, FSU Food 0.44 Manufactures 0.10 Services 0.47

0.44 0.10 0.47

0.26 0.27 0.47

0.26 0.27 0.47

0.26 0.27 0.47

0.26 0.27 0.47

0.43 0.07 0.50

0.43 0.07 0.50

Japan Food Manufactures Services

0.18 0.07 0.75

0.18 0.07 0.75

0.10 0.18 0.72

0.10 0.18 0.72

0.10 0.18 0.72

0.10 0.18 0.72

0.17 0.05 0.78

0.17 0.05 0.78

China Food Manufactures Services

0.47 0.13 0.40

0.47 0.13 0.40

0.26 0.35 0.39

0.26 0.35 0.39

0.26 0.35 0.39

0.26 0.35 0.39

0.47 0.10 0.43

0.47 0.10 0.43

Indonesia Food Manufactures Services

0.46 0.07 0.48

0.46 0.07 0.48

0.30 0.26 0.44

0.30 0.26 0.44

0.30 0.26 0.44

0.30 0.26 0.44

0.45 0.05 0.50

0.45 0.05 0.50

Other East Asia Food Manufactures Services

0.30 0.10 0.60

0.30 0.10 0.60

0.17 0.35 0.47

0.17 0.35 0.47

0.17 0.35 0.47

0.17 0.35 0.47

0.29 0.08 0.63

0.29 0.08 0.63

India Food Manufactures Services

0.57 0.08 0.35

0.57 0.08 0.35

0.37 0.31 0.32

0.37 0.31 0.32

0.37 0.31 0.32

0.37 0.31 0.32

0.56 0.06 0.38

0.56 0.06 0.38

Other South Asia Food Manufactures Services

0.54 0.07 0.39

0.54 0.07 0.39

0.37 0.27 0.36

0.37 0.27 0.36

0.37 0.27 0.36

0.37 0.27 0.36

0.54 0.05 0.41

0.54 0.05 0.41

(continued)


359

Table 13.6 (continued) 0–14

15–39

40–59

60+

Male

Female

Male

Female

Male

Female

Male

Female

0.36 0.10 0.53

0.36 0.10 0.53

0.21 0.36 0.43

0.21 0.36 0.43

0.21 0.36 0.43

0.21 0.36 0.43

0.36 0.08 0.57

0.36 0.08 0.57

Mid East Nth Africa Food 0.39 Manufactures 0.07 Services 0.54

0.39 0.07 0.54

0.25 0.27 0.48

0.25 0.27 0.48

0.25 0.27 0.48

0.25 0.27 0.48

0.38 0.05 0.57

0.38 0.05 0.57

Sub-Saharan Africa Food 0.46 Manufactures 0.07 Services 0.47

0.46 0.07 0.47

0.30 0.28 0.42

0.30 0.28 0.42

0.30 0.28 0.42

0.30 0.28 0.42

0.45 0.05 0.50

0.45 0.05 0.50

Rest of World Food Manufactures Services

0.36 0.10 0.54

0.20 0.27 0.53

0.20 0.27 0.53

0.20 0.27 0.53

0.20 0.27 0.53

0.35 0.07 0.58

0.35 0.07 0.58

South America Food Manufactures Services

0.36 0.10 0.54

Source: Constructed with guidance from the results presented by: Abdel-Ghany and Sharpe (1997), Blisard (2001a and b), Blisard et al. (2003), Case and Deaton (2002), Paulin (2000), Regmi et al. (2001) and Weber et al. (2002). The shares are then modified using a RAS process to conform with aggregate expenditures by product in the GTAP Data Base.

3. Constructing the Baseline Scenario The baseline scenario represents a “best judgment” projection of the global economy through 2030. Although policy analysis can be sensitive to the content of this scenario, our focus is on the extent of departures from it that would be caused by changes in demographic scenarios. Nonetheless, it is instructive to describe the baseline for two reasons. First, all scenarios examined have in common a set of assumptions about future trends in productivity, and second, some exposition of the baseline makes the construction of departures from it clearer.

3.1 Exogenous Factor Productivity Growth Exogenous sources of growth enter the model as factor productivity growth shocks, applied separately for each of the model’s five factors of production (land, physical capital, natural resources, production labor, and professional labor). Simulated growth rates are very sensitive to productivity growth rates because the larger they are for a particular region, the larger is that region’s marginal product of capital. The region therefore enjoys higher

360


levels of investment and hence a double boost to its per capita real income growth rate. The importance of productivity notwithstanding, the empirical literature is inconsistent as to whether productivity growth has been faster in agriculture or in manufacturing and whether the gains in any sector have enhanced all primary factors or merely production labor. The single set of factor productivity growth rates assumed in all scenarios are drawn from a new survey of the relevant literature (Tyers et al. 2005). Agricultural productivity grows more rapidly than that in the other sectors in China, Australia, Indonesia, Other East Asia, India, and Other South Asia. This faster growth is caused by continued increases in labor productivity in agriculture and the associated shedding of labor to the other sectors. In the other industrialized regions, the process of labor relocation has slowed down, and labor productivity growth is slower in agriculture. In the other developing regions, the relocation of workers from agriculture has tended not to be so rapid.

3.2 Investment Interest Premia In addition to exogenous productivity growth, a key aspect of the baseline projection is its allocation of investment across regions. The model takes no explicit account of investment risk and the segmented capital markets that are prevalent in developing countries, and so it tends to allocate investment to regions that have high marginal products of physical capital. These regions tend to include labor-abundant developing countries whose labor forces are still expanding rapidly, yet we know that capital market underdevelopment, capital controls, and risk considerations limit the flow of foreign investment into these regions at present and that these considerations are likely to remain important in the future. To account for investment risk and the segmented capital markets prevalent in developing countries we have constructed a “pre-baseline” simulation in which we maintain the relative growth rates of investment across regions. In this simulation, global investment rises and falls, but its allocation between regions is thus controlled. To do this an interest premium variable (GDyn variable SDRORT) is made endogenous. This creates wedges between the international and regional interest rates, the scale of which is indicated in Figure 13.2. It shows high premia for the populous developing regions of Sub-Saharan Africa, the Middle East and North Africa, and India. Premia tend to fall in regions in which labor forces are falling or growing more slowly, such as Japan and the East Asian regions. Most spectacular is the secular fall in the Chinese premium, which occurs because investment growth is maintained in China despite the eventual decline in


361

8

6 North America

4

Sub-S Africa Mid-East N Africa

2

Western Europe

0

Central Europe FSU India

-2

Japan China

-4

Other East Asia

-6

-8 1995

2000

2005

2010

2015

2020

2025

2030

2035

Figure 13.2. Baseline interest premia. These are simulated cumulative percentage point wedges between regional and global rates of return (DROR-DRORW), the global rate being that offered by the global trust. They are derived from a pre-base simulation in which investment is constrained to grow at the same rate in all regions.

its labor force.25 The final baseline simulation then frees up investment. It and the subsequent simulations not only maintain endogenous investment, but to do this they have the time paths of the regional interest premia set as exogenous to the levels indicated in the figure.

3.3 The Baseline Population Projection Notwithstanding their dependence on a comparatively simple four-agegroup demographic model, regional population levels and age structures 25

A possible consequence of this is that our baseline investment in China, and therefore China’s projected economic growth rate, is optimistic. A more detailed study of this issue is offered by Tyers and Golley (2006).

362

Rod Tyers and Qun Shi Table 13.7. Baseline projections of labor force size and structure Labor forcea

Australia North America Western Europe Central Europe Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World

% Female

% 40+

Initial

2030

Initial

2030

Initial

2030

8 182 184 181 61 570 87 127 356 134 123 103 150 79

10 250 165 148 55 592 130 178 594 265 193 176 349 131

37 40 40 47 37 37 38 41 27 28 38 24 28 36

40 42 44 46 37 36 38 40 28 28 39 23 29 34

42 42 47 44 58 34 40 37 36 32 33 30 29 38

48 47 55 53 65 47 54 51 47 44 48 42 36 48

a Measured in FTE workers (in millions). Source: Projection using the demographic model described in the text, as presented in detail by Chan and Tyers (2006).

follow closely corresponding United Nations projections.26 Baseline projections of labor force levels and age structures are summarized in Table 13.7, which shows substantial aging of labor forces in all regions. Indeed, the extent of the widespread aging is especially clear from the trends in nonworking aged dependency ratios listed in Table 13.8. Western Europe, Australia, and Japan are the regions with the “oldest” population by 2030 in terms of the aged dependency ratio. China’s dependency ratio is the highest among the developing economies. In particular, it rises fairly rapidly during this period, suggesting a significant demographic transition for the economy. Finally, baseline projections of total populations and labor forces for a selection of regions are displayed in Figure 13.3. These illustrate divergences in the growth paths of populations and labor forces that are important for economic performance. Particularly noteworthy are the declines in the labor forces of Japan, Western Europe, and China, all of which precede their associated population declines as their populations age. 26

See United Nations (2003) and the detailed comparison provided in Chan and Tyers (2006).


363

Table 13.8. Baseline nonworking aged dependency ratios Nonworking aged/working


Initial

2030

0.35 0.24 0.42 0.29 0.32 0.19 0.09 0.09 0.12 0.09 0.16 0.15 0.13 0.15

0.54 0.36 0.61 0.42 0.48 0.44 0.16 0.23 0.23 0.18 0.29 0.33 0.15 0.27

Source: Base period statistics constructed from population statistics from United Nations (2003) and simulation results from the demographic model described in the text.

3.4 The Baseline Economic Projection Overall baseline economic performance is suggested by Table 13.9, which lists the projected increments to regional real per capita incomes by 2030. In part because of its comparatively young population and hence its continuing rapid labor force growth, India attracts substantial new investment and is projected to take over from China as the world’s most rapidly expanding region. This investment, combined with exogenous factor productivity growth, ensures that India is also projected to deliver the largest improvement in real per capita income through 2030. China’s growth is slower in aggregate because of its declining labor force, but its declining interest premium maintains a high level of investment growth sufficient to deliver the second largest proportional change in real per capita income. Indonesia and Other East Asia are also strong performers, while the older industrial economies continue to grow more slowly. The African regions enjoy good GDP growth performance, but their high population growth rates limit their performance in per capita terms. In the GDyn model the current account is not forced to converge on any particular steady state. As a result, it is possible for the interregional distribution of asset ownership to change so as to cause current account

364

1.05

Rod Tyers and Qun Shi Western Europe

1

1

0.95

0.95

0.9

0.9

0.85 1995

Japan

1.05

population

population

labor force

labor force

2005

2015

2025

2035

China

1.1

0.85 1995

2005

2015

2025

2035

India

1.75 1.65

1.05

1.55 1 1.45 0.95

1.35 1.25

0.9

1.15

population 0.85

0.8 1995

population

labor force

2005

2015

1.05

2025

2035

0.95 1995

labor force

2005

2015

2025

2035

Figure 13.3. Baseline population and labor force projections.

deficits to widen relative to GDP. Some imbalances do widen, although the deficits and surpluses expand more slowly than the trend of the past two decades. Most notable are expanding deficits in Japan and China, in which the aging populations reduce saving through time and deficits are required to finance continued investment. North America trends toward widening trade surplus, although it belies comparatively large changes in the net factor income component of its current account, which actually goes


365

Table 13.9. Baseline real per capita income projection to

2030, percent change over 1997 Ranked by performance India China Indonesia Other East Asia Central Europe FSU Japan Australia Western Europe Other South Asia South America Rest of World North America Sub-Saharan Africa Mid-East Nth Africa

369 361 356 341 218 207 185 171 156 142 141 139 125 88

Source: Baseline simulation using the augmented GDyn, as described in the text.

into deficit; therefore North America’s overall current account deficit does not diverge significantly. Net factor income flows are less significant for the other regions. Finally, the product composition of global output in the baseline projection is suggested by Figure 13.4. Engel’s Law drives increased resources into the industrial and services sectors relative to the food and agricultural sector. However, this change emerges despite a relative increase in food prices, which is also shown in the figure. This result is not without controversy. Ever since the work of Lewis (1952), a measured trend toward declining prices of food commodities relative to traded manufactures has been prominent in the commodity trade literature. Grilli and Yang (1988) confirmed and updated the trend identified by Lewis, indicating a decline in relative commodity prices of 0.5% per year. This work was flawed, however, by its lack of consideration of improvements in the quality of manufactures over time. Lipsey (1994), after adjusting for quality changes, found that primary commodity prices actually increased by 0.5% per year in the last half of the 20th century. Our baseline simulation reflects this result. However, it is worth recalling that the purpose of the baseline is to provide a noncontroversial reference against which to compare the alternative scenario with different demographic behavior. Our

366

Rod Tyers and Qun Shi Global output

4

World trading prices 1.15

Food Industrial goods

3.5

Services

1.10

3

2.5

1.05

2

1.00 1.5

1

Food

0.95

Industrial goods

0.5

Services 0 1995 2000 2005 2010 2015 2020 2025 2030 2035

0.90 1997

2002

2007

2012

2017

2022

2027

Figure 13.4. Baseline global output and world trading product prices. World trading prices are relative to a common global numeraire.

emphasis is therefore on the departures of the alternative scenario from the baseline.

4. The Impact of Accelerated Population Aging The analysis presented here is centered on the baseline projection of populations, labor forces, and their structures described in the previous section. This baseline is compared with an alternative scenario that is identical in all respects except that aging is accelerated in all regions via increases in the life expectancy at 60. In particular, life expectancies at 60 grow faster than in the baseline case, by 2% per year. This alternative “accelerated aging” scenario is prompted by the conjecture of Booth (2004) that the standard demographic projections tend to ignore many new and potential developments in health science, which makes them pessimistic about future longevity.27 The effect of accelerated aging is to increase the aged population in every region, while leaving younger populations the same as in the baseline. This has two key implications. First, it raises projected overall populations so 27

An alternative approach to the measurement of the effects of aging is to compare the baseline, which embodies projected changes in age distributions, with a hypothetical scenario in which aggregate populations grow at the same rates but age distributions do not change. This approach is adopted in Tyers and Shi (2006).


367

Table 13.10. Demographic effects in 2030 of accelerated aging,a percent departures

from the baseline scenario Population


Labor force

Nonworking aged dep ratio

Total

60+

Total

60+

Aged NW

5.7 5.1 7.5 7.8 7.4 5.9 4.5 4.4 3.8 3.1 4.1 3.3 2.4 3.6

23.7 23.6 25.8 31.4 21.4 27.4 31.0 25.8 26.6 26.6 24.4 25.3 34.5 20.1

1.3 2.2 2.3 3.6 5.5 1.4 4.7 3.5 3.2 3.0 3.4 2.3 2.8 3.3

24.6 24.0 26.3 32.5 22.1 29.0 31.2 26.1 27.5 26.7 24.6 25.8 36.2 21.0

22.0 20.9 22.9 26.6 14.8 25.5 25.0 21.4 22.2 22.7 20.1 22.3 30.0 15.8

a Growth in target life expectancies at 60 by 2% per year. Source: Simulation results from the model described in the text.

that it raises aggregate demand, and second, to the extent that the aged continue to work it raises labor forces and so also bolsters the supply side. Because the departure from the baseline only affects the aged population in each region, it also causes nonworking aged dependency ratios to rise in all regions.28 These demographic and labor supply effects are summarized in Table 13.10. The absolute increase in each region’s population and labor force depends on the demographic structure of its population in the base year and the aged labor force participation rate at the time. As shown in Figure 13.5, Japan enjoys the largest labor force increase. This is because Japan began with the largest aged share of its population and also the largest aged labor force participation rate.29 Because of its comparatively large aged population, Western Europe also enjoys substantially higher labor force growth. Among 28

29

Our modeling to date ignores the fiscal implications of this change because governments are modeled as maintaining patterns of expenditure that do not depend on the age distributions of populations. Moreover, governments balance their budgets in all simulations. To capture fiscal implications it will be necessary to separate out such sectors as health and retirement services on the one hand and educational services on the other. Almost half of Japan’s men aged 60+ are in the labor force. This is substantially higher than for other industrialized regions, although retirement is not a luxury enjoyed in most

368

Rod Tyers and Qun Shi Labor Force

Average Saving Rate

6

2

Japan Indonesia 5

0

Central Europe FSU India Western Europe

4

-2

North America China

3

-4

2

-6

1

-8

Japan Australia Central Europe FSU

-10

0

Western Europe North America

-12

-1 1997

2002

2007

2012

2017

2022

2027

1997

2002

2007

2012

2017

2022

2027

Figure 13.5. Accelerated aging–labor force and average saving rate, departure from baseline, %.

the regions at the other extreme is China, which has a proportionally smaller aged population and, compared with all the other regions, smaller aged labor force participation rates.30 The economic implications of accelerated aging stem not only from expanded consumption demand on the one hand and larger labor forces on the other. In North America, Western Europe, Central Europe, the former Soviet Union, and Australia, accelerated aging also raises the share of income in the hands of the over 60s, who tend to have negative saving rates. In these regions, therefore, average saving rates decline, as also shown in Figure 13.5. Indeed, the savings effect of this demographic change is so significant that real savings in most regions fall, despite an increase in real GDP and hence real income of the private households. Capital returns are raised by the population-driven demand increases and the expanded labor

30

of the developing world. A full listing of participation rates is provided by Chan and Tyers (2006). In China, formal retirement was a tradition in the state-owned enterprises that dominated the economy until recently. Recorded aged labor force participation rates actually declined during the 1990s. With the rise in private sector employment, however, it is expected that aged labor force participation rates will rise, rather than fall, as is assumed in the analysis presented.


369 Real GDP

Investment 4

2

0

3

-2

2

-4

1

North America South America Sub-Saharan Africa Mid-East Nth Africa Western Europe India Japan China Australia

Japan -6

0

Indonesia North America India

-8

-1

Western Europe Central Europe FSU China

-10

-2

1997

2002

2007

2012

2017

2022

2027

1997

2002

2007

2012

2017

2022

2027

Figure 13.6. Accelerated aging–investment and real GDP, departure from baseline, %.

supplies, yet real financing costs also rise because proportionally less is saved worldwide. The net effect is slower growth in global investment, with the level achieved in 2030 emerging lower by more than 3% compared to the baseline. The global investment slowdown notwithstanding, GDP tends to increase everywhere because of faster growing, albeit aged, labor forces (Table 13.11). The gains are not evenly distributed, however. Regions with the most labor force expansion, and particularly Japan, enjoy higher capital returns relative to other regions. This raises their shares of global investment compared to that in the baseline projection, particularly later in the simulation, and hence it accelerates their GDP growth (Figure 13.6). This investment is reassigned mainly from China, where investment falls by more than 8% by 2030, and from Australia, where investment falls close to 3%. In both regions, labor forces are boosted the least because their aged populations have lower labor force participation rates. The redistribution of investment expands the mismatch between the location of physical capita and its ownership, causing a corresponding divergence between GDP and GNP growth. The regions that expand more rapidly as a consequence of the accelerated aging tend also to be those with slower saving growth, and so these regions experience outflows of factor income on their current accounts. When this is combined with the tendency, due to diminishing returns, for labor expansions to reduce per capita income,

370 Table 13.11. Accelerated aging – product and factor prices, percent departure from the baseline in 2030 Labor Average Real Real GDP Real per Real production Land Resource Food Manufactures Services force saving rate investment GDP pricea capita income wageb rentb rentb output output output Australia North America Western Europe Central Eur, FSU Japan China Indonesia Other East Asia India Other South Asia South America Mid East Nth Africa Sub-Saharan Africa Rest of World a

1.3 2.2 2.3 3.6 5.5 1.4 4.7 3.5 3.2 3.0 3.4 2.3 2.8 3.3

−5.1 −11.7 −11.5 −6.7 −0.5 0.3 −0.2 0.1 0.0 −2.3 −1.1 −0.1 0.6 −3.3

−2.7 −0.5 −1.6 −2.0 1.4 −8.2 0.7 −1.8 −0.9 −0.7 0.9 −2.5 1.0 0.0

−0.1 1.3 0.8 0.9 3.1 −1.9 1.4 0.8 0.6 1.0 1.7 0.3 1.9 1.1

0.3 −0.1 0.2 −0.1 0.5 −0.5 0.0 0.0 0.2 −0.1 −0.2 0.0 −0.2 −0.2

The GDP price is measured relative to the common numeraire in GDyn. Note relative values only, indicative of real exchange rate changes. Source: Simulations of the model described in the text.

b

−5.2 −4.6 −6.8 −6.6 −2.7 −6.2 −2.4 −2.1 −2.4 −1.9 −2.4 −2.1 −0.5 −2.6

−0.9 −0.3 −1.0 −1.7 −1.4 −2.1 −2.1 −1.7 −2.1 −1.6 −1.1 −1.3 −0.5 −1.4

2.7 5.4 7.9 6.6 7.8 2.9 4.3 3.6 2.8 3.0 5.3 5.4 6.2 4.3

−1.2 0.6 −0.6 0.3 0.9 −1.1 0.5 −0.1 −0.1 0.6 1.1 −0.5 1.0 0.7

1.5 2.8 4.9 3.8 4.6 2.1 2.1 1.8 1.3 1.2 2.7 3.2 3.4 2.3

−1.3 0.7 −0.3 0.4 1.1 −1.1 0.6 −0.1 0.2 0.8 1.3 −0.7 0.9 0.7

0.0 1.3 0.9 0.8 3.4 −3.1 1.8 1.1 0.6 1.0 1.8 0.5 2.1 1.2

Global Demographic Change Real per Capita Income

371 Global Output

1

3

0

2.5

Food Industrial products Services

-1

2 -2

1.5 -3

1 -4 Sub-Saharan Africa India -5

0.5

Japan North America

-6

0

China Western Europe

-7 1997

2002

2007

2012

2017

2022

2027

-0.5 1995 2000 2005 2010 2015 2020 2025 2030 2035

Figure 13.7. Accelerated aging real per capita GNP and global product output, departures from baseline, %.

declines in real per capita GNP are observed in all regions (Fig. 13.7 and Table 13.11). Turning finally to the effects on the global product mix, as Table 13.6 demonstrates, when the array of products and services is aggregated into just three groups, the differences between the consumption preferences of the old and the young are not substantial. Overall, the effect of accelerated aging is to raise the production of food and agricultural products relative to the others and that of services relative to industrial products (Fig. 13.7). The rise in food production is due primarily to the declines in per capita incomes and Engel’s Law, whereas the relative rise in services production does appear to be a response to aging preferences. Supply-side effects, via declines in real wages (Table 13.11), advantage comparatively labor-intensive manufacturing, but these seem insufficient for the change in manufacturing output to outpace that in services.

5. Conclusion Global population and labor force changes that are already built into regional age distributions will cause the populations in several key regions, including Western Europe, Japan, and China, to decline in the near future and their labor forces to decline sooner and more dramatically. Associated with these changes will be the aging of the populations in all regions. This will not

372


only affect their economic performance but also the regional distribution of global investment and saving, and hence capital account flows, as well as regional production costs, comparative advantage, and therefore the pattern of global trade. To address these issues, standard GDyn is modified to accommodate eight age-gender groups within each regional household. A full demographic submodel is then incorporated and a baseline projection constructed through 2030. To illustrate the use of this model in the analysis of demographic change, an alternative scenario is constructed in which advances in health science cause life expectancies at 60 to rise more quickly in all regions. This accelerates the aging of all populations, and because the aged also contribute to the labor force, there is a relative expansion in the global labor supply. This expansion is largest in regions with either large aged shares of their populations at the outset, or high aged labor force participation rates, or both. It is therefore particularly strong for Japan. The labor supply expansions tend to cause some redistribution of global investment in favor of aging regions with high labor force participation rates. The volume of that investment is made smaller, however, by reduced global saving. Nonetheless, most regions enjoy net expansions in GDP. Because the capital attracted to the expanding regions tends to be owned elsewhere, the boost to regional GNP levels is more modest. Moreover, diminishing returns ensure that the effects on real per capita regional incomes are consistently negative. Hardest hit are Western and Central Europe and China, where aged participation rates are low, so that their populations age and expand, yet this does not increase their labor supplies and therefore their capital growth. Reduced per capita incomes ensure that global food output is boosted relative to that of both industrial products and services. In addition, because the consumption preferences of the old become more prominent, service output is accelerated relative to that of industrial products. In the form presented here, the model offers little role for governments in the aging process and its consequences. Governments balance their budgets, and age-related service expenditures are not identified explicitly. A number of potential improvements to the model therefore arise. First, increased product variety would make it possible to better reflect differences in consumption preferences across age groups. Second, health and retirement services could be identified in the model, with contributions to their supply from both governments and regional private sectors. These improvements would facilitate the analysis of the fiscal implications of aging and of changes in health and retirement sector productivity.


373

References Abdel-Ghany, M. and D. L. Sharpe. 1997. “Consumption Patterns among the Young-Old and Old-Old.” Journal of Consumer Affairs 31(1), 90–112. Attanasio, O. P. and J. Banks. 1998. “Trends In Household Saving Don’t Justify Tax Incentives to Boost Saving.” Economic Policy: A European Forum 10(27), 547– 83. Attanasio, O. P., J. Banks, C. Meghir, and G. Weber. 1999. “Humps and Bumps in Lifetime Consumption. Journal of Business and Economic Statistics 17(1), 22–35. Attanasio, O. P. and M. Szekely. 1998, December. Household Savings and Income Distribution in Mexico. Office of the Chief Economist, Documento de Trabajo No. 390. Washington, DC: Inter-American Development Bank. Barro, R. J., and G. S. Becker. 1989. “Fertility Choice in a Model of Economic Growth.” Econometrica 57(2), 481–501. Blisard, N. 2001a. Food Spending in American Households, 1997–98. Economic Research Service Statistical Bulletin 972. Washington, DC: U.S. Department of Agriculture. Blisard, N. 2001b. Income and Food Expenditures Decomposed by Cohort, Age and Time Effects. Electronic Report from the Economic Research Service. Washington, DC: U.S. Department of Agriculture. Available at www.ers.usda.gov. Blisard, N., J. N. Variyam, and J. Cromartie. 2003. Food Expenditures by U.S. Households: Looking Ahead to 2020. Electronic Report from the Economic Research Service. Washington, DC: U.S. Department of Agriculture. Available at www.ers.usda.gov. Bloom, D. E. and J. G. Williamson. 1997. “Demographic Transitions, Human Resource Development and Economic Miracles in Emerging Asia.” In J. Sachs and D. Bloom (eds.), Emerging Asia (pp. 419–455). Manila: Asian Development Bank. Booth, H. T. 2004. “On the Importance of Being Uncertain: Forecasting Population Futures for Australia.” People and Place 12(2), 1–12. Booth, H. T., J. Maindonald, and L. Smith. 2002. “Applying Lee-Carter under Conditions of Variable Mortality Decline.” Population Studies 56(3), 325–36. Booth, H. T. and L. Tickle. 2003. “The Future Aged: New Projections of Australia’s Elderly Population.” Australasian Journal of Ageing 22(4), 196–202. Bryant, R. C. 2001. Incorporating Demographic Change in Multi-Country Demographic Models: Some Preliminary Results. Available at http://www.sensiblepolicy.com/wmhp/ home1.htm. Bryant, R. C., H. Faruqee, D. Veculescu, and E. Arbatli. 2003. Fertility Declines and Youth Dependency: Implications for the Global Economy. Washington, DC: Brookings Institution. Bryant, R. C. and W. J. McKibbin. 1998. Issues in Modelling the Global Dimensions of Demographic Change. Available at http://www.sensiblepolicy.com/wmhp/home1.htm. Case, A. and A. Deaton. 2002, May. Consumption, Health, Gender and Poverty. Research Program in Development Studies. Princeton: Princeton University. Chan, M. M. and R. Tyers. 2006, December. Global Demographic Change and Labour Force Growth: Projections to 2020. Centre for Economic Policy Research Discussion Paper, Research School of Social Sciences (RSSS). Canberra: Australian National University. Dixon, P. B. and M. Rimmer. 2002. Dynamic General Equilibrium Modelling for Forecasting and Economic Policy. No. 256 in the Contributions for Economic Analysis series. Amsterdam: Elsevier North Holland.

374


Duncan, R., Q. Shi, and R. Tyers. 2005. Global Demographic Change and Demand for Food in Australia. Rural Industries Research and Development Corporation Report No 05/014. Canberra: RIRDC. Faruqee, H. and M. Muhleisen. 2002. “Population Ageing in Japan: Demographic Shock and Fiscal Sustainability.” Japan and the World Economy 15: 185–210. Gibson, J. and G. Scobie. 2001. “A Cohort Analysis of Household Income, Consumption and Savings.” New Zealand Economic Papers 35(2), 196–216. Grilli, E. and M. C. Yang. 1988. “Primary Commodity Prices, Manufactured Goods Prices and the Terms of Trade of Developing Countries: What the Long Run Shows.” World Bank Economic Review, 2(1). Harrigan, J. 1995. “The Volume of Trade in Differentiated Products: Theory and Evidence.” Review of Economics and Statistics 77(2): 283–93. Hertel, T. W. (ed.). 1997. Global Trade Analysis Using the GTAP Model. Cambridge: Cambridge University Press. Huff, K. M., K. Hanslow, T. W. Hertel, and M. E. Tsigas. 1997. “GTAP Behavioral Parameters.” In T. W. Hertel (ed.), Global Trade Analysis Using the GTAP Model (pp. 124–48). Cambridge: Cambridge University Press. IMF. 2004, September. World Economic Outlook. Washington, DC: International Monetary Fund. Kitamura, U., N. Takayama, and F. Arita. 2001, September. Household Savings and Wealth Distribution in Japan. Discussion Paper No. 38, Project on Intergenerational Equity, Institute of Economic Research. Tokyo: Hitotsubashi University. Lee, R. D. 2003. “The Demographic Transition: Three Centuries of Fundamental Change.” Journal of Economic Perspectives 17(4), 167–90. Lewis, W. A. 1952. “World Production, Prices and Trade, 1870–1960.” Manchester School of Economic and Social Sciences 20(2), 105–38. Lipsey, R. E. 1994. Quality Change and Other Influences on Measures of Export Prices of Manufactured Goods. World Bank Policy Research Working Paper 1348. Washington, DC: World Bank. Liu, J., N. Van Leeuwen, T. T. Vo, R. Tyers, and T. W. Hertel. 1998, September. Disaggregating Labor Payments by Skill Level in GTAP. Technical Paper No. 11. West Lafayette, IN: Center for Global Trade Analysis, Purdue University, West Lafayette. Available at http://www.agecon.purdue.edu/gtap/techpapr/tp-11.htm. Mason, A. (ed.). 2003. Population Change and Economic Development in East Asia: Challenges Met and Opportunities Seized. Stanford, CA: Stanford University Press. McDonald, P. and R. Kippen. 2001. “Labour Supply Prospects in 16 Developed Countries.” Population and Development Review 27(1): 1–32. Obstfeld, M. and K. Rogoff. 2000. “The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?” In B. Bernanke and K. Rogoff (eds.), NBER Macroeconomics Annual 2000 (pp. 339–412). Cambridge: NBER and the MIT Press. OECD. 1996. Ageing in OECD Countries. Social Policy Studies No. 20. Paris: Organization for Economic Cooperation and Development. OECD. 1998. Maintaining Prosperity in an Ageing Society. Paris: Organization for Economic Cooperation and Development. OECD. 1999. Employment Outlook. Paris: Organization for Economic Cooperation and Development.


375

OECD. 2002. Employment Outlook. Paris: Organization for Economic Cooperation and Development. Paulin, G. D. 2000. “Expenditure Patterns of Older Americans, 1984–97.” Monthly Labour Review 123(5), 3–28. Productivity Commission. 1999, November. Microeconomic Reforms and Australian Productivity: Exploring the Links. Commission Research Paper. Melbourne: Government of Australia. Productivity Commission. 2001, April. Resource Movements and Labour Productivity, An Australian Illustration 1994–95 to 1997–98. Staff Research Paper. Melbourne: Government of Australia. Regmi, A., M. S. Deepak, J. L. Seale Jr., and J. Bernstein. 2001. Cross Country Analysis of Food Consumption Patterns. Economic Research Service WRS-01–1. Washington, DC: U.S. Department of Agriculture. Shi, Q. and R. Tyers. 2004. Global Population Forecast Errors, Economic Performance and Australian Food Export Demand. Working Papers in Economics and Econometrics No.439. Canberra: Australian National University. Available at http://ecocomm.anu. edu.au/ecopapers. Trefler, D. and H. Lai. 1999. Gains from Trade: Standard Errors with the CUS Monopolistic Competition Model. Toronto: University of Toronto. Tyers, R. and J. Golley. 2006. China’s Growth to 2030: The Roles of Demographic Change and Investment Risk. Paper presented at the conference, WTO, China and the Asian Economies IV: Economic Integration and Development, University of International Business and Economics, June 24–5, Beijing. Tyers, R. and Q. Shi. 2007. “Global Demographic Change, Policy Responses and Their Economic Implications.” World Economy 30(4), 537–66. Tyers, R., Q. Shi, and M. M. Chan. 2005, September. Global Demographic Change and Economic Performance: Implications for the Food Sector. Report to the Rural Industries Research and Development Corporation. Canberra: RIRDC. United Nations. 2000, March. Replacement Migration: Is It a Solution to Declining and Ageing Populations? UN Population Division. New York: UN Secretariat. United Nations. 2003, February. World Population Prospects: The 2002 Revision. UN Population Division. New York: UN Secretariat. Available at www.un.org/esa/population/ publications/wpp2002. Vedi, J. 2005. The Global Economic Effects of Expanded Skilled Migration. Honours dissertation, School of Economics, College of Business and Economics, Australian National University. Weber, G., R. Miniaci, and C. Monfardini. 2002. “Changing Consumption Patterns.” In H. Siebert (ed.), Economic Policy for Ageing Societies (pp. 53–76). Berlin: Springer Verlag. WHO. 2003. Mortality Database: Table One: Number of Registered Deaths. Geneva: World Health Organization. Available at http://www3.who.int/whosis/menu.cfm?path= whosis,inds,mort&language=english. Williamson, J. G. 1998. “Growth, Distribution and Demography: Some Lessons from History.” Explorations in Economic History 35(3), 241–71.

PART IV

EVALUATION OF THE DYNAMIC GTAP FRAMEWORK



FOURTEEN

Household Saving Behavior in the Dynamic GTAP Model Evaluation and Revision Alla Golub and Robert A. McDougall 1. Introduction The GDyn model presented in Chapter 2 inherits from the standard GTAP model its specification of the regional household demand system and, in particular, the treatment of saving. As in the standard GTAP model, regional households in GDyn spend their income according to a Cobb-Douglas per capita utility function specified over three sources of utility: private consumption, government consumption, and real saving. Because the model is not forward-looking but is recursively dynamic, the utility function is static – it represents utility from present but not future consumption. The practice of including saving in the static utility function derives from Howe (1975) and allows regional households to value saving in atemporal settings.1 Because of the Cobb-Douglas functional form, the average propensity to save is fixed, and saving is a fixed proportion of income in each region.2 There are several unwelcome implications of this assumption. Because propensities to save are fixed and incomes are rising over time, countries in which saving substantially exceeds investment, like Japan, accumulate unrealistically large stocks of foreign assets. If like China they also exhibit high rates of growth in income, at the end of long-run GDyn simulations that span many decades, such countries may end up owning a large part of the wealth of the whole world. Although such outcomes cannot altogether 1 2

See Hertel (1997) for a discussion of this issue. In fact, according to McDougall (2002) the propensity to save is not quite fixed, but it is close enough to fixed.

We thank Elena Ianchovichina and Terrie Walmsley for many useful comments. The application related to this chapter is Ch14_gdyns_34_97.zip and is available for download on the Web site at https://www.gtap.agecon.purdue.edu/models/Dynamic/applications.asp

379

380


be ruled out a priori, they are very strong predictions resting on a very weak empirical basis. In the real world, saving and investment are highly correlated across countries (Feldstein and Horioka 1980) and net international capital flows, and as a result, net foreign positions are much smaller. Another problem is that as economies with high savings rates, like China, grow there is a glut of global savings and, as a result, of investment and capital in the world. Because of excessive investment, rates of return to capital fall without bound. This prevents us from running simulations with the GDyn model over very long time horizons. This chapter evaluates the performance of GDyn presented in Chapter 2 by comparing the model’s projected outcomes for gross and net foreign assets and liabilities with empirical data and proposes a change to the household saving behavior to address the problems associated with the saving behavior in GDyn. We start from a review of theories explaining household saving behavior and their implementation in CGE models. In forwardlooking models, saving rates are determined by the tradeoff between utility from present consumption and utility from future consumption. However, this approach cannot be implemented in a recursive model like GDyn. In this work we adopt an approach that supports a balanced growth scenario in which regional income, wealth, and saving have the same growth rate. The approach has no particular theoretical foundation, but is practical and motivated by the stylized fact that gross foreign assets and liabilities do not diverge through time in reality nearly as much as in standard GDyn. We modify the theoretical structure of GDyn so that the saving rate in each region is endogenous and is a function of the ratio of wealth to income.

2. Theories of Household Saving Behavior There are several theories explaining people’s desire to save. The most influential theory of household saving behavior is the life-cycle hypothesis developed in Modigliani and Brumberg (1954) and further tested in Ando and Modigliani (1963). This model starts from a consumer who maximizes his or her utility, which is a function of current and future consumption, subject to a resource constraint: “As a result of this maximization the current consumption of the individual can be expressed as a function of his resources and the rate of return on capital with parameters depending on age” (Ando and Modigliani 1963, p. 56). This theory suggests that people try to smooth their consumption over their lifetimes, saving little when they are young but much more in middle age and dissaving in retirement.

Household Saving Behavior in the Dynamic GTAP Model

381

Another theory of household saving behavior suggests that people save for precautionary reasons. If people are uncertain about their future income, and access to credit is limited, people tend to save more. At the same time, stock market and real estate capital gains reduce saving rates because people feel richer. Finally, people may adjust their saving in response to fiscal policies, hence the proposition of Ricardian equivalence. Despite the fact that data support these different theories, more or less, in different regions, it would be difficult to implement them in a CGE model. In the forward-looking G-Cubed CGE model (McKibbin and Wilcoxen 1995), household behavior is modeled by infinitely lived representative agent maximizing intertemporal utility subject to intertemporal budget constraint. However, only a portion of consumption – and, as it follows, saving – is determined by these intertemporally optimizing consumers. The remainder is determined by after-tax current income and the fixed marginal propensity to save. Because at least part of the supply of savings in the G-Cubed model is determined by the tradeoff between utility from present consumption and utility from future consumption, current saving is implausibly sensitive to remote future events. In the forward-looking CGE model described in Benjamin (1994), households choose current consumption and saving based on income level, expected future price level, and the interest rate. The household choice between present and future consumption is expressed with a Cobb-Douglas function, and thus, the budget share spent on saving is fixed. In the GREEN (Burneaux et al. 1992), recursively dynamic model, like in GDyn, the saving enters atemporal extended linear expenditure system utility function (Howe 1975), and the marginal propensity to save is constant and independent of the rate of reproduction of capital. The linkages between aging profiles and saving rates in a CGE model context are explored in Tyers (2005). To investigate the economic implications of changes in population sizes and age profiles, determined by changes in fertility, mortality, and migration, this work builds a demographic model into the GDyn model presented in Chapter 2. Tyers (2005) relaxes the assumption of fixed propensity to save in GDyn by introducing endogenous age-gender specific propensities to save, which depend on real disposable income and the real interest rate. These group-specific saving rates then determine regional saving rates in each period.

3. Comparison of Historic and Simulated Assets and Liabilities How large are gross and net foreign assets and liabilities in GDyn simulations? Our basis for comparison is a country portfolio database constructed

382


by Kraay et al. (2000) that was used in Chapter 3 of this book.3 We construct three indicators to evaluate foreign assets and liabilities in GDyn simulations: gross foreign assets as a share of household wealth, gross foreign liabilities as a share of domestic capital, and the ratio of net foreign assets to GDP. The choice of the first two indicators is driven by our interest in regional wealth allocation and the composition of capital in GDyn. The third indicator is of more general interest and is often used in discussion in the financial press and among policymakers about what size of net foreign liabilities is sustainable (Kouparitsas 2004). We compare the development through time of gross foreign positions in the Kraay et al. (2000) database and GDyn simulations. For this illustration, the GTAP 5.4 Data Base was aggregated to 22 regions and 3 sectors.4 To make the figures clear, we show results for just 11 of the 22 regions, representing 11 individual countries. Figure 14.1 shows gross foreign assets as a share of wealth in 1970 and 1997 from the Kraay et al. (2000) database together with gross foreign assets as a share of wealth in 1997 and 2024 simulated with GDyn. Years 1970 and 1997 from the Kraay et al. (2000) sample are chosen to capture changes over 27 years in gross foreign positions and because for some countries data before 1970 are not available. For GDyn simulations, the year 1997 is chosen because it is the initial year in our simulations and the benchmark year in the GTAP 5.4 Data Base, and it allows comparison of foreign positions in the GTAP and Kraay et al. (2000) databases. Year 2024 is chosen to show changes in gross foreign assets in the period of 27 years in the simulation. Similarly, Figure 14.2 shows gross foreign liabilities as a share of capital in 1970 and 1997 of the Kraay et al. (2000) database, together with gross foreign liabilities as a share of capital in 1997 and 2024 of GDyn simulations. First, there are differences in the 1997 gross foreign positions between Kraay et al. (2000) and the initial GDyn Data Base. This is because they adopt different approaches to construct wealth and gross foreign positions. Kraay et al. use historical information on investment accumulation, whereas in GDyn, regional capital is the starting point in the construction of wealth 3

4

Table 3.4 shows variation in historical gross foreign assets and liabilities across regions and over time. Figures 3.8 and 3.9 show the distribution of gross foreign assets as a share of wealth and gross foreign liabilities as a share of capital, respectively. On average, gross foreign assets and liabilities are small (see discussion of the size of gross foreign assets and liabilities in Chapter 3). Only the time variable is shocked. The sectors are agriculture, manufacturing, and services. The 22 countries/regions are Australia, New Zealand, China, Japan, Korea, South Asia, Canada, United States, South America, Austria, Denmark, Finland, United Kingdom, France, Ireland, Italy, Netherlands, Portugal, Sweden, Turkey, Rest of Europe, and Rest of World.


383

1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 Korea

Japan

Turkey

China Kraay 1970

US

France Netherlands Italy

Kraay 1997

GDyn 1997

Sweden

UK

Ireland

GDyn 2024

Figure 14.1. Gross foreign assets as share of wealth in Kraay et al. (2000) database and in GDyn simulation. For Japan, the first observation in the Kraay et al. database is for 1971. For Sweden, the last observation is for 1996. For Japan and Sweden, we show 1971 and 1996 gross foreign assets, respectively. Source: Authors’ simulation and Kraay et al. (2000) database.

and its components. The ownership shares in regional capital of domestic and foreign residents are then determined by foreign income receipts and payments. Although the differences between the initial starting point of simulation and the corresponding gross foreign positions recorded in Kraay et al. (2000) are important, they are not the focus of this discussion. Figures 14.1 and 14.2 reveal important differences in the development of gross foreign positions over time between the historical data and the GDyn simulation data. First, note that, although on average gross foreign positions are small, in some countries, like the United Kingdom and Ireland, gross foreign positions can be large. Growth in gross foreign positions in the United Kingdom and Ireland was very significant between 1970 and 1997 (Figs. 14.1 and 14.2). The large size of gross foreign positions in GDyn simulations need not be of concern as long as their size is comparable with historical data.

384


1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 Korea

Japan

Turkey

China

Kraay 1970

US

France Netherlands

Kraay 1997

GDyn 1997

Italy

Sweden

UK

Ireland

GDyn 2024

Figure 14.2. Gross foreign liabilities as share of capital in Kraay et al. (2000) and in GDyn simulation. For Japan, the first observation in the Kraay et al. database is 1971. For Sweden, the last observation is in 1996. For Japan and Sweden, we show 1971 and 1996 gross foreign liabilities, respectively. Source: Authors’ simulation and Kraay et al. (2000) database.

Second, as can be seen from Figures 14.1 and 14.2, if historical gross foreign assets and liabilities grow over time, they grow together, which results in small net foreign positions. In contrast, in GDyn simulations, growth in gross foreign assets corresponds to decline in gross foreign liabilities, and vice versa. The faster the growth in gross foreign assets, the more rapid the decline in gross foreign liabilities, as in Korea and Japan. The faster the growth in gross foreign liabilities, the faster the decline in gross foreign assets, as in Turkey and the United States. This relationship results in very large net foreign positions in GDyn simulations. To further illustrate this statement, we compare historical and simulated ratios of net foreign positions to GDP. The net foreign asset to GDP ratios calculated from the Kraay et al. (2000) database are shown in Figure 14.3. Kraay et al. (2000) provide a detailed discussion and presentation of net foreign asset positions. To save space, we show ratios for just 12 of 68 countries covered in Kraay et al. (2000). These 12 countries are chosen to represent all 6 regions of the Kraay et al. database (see Table 3.3) and to present extremes as well as common tendencies in


385

1.5

1

0.5

0

-0.5

-1

19

66 19 67 19 68 19 69 19 70 19 71 19 72 19 73 19 74 19 75 19 76 19 77 19 78 19 79 19 80 19 81 19 82 19 83 19 84 19 85 19 86 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97

-1.5

USA India

China New Zealand

Japan Switzerland

Korea Jamaica

Turkey Brazil

France Singapore

Figure 14.3. Net foreign position to GDP ratio for selected 12 countries from 1966 to 1997. Source: Kraay et al. (2000).

the database. Switzerland’s net foreign assets and New Zealand’s net foreign liabilities are two extreme cases among the industrialized countries. Net foreign position to GDP ratios of all countries in this group are in the range from −0.7 to 0.5. For the East Asia and Pacific countries the net foreign assets to GDP ratio ranges from −0.34 (Malaysia) to 0.86 (Singapore), with most observations in the range from −0.34 to 0.02. Among Latin American and Caribbean countries, Jamaica is an outlier, with relatively large net foreign liabilities (see Fig. 14.3). Net foreign positions relative to GDP of other Latin American and Caribbean countries range from −0.53 to 0.07. In the Middle East and North Africa (MENA) region, countries’ net foreign positions relative to GDP are in the range from −0.6 to 0.25, with Saudi Arabia’s net foreign positions as high as 1.34. In South Asia, net foreign assets to GDP ratios are in the range from −0.18 (Pakistan) to 0.02 (Bangladesh). Finally, net foreign position to GDP ratios of SubSaharan Africa (SSA) countries are in the range from −0.92 (Congo) to 0.15 (Lesotho). To summarize, historical net foreign assets are small. Simulated net foreign assets as a share of GDP are shown in Figure 14.4. For this simulation we use the same 22 × 3 aggregation described earlier. To make the figure clear, only 10 regions are shown. Net foreign assets as

386


15

10

5

0

-5

19

97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85

-10

China

Japan

Korea

SA

US

SAM

Turkey

France

Sweden

Italy

Figure 14.4. Net foreign position to GDP ratios in a GDyn simulation. For this simulation the GTAP 5 Data Base was aggregated to 22 regions. To make the figure clear only 10 regions are shown. The ratios of other 12 regions are in the range between ratios of Japan and Turkey. Source: Authors’ simulation.

a share of GDP of the other 12 regions in this aggregation are between the ratios of Japan and Turkey. As can be seen from a comparison of Figures 14.3 and 14.4, net foreign positions in GTAP simulations are unrealistically large. In less than 10 years of the simulation, the ratios become outside the ( −1, 1) range and continue to grow in absolute value. Fixed propensities to save in the model coupled with rising incomes lead to unrealistically large international capital flows. They create too much saving and, as a result, too much investment and capital in the world; they allow chronic rapid saving or dissaving and, as a result, too great foreign assets or liabilities in individual regions. To overcome this problem we modify household saving behavior by endogenizing saving rates. The development of wealth of a region, which is a sum of its capital stock and net foreign assets, is determined by its saving behavior. For illustrative purposes, consider a country where capital and income grow at the same rate. Rapid growth in net foreign liabilities relative to regional income will lead to a decrease in wealth relative to income, whereas fast-growing net foreign assets relative to income increase regional wealth-to-income ratios. Our treatment is based on a simple historical observation that country


387

20 18 16 14

ratio

12 10 8 6 4 2 0

1997 2002 2007 2012 2017 2022 2027 2032 2037 2042 2047 2052 2057 2062 2067 2072 2077 2082 2087 year China

Japan

Korea

US

France

Italy

Netherland

Turkey

Figure 14.5. Wealth-to-income ratios in a long-run GDyn simulation. Source: Authors’ simulation.

wealth-to-income ratios are in a specific range and do not change rapidly over time. As an illustration, we compare historical wealth-to-income ratios calculated from the Kraay et al. (2000) database with ratios in GDyn simulations. Rapidly growing wealth-to-income ratios in GDyn simulations are shown in Figure 14.5. By comparison, the historical wealth-to-income ratios in Figure 14.6 are very stable. To show this, we regress historical wealthto-income ratios on a time trend. The regression results are reported in Table 14.1. The time trend is not statistically significant in the majority of countries, with the exception of countries in the East Asia and Pacific region and a few Latin American and Caribbean countries. Figure 14.6 shows historical wealth-to-income ratio over time for selected countries and illustrates the regression result.

4. New Household Saving Behavior in GDyn We assume that regional households try to maintain target wealth-toincome ratios WYRT(r). The actual wealth-to-income ratio is given by WYRA(r) =

WQHHLD(r) , INCOME(r)

(14.1)

388


6

5

4

3

2

1

0

1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 China

Korea

US

Turkey

Finland

Singapore

Brazil

Germany

Figure 14.6. Historical wealth-to-income ratios of selected countries constructed using the Kraay et al. (2000) database.

where WQHHLD(r) and INCOME(r) denote household wealth and income. Differentiating (14.1), we obtain the percentage change equation: wyra(r) = wqh(r) − y(r),

(14.2)

where wyra(r) denotes the percentage change in the actual wealth-to-income ratio, wqh(r) the percentage change in the wealth of the regional household, and y(r) the percentage change in income. If the wealth-to-income ratio is too low or too high compared to the target wealth-to-income ratio in a region, then the regional household gradually adjusts its actual wealth-toincome ratio toward the target. Similar to the lagged adjustment mechanism in the rates of return presented in Chapter 2, we define a lagged adjustment mechanism for the wealth-to-income ratio: RRG WYR(r) = LAMBWYR(r)∗ log

WYRT(r) , WYRA(r)

(14.3)

where RRG_RORG(r) denotes the required rate of growth in the wealthto-income ratio, and LAMBWYR(r) is a coefficient of adjustment.


389

Table 14.1. The wealth-to-income ratio regressed on a time trend Country 1 INDC Australia Austria Canada Switzerland Germany Denmark Spain Finland France United Kingdom Greece Ireland Italy Japan Netherlands Norway New Zealand Portugal Sweden USA LAC Bolivia Brazil Colombia Costa Rica Dominican Republic Ecuador Guatemala Honduras Jamaica Mexico Nicaragua Peru Slovenia Trinidad and Tobago Uruguay

Intercept 2

Time trend 3

First lag 4

Second lag 5

R-square 6

Obs. 7

0.588∗∗ 0.902∗∗∗ 0.558∗∗∗ 2.231∗∗ 0.501∗∗ 0.979∗∗ 0.661∗∗ 1.026∗∗ 0.900∗∗ 0.796∗∗ 0.564∗∗ 0.688∗∗ 1.248∗∗∗ 1.655∗∗ 0.845∗∗ 0.903 0.921∗ 1.472∗∗∗ 0.701∗∗ 0.431∗∗

−0.007 0.013 0.000 0.146∗∗∗ −0.001 −0.010 0.008 −0.000 0.006 −0.004 −0.002 −0.002 −0.003 0.000 −0.000 −0.005 −0.002 −0.010 −0.005 −0.005∗∗∗

1.326∗∗∗ 1.096∗∗∗ 1.377∗∗∗

−0.495∗∗∗ −0.462∗∗ −0.669∗∗∗

1.051∗∗∗ 1.173∗∗∗ 1.163∗∗∗ 1.270∗∗∗ 1.015∗∗∗ 1.161∗∗∗ 1.199∗∗∗ 1.155∗∗∗ 0.966∗∗∗ 0.534∗∗ 1.165∗∗∗ 0.754∗∗∗ 0.593∗∗∗ 0.878∗∗∗ 1.351∗∗∗ 0.855∗∗∗

−0.200 −0.425∗∗ −0.442∗∗ −0.528∗∗∗ −0.338∗ −0.470∗∗ −0.428∗∗ −0.451∗∗ −0.329∗

0.895 0.788 0.819 0.666 0.909 0.779 0.792 0.781 0.713 0.724 0.771 0.718 0.610 0.356 0.755 0.762 0.361 0.618 0.843 0.904

30 29 30 14 28 27 30 30 28 30 29 30 28 15 30 22 24 26 29 28

0.909∗∗ 0.524∗∗∗ 0.268 0.514∗∗∗ 0.300∗∗∗ 0.575∗∗ 0.307∗∗ 0.312∗ 0.672∗∗ 0.667∗∗∗ 0.694∗∗ 0.692∗ 0.197∗∗ 0.576∗∗ 0.125

0.035∗∗ 0.024∗∗∗ −0.002 0.010∗∗ 0.067∗∗∗ −0.005 −0.002 −0.002 0.003 0.007∗ 0.002 0.063∗∗∗ 0.002 −0.004 0.021∗∗

0.761 0.822 0.763 0.758 0.941 0.653 0.490 0.680 0.479 0.646 0.030 0.393 0.680 0.364 0.708

19 30 26 29 16 30 28 31 27 30 15 19 31 20 16

0.428∗∗ 0.878∗∗∗ 1.055∗∗∗ 0.425∗∗ 0.754∗∗∗ 0.687∗∗∗ 0.749∗∗∗ 0.621∗∗∗ 0.894∗∗∗ 0.180 0.729∗∗∗ 0.568∗∗∗ 0.487∗∗

−0.412∗∗ −0.432∗∗ −0.590∗∗∗

−0.414∗∗ −0.243

−0.449∗∗

(continued)

390

Alla Golub and Robert A. McDougall Table 14.1 (continued)

Country 1 EAP China Indonesia Korea Malaysia Philippines Singapore Thailand MENA Algeria Iran Israel Jordan Morocco Oman Syria Tunisia Turkey SA India Sri Lanka Pakistan SSA Cote d’Ivoire MUS South Africa

Intercept 2

Time trend 3

First lag 4

0.018∗∗ 0.013∗∗ 0.018∗∗ 0.014∗ 0.010∗∗ 0.009∗ 0.003

0.330 1.019∗∗∗ 1.295∗∗∗ 0.410∗ 1.022∗∗∗ 1.378∗∗∗ 1.232∗∗∗

0.518∗∗ −0.065 0.710∗∗ 0.237∗ 0.354∗∗ 0.370∗ −0.927∗∗ 0.315∗ 0.400∗

0.015 0.067∗ −0.003 −0.009∗∗∗ 0.001 0.021 0.019∗∗ −0.007∗∗ 0.003

0.572∗∗∗ 0.960∗∗ 0.681∗∗∗ 0.870∗∗∗ 0.842∗∗∗ 0.538∗∗ 1.305∗∗∗ 0.835∗∗∗ 0.761∗∗∗

0.300∗∗ −0.034 0.151

−0.004∗∗ 0.019∗ −0.000

1.196∗∗∗ 0.533∗∗ 1.176∗∗∗

−0.396∗∗

0.006 −0.002 0.001

1.165∗∗∗ 0.710

−0.336

0.441 0.305∗∗∗ 0.291∗∗∗ 0.703∗∗ 0.311∗∗ 0.332∗∗ 0.175

0.877∗∗∗ 0.218 0.543∗∗

Second lag 5

−0.464∗∗ −0.634∗∗∗ −0.416∗ −0.554∗∗∗ −0.384∗ −0.414 −0.365 0.384

−0.379∗∗

R-square 6

Obs. 7

0.533 0.908 0.965 0.582 0.883 0.961 0.870

16 27 27 18 29 30 27

0.762 0.917 0.537 0.761 0.447 0.814 0.816 0.865 0.662

25 15 28 23 23 16 19 31 31

0.885 0.946 0.834

30 17 22

0.013 0.817 0.541

16 22 29

∗∗∗ , ∗∗ ,

and ∗ denote significance levels at 0.01, 0.05, and 0.1, respectively. Number of lags is chosen to whiten the errors. Source: Authors’ calculations using Kraay et al. (2000) data.

Differentiating (14.3) we obtain rrg wyr(r) = LAMBWYR(r)∗ (wyrt(r) − wyra(r)),

(14.4)

where rrg_wyr(r) denotes the (absolute) change in the required percentage point rate of growth in the wealth-to-income ratio WYRA(r), and wyrt(r) is the percentage change in the target wealth-to-income ratio. The latter is typically exogenous and zero. If the actual wealth-to-income ratio is higher or lower than the target ratio in a region, then the regional household will decrease or increase,


391

respectively, its saving rate to move toward the target ratio. The actual rate of growth in the wealth-to-income ratio is defined as ARG WYR(r) =

dWQHHLD(r)/dTIME dWYRA(r)/dTIME = WYRA(r) WQHHLD(r) −

dINCOME(r)/dTIME . INCOME(r)

(14.5)

Expression (14.5) says that the actual rate of growth in the wealth-to-income ratio is the difference between the rate of growth in household wealth and the rate of growth in income. From Chapter 2, change in household wealth is given in the model by equation (14.6): WQHHLD(r)∗ wqh(r) = WQHFIRM(r)∗ pcgds(r) +WQHTRUST(r)∗ pqtrust(r) + 100∗ SAVE(r)∗ time,

(14.6)

where WQHFIRM(r) denotes household equity in domestic firms; pcgds(r), the percentage change in the price of equity in domestic firms; WQHTRUST(r), the household equity in the global trust; pqtrust(r), the percentage change in the price of equity in the global trust; and SAVE(r), the time rate of (net of depreciation) household saving. Expression (14.6) says that change in household wealth is determined by changes in the prices of old assets plus the rate of new saving both in domestic assets and abroad. We assume that in forming its expectation of rate of growth in its wealth, the regional household takes into account only changes in wealth that are due to new saving, ignoring current changes in asset prices (or, equivalently, assuming that asset prices are constant). Under this assumption, expected change in household wealth is dWQHHLD(r) = SAVE(r)∗ dTIME.

(14.7)

From (14.7), the expected rate of growth in household wealth is dWQHHLD(r)/dTIME SAVE(r) = . WQHHLD(r) WQHHLD(r)

(14.8)

In estimating the rate of growth in the wealth-to-income ratio, the regional household takes the perceived long-run economic growth of its own region into account. The normal rate of growth in income, which we denote as YHAT(r), is a long-run concept and is different from the actual percent change in income in every period of simulation. Using (14.8) and

392


YHAT(r) we now can define the expected rate of growth in the wealth-toincome ratio: ERG WYR(r) =

SAVE(r) − YHAT(r). WQHHLD(r)

(14.9)

When the long-run steady state is known, we can set YHAT(r) to its actual long-run value at the beginning of the simulation. For example, if a simulation is characterized by the static steady state with no growth in endowments and no technological progress in the long-run, then we can set YHAT(r) equal to zero in the initial period and keep it equal to zero during the simulation. In scenarios where the long-run economic growth is not known precisely, the household is allowed to adjust YHAT(r) so that as the economy converges toward a steady growth equilibrium, YHAT(r) converges toward the actual steady equilibrium growth rate. We postulate a gradual adjustment mechanism in YHAT(r): DYHAT(r) = LAMBYHAT(r)∗ (y(r) − 100∗ YHAT(r)∗ time),

(14.10)

where DYHAT(r) denotes absolute percentage point change in YHAT(r), and LAMBYHAT(r) is a coefficient of adjustment. Totally differentiating (14.9), we obtain erg wyr(r) =

SAVE(r) (psave(r) + qsave(r) − wqh(r)) − DYHAT(r), WQHHLD(r) (14.11)

where erg_wyr(r) denotes the absolute percentage point change in the expected rate of growth in wealth-to-income ratio; psave(r), the percentage change in the price of saving; and qsave(r), the percentage change in real saving. To achieve the required rate of growth in wealth-to-income ratio, RRG_WYR(r), the regional household should make an appropriate savings decision. Each period, the regional household decides how much of its income should be saved to achieve the required rate of growth in the wealth-to-income ratio. When making the decision, the regional household adjusts its propensity to save through the variable dpsave(r), which is exogenous in the standard GTAP treatment of regional household demand, but endogenous with the new structure. The expected rate of growth in the wealth-to-income ratio is equal to the required wealth-to-income ratio, both in levels: ERG WYR(r) = RRG WYR(r),

(14.12)


393

and in differentials: erg wyr(r) = rrg wyr(r).

(14.13)

Substituting (14.3) and (14.9) into (14.12), we see that the rate of saving is given by the condition that the expected rate of growth is equal to the required rate of growth in the wealth-to-income ratio: WYRT(r) SAVE(r) − YHAT(r) = LAMBWYR(r)∗ log . WQHHLD(r) WYRA(r)

(14.14)

The four equations in the following box comprise the new savings module in the model. Box 14.1. New savings module wyra(r) = wqh(r) − y(r)

(14.2)

erg wyr(r) = LAMBWYR(r)∗ (wyrt(r) − wyra(r))

(14.4)

DYHAT(r) = LAMBYHAT(r)∗ (y(r) − 100∗ YHAT(r)∗ time) (14.10) WQHHLD(r) (erg wyr(r) + DYHAT(r)) = psave(r) + qsave(r) − wqh(r) SAVE(r) (14.11)

One new exogenous variable, wyrt(r) denotes the percentage change in the target wealth-to-income ratio. New endogenous variables include the percentage change in the actual wealth-to-income ratio, wyra(r), determined by equation (14.2); the percentage change in the propensity to save, dpsave(r), determined by equation (14.11); and the change in the expected percentage point rate of growth in the wealth-to-income ratio, erg_wyr(r), determined by equation (14.4). We also add a variable representing the absolute change in the normal percentage point rate of income growth, DYHAT(r), determined by equation (14.10).

394


5. Data and Parameters To implement the new household saving treatment in the model, we need to set the initial value of the coefficient YHAT(r), the parameters LAMBWYR(r) and LAMBYHAT(r), and the target wealth-to-income ratio WYRT(r). We set the initial perceived long-run economic growth rate YHAT(r) equal to a geometric average of historical annual growth rates. Time-series data for gross national income (GNI) at constant local currency units are obtained from the World Development Indicators database, which covers the period from 1960 to 2004. However, we use just the last 15 years (1990–2004), because for many countries observations for earlier years are missing. Another reason to restrict the series to later years is to avoid incorporating past episodes of rapid economic growth into our current assessments. For example, annual growth rates in the Japanese economy between 1960 and 1973 were in the range of 6 to 13%, but became much lower in recent years. Having the GNI time series from 1990 to 2004, we aggregate over countries according to an aggregation scheme and then calculate the geometric average GNI growth rate, which is a proxy for YHAT(r). Note that precision in estimating YHAT(r) is not very important because it is updatable. However, as we see later, the initial YHAT(r) setting plays a role in setting the regional target wealth-to-income ratios. The parameter LAMBWYR(r) determines the initial distance between the target and actual wealth-to-income ratios and how fast the actual wealthto-income ratio moves toward the target ratio. From (14.14), the initial distance between the wealth-to-income target and actual ratios is given by the formula WYRT(r) 1 SAVE(r) = exp − YHAT(r) . WYRA(r) LAMBWYR(r) WQHHLD(r) (14.15) Other things being equal, smaller LAMBWYR(r) leads to a larger gap between target and actual wealth-to-income ratios in the initial database. If growth in regional wealth is faster than perceived normal economic growth (YHAT(r)), then the expression in parentheses is positive and WYRT(r) is greater than WYRA(r). Similarly, when growth in regional wealth is slower than perceived normal economic growth, then the expression in parentheses is negative and WYRT(r) is less than WYRA(r).


395

To achieve desired wealth-to-income ratios, regions adjust their propensities to save. Consider a simple scenario with no growth in endowments and no technological progress over the long-run. For such a scenario, we can set YHAT(r) equal to zero in the initial period and keep it equal to zero during the simulation, removing the adjustment mechanism equation (14.10). The region accumulates more wealth and increases its wealth-toincome ratio, moving it closer to the target. As the distance between the target and actual wealth-to-income ratios diminishes, less growth in wealth is needed, and the saving rate declines. When WYRA(r) = WYRT(r), the regional household saves only to cover its capital depreciation, and the net saving rate SAVE(r) is zero. How much adjustment should we allow in wealth-to-income ratios? An answer to this question will help us set parameter LAMBWYR(r). As discussed in Section 3 and as shown by the insignificant time trends in wealth-to-income ratios in Table 14.1, these ratios are very stable for most countries. This suggests that the target wealth-to-income ratios should be set close to the actual wealth-to-income ratios calculated in the initial database. The size of historical wealth-to-income ratios is also helpful in setting target wealth-to-income ratios. The ratio is larger for industrialized countries than for other countries in the sample. The ratios for industrialized countries in the period 1966–97 are in the range from 1.5 to 5.5, with the exception of Switzerland’s, which is in the range from 4.35 to 6.72.5 For other countries the ratios are in the range from 0.65 to 4 in the East Asia and Pacific region, from 0.5 to 3.7 in Latin America and the Caribbean region, from 0.5 to 3.7 in the Middle East and North Africa, from 0.5 to 1.6 in South America, and from 0.5 to 2.7 in Sub-Saharan Africa. From these observations, it appears that a reasonable upper bound on the target wealth-to-income ratio is 5.5. We now present a specific example illustrating how we set LAMBWYR(r). We use the 22 × 3 aggregation of the GTAP 5 Data Base mentioned in Section 3. Following equation (14.15), the target wealth-to-income ratio is given as ∗ 1 SAVE(r) − YHAT(r) . WYRT(r) = WYRA(r)∗exp LAMBWYR(r) WQHHLD(r) (14.16) 5

This range is for 1983–96 because the data for Switzerland in the Kraay et al. (2000) database are available only for those years.

396


Table 14.2. Calculation of target wealth-to-income ratios for 22 × 3 aggregation of

the GTAP 5.4 Data Base when LAMBWYR(r) = 0.05

Region Australia New Zealand China Japan Korea South Asia Canada United States South America Austria Denmark Finland France United Kingdom Ireland Italy Netherlands Portugal Sweden Europe Turkey ROW

WYRAa

YHATb

SAVE/ WQHHLDc

SAVE/ WQHHLD − YHAT

WYRT

2.842 2.476 2.657 4.013 3.296 2.894 2.528 2.626 3.013 4.064 2.356 3.325 3.531 3.050 2.111 3.534 3.593 2.769 2.286 3.676 3.075 3.770

0.037 0.030 0.073 0.013 0.053 0.050 0.028 0.032 0.029 0.021 0.025 0.019 0.020 0.027 0.060 0.015 0.024 0.027 0.017 0.024 0.030 0.034

0.031 0.024 0.100 0.051 0.081 0.065 0.039 0.028 0.022 0.022 0.046 0.027 0.027 0.026 0.047 0.021 0.051 0.012 0.040 0.027 0.031 0.013

−0.007 −0.006 0.027 0.039 0.028 0.014 0.010 −0.003 −0.007 0.001 0.021 0.009 0.007 −0.002 −0.013 0.006 0.027 −0.014 0.023 0.003 0.001 −0.021

2.489 2.187 4.548 8.688 5.737 3.866 3.107 2.470 2.614 4.140 3.588 3.942 4.099 2.946 1.632 3.973 6.179 2.077 3.613 3.918 3.132 2.475

a

Wealth-to-income ratio, based on GTAP 5.4 Data Base. Geometric average growth calculated using WDI database. c Wealth growth rate, based on GTAP 5.4 Data Base. Source: Authors’ calculations using sources indicated in the table. b

The inputs into the calculation, together with the resulting target wealth-toincome ratios when LAMBWYR(r) = 0.05, are given in Table 14.2. For most of the regions in this aggregation, the calculated target wealth-to-income ratios are within the plausible range. The exceptions are Japan, Korea, and the Netherlands. In those countries unusually high target wealth-to-income ratios result from a combination of high initial ratios (WYRA(r)) and rapid initial growth in the ratios, calculated as SAVE(r)/WQHHLD(r) − YHAT(r). The latter is driven by slow growth in income in Japan, rapid growth in wealth in Korea, and both slow growth in income and rapid growth in wealth in the Netherlands. It is possible that for these countries the 1997


397

savings-to-wealth ratio recorded in GTAP 5.4 Data Base is abnormally high. Yet what is really important is that even these relatively high target wealthto-income ratios impose an upper bound on actual wealth-to-income ratios in GDyn simulations, and these bounds are much lower than the ratios that would be observed with fixed saving rates, as illustrated in Figure 14.5. How sensitive are the target ratios with respect to LAMBWYR(r)? As can be seen from equation (14.16), as LAMBWYR(r) becomes very small, WYRT(r) approaches infinity. Large LAMBWYR(r) leads to WYRT(r) = WYRA(r). This is supported by Figures 14.7 and 14.8. With LAMBWYR(r) below 0.05 (Fig. 14.7), WYRT(r) becomes very large. In contrast, with LAMBWYR(r) larger than 0.1 (Fig. 14.8), target wealth-to-income ratios are not sensitive to LAMBWYR(r), but remain very close to initial actual ratios WYRA(r). Taking into account the historical fact that wealth-to-income ratios are stable for most countries, setting LAMBWYR(r) to 0.1 or larger seems plausible. However, a high setting of the parameter would lead to rapid changes in savings rates in early years of the simulations; for example, rapid reductions in saving in Japan, Korea, China, and the Netherlands and rapid increases in the residual Rest of World region. However, although it is plausible that such changes will occur eventually, there is little reason to expect them to take place rapidly and in the near future. We therefore prefer to avoid such obtrusive phenomena in our scenario results. It is also possible that, for some scenarios, setting the adjustment parameter too high may make the model unsolvable. Based on these considerations, we conclude that the acceptable range for LAMBWYR(r) is around 0.05 to 0.1. Finally, we need to set LAMBYHAT(r). This plays the same role in the formation of expected growth in income in our new saving treatment as LAMBKHAT(r) plays for expected growth in the capital stock in GDyn investment theory, presented in Chapter 2. Because LAMBYHAT(r) has a similar role to LAMBKHAT(r), we set LAMBYHAT(r) equal to LAMBKHAT(r), that is, to 0.2. As noted earlier, the introduction of the adjustment mechanism for the expected growth in income (equation [14.10]) is for convenience only. In some cases, it may prove more convenient to delete the adjustment mechanism and provide a fixed expected growth rate. In particular, high settings of the adjustment parameter lead to faster adjustment toward equilibrium and larger changes in the model and may make the model unsolvable. Specifically, large changes in saving may lead to negative utility from savings in the top-level Cobb-Douglas regional household demand system. In such cases, the solution is to calibrate YHAT(r) for the specific simulation and set it consistent with long-run trends in the scenario.

398


250

target wealth-to-income ratio

200

150

100

50

0 0.01

0.02

0.03

0.04

0.05

LAMBWYR Japan

Korea

United States

France

United Kingdom

Netherlands

Europe

Figure 14.7. Target wealth-to-income ratios in GDyn simulations when LAMBWYR(r) is relatively small. Source: Authors’ calculation. 10 9

target wealth-to-income ratio

8 7 6 5 4 3 2 1 0 0.05 0.07 0.09 0.11 0.13 0.15 0.17 0.19 0.21 0.23 0.25 0.27 0.29 0.31 0.33 0.35 0.37 0.39

LAMBWYR Japan

Korea

United States

France

United Kingdom

Netherlands

Europe

Figure 14.8. Target wealth-to-income ratios in GDyn simulations when LAMBWYR(r) is relatively large. Source: Authors’ calculation.


399

6. Two Simulations and Final Notes We conclude this chapter with two illustrative simulations based on the 22 × 3 aggregation of the GTAP Data Base, followed by a discussion of directions for future research to improve the realism of the GDyn model. In the first illustrative simulation there is no growth in endowments and productivity, and only the time variable is shocked. Here, we assume that the long-run steady state is static; that is, there is no growth in productivity or endowments. Under these assumptions net saving tends toward zero over the long-run, as does YHAT(r), resulting in relatively high target wealth-toincome ratios. In terms of the saving rate adjustment, it is an extreme case because it requires a large adjustment in such countries as China, Japan, and Korea in this 90-year simulation. Figure 14.9 shows convergence of wealth-to-income ratios toward their targets in the long-run simulation. Figure 14.10 contrasts these wealth-to-income ratios with the unbounded ratios when the saving rates are constant. Figure 14.11 shows endogenously declining saving rates. Figure 14.12, contrasted with Figure 14.4, illustrates how ratios of net foreign assets to GDP are now bounded. These ratios are still high, however, compared to the historically observed ratios in Figure 14.3. In the second illustrative simulation, we use population, labor, and GDP growth given in the baseline provided in Chapter 5. In the longrun we assume that GDP growth rates gradually converge to levels given in Table 14.3 and that labor and population growth rates remain after 2020 as in 2020. Wealth-to-income ratios, saving rates, and the share of net foreign asset in GDP in this simulation are shown in Figures 14.13–14.15. Figure 14.16 shows the global rate of return to capital in the long-run GDyn simulations at constant and endogenous saving rates. New household saving behavior allows reduction in the rate of growth in world capital and stabilization of the global rate of return to capital. The proposed modification enforces long-run stabilization in regional wealth-to-income ratios, which is critical to obtaining a sensible baseline in light of currently divergent savings rates across countries. As a direction for future research, it would be interesting to compare investmentsaving correlations in the simulations from the model with and without new household saving behavior, and with correlations reported in the studies following Feldstein and Horioka (1980). Finally, it would be interesting to validate projections of the saving rates generated by the new mechanism against those tied to regional aging profiles and reported in Tyers (2005).

400

Alla Golub and Robert A. McDougall Table 14.3. Long-run growth rates assumed in the second

illustrative simulation Region

YHAT

Region

YHAT

Australia New Zealand China Japan Korea South Asia Canada US South America Austria Denmark

0.031 0.030 0.035 0.023 0.035 0.039 0.023 0.026 0.035 0.028 0.025

Finland France United Kingdom Ireland Italy Netherlands Portugal Sweden Europe Turkey ROW

0.032 0.024 0.026 0.035 0.025 0.027 0.027 0.028 0.026 0.035 0.036

Source: Authors’ calculations based on time-series data (1990–2004) for gross national income (GNI) at constant local currency units obtained from the World Development Indicators database.

8

7

6

5

4

3

2

1

19

97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85

0

China

Japan

Korea

SA

US

SAM

Turkey

France

Sweden

Italy

Figure 14.9. Wealth-to-income ratio in the GDyn simulations with new household saving behavior, LAMBWYR(r) set to 0.1, and long-run growth assumed to be zero. Source: Authors’ simulation.


401

20 18 16 14 12 10 8 6 4 2 0 China

Japan

Korea

US

France

constant saving rates

Italy

Netherlands

Turkey

endogenous saving rates

Figure 14.10. Comparison of wealth-to-income ratios at the end of 90-year GDyn simulations. Source: Authors’ simulations. 0.3

0.25

0.2

0.15

0.1

0.05

09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85

01

05

20

20

20

19 97

0

China

Japan

Korea

SA

US

SAM

Turkey

France

Sweden

Italy

Figure 14.11. Saving rates in GDyn simulations with new household saving behavior, with long-run growth assumed to be zero. Source: Authors’ simulation.

402


15

10

5

0

-5

19 97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85

-10

China

Japan

Korea

SA

US

SAM

Turkey

France

Sweden

Italy

Figure 14.12. Net foreign assets as a share of GDP in GDyn simulations with new household saving behavior, with long-run growth assumed to be zero. Source: Authors’ simulation. 6

5

4

3

2

1

19 97 20 01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85

0

China

Japan

Korea

SA

US

SAM

Turkey

France

Sweden

Italy

Figure 14.13. Wealth-to-income ratio in GDyn simulations with new household saving behavior, LAMBWYR(r) set to 0.1, and long-run growth rates as given in Table 14.3. Source: Authors’ simulation.


403

0.3

0.25

0.2

0.15

0.1

0.05

01 20 05 20 09 20 13 20 17 20 21 20 25 20 29 20 33 20 37 20 41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85

20

19

97

0

China

Japan

Korea

SA

US

SAM

Turkey

France

Sweden

Italy

Figure 14.14. Saving rates in GDyn simulations with new household saving behavior, with long-run growth rates as given in Table 14.3. Source: Authors’ simulation. 15

10

5

0

-5

China

Korea

Sweden

81

85

20

20

73

77

20

20

65

69

20 France

20

61

57

Turkey

20

53

20

49 20

SAM

20

45

41

US

20

37

20

33 SA

20

20

25

29 20

21

Japan

20

17

20

20

09

13

20

20

05 20

01 20

19

97

-10

Italy

Figure 14.15. Net foreign assets as a share of GDP in GDyn simulations with new household saving behavior, with long-run growth rates as given in Table 14.3. Source: Authors’ simulation.

404


0.11

0.1

0.09

0.08

0.07

0.06

0.05

Net rate of return at constant saving rate

41 20 45 20 49 20 53 20 57 20 61 20 65 20 69 20 73 20 77 20 81 20 85

20

33

29

25

37

20

20

20

21 20

20

13

09

05

01

17 20

20

20

20

20

19

97

0.04

Net rate of return at new household saving behavior

Figure 14.16. Global net rate of return to capital at constant and endogenous saving rates in the long-run GDyn simulations. Source: Authors’ simulation.

References Ando, A. and F. Modigliani. 1963. “The ‘Life Cycle’ Hypothesis of Saving: Aggregate Implication and Tests.”American Economic Review 53(1), 55–84. Benjamin, N. 1994. “Investments, Expectations, and Dutch Disease: A Comparative Study (Bolivia, Cameroon, Indonesia).” In J. Mercenier and T. N. Srinivasan (eds.), Applied General Equilibrium and Economic Development (pp. 235–51). Ann Arbor: University of Michigan Press. Burniaux, J., G. Nicoletti, and J. Olivera-Marins, 1992. GREEN: A Global Model for Quantifying the Costs of Policies to Curb CO2 Emissions. OECD Economic Studies 19. Paris: OECD. Feldstein, M., and C. Horioka. 1980. “Domestic Savings and International Capital Flows.” Economic Journal 90, 314–29. Hertel, T. W. and M. E. Tsigas. 1997. “Structure of GTAP.” In T. W. Hertel (ed.), Global Trade Analysis: Modeling and Applications (pp. 13–73). Cambridge: Cambridge University Press. Howe, H. 1975.“Development of the Extended Linear Expenditure System from Simple Saving Assumptions.”European Economic Review 6, 305–10. Ianchovichina, E. and R. McDougall. 2001. Structure of Dynamic GTAP. GTAP Technical Paper 17. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Available at https://www.gtap.agecon.purdue.edu/resources/download/160 .pdf.


405

Kouparitsas, M. 2004. How Worrisome is the U.S. Net Foreign Debt Position? Chicago Federal Letter, No. 2002. Available at http://www.chicagofed.org/publications/fedletter/ cflmay2004 202.pdf. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. National Bureau of Economic Research Working Paper Series No. 7795:1–61. Washington, DC: NBER. McDougall, R. 2002. A New Regional Household Demand System for GTAP. GTAP Technical Paper No. 20. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Available at https://www.gtap.agecon.purdue.edu/resources/download/1593.pdf. McKibbin, W. J. and P. J. Wilcoxen. 1995. “The Theoretical and Empirical Structure of G-Cubed.” Mimeo, Brookings Institution. Modigliani, F. and R. Brumberg 1954. “Utility Analysis and the Consumption Function: An Interpretation of Cross-Section Data.” In K. K. Kurihara, ed., Post-Keynesian Economics (pp. 388–436). New Brunswick, NJ: Rutgers University Press. Tyers, R. 2005. Aging and Slower Population Growth: Effects on Global Economic Performance. Paper presented to the Experts’ Meeting on Long Term Scenarios for Asia’s Growth and Trade, November 10–11, Manila. Available at http://www.adb .org/Documents/Events/2005/Experts-Meeting/agenda.asp.

FIFTEEN

Implications for Global Economic Analysis Elena I. Ianchovichina and Terrie L. Walmsley

With the launch of the GTAP project two decades ago it became possible for economists who are not specialists in AGE modeling to start using the general equilibrium approach for economic policy analysis, while at the same time avoiding the excessive costs of collecting data and programming. Not only the large number of citations of the first GTAP book documenting the standard GTAP model (Hertel 1997) but also the large number of individuals around the world using different versions of the framework and the database for policy research and analysis are testimony to the success of this project. The GTAP network of researchers and analysts has been a key asset to the project: Members of the team not only benefited from the availability of data and advances in the modeling framework but also contributed models and data to the project. Knowledge generated by network members has been disseminated at annual conferences, and new ideas have been implemented quickly as the initial data and model lowered the costs of each new extension. In this respect GDyn is no different from other extensions of the GTAP model, although it did benefit substantially from the fact that there was a global database and standard model to start from. However, just like the GTAP model, GDyn will make it possible for others to use dynamic CGE modeling techniques to analyze specific issues of interest without incurring the costs of building a global dynamic CGE model and corresponding databases. Moreover, like GTAP, the GDyn model and database provide a standard dynamic modeling platform on which others can build. The book presents a number of applications developed by such users of GDyn. We hope that, just as did the GTAP book (Hertel 1997), this book will support the development of new applications and in the process result in improvements to the data, parameters, and economic behavior of GDyn. 406

Implications for Global Economic Analysis

407

Synopsis This book documents GDyn, including its model structure, database, parameters, construction of baseline, and software. The GDyn model represents an important extension of the GTAP model that takes into account capital accumulation and ownership over time. First, the GDyn model introduces the theory of adaptive expectations and the gradual movement toward a steady-state equilibrium into the theory determining regional investment. Second, this investment adds to capital stocks available for production over time, thereby providing the dynamic effects of a policy shock from the accumulation of capital. Third, one of the key drawbacks of the GTAP model was that all capital was assumed to be owned by the region in which it was located, and therefore all income earned on capital located in the region accrued to that region’s households. This assumption meant that the GTAP model failed to take account of the implications of foreign ownership on welfare and the trade balance. As the importance of foreign investment grows and we seek to investigate the long-run impact of policies, this assumption becomes more contentious. By incorporating capital accumulation and foreign ownership, the GDyn model addresses this drawback of the GTAP model. Finally, the GDyn model allows the user to show the transition path by which a policy affects the global economy. In undertaking these extensions to the underlying model, the GTAP Data Base has also been enhanced with additional data on foreign and domestic income flows, as well as a number of new parameters required by the theory; for instance, measuring the rate of convergence in rates of return. Underlying the GTAP project are a number of core values inherited from the Impact Project/Center of Policy Studies in Australia – public availability, documentation, replication, and ease of use. We have continued to hold onto these values throughout the development of the GDyn model and database, because we believe that the benefits to trained users far exceed the costs. With the inclusion of dynamics into the GTAP model, however, the complexity of these models has the potential to increase dramatically. In developing the GDyn model, the key developers have been acutely aware of the potential problems of building such a complex dynamic model and have to the extent possible tried to retain many of the benefits of the standard GTAP model. The standard GDyn model therefore keeps many of the features of the standard GTAP model, imposing complexity only where required. Despite these efforts the model does add additional complexity, not simply because of the new dynamic equations added to the model but also because of the need to understand

408

Elena I. Ianchovichina and Terrie L. Walmsley

how to develop more complex baseline and policy scenarios, any additional software, and the more intricate methods required to measure welfare in a dynamic setting. All these new complications are examined in this book. The book also presents a collection of major and diverse GDyn applications, along with the tools needed to replicate them. These are needed because we believe that replication is an essential ingredient of empirical research. We also hope that the book will serve to motivate others to improve the model and data and use them to expand the range of issues one can address with GDyn. The book will also be used as a principal textbook for GDyn courses. This book project is a culmination of a truly collective effort that has been sustained over a number of years. The team has benefited from the availability of the GDyn model and data (Ianchovichina and McDougall 2001), as well as a standard platform for implementing applications that will facilitate the documentation and replication of future GDyn applications. The experience with the first GTAP book (Hertel 1997) shows that standardization helps the pedagogy and replication processes, while imposing no limits on innovations related to modeling and applications. The time saved building a new model and data for dynamic CGE models is significant, allowing for the development of more complex model versions tailored to different tasks, for careful attention to the nature of the simulation design being undertaken, and to additions to the database. Experienced modelers may find the database itself of great interest. It can provide the starting point for applications using different models.

Evaluation Many of the types of issues that GDyn is especially well suited for are global and long term in nature. These are issues that cannot be analyzed to one’s satisfaction without paying attention to the transition path along the way and without taking account of capital accumulation and ownership; providing a model to take those factors into account was the primary motivation behind the development of the GDyn model. In Chapter 8, Hertel, Walmsley, and Ianchovichina use GDyn to compare alternative time paths of investment and ownership in the Chinese economy and quantify the benefits from China’s WTO accession when increases in investment flows into China – a result of accession reforms including in the services sector – are taken into account. They find much larger gains from WTO accession than those predicted by earlier studies, which ignored the impact of accession


409

on productivity in services and abstracted from capital accumulation and foreign investment. Ex-post analyses of the impact of China’s WTO accession suggest that China’s actual gains from the reforms were large and close to the gains suggested in Chapter 8. In Chapter 9, Hertel, Walmsley, and Itakura use GDyn to capture the dynamic effects of the “new-age” FTA between Japan and Singapore, as well as the potential impacts on international investment flows and wealth. Taking account of foreign capital ownership is especially important for this analysis because international investment has boomed in recent decades in East Asia, as Japan and other high-wage economies started outsourcing production to low-wage economies to cut production costs. Explicit modeling of the ownership of regional investment in Japan and other Asian economies allows the authors to track the accumulation of Japanese wealth by foreigners and Japan’s ownership of domestic and foreign assets. Income accruing from the ownership of these foreign and domestic assets can then be appropriately incorporated into total regional income and, hence, the computation of welfare for Japan, Singapore, and the Rest of World. The authors find that the impact of the FTA on investment, capital accumulation, and economic growth is significant – particularly in Singapore. It is critical to have a global dynamic framework that captures the accumulation of capital and other stock variables of importance in the chapters on environmental issues. In Chapter 10 Ianchovichina, Darwin, and Shoemaker extend GDyn to study the global effects of economic and population growth and the impact of a slowdown in agricultural TFP growth on farm and forest resources. In this extension land is divided into six land classes based on the length of the growing season, and land is allocated to sectors without losing sight of its inherent productivity differences. They find that slower agricultural productivity growth could have negative environmental effects because it is associated both with reductions in forestland and increases in environmental or ecological damages on remaining forestlands. Chapter 11 aims to shed light on the role of global economic integration for land use change. Golub and Hertel add to GDyn the most important economic features driving global land demand and supply, including an econometrically estimated, international demand system for commodities and an additional database and equations characterizing land use by agroecological zones. This application is a case study showing the interplay between economic and environmental issues in a global context. It is also an example of how GDyn can be extended to address environmental issues

410

Elena I. Ianchovichina and Terrie L. Walmsley

that have grown in importance in the past few years and will become even more prominent in the future. In Chapter 12 Itakura, Hertel, and Reimer investigate whether increases in productivity associated with increases in trade and investment as part of FTAs have a significant impact on the CGE results of FTAs. To test their hypothesis they incorporate econometric evidence into GDyn, which allows the exploration of FDI issues in a CGE framework. They found that the conventional, dynamic AGE model captures more than half of the ensuing GDP and trade changes. The rest could be attributed to the procompetitive and FDI-productivity linkages, whereas the export-productivity linkage played a minor role. Their application is important because it provides a good example of a simulation exercise that sets priorities for future policyoriented econometric work aimed at refining the estimates used in this chapter. In Chapter 13, Tyers and Shi focus on an important long-run issue – the global population and labor force changes that will affect not only the performance of regional economies but also the distribution of global investment and savings and, hence, capital account flows and regional production costs, and therefore the pattern of global trade. They add a demography submodule to GDyn and relax the assumption of a fixed private saving rate. Their extension offers a possibility to expand the treatment of the government sector in GDyn and to undertake a number of potential improvements to facilitate the analysis of the fiscal implications of aging. Finally, Golub and McDougall undertake a major validation exercise in Chapter 14. Their backcasting work is important because the quality of the insights generated by GDyn depends on the validity of the modeling framework, including model, data, and parameters. Golub and McDougall compare the gross and net foreign assets and liabilities in GDyn simulation to those in a country portfolio database constructed by Kraay et al. (2000) for the period between 1966 and 1997. Their results show that the fixed propensities to save in the model coupled with rising incomes lead to unrealistically large international capital flows in the very long run, which is defined as a period longer than 20 years. They create too much saving and, as a result, too much investment and capital in the world; they allow chronic rapid saving or dissaving, and therefore too great foreign assets or liabilities in individual regions. When Golub and McDougall modify household saving behavior by endogenizing saving rates, the model is able to enforce long-run stabilization in a regional wealth-to-income ratio, which is observed in the data and is critical to obtaining a sensible baseline in light of the currently divergent


411

saving rates across countries. Their work elevates high on the list of priorities for future work with GDyn comparisons of investment-saving correlations in GDyn simulations with fixed and endogenous savings, with those in the studies following Feldstein and Horioka (1980), and those in Chapter 13 of Tyers and Shi.

Future Directions and Conclusions GDyn is an important tool for economic policy analysis because the model and data enable economists to investigate the impact of a number of global issues that will continue to grow in importance and dominate economic policy debate in years to come: climate change; globalization, including the movement of goods, services, capital, and labor across regions; and the rapid changes in the profile of the global workforce and population. The Center for Global Trade Analysis, the home of the GTAP project, will continue to rely on the GTAP network to assist in advancing the GTAP Data Base and models to ensure that tools are publicly available to the network to investigate the important policy issues of the day. These tools, including the GDyn model and Data Base, will continue to be made publicly available, along with essential training on these tools for policy analysts and researchers around the world. It is hoped that GDyn will support the development of new applications and in the process result in improvements to the data, parameters, and economic behavior of GDyn, as has been the case with the GTAP model. References Feldstein, M. and C. Horioka. 1980. “Domestic Savings and International Capital Flows.” Economic Journal 90, 314–29. Hertel, T. W. 1997. Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press. Ianchovichina, E. and R. McDougall. 2001. Structure of Dynamic GTAP. GTAP Technical Paper No. 17. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2000, July. Country Portfolios. National Bureau of Economic Research World Paper Series No. 7795: 1–61. Washington, DC: NBER.

APPENDIX

Negative Investment: Incorporating a Complementarity into the Dynamic GTAP Model Terrie L. Walmsley and Robert A. McDougall It has been found that in some cases the investment mechanisms in the GDyn model can result in gross investment falling below zero. This will occur when the investment mechanisms cause the actual rate of return to rise considerably, lowering investment to unacceptable levels; for example, if there is a large error in expectations (the expected rate of return far exceeds the actual rate) or the expected and actual rates of return are far below the target rate of return. In this appendix a complementarity is introduced to ensure that in these cases investment does not fall below a minimum level. We use the method developed by Harrison, Horridge, Pearson and Wittwer (2002) for introducing complementarities. The method is designed specifically for models implemented using the GEMPACK software. First the equations from the GDyn model are used to demonstrate the negative investment problem in the GDyn model. Following this we introduce the complementarity and show, using the example, how the complementarity works.

1. Negative Investment Here we outline an example of the case where investment falls to an unacceptable level. The example is based on a small 7 commodities1 by 7 region2 aggregation of the GDyn Data Base.3 1 2 3

Primary and Agricultural commodities; Processed Food; Natural Resources; Textiles; Manufacturing; Transport, Machinery and Equipment; and Services. US and Canada; EU; Japan; Newly Industrializing economies; South East Asia; China and Hong Kong; and Rest of World. The application related to this Appendix is Ch7HO7 × 7 gdyn v31c 97.zip and is available for download on the Web site at: https://www.gtap.agecon.purdue.edu/models/Dynamic/ applications.asp.

413

414

Terrie L. Walmsley and Robert A. McDougall

Actual Expected

Target

ROW

CHN_HKG

SEA

NIC

JPN

EUN

USC

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

Figure 1. Rates of return.

Figure 1 shows the target and actual, expected rates of return in this 7 × 7 aggregation. Expected rates of return hover around 13 and 14%, with the target rate of return equal to 13.5%. There is not a great deal of variation between the expected and target rates of return (Figure 1). In contrast, actual rates of return vary greatly across regions, with a very low 9% in Japan and 19% in South East Asia. Let us consider these two extreme cases. In South East Asia the actual rate of return is very high; it is much higher than the expected rate of return. The actual rates of return are based on returns to capital relative to the price of capital goods, whereas expected rates of return are implied or calibrated from the data. In countries where rates of return are high, theory tells us that investment should also be high as investors seek high returns. If the data agree, then the expected and actual rates of return will be equal. However, if the data show small levels of investment in South East Asia, then either the data or the theory is deemed to be incorrect. In the GDyn model, the theory is adjusted to include errors in expectations. Expected rates of return therefore reflect the rates of return implied by the

0.2

Appendix: Negative Investment

415

2,500,000

2,000,000

1,500,000

1,000,000

500,000

0 1997

2002

2007

2012

2017

2020

Figure 2. Investment in Japan over time ($US).

data, and the difference between the actual and expected rates of return are errors in expectations. Hence although rates of return are high, investment is low in South East Asia, and hence the expected rates of return are much lower than the actual rates. In the case of Japan the reverse is true: The low rates of return suggest low levels of investment, yet investment is very high, suggesting that expected rates of return are much higher. As a result of Equation 1 in Box 1, expected rates of return in South East Asia will rise, and those in Japan will fall (relative to the target). Assuming no change in the target, the rate of growth in the expected rate of return (Equation 2) is therefore negative for South East Asia and positive for Japan as errors in expectations are eliminated and the expected rate of return is also drawn toward the target. If the expected rate of return is to rise in South East Asia, then so too must investment, as investors choose to invest where returns are highest (Equation 3). The rise in investment also causes actual rates of return to fall as more investment leads to more capital stocks and lower returns. In Japan the reverse is true. Lower expected rates of return lead to lower investment (Equation 3), and hence actual rates of return rise. The fall in expected rates of return and the rise in actual rates cause the errors in expectations to be eliminated faster.

416


Box 1: Investment theory of GDyn 1. Rˆ E = −φ(Kˆ − dt) − μ log(R E /R A )dt 2. E = [Rˆ T − Rˆ E ] I 3. E = φ [Iˆ − Kˆ ] + φd K Rˆ A 4. ω = Kˆ + − dt φ 5. Kˆ = Idt where: R E is the expected rate of return R A is the actual rate of return R T is the target rate of return K is the quantity of capital stocks I is Investment is the normal growth rate of capital – that rate of growth consistent with no change in rates of return dt is change in years log(R E /R A ) is a measure of the errors in expectations φ is the elasticity of the rate of return with respect to capital stocks μ is the rate at which errors in expectations are eliminated E is expected rate of growth in the expected rate of return is the rate at which differences in the expected and actual rate of return are eliminated ω is the change in the normal growth rate of capital ˆ the hat represents proportionate change. Note: All variables except R T are indexed by region. The problem in the case of Japan is that the level of investment required to eliminate errors in expectations and ensure convergence to the target rate of return (at the rates given in the database) may require the level of gross investment to fall below zero. This creates problems because it is assumed that capital (to which investment is added, Equation 5) cannot be discarded, but will depreciate gradually over time (at a rate of 4% in the GDyn Data Base). In our small 7 × 7 aggregation, this occurs in Japan. Rates of return are so low that the gradual elimination of errors results in investment falling too quickly.


417

2. Complementarities To overcome this we introduce a complementarity into the model that stops gross investment falling below some minimum investment-to-capital ratio (MINIKRAT: currently set at 0.0054 ). A full list of the equations incorporated are provided in Box 2. Coefficient (parameter) MINIKRAT #Minimum ratio of Investment to Capital stock #; Read MINIKRAT from file GTAPPARMK header “MIKR”; To implement this we define the level of real gross investment QREGINV and the level of real gross investment (QREGINV_D) determined by the dynamic model Equations (1–5) and the minimum level of real investment (QMININV), which is related to the minimum investment-to-capital ratio (MINIKRAT). Variable (levels)(all,r,REG) QREGINV(r) # real GROSS investment in r (qty of “cgds” output) #; Variable (levels)(all,r,REG) QREGINV D(r) # real GROSS investment in r (qty determined by dynamics) #; Variable (levels)(all,r,REG) QMININV(r) # Minimum GROSS investment allowed in r #; Formula (initial)(all,r,REG) QMININV(r) = MINIKRAT∗ VK(r); The complementarity is then introduced in the following way: Complementarity (variable = QREGINV, lower bound = QMININV) ! Ensures QREGINV = QREGINV D when QREGINV D > QMININV else QREGINV = QMININV ! C QREGINV (all,r,REG) QREGINV(r) − QREGINV D(r); 4

Note that in older versions of the GDyn Data Base this may be set equal to 1. You must change it to 0.005 as it will not work with a value of 1.

418


This equation simply states that when QREGINV D < or = QMININV, QREGINV = QMININV QREGINV D > QMININV, QREGINV = QREGINV D. In most cases the dynamic equations result in a level of investment greater than the minimum, and hence the value determined by the dynamic equations is used. However, on rare occasions where the level of investment is too low, investment will be set equal to the minimum. Box 2: Equations Variable (levels)(all,r,REG) QREGINV(r) # real GROSS investment in r (qty of “cgds output”) #; Formula (initial)(all,r,REG) QREGINV(r) = sum(k,CGDS COMM, VOA(k,r)); Equation E QREGINV(all,r,REG) p QREGINV(r) = qcgds(r); Variable (levels)(all,r,REG) QREGINV D(r) # real GROSS investment in r (qty det er min ed by dynamics) # ; Formula (initial)(all,r,REG) QREGINV D(r) = sum(k,CGDS COMM, VOA(k,r)); Equation E QREGINV D(all,r,REG) p QREGINV D(r) = qcgds d(r); Coefficient (parameter) MINIKRAT # Minimum ratio of Investment to Capital stock #; Read MINIKRAT from file GTAPPARMK header “MIKR”; Variable (levels)(all,r,REG) QMININV(r) # Minimum GROSS investment allowed in r #; Formula (initial)(all,r,REG) QMININV(r) = MINIKRAT∗ VK (r); (continued)


419

Box 2 (continued) Equation E MININV(all,r,REG) p QMININV(r) = qk(r); Complementarity (variable = QREGINV, lower bound = QMININV) ! Ensures QREGINV = QREGINV D when QREGINV D > QMININV else QREGINV = QMININV ! C QREGINV (all,r,REG) QREGINV(r) – QREGINV D(r);

In the 7 × 7 application outlined earlier, Japan’s real investment is set equal to the minimum level. The dynamic equations push Japan’s investment down by 99.94%, or from 1222350 in the initial database to 66334.04, which is lower than the minimum of 75812.88. As a result, real investment is set equal to the minimum and falls by only 94.5%.5 Figure 2 shows the resulting changes in investment over time. Although investment falls in the initial period, the target rate of return also falls over time, causing investment in Japan to rise in later years. References Elbehri, A. and K. Pearson. 2005. Implementing Bilateral Tariff Rate Quotas in GTAP Using GEMPACK. GTAP Technical Paper No. 18. West Lafayette, IN: Center for Global Trade Analysis, Purdue University. Harrison, J., M. Horridge, K. Pearson, and G. Wittwer. 2002. “A Practical Method for Explicitly Modeling Quotas and Other Complementarities.” Computational Economics 23, 325–41. Hertel T. W. (ed.). 1997, Global Trade Analysis: Modeling and Applications. Cambridge: Cambridge University Press.

5

Note that real investment is affected by this complementarity within the simulation, however at the beginning of each period the value of investment and capital are used to determine the real quantities.

Glossary of GDyn Notation Terrie L. Walmsley

1. Overview Specification of the GDyn Data Base and model requires a vast amount of notation. This notation has been carefully chosen to be brief, yet descriptive. Since many of the variables used in the GDyn model are also in the standard GTAP model, this glossary restricts itself to the additional variables in the GDyn model used throughout this book. The variables correspond to the GEMPACK representation of the model, GDynv34.tab, as specified in February 2009. It lists the sets and subsets, base data, derivatives of the base data, and variables used in the model. We hope that the glossary will be useful for both new and experienced users of the GTAP framework. Some important conventions follow: a) Sets and parameters are denoted in uppercase. b) The levels form of the variables in GDyn is denoted in uppercase. Percentage changes in variables are denoted in lowercase (linearized form of the variables). For instance, QK(r) is the quantity of capital located in region r in the levels form, and qk(r) = [d QK(r)/QK(r)]∗ 100% is the linearized form of this variable. c) The GDyn Data Base comprises only value flows (in their levels form). Database values are accordingly written in uppercase. These are declared as coefficients in the GEMPACK code and are updated using percentage changes in the component prices and quantities, after each step in the solution. The database stores the minimal amount of information. No redundancies are permitted. d) Derivatives of the database variables are also in levels form. There are two types of derivatives: value flows and shares. The derivative variables naturally get updated following each update of the database. 421

422

Terrie L. Walmsley Table 1. Aide memoire for the naming conventions used in the GDyn model

Variable

Longnamea

Definition

First letter: income or wealth Y

Income

W

Wealth

Second letter: type of ownership Q Equity Third letter: owner and receiver of income H HHLD Household

T

TRUST

_ (underscore)

Trust

All owners

Last letter: investment or payer of income T TRUST Trustc F

FIRM

None

a b c

Firm

All investments

Exampleb YQTRUST (or yqt) − Income on equity owned by the trust WQHHLD(r) (or wqh(r)) − Wealth in equity owned by households in region r YQ_TRUST (or yq_t) − Income on equity paid by the trust YQHHLD(r) (or yqh(r)) − Income earned on equity owned by the regional household r YQTFIRM(r) (or yqtf(r)) − Income on equity paid to the trust by firms in region r WQ_TRUST (or wq_t) − Total wealth of trust − Sum of all households’ ownership in the trust YQHTRUST(r) (or yqht(r)) − Income on equity paid to household r by the trust YQHFIRM(r) (or yqhf(r)) − Income on equity paid to household r by domestic firms WQTRUST (or wqt) − Total wealth of trust − Sum of all investments made by the trust

In order to distinguish variables from coefficients, the last letter of a coefficient is always extended to its longname. Coefficients are denoted in upper case and percentage change variables are denoted in lower case. Note that the trust is an owner of wealth invested in regional firms and it is also owned by regional households.

e) The GDyn model consists of a system of linearized equations with all variables appearing in percentage change form. GEMPACK solves for percentage changes in (endogenous) prices and quantities, thereupon relinearizing the model and solving it once again. Before commencing with the glossary, the user may wish to review Table 1, which is provided to assist the user in understanding the naming conventions used in introducing the new income and wealth data.

Glossary of GDyn Notation

423

2. Glossary Sets and subsets Sets CGDS_COMM DEMD_COMM ENDW_COMM ENDWC_COMM ENDWM_COMM ENDWS_COMM NSAV_COMM PROD_COMM TRAD_COMM

Capital Goods Commodities (“cgds”) Demanded Commodities Endowment Commodities Capital Endowment Commodity (“capital”) Mobile Endowment Commodities Sluggish Endowment Commodities Non-Savings Commodities Produced Commodities Traded Commodities

Subsets CGDS_COMM CGDS_COMM DEMD_COMM ENDW_COMM ENDWC_COMM ENDWC_COMM ENDWM_COMM ENDWS_COMM PROD_COMM TRAD_COMM TRAD_COMM

in NSAV_COMM in PROD_COMM in NSAV_COMM in DEMD_COMM in ENDW_COMM in NSAV_COMM in ENDW_COMM in ENDW_COMM in NSAV_COMM in DEMD_COMM in PROD_COMM

Example: The 3 × 3 (3 regions × 3 sectors) economy CGDS_COMM DEMD_COMM ENDW_COMM ENDWC_COMM ENDWM_COMM ENDWS_COMM NSAV_COMM PROD_COMM REG TRAD_COMM

= {capital goods} = {land, labor, capital, food, manufacturing, services} = {land, labor, capital} = {capital} = {labor} = {land, capital} = {land, labor, capital, food, manufacturing, services, capital goods} = {food, manufacturing, services, capital goods} = {usa, eu, row} = {food, manufacturing, services}

424

Terrie L. Walmsley

BASE DATA KHAT(r)

RORGEXP(r) RORGTARG YQHFIRM(r) YQHTRUST(r) YQTFIRM(r)

Price-neutral rate of growth in the capital stock. The rate of growth in the capital stock is, agents believe, consistent with a constant rate of return on capital. Expected gross rate of return. Target gross rate of return. Income on equity paid to households r by domestic firms. Income on equity paid to regional household r by the trust. Income on equity paid to the trust by the regional firm r.

∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG

PARAMETERS INC(r)

LAMBKHAT(r) LAMBRORG(r)

LAMBRORGE(r)

RIGWQ_F(r)

RIGWQH(r)

RORGFLEX(r)

Initial equilibrium regional expenditure data INC is set equal to INCOME and does not change during a simulation. Coefficient of adjustment for the normal rate of growth. Coefficient of adjustment for the rate of return. Controls the rate at which agents aim to adjust rates of return in response to differences between expected and target rates. If the target rate exceeds the expected rate by 1%, agents aim to adjust the rate by LAMBRORG% per period. Coefficient of adjustment for the expected rate of return. Controls the rate at which agents adjust expectations of rates of return in response to differences between expected and actual rates. If the actual rate exceeds the expected rate by 1%, agents adjust the expected rate by LAMBRORGE% per period. Rigidity on total wealth in equity invested by both the trust and the regional household in the regional firm r. Rigidity on total wealth in equity owned by the regional household r and invested in both the trust and domestic firms. Flexibility of the gross rate of return. Agents expect each 1% expansion in the capital stock to reduce the gross rate of return by RORGFLEX %.

∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG

∀ r ∈ REG


Glossary of GDyn Notation

425

DERIVATIVES OF BASE DATA RORGROSS(r) WQ_FHHLDSHR(r) WQ_FIRM(r) WQ_FTRUSTSHR(r) WQ_THHLDSHR(r) WQ_TRUST WQHFIRM(r) WQHHLD(r)

WQHTRUST(r) WQT_FIRMSHR(r) WQTFIRM(r) WQTRUST YQ_FIRM(r) YQ_FHHLDSHR(r)

YQ_FTRUSTSHR(r)

YQ_TRUST YQ_THHLDSHR(r)

YQHHLD(r) YQT_FIRMSHR(r)

YQTRUST

Gross rate of return. Share of total wealth on equity of regional firm r which is owned by domestic households. Total wealth in equity invested by the domestic household and by the trust in regional firms. Share of wealth on equity in regional firm r which is owned by the trust. Share of wealth in the equity of the trust which is owned by the regional household r. Total wealth in equity invested by all regional households in the trust. Wealth in equity owned by the regional household r and invested in domestic firms. Total wealth in equity owned by regional household r and invested in both the trust and domestic firms. Wealth in equity owned by household r and invested in the trust. Share of wealth in equity owned by the trust which is invested in regional firms r. Wealth in equity owned by the trust and invested in regional firms r. Total Wealth in equity owned by the trust and invested in all regional firms. Total Income on equity paid to both the trust and the regional household by regional firms r. Share of total income on equity paid by regional firm r which is paid to domestic households (domestic share of firm income). Share of total income on equity paid by regional firms which is paid to the trust (trust’s share of firm income). Total income paid by the trust (i.e., to all regional households). Share of total income paid by the trust which is paid to the regional household r (regional household’s share of trust income). Total income on equity paid to the household r by both the domestic firms and by the trust. Share of total income in equity paid to the trust which was paid by the regional firm r (i.e., share of trust income paid by firm r). Total income in equity paid to the trust by all regional firms.





426

Terrie L. Walmsley

VARIABLES DKHAT(r) DROR(r) DRORT DRORW(r) erg_rorg(r) ERRRORG(r) pqtrust qk(r) rorga(r) rorge(r) rorgt rorwqht(r) rorwqtf (r) SDROR(r) SDRORT(r) SDRORTW SDRORW sqk(r) sqkworld srorge(r) time wq_f (r)

wq_t wqh(r)

wqhf (r)

Change in the normal rate of growth in capital. Change in the rate of return − net or gross. Change in the target rate of return − net or gross. Change in the world-wide average rate of return − net or gross. Change in the expected rate of growth in the gross rate of return. Measure of error in the rate of return. Percentage change in the price of equity invested in the trust. Percentage change in the capital stock. Percentage change in the actual gross rate of return. Percentage change in the expected gross rate of return. Percentage change in the target gross rate of return. ROR on wealth in equity owned by regional household r and invested in the trust. ROR on wealth in equity owned by the trust and invested in regional firms. Change in the region-specific shift in the rate of return. Change in the region-specific shift in the target rate of return. Change in the world-wide shift in the target rate of return. Change in the world-wide shift in the rate of return. Percentage change in arbitrary region-specific shock to capital stock. Percentage change in arbitrary region-generic shock to capital stock. Percentage change in the exogenous shift in the expected gross rate of return. Absolute change in time, measured in years. Percentage change in total wealth in equity invested by both the regional household and the trust invested in regional firm r. Percentage change in total wealth in equity invested by all regional households in the trust. Percentage change in total wealth in equity owned by regional household r and invested in both domestic firms and in the trust. Percentage change in wealth in equity owned by the regional household r and invested in domestic firms.

∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG

∀ r ∈ REG

∀ r ∈ REG ∀ r ∈ REG


Glossary of GDyn Notation wqht(r) wqt(r) wqtf (r) wtrustslack xwq_f (r)

xwqh(r)

yq_f (r)

yqh(r)

yqhf (r) yqht(r) yqt yqtf (r)

Percentage change in wealth in equity owned by the regional household r and invested in the trust. Percentage change in wealth in equity owned by the trust. Percentage change in wealth in equity owned by the trust and located in the regional firm r. Percentage change in the slack between wealth of trust. Percentage change in the shift variable for the total wealth in equity invested by both the domestic household and the trust in the regional firm r. Percentage change in the shift variable for the total wealth in equity owned by the regional household r from investments in both domestic firms and the trust. Percentage change in total income on equity paid to the domestic household and the trust by regional firms r. Percentage change in total income on equity paid to regional households r by both domestic firms and the trust. Percentage change in income on equity paid to regional households r by domestic firms. Percentage change in income on equity paid to regional household r by the trust. Percentage change in total income on equity paid to the trust by all regional firms. Percentage change in income on equity paid to the trust by regional firms r.

427 ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG


∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG ∀ r ∈ REG

∀ r ∈ REG

Index

accession: China’s WTO, 7, 207–9, 225, 233, 326, 408–9; scenario, 208–9, 212, 229 accumulation: asset, 5; capital, 7, 13–14, 16, 18, 20, 26, 57–8, 68, 89, 161, 217, 220, 228, 263, 343, 407–9; equation, 14–20, 26–7, 36, 56–8, 68; wealth, 14–15, 18, 65, 70 actual investment schedule, 46 actual rate of return, 48–50, 57–8, 65, 68, 81–2, 87, 117, 128, 132, 156, 219, 413–4, 416 Adams, 291, 309 adaptive expectations: theory of investment, 13, 41; treatment, 65 additional shock (ashock), 52, 188 Africa, 5, 98–9, 102, 109, 111, 116, 123, 125–6, 130, 136, 142, 144–5, 147–8, 152, 226–7, 231, 244, 266, 269, 271, 281–2, 285–6, 299, 301–2, 305–7, 309, 339, 350–2, 355, 359–63, 365, 367, 369, 371, 385, 390, 395 aggregation procedure, 130–1, 133 aging: population, 364; process, 344, 372 Agreement: Free Trade (FTA), 5, 7, 136, 148, 245, 254, 261, 312, 326, 332; Multi-Fibre (MFA), 149, 150, 212, 233; North American Free Trade (NAFTA), 3; on Textiles and Clothing (ATC), 151, 207; Uruguay Round, 150, 247 agricultural productivity, 273, 280–1, 284–5, 288, 304, 409 agricultural research expenditure, 280–1 Agro-Ecological Zone (AEZ), 291, 295, 298–300, 304–5, 307–8 Ahammad, 291, 309 Allen partial elasticities (APE), 273, 276 allocation rule, 30

Alston, 269, 281, 289 alternative closure, 5 An Implicit Directly Additive Demand System (AIDADS), 292–3 AnalyseGE program, 191 Anderson, 157, 207, 232, 236, 267 Ando, 125, 380, 404 apparent current normal growth rate (KHAPP), 53–6 Armington, 315 assets: domestic, 27, 30, 92, 100, 116, 134, 245, 391, 409; financial, 5, 6, 13, 14, 20–6, 37, 101; foreign, 4, 8–9, 22–3, 29–30, 35, 69, 91, 95, 97, 99–102, 112, 116, 223, 225, 245, 379, 380–6, 399, 402, 403, 409–10; local, 22–3, 93–4, 96, 112 automatic accuracy, 186, 199, 201 autoregressive conditional heteroskedasticity (ARCH), 105–6 backcasting, 410 balance of trade/trade balance, 223–4, 246, 250–1, 254, 260, 301–3, 407 balance sheet identity, 23, 25 base rerun, 185, 190–1, 199, 200 base time, 15–16, 46, 51 basecase shock, 188 baseline: assumptions, 296; projection, 239, 281, 291, 301, 314, 343, 360, 362, 365, 366, 369, 372; scenario, 4, 136–7, 148–9, 151–2, 156, 162, 164, 208, 210, 211, 220, 247, 273, 282, 297, 300–1, 303, 325, 327–9, 332, 335, 344, 359, 367; simulation, 160, 238, 249–51, 270, 276, 291, 297–8, 300, 305, 327, 342, 357, 361, 365

429


430

Index

behavioral parameters, 128 Benjamin, 381, 404 Bernard, 79, 118, 279, 289, 313, 321–2, 340 Blalock, 340 Booth, 342, 366, 373 Brown-Kojima-Drysdale export intensity, 235 Brumberg, 380, 405 Bryant, 342–3, 373 Calderon, 97 calibration procedure, 71, 80, 87, 128–9, 279 capital: account volatility, 59; accumulation, 7, 13–4, 16, 18, 20, 26, 57–8, 68, 89, 161, 217, 220, 228, 263, 343, 407–9; fixed, 26, 38, 73, 76; flows, 69, 80, 223–4, 380, 386, 410; gain effect, 43; rental price, 42; services, 38; value of, 30, 38, 76, 127–8, 131, 160 capital mobility: degree of, 72, 80–2, 84–5, 117; international, 4, 7, 13, 21–3, 69–71, 101, 117, 208; perfect, 43, 56, 249 capital stock: 4, 13–14, 20, 27, 43, 56, 65, 74–7, 80–1, 114, 123, 125–8, 131, 137–8, 153, 166, 220–1, 228, 247, 251–2, 326, 332, 407, 415–16; rate of growth in, 81; series, 73, 76; value of, 127–8, 131 Center for Global Trade Analysis (CGTA), 6 Central Europe, 231–2, 266, 350–2, 355–6, 358, 361–3, 365, 367–9, 372 Centre of Policy Studies (CoPS), 233 China’s WTO Accession, 7, 207–9, 225, 233, 326, 408–9 Chuang, 314, 323, 340 Chung, 313, 340 closure: alternative, 5; file, 183, 186, 189–90, 199; GDyn, 189; policy, 185; standard, 152, 155; swap, 69, 166 Cobb-Douglas per capita utility function, 379 Coefficient of adjustment: 424; in estimated normal growth rate (LAMBKHAT), 53–4, 56, 72, 81–2, 89, 121, 128, 133–4, 397, 424; in expected rate of return (LAMBRORGE), 51–2, 56, 72, 81–5, 88, 121, 133–4, 195, 424; in normal rate of growth (LAMBYHAT), 392–4, 397; in rate of return (LAMBRORG), 44, 50–2, 56, 72, 81–5, 88–9, 121–2, 128–9, 133, 134, 195, 424; in wealth-to-income ratio (LAMBWYR), 388, 390, 393–8, 400, 402

commercial presence, 210, 215–6, 222, 228 commodity aggregation, 241, 272 Common Effective Preferential Tariff (CEPT) reduction program, 326 comparative advantage, 260, 324, 372 comparative static: GTAP model, 158–9; simulation, 158, 165–8, 199, 225 composite regions, 142 composition of wealth, 105, 109, 111–12, 114–15, 135 Constant Difference Elasticity (CDE) demand system, 356 constant elasticity: of substitution (CES), 86–8, 356; of transformation (CET), 271–3, 276, 295 consumer demand, 291–2, 299–300, 304 consumption analysis literature, 356 consumption expenditure, 353–6 consumption of services abroad, 215 consumption preferences, 344, 353, 356, 371–2 continuous time, 16 convergence: of rates of return, 73, 80; speed of, 4, 72, 81–4, 117 cost disadvantage ratio (CDR), 319 counterfactual policy simulations, 326 Cournot, 315, 319 Cranfield, 310 Cross border supply of services, 215, 228 cumulative differences, 162, 164, 193, 195, 200 cumulative results, 61–3, 191, 193 current time, 20, 46, 50 Darwin, 7, 269–70, 273, 276, 282, 289, 291, 310, 409 data: construction, 4, 6, 22; foreign income, 124; GTAP, 3, 120–1, 125, 177, 323–4; investment, 23, 75, 125–7, 211; population, 138; trade, 316; updated, 30, 167, 193, 197–9 Data Base: GDyn, 120–3, 125, 128, 130–1, 382, 413, 416–17, 421; GTAP 70, 118, 120, 122, 129–30, 135, 138, 149, 157, 201, 209, 212, 233, 237–8, 245, 268, 276, 289, 292–3, 310, 323, 340, 386, 395; updated policy, 165–6 Database: GATT/WTO Integrated (IDB), 149; updated, 30, 167, 197–8 Davis, 269, 289, 342 debt, 23, 26, 99


Index demand: consumer, 291–2, 299–300, 304; derived, 290, 293, 295, 297–9; equations, 246; final, 353; import, 244, 246 demand-side simulation, 297 demographic change, 4, 6–8, 343, 353, 368, 372 dependency ratio, 344, 352–3, 362–3, 367 depreciation, 24, 38, 71–2, 74–9, 82, 127, 160, 391, 395 derived coefficient, 182 Devarajan, 340 Dimaranan, 118, 120, 135–6, 157, 232–3, 268, 289, 310, 340 direct trade in services, 209, 215, 263 discount factor, 161, 168 disequilibrium approach, 69 domestic assets, 27, 30, 92, 100, 116, 134, 245, 391, 409 domestic capital stock, 97, 117, 135, 227 domestic equity, 30, 34, 91, 97, 252 domestic firms, 26–7, 92–3, 105, 207, 313, 319, 322, 337, 391, 422, 424–7 domestic wealth, 31, 97, 99 Drysdale, 236, 267 Durbin-Watson test, 105 duty drawback regime, 7, 208 dynamic baseline, 323 dynamic effects, 236, 245, 270, 407, 409 dynamic equations, 407, 418–9 dynamic FARM model (D-Farm), 270 dynamic simulation, 16, 43, 162, 166–8, 199–200 East Asia, 8, 98, 102, 106, 109, 111, 116, 129–30, 136, 142, 144–5, 147–8, 152, 225–8, 232–3, 244–5, 272, 281, 285, 288, 312, 314, 340, 343, 350–1, 355, 358, 360–3, 365, 367, 374, 385, 387, 395, 409, 413–15 Eastern Europe, 130, 211, 306 Eastern Europe and the Commonwealth of Independent States region (EIT), 299, 301–3, 305, 306 ecological-economic information, 280 econometric: analysis, 101, 117; investigation, 73, 118; standard errors, 336 Economies in Transition, 302, 305, 309 Edwards, 289 effective price, 240, 242–4, 246, 250, 262 efficiency losses, 322

431

elasticity: of the expected rate of return with respect to capital stocks (RORGFLEX), 45, 47–8, 52–4, 72, 85–90, 121, 129, 131–2, 195, 424; markup, 319, 320–1; of substitution, 86, 246–7, 295, 356; of transformation, 271, 295 Elbehri, 23, 70, 149, 157, 209, 233, 268, 276, 289, 419 Electronic Trade Document Exchange System (ETDS), 240, 242 empirical trade studies, 335 employment, 153, 343, 368 endowment: non-accumulable, 171; ownership of capital, 159 Engel, 105, 118, 293, 365, 371 entropy: parameters, 6, 117; theory, 30, 92 environmental: or ecological damages, 8, 288, 409; effects, 288, 409; issues, 409–10 equilibrium: long-run, 14, 59–60, 71–2, 78, 81, 117, 122; partial, 5, 245, 270, 300, 312, 320; steady growth, 392; steady-state, 407 equity: domestic, 34, 91, 97, 252; in firms, 23–4, 28; foreign, 28, 31, 33–4, 91, 97, 99, 252, 323; income, 37, 40, 181, 183; income earned on, 160; income from, 26, 41, 122, 181; negative, 30; portfolio, 31; price of, 25, 27, 37, 391, 426; shareholder, 23; total, 23, 26, 37, 39; wealth in, 26, 29, 424–7 equity-for-debt substitution, 70 equivalent variation (EV), 62–4, 160, 163–4, 168–71, 200, 225, 261–2, 377 errors in expectation, 60, 155–6, 414–6 Evenson, 269–70, 283, 288–9 exogenous productivity, 360 expectation time, 46, 48 expected investment schedule, 54–5, 60 expected rate of growth in the rate of return, 45, 47–8, 50, 84–5, 88–9 expected rate of return (rorge), 45–6, 49–53, 56–8, 65, 68–9, 72, 81, 84–7, 89, 117, 121, 132–4, 166, 193–5, 200, 413, 415–6, 426 expenditure: agricultural research, 280–1; consumption, 353–6; matrix of, 357; private, 280; public, 280; regional, 122, 424 export: subsidies, 187; tax equivalents, 149 extended linear expenditure system, 381, 404 Faruqee, 343, 373–4


432

Index

FDI-productivity effect, 327–8, 333–4 Feldstein, 70, 91, 118, 267, 399, 404, 411 final demand, 353 financial assets, 5–6, 13–14, 20–6, 37, 101 financial sector, 23, 101 Findlay, 215–6, 234 firms: domestic, 26–7, 92–3, 105, 207, 313, 319, 322, 337, 391, 422, 424–7; foreign, 23, 92–3, 116, 118, 206; liabilities of, 5; local, 23–4, 27, 31, 33, 37, 39, 92, 112, 116, 118, 121–2, 182–3, 314, 327; regional, 23, 26, 29, 92, 422, 425–7 fixed capital, 26, 38, 73, 76 foreign assets, 4, 8–9, 22–3, 29–30, 35, 69, 91, 95, 97, 99–102, 112, 116, 223, 225, 245, 379, 380–6, 399, 402–3, 409–10 Foreign Direct Investment (FDI), 23, 156, 205–7, 211, 215, 232, 252, 261, 313–14, 322–3, 327–36, 410 foreign equity, 28, 31, 33–4, 91, 97, 99, 252, 323 foreign income: data, 124; payments, 95, 123–5, 250, 254; receipts, 121–2, 124, 383 foreign investment, 7, 23, 91, 93, 112, 118, 156, 205–6, 208–11, 215–16, 221, 223, 225, 227–9, 235, 252, 261, 327, 330, 333, 360, 407, 409 foreign liabilities, 8, 9, 22, 30, 69, 99–102, 112, 382, 384–6 foreign ownership, 7, 29, 30, 35, 91, 123–4, 156, 160, 205, 209–11, 217, 221–2, 225–8, 251, 261, 323, 328, 330, 332–3, 407 foreign wealth, 250 Forest and Agricultural Sectors Model (FASOM), 291, 309 forward-looking behavior, 70 Francois, 149–50, 157, 211, 213–15, 232, 243–4, 267, 289, 312–13, 315, 319, 340 Free Trade Agreement (FTA), 5, 7–8, 136, 148, 235–40, 242–7, 249–54, 259–64, 268, 312–20, 323, 325–36, 339, 409–10 Freebairn, 270, 289 Frisvold, 270, 289 Future Agricultural Resources Model (FARM), 270, 273, 282, 284, 286–7, 289, 291 future time, 45–6, 133 Garbaccio, 232

GATT/WTO Integrated Database, 149 G-Cubed Model, 381 GDyn: closure, 189; Data Base, 120–3, 125, 128, 130–1, 382, 413, 416–17, 421 GDYNView, 197–9 GDYNVol, 197–9 gender, 8, 343–5, 348–9, 351–4 General Agreement on Tariffs and Trade (GATT), 149 General Agreement on Trade in Services (GATS), 206 General Equilibrium Modelling Package (GEMPack), 5, 10, 15, 62, 66–7, 70–1, 167, 173, 198, 201, 267, 289, 413, 419, 421–2 global climate change, 6 global investment, 36, 59, 60, 344, 360, 369, 372, 410 global rate of return, 220, 399 global savings, 69, 281, 380 global trust, 23–4, 26–31, 35, 37–40, 91–3, 112, 118, 121–2, 160, 182, 323, 361, 391 globalization, 8, 308 Golub, 8–9, 71, 120, 290–3, 295–7, 310, 379, 409–10 gravity model, 243–4 GREEN, 381, 404 greenhouse gas emissions, 304 Grilli, 365, 374 Gross Domestic Investment (GDI), 58, 144, 155, 279 Gross Domestic Product (GDP), 7, 67, 73–4, 81, 86–7, 122, 124–5, 127, 136, 138–42, 152–3, 155–6, 158, 164, 185, 187, 189, 192, 205, 216–20, 223–5, 229, 247–8, 251, 253–4, 260–1, 273, 279, 286, 296, 324, 326–30, 336, 343, 363–4, 368–70, 372, 382, 384–6, 399, 402–3, 410 gross investment, 47, 58, 85, 127, 247, 326, 413, 416–17 Gross National Product (GNP), 369, 371–2 gross rate of return, 42, 45–6, 53, 72, 121–2, 127, 424, 426 growth: in capital stocks, 81, 126; labor force, 142, 343–4, 363, 367; per capita income, 290, 297, 301; population, 8, 139–40, 187, 269–70, 280–2, 284, 286, 288, 296, 297, 299–301, 308, 342–3, 363, 399, 409; productivity, 216–17, 219, 279–81, 296, 298, 304–5, 321, 359–60, 363, 409; in the


Index growth (cont.): rate of return, 44–5, 47–8, 50, 54, 57, 81, 84–5, 88–9, 133; rates, 146, 155, 281; in real GDP, 153; real income, 8, 308; Total Factor Productivity (TFP), 7–8, 279–80, 282–3, 285, 288, 310, 409 GTAP: 4 Data Base, 70, 149, 157, 209, 212, 233, 237–8, 245, 268, 276, 289; 5 Data Base, 118, 157, 201, 292–3, 310, 323, 340, 386, 395; 6 Data Base, 120, 122, 129–30, 135, 138, 157; model, 3, 5, 13–14, 24, 69, 72, 87, 120, 158–9, 162, 173, 177, 207, 270, 353, 379, 406–7, 411, 421; parameters, 121 Hanslow, 158–9, 162, 171, 267, 374 Harrigan, 357, 374 Harris, 314 Harrison, 10, 70, 201, 207, 232, 267, 289, 413, 419 header array file (HAR), 71, 177, 180–1, 198 Hertel, 3, 7–8, 10, 13, 64, 70–1, 119, 137, 157–8, 162, 172, 193, 201, 205, 207–8, 213, 225, 232–3, 235, 267–8, 270, 273, 289–93, 295–7, 310–13, 316, 340–1, 353, 374, 379, 404, 406, 408–11, 419 Home bias effect, 92 Horioka, 69–70, 92, 118, 267, 380, 399, 404, 411 Horridge, 193, 201, 216, 233, 267, 413, 419 household: local, 112, 122; regional, 8, 21, 23–4, 26–9, 37, 39–40, 69, 72, 91–4, 105, 117–18, 121–2, 129, 134, 159, 181, 225, 327, 343–4, 372, 379, 387–8, 390–2, 395, 422, 424–7; wealth, 32–3, 35, 92, 94, 225, 382, 388, 391 Howe, 381, 404 Huang, 232 Huff, 70, 158, 162, 172, 225, 232, 353, 356, 374 Hummels, 240–2, 262, 268 Ianchovichina, 3, 7, 9, 13, 60–4, 70–1, 81, 88, 118, 120, 157–9, 172–3, 205, 207–9, 213, 227, 233, 269–70, 289, 313, 320–1, 341, 379, 404, 406, 408–9, 411 Impact Project/Center of Policy Studies, 407 imperfect competition, 314–15

433

import demand, 244, 246 income: earned on equity, 160; from equity, 26, 41, 122, 181; flows, 5–6, 14, 21, 37, 69, 91, 158–9, 365, 407; tax, 38, 205 income growth: per capita, 290, 297, 301; real, 8, 308 interest: premium, 360, 363; rate, 353–5, 360, 381 intermediate goods, 213, 241 international capital: flow, 69, 380, 386, 410; mobility, 4, 7, 13, 21–3, 69–71, 101, 117, 208; movement, 280 international demand system for commodities, 308, 409 international investment flows, 237, 245, 409 International Monetary Fund (IMF), 206, 211, 342, 374 international spillovers, 270 intra-industry trade, 239 investment: allocation of, 58, 360; data, 23, 75, 125–7, 211; flows, 7, 237, 245, 249, 408–9; foreign, 7, 23, 91, 93, 112, 118, 156, 205–6, 208–11, 215–16, 221, 223, 225, 227–9, 235, 252, 261, 327, 330, 333, 360, 407, 409; Foreign Direct (FDI), 23, 156, 205–7, 211, 215, 232, 252, 261, 313–14, 322–3, 327–36, 410; global, 36, 59, 60, 344, 369, 372, 410; gross, 47, 58, 85, 127, 247, 326, 413, 416–17; Gross Domestic (GDI), 58, 144, 155, 279; level of, 43–5, 47–9, 56–7, 65, 84–5, 363, 416, 418; model of, 41, 409; net, 19, 20, 166; value, 127, 419 investment schedule: actual, 46; expected, 54–5, 60 Itakura, 7–8, 120, 173, 235, 312–14, 340, 409–10 Japan-Singapore Free Trade Agreement, 235–6, 239–40, 247, 249, 251, 254, 260, 326, 339 Jensen, 313, 321–2, 340 Jones, 78–9, 118, 279, 289 Kapur, 31, 70 Kehoe, 313, 341 Kesavan, 31, 70 Kets, 296, 310 Kmenta, 109, 119 Kouparitsas, 405


434

Index

Kraay, 9–10, 92, 97–102, 104, 106, 109, 111, 115, 118–19, 123–4, 135, 382–5, 387–8, 390, 395, 405, 410–11 labor: force growth, 142, 343–4, 363, 367; productivity, 296, 360; projections, 138, 140, 142, 146–7; skilled, 137–8, 140, 144–7, 160, 273, 326; unskilled, 4, 137, 140, 146–7, 152–3, 162, 164, 187, 247, 273, 288, 296–7, 326 lagged adjustment: approach, 44; of capital stocks, 14; parameters, 71–2, 81, 117, 128 Lagrangean, 33 Lai, 357, 375 land-intensive: crops, 298; goods, 293, 303 Larson, 75–8, 117, 119 Latin America, 98, 102, 109–11, 116, 129, 140, 142, 144–5, 147–8, 152, 226–7, 244, 271, 279–82, 301, 309, 355, 385, 387, 395 Lejour, 233, 296, 310 Levin, 79, 119, 313, 341 Levinsohn, 313, 341 Lewis, 92, 119, 365, 374 liberalization of services, 210, 215, 243 life expectancy, 342–3, 345, 348–9, 352, 366 life-cycle hypothesis, 380 Lin, 70, 79, 118–19, 172, 312–15, 319, 321, 323, 325, 327, 329, 331, 333, 335, 337, 339, 340, 341, 375, 404 Lipsey, 365, 374 local assets, 22–3, 93–4, 96, 112 local capital, 30–2, 92, 112 local firms, 23–4, 27, 31, 33, 37, 39, 92, 112, 116, 118, 121–2, 182, 183, 314, 327 local household, 112, 122 logical file name, 178, 180–2 long-run equilibrium, 14, 59–60, 71–2, 78, 81, 117, 122 long-run stabilization, 399, 410 low-wage economies, 409 Ludena, 296, 310 macroeconomic projections, 136–8, 156, 157 Maddison, 233 Mai, 214, 217, 227, 229 marginal propensity to save, 381 markup elasticity, 319, 320–1 Markusen, 313, 315–16, 341

Martin, 126, 157, 207, 209, 212, 227, 232–3, 268, 289 Mattoo, 206 maximum likelihood estimation method, 106, 108 McDougall, 8–9, 13, 23, 60, 70–1, 81, 118, 120, 136, 149, 157–8, 209, 233, 268, 273–6, 289, 293, 296, 310, 323, 340, 342, 379, 404–5, 408, 410, 411, 413 McKibbin, 207–8, 342–3, 373, 405 Mendelsohn, 296, 310, 311 Mi, 291, 309 Middle East and North Africa, 98, 102, 109, 116, 244, 299, 301–2, 305, 309, 360, 385, 395 model: comparative static GTAP, 158–9; discrete choice, 241; dynamic FARM (D-Farm), 270; Future Agricultural Resources Model (FARM), 270, 273, 282, 284, 286–7, 289, 291; G-Cubed, 381; of investment, 41, 409; partial equilibrium, 270, 320; path-dependent, 64; recursive dynamic, 61, 161, 219; regression, 111 movement of natural persons, 215 Mrongowius, 216, 233 Muhleisen, 343, 374 Multi-Fibre Agreement (MFA), 149, 150, 212, 233 negative equity, 30 Nehru, 75, 76, 119 net investment, 19, 20, 166 net rate of return, 41, 46, 72, 404 Newly Industrialized Economies (NIEs), 152, 225 non-accumulable endowments, 171 nontariff barriers, 215, 235 nontariff trade costs, 239, 245 Norheim, 236, 267 normal growth rate: apparent current (KHAPP), 53–56; of the capital stock (KHAT), 46–9, 52–7, 59–60, 63, 81, 84, 121, 125–7, 131–2, 195, 424 normal rate of growth in income (YHAT), 391 North America, 3, 82, 130, 140, 144, 146–7, 151–3, 164, 167–8, 207, 213, 226–8, 235, 244, 249, 267, 295, 301, 304, 306, 309, 324, 326, 340, 350–2, 355–6, 358, 361–9, 371


Index North American Free Trade Agreement (NAFTA), 3, 312–13, 341 Norton, 269, 289 Numeraire shock, 166 oligopolistic competition, 316 omitted variables problem, 109, 111 Ordinary Least Squares (OLS), 79, 105–6 Organization for Economic Cooperation and Development (OECD), 74–80, 83, 117, 119, 226, 233, 310, 352, 374–5, 404 ownership: of capital endowment, 159; foreign, 7, 29, 30, 35, 91, 123–4, 156, 160, 205, 209–11, 217, 221–2, 225–8, 251, 261, 323, 328, 330, 332–3, 407 Pangestu, 233 parameters: behavioral, 128; entropy, 6, 117; GTAP, 121; lagged adjustment, 71–2, 81, 117, 128; rigidity, 32, 92–5, 97, 112–16, 118, 129–30 partial equilibrium, 5, 245, 270, 300, 312, 320; model, 270, 320; simulation, 300, 320 path dependent/dependency, 59, 64–7, 161–9, 185 Pearson, 5, 10, 66, 70, 173, 193, 201, 267, 289, 413, 419 perfect adjustment model of investment, 41 perfect capital mobility, 43, 56, 249 Perkins, 216, 233 Perpetual Inventory Method (PIM), 74–5 physical capital, 13, 21–4, 36, 223, 279, 343, 359–60 policy: closure, 185; projections, 4, 5, 137, 148–50, 273, 326; shock, 4, 13, 158, 164, 166, 169, 183, 185, 187, 188, 191, 407; simulation, 136, 160, 164–5, 185, 188–92, 197, 200, 210, 222, 279, 283, 325–6; variables, 200 pooling technique, 109–11 population: data, 138; growth, 8, 139–40, 187, 269–70, 280–2, 284, 286, 288, 296, 297, 299, 300–1, 308, 342–3, 363, 399, 409 Powell, 292, 310 price: of capital goods, 20, 26–7, 43, 87, 93, 166, 219, 220, 226, 414; effective, 240, 242–4, 246, 250, 262; of equity, 25, 27, 37, 391, 426; food, 365; food and resource, 269; resource, 269; of saving, 226, 392

435

private expenditure, 280 procompetitive effect, 313, 315, 316, 319–22, 327–9, 333–4, 336 product differentiation, 315–16, 357 productivity: agricultural productivity, 273, 280–1, 284–5, 288, 304, 409; gains, 208–10, 212, 216–17, 219, 221, 226–9; growth, 216–17, 219, 279–81, 296, 298, 304–5, 321, 359–60, 363, 409; linkages, 313, 327, 335; in services, 229, 408; shock, 216–17, 220, 226, 229 projections: baseline, 138, 146–7; labor, 138, 140, 142, 146–7; macroeconomic, 136–8, 156, 157; policy, 4, 5, 137, 148–50, 273, 326; skilled labor, 138, 146–7; unskilled labor, 147 propensity to save, 9, 17, 58–9, 69, 379, 381, 392–3, protection, 198, 208–9, 245, 248, 320, 324–5 public expenditure, 280 rate of growth in capital stocks, 81 rate of growth in the rate of return: expected, 45, 47–8, 50, 84–5, 88–9; required, 44, 48, 133 rate of return: actual (rorga), 48–50, 57–8, 65, 68, 81–2, 87, 117, 128, 132, 156, 219, 413–14, 416; expected (rorge), 45–6, 49–53, 56–8, 65, 68–9, 72, 81, 84–7, 89, 117, 121, 132–4, 166, 193–5, 200, 413, 415–16, 426; global, 220, 399; net, 41, 46, 72, 404; target, 44, 81–2, 128, 133, 156, 413–14, 416, 419, 426 realization time, 46–8, 51 recursive dynamic model, 61, 161, 219 recursive solution procedure, 69 regional firms, 23, 26, 29, 92, 422, 425–7 regional household, 8, 21, 23–4, 26–9, 37, 39–40, 69, 72, 91–4, 105, 117–18, 121–2, 129, 134, 159, 181, 225, 327, 343–4, 372, 379, 387–8, 390–2, 395, 422, 424–7 regional trade agreement, 235, 237, 245, 312 regional wealth, 8, 27, 65, 71, 91, 382, 394 regression model, 111 Reimer, 8, 293, 310, 312, 410 research and development (R&D), 340 resource price, 269 Rigidity: of allocation of wealth by regional household (RIGWQH), 72, 121; of source of funding of enterprises (RIGWQ F), 72, 121, 129


436

Index

rigidity parameters, 32, 92–5, 97, 112–16, 118, 129–30; recommendations for setting, 114 Rimmer, 211, 232, 292, 310, 373 Roberts, 313, 340 Rodrik, 315, 322, 340, 341 Roland, 315, 319, 340 Rose, 289, 310–11 rules-based economy, 206 RunDynam program, 173 secondary education, 145 sensitivity analysis, 276, 336 services: capital, 38; cross border supply of, 215, 228; direct trade in, 209, 215, 263; trade in, 7, 209–10, 215, 217, 219–20, 222, 244, 263; trade liberalization, 210, 244, 261 set information, 178–9 shareholder equity, 23 Shi, 8, 342, 344, 366, 374–5, 410–11 Shiells, 245, 267, 312–13, 340 Shock: additional (ashock), 52, 188; basecase, 188; numeraire, 166; policy, 4, 13, 158, 164, 166, 169, 183, 185, 187, 188, 191, 407; productivity, 216–17, 220, 226, 229; total (tshock), 66 Shoemaker, 7, 269, 409 Short run dynamics, 65 simulation: baseline, 160, 238, 249–1, 270, 276, 291, 297–8, 300, 305, 327, 342, 357, 361, 365; demand-side, 297; dynamic, 16, 43, 162, 166–8, 199–200; partial equilibrium, 300, 320; policy, 136, 160, 164–5, 185, 188–92, 197, 200, 210, 222, 279, 283, 325–6 skilled labor: 137–8, 140, 144–7, 160, 273, 326; projections, 138, 146–7 Sohngen, 291–2, 295–7, 310–11 Sokoloff, 314, 340 Solow-Swan, 343 solution: file, 193, 200; method, 15, 42, 184, 186; procedure, 17–19, 66, 69, 270–1 South Asia, 99, 102, 109, 111, 116, 130, 142, 144–5, 147–8, 152, 225–7, 231, 254, 261, 266, 293, 295–6, 299–304, 309, 339, 350–2, 355–6, 358, 360, 362–3, 365, 367, 382, 385–6, 400 speed of convergence, 4, 72, 81–4, 117 Spinanger, 213–15, 228, 232–3, 249, 267 standard closure, 152, 155

steady growth equilibrium, 392 steady-state equilibrium, 407 stock-flow dynamics, 69 structural adjustment, 161, 166 Strutt, 149–50, 157, 173, 232, 267, 289 Tang, 207–8, 233 target rate of return, 44, 81–2, 128, 133, 156, 413–14, 416, 419, 426 tariff: and quotas, 152, 210, 217, 228; rate, 149–51, 212–13, 226, 273, 324; revenue, 149–50, 209 tax: income, 38, 205; revenue replacement, 166 technological progress in agriculture, 282 terms of trade, 97, 159, 162, 170–1, 225, 227, 261, 263, 326–7 tertiary education, 145 TFP growth, 7–8, 279–80, 282–3, 285, 288, 310, 409 time: base, 15, 16, 46, 51; continuous, 16; current, 20, 46, 50; dependent variables, 166; expectation, 46, 48; future, 45–6, 133; realization, 46–8, 51; treatment of, 5, 14–15; as a variable (TIME), 20, 46 total equity, 23, 26, 37, 39 total factor productivity (TFP), 7, 183, 214, 216–17, 270 total shock (tshock), 66 trade: balance, 223–4, 246, 250–1, 254, 260, 301–3, 407; data, 316; in services, 7, 209–10, 215, 217, 219–20, 222, 244, 263; and transport, 226, 244; volume effects, 254 traded commodities, 180 transport margins, 323 treatment: of saving, 58, 69, 379; of time, 5, 14–15 Trefler, 357 Truong, 23, 70, 149, 157, 209, 268, 276, 289 Tyers, 8, 342, 344–5, 348, 351–3, 356, 360–2, 366, 368, 373–5, 381, 399, 405, 410–11 UNCTAD, 239 United Nations (UN), 239, 342, 350–2, 362–3, 375 unskilled labor, 4, 137, 140, 146–7, 152–3, 162, 164, 187, 247, 273, 288, 296–7, 326; projections, 147 updated data, 30, 167, 193, 197–9 updated policy data, 165–6


Index Uruguay Round (UR), 3, 98, 107, 113, 148–50, 157, 213, 232, 238–9, 247–8, 267–8, 273, 289, 326, 339; agreement, 150, 247 utility function, 353, 379, 381 validation exercise, 410 Venables, 315–16, 341 Ventura, 10, 97, 118–19, 135, 405, 411 ViewHAR, 179, 191, 195, 198 Walmsley, 7, 9–10, 71, 120, 136–7, 157–8, 173, 205, 207–8, 213, 233, 235, 313, 340, 342, 379, 406, 408–10, 413, 421 Walton, 72–3, 119

437

Wang, 207, 209, 233 wealth: accumulation, 14–15, 18, 65, 70; in equity, 26, 29, 424–7 wealth-to-income ratio, 386–99, 401, 410 Weber, 357, 359, 373, 375 welfare: analysis, 6, 22, 225; decomposition, 158–9, 161–2, 164–5, 168, 173, 199, 200 Wenping, 234 Western Europe, 144, 146, 151–3, 226–7, 244, 254, 260, 301–3, 305–6, 309, 350–1, 355, 358, 361–5, 367–9, 371 Wilcoxen, 381, 405 Yang, 207, 234, 365, 374


Dynamic Modeling and Applications for Global Economic Analysis

Dynamic system modeling and control

Real Analysis with Economic Applications

Dynamic system modeling and control

Dynamic System Modeling and Control

Dynamic Systems, Economic Growth, and the Environment (Dynamic Modeling and Econometrics in Economics and Finance)

Dynamic Social Network Modeling and Analysis: Workshop Summary and Papers

Modeling Dynamic Climate Systems

Modeling of dynamic systems

Modeling and Simulation for Analyzing Global Events

Dynamic Modeling for Business Management: An Introduction

Global Analysis - Studies and Applications I

Modeling Dynamic Climate Systems

Manifolds, Tensor Analysis and Applications (Global analysis, pure and applied)

Manifolds, Tensor Analysis and Applications (Global analysis, pure and applied)

Dynamic Response of Linear Mechanical Systems: Modeling, Analysis and Simulation

Discrete Time Dynamic Economic Models: Theory and Empirical Applications

Dynamic Economic Theory

Information Modeling for Internet Applications

Hierarchical Modeling and Analysis for Spatial Data

MOSFET modeling for circuit analysis and design

Simulation Modeling and Analysis

Dynamic games theory and applications

Dynamic Laser Speckle and Applications

Economic Analysis for Property and Business

Computational Economic Analysis for Engineering and Industry

Economic Analysis for Property and Business

Handbook of dynamic system modeling

Dynamic Games: Theory and Applications

Dynamic Mechanical Analysis: A Practical Introduction to Techniques and Applications

Economic Theory and Global Warming

Dynamic Modeling and Applications for Global Economic Analysis

Dynamic system modeling and control

Real Analysis with Economic Applications

Dynamic system modeling and control

Dynamic System Modeling and Control

Dynamic Systems, Economic Growth, and the Environment (Dynamic Modeling and Econometrics in Economics and Finance)

Dynamic Social Network Modeling and Analysis: Workshop Summary and Papers

Modeling Dynamic Climate Systems

Modeling of dynamic systems

Modeling and Simulation for Analyzing Global Events

Dynamic Modeling for Business Management: An Introduction

Global Analysis - Studies and Applications I

Modeling Dynamic Climate Systems

Manifolds, Tensor Analysis and Applications (Global analysis, pure and applied)

Manifolds, Tensor Analysis and Applications (Global analysis, pure and applied)

Dynamic Response of Linear Mechanical Systems: Modeling, Analysis and Simulation

Discrete Time Dynamic Economic Models: Theory and Empirical Applications

Dynamic Economic Theory

Information Modeling for Internet Applications

Hierarchical Modeling and Analysis for Spatial Data

MOSFET modeling for circuit analysis and design

Simulation Modeling and Analysis

Dynamic games theory and applications

Dynamic Laser Speckle and Applications

Economic Analysis for Property and Business

Computational Economic Analysis for Engineering and Industry

Economic Analysis for Property and Business

Handbook of dynamic system modeling

Dynamic Games: Theory and Applications

Dynamic Mechanical Analysis: A Practical Introduction to Techniques and Applications

Economic Theory and Global Warming

Recommend Documents