3 - Lecture Notes

Contents
List of Figures 5
List of Tables 11
1 Introduction 13
2 The Middle Income Trap 19
2.0.1 Agglomeration: A way out of the middle income trap? . . . . . . 21
2.1 Internal Migration and economies of scale effects . . . . . . . . . . . . . 25
2.1.1 Trade liberalization . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.2 Agglomeration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2 Agglomeration and migration . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2.1 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3 Economic Growth 55
3.1 The Solow (1956) model of economic growth . . . . . . . . . . . . . . . . 58
3.1.1 Basic model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.1.2 Dynamics in the short run . . . . . . . . . . . . . . . . . . . . . . . 66
1
2 CONTENTS
3.1.3 Model solution under Cobb Douglas . . . . . . . . . . . . . . . . 84
3.2 Population growth in the Solow model . . . . . . . . . . . . . . . . . . . 89
3.3 Economic growth and technological change . . . . . . . . . . . . . . . . . 93
3.3.1 Dynamics with technological change . . . . . . . . . . . . . . . . 99
3.4 Growth and internal migration . . . . . . . . . . . . . . . . . . . . . . . . 102
3.5 Community enforcement, migration and development . . . . . . . . . . 107
3.5.1 Empirical relevancy . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.6 Human Capital in the Solow model . . . . . . . . . . . . . . . . . . . . . 117
3.7 Empirical application on institutions and growth . . . . . . . . . . . . . 125
A An introduction to linear regression . . . . . . . . . . . . . . . . . . . . . 133
A-1 Dependent Variable, Independent variable and Regressions co-

efficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
A-2 From the theoretical to the empirical model: the error term . . . 135
A-3 Standard errors and statistical significance . . . . . . . . . . . . . 137
A-4 Linear regression with many variables . . . . . . . . . . . . . . . 139
4 Canonical Trade Models 141
1 The Ricardo Model of Comparative Advantage . . . . . . . . . . . . . . . 142
1.1 The model basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
1.2 Gains from trade . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
1.3 Trade liberalization and income . . . . . . . . . . . . . . . . . . . 153
1.4 Equilibrium goods prices . . . . . . . . . . . . . . . . . . . . . . . 154

CONTENTS 3
2 The Heckscher-Ohlin model . . . . . . . . . . . . . . . . . . . . . . . . . . 156
2.1 The 2 × 2 × 2 model . . . . . . . . . . . . . . . . . . . . . . . . . . 157
2.2 Production, consumption and the gains from trade . . . . . . . . 161
2.3 Factor prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
2.4 Trade liberalization and factor prices . . . . . . . . . . . . . . . . 168
2.5 Stolper Samuelson Theorem . . . . . . . . . . . . . . . . . . . . . 170
3 FDI and outsourcing in a multi-industry framework . . . . . . . . . . . . 179
3.1 The Feenstra and Hanson model . . . . . . . . . . . . . . . . . . . 183
3.2 Capital market liberalization, offshoring and wage inequality . . 191
3.3 The role of the continuum of industries . . . . . . . . . . . . . . . 193
3.4 Empirical tests of the Feenstra and Hanson model . . . . . . . . 200
4 Migration in the 2 × 2 × 2 model . . . . . . . . . . . . . . . . . . . . . . . 209
4.1 Endowments in the 2 × 2 × 2 model . . . . . . . . . . . . . . . . . 210
4.2 The Migration Shock . . . . . . . . . . . . . . . . . . . . . . . . . . 216
4.3 Formal proof of the Rybzynski theorem . . . . . . . . . . . . . . . 220
5 The Mariel boat lift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
5 International migration 227
1 International migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
2 Why do people migrate? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
2.1 The determinants of the migration decision . . . . . . . . . . . . 232

4 CONTENTS
2.2 The determinants of international migration: empirical evidence 252
3 Who migrates? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
3.1 The Borjas 1987 Model . . . . . . . . . . . . . . . . . . . . . . . . . 256
4 The brain drain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
4.1 The brain drain: theoretical framework . . . . . . . . . . . . . . . 276
4.2 Model simulation in Excel . . . . . . . . . . . . . . . . . . . . . . . 293
4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
4.4 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
4.5 The brain drain: empirical evidence . . . . . . . . . . . . . . . . . 313
5 The effects of immigration on the host country . . . . . . . . . . . . . . . 317
6 Trade and growth 327
1 Economic growth in open economy models . . . . . . . . . . . . . . . . . 328
1.1 Consumption of final goods . . . . . . . . . . . . . . . . . . . . . 333
2 Trade and innovation in a North-South framework . . . . . . . . . . . . 344
2.1 The specialization equilibrium . . . . . . . . . . . . . . . . . . . . 347
2.2 Equalization equilibrium . . . . . . . . . . . . . . . . . . . . . . . 349
2.3 Endogenous innovation and imitation . . . . . . . . . . . . . . . . 352
2.4 Comparative statics . . . . . . . . . . . . . . . . . . . . . . . . . . 355
3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
List of Figures
2.1 The Middle Income Trap Source: Gill and Kharas (2007) . . . . . . . . . 20
2.2 Urbanization in East Asia Gill and Kharas (2007) . . . . . . . . . . . . . 21
2.3 Regional trade integration in East Asia Gill and Kharas (2007) . . . . . . 23
2.4 Average and marginal cost curve . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 General Equilibrium in the Krugman Model . . . . . . . . . . . . . . . . 33
2.6 Trade liberalization in the Krugman (1979b) model . . . . . . . . . . . . 35
2.7 Separation equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.8 Spreading and agglomeration Source: van Marrewijk (2012) . . . . . . . 52
2.9 Low transportation costs Source: van Marrewijk (2012) . . . . . . . . . . 53
2.10 Dynamics with high transport costs Source: van Marrewijk (2012) . . . 54
3.1 The per capita GDP distribution over time . . . . . . . . . . . . . . . . . 56
3.2 Production functions in line with the Inada conditions . . . . . . . . . . 66
3.3 Fundamental law of motion . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.4 Non-existent or not-unique steady states . . . . . . . . . . . . . . . . . . 72
3.5 Hump shaped relationship between s and c . . . . . . . . . . . . . . . . . 82
5
6 LIST OF FIGURES
3.6 Estimated coefficients of a production function Source: Olley and Pakes

(1992) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.7 Steady state equilibrium with population growth . . . . . . . . . . . . . 93
3.8 Shifts of the production function due to population growth . . . . . . . 96
3.9 Steady state equilibrium with technological change . . . . . . . . . . . . 100
3.10 Dynamics due to technological change . . . . . . . . . . . . . . . . . . . 101
3.11 Migration dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.12 Livability in East Asia and other Middle-Income Regions Source: Gill,
I. and Kharas, H. (2007) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.13 Urban Transport and Road Safety Source: Gill, I. and Kharas, H. (2007) 115
3.14 Livability in China vs. G-7 Source: Gill, I. and Kharas, H. (2007) . . . . 115
3.15 Livability in cities: East Asia vs. Latin America . . . . . . . . . . . . . . 116
3.16 Dynamics of physical capital per effective unit of labor. . . . . . . . . . . 120
3.17 Dynamics of human capital per effective unit. . . . . . . . . . . . . . . . 121
3.18 Steady state equilibrium in the Solow model with human capital . . . . 125
3.19 Institutions and economic growth . . . . . . . . . . . . . . . . . . . . . . 127
3.20 Geography and Economic Growth . . . . . . . . . . . . . . . . . . . . . . 131
3.21 Settler mortality and average protection against the risk of expropriation 132
3.22 Settler mortality and per capita GDP . . . . . . . . . . . . . . . . . . . . . 132
A-1 Regression line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
A-2 Regression line and real data . . . . . . . . . . . . . . . . . . . . . . . . . 137

LIST OF FIGURES 7
4.1 The production possibility frontier at Home and Foreign . . . . . . . . . 145
4.2 Optimal production/consumption under autarky . . . . . . . . . . . . . 149
4.3 Optimal production/consumption under free trade . . . . . . . . . . . . 152
4.4 Optimal world market prices . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.5 Cost minimization in sector 1 . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.6 Cost minimization in sector 2 . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.7 The Gains from Trade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.8 Factor prices and trade liberalization . . . . . . . . . . . . . . . . . . . . . 169
4.9 The zero profit conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.10 Wage inequality among males in Germany (Source: Dustmann, Lud-

steck, and Schönberg (2009, QJE) . . . . . . . . . . . . . . . . . . . . . . . 182
4.11 Wage inequality among females in Germany (Source: Dustmann, Lud-

steck, and Schönberg (2009, QJE) . . . . . . . . . . . . . . . . . . . . . . . 182
4.12 Labor demand shifts in Germany (Source: Dustmann, Ludsteck, and

Schönberg (2009, QJE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
4.13 The comparative cost advantage . . . . . . . . . . . . . . . . . . . . . . . 188
4.14 Labor market clearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.15 The effects of FDI on the comparative cost advantage . . . . . . . . . . . 192
4.16 Labor market clearing after FDI . . . . . . . . . . . . . . . . . . . . . . . . 193
4.17 Autarky and trade equilibrium . . . . . . . . . . . . . . . . . . . . . . . . 197
4.18 Zero profit conditions and the effects on factor prices . . . . . . . . . . . 199
4.19 Regression results to equation (4.56) . . . . . . . . . . . . . . . . . . . . . 204

8 LIST OF FIGURES
4.20 Regional variation in the wage gap Source: Feenstra and Hanson (1995) . 207
4.21 The evolution of the wage gap in comparison to the US Source: Feenstra
and Hanson (1995) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.22 FDI, remittances and aid Source: The Economist . . . . . . . . . . . . . . . 209
4.23 Economic development and infant industry protection . . . . . . . . . . 211
4.24 Annual wage growth in China Source: Li, Li, Wu, and Xiong (2012) . . . . 211
4.25 Industry specific labor and capital demand . . . . . . . . . . . . . . . . . 214
4.26 Changes of capital and labor input due to migration . . . . . . . . . . . 217
4.27 Cone of diversification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
4.28 Difference in differences treatment effect . . . . . . . . . . . . . . . . . . 222
4.29 Estimation results from Card (1990) . . . . . . . . . . . . . . . . . . . . . 223
4.30 Further estimation results from Card (1990) . . . . . . . . . . . . . . . . . 225
5.1 Permanent migration flows to OECD countries, 2006-2014. . . . . . . . 228
5.2 Foreign-born individuals as percentage of the total population. 2000

and 2013. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.3 The intention to stay abroad Source: Dustmann and Görlach (2016) . . . . 236
5.4 The wage profile for different α . . . . . . . . . . . . . . . . . . . . . . . . 238
5.5 Return migration due to home bias (scenario II) . . . . . . . . . . . . . . 247
5.6 Savings program with home bias or differences in purchasing power

(scenario II and III) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
5.7 Return migration and human capital accumulation (scenario IV) . . . . 251
LIST OF FIGURES 9
5.8 Distribution of wages in both countries . . . . . . . . . . . . . . . . . . . 258
5.9 Graphical illustration of the selection cases (I) . . . . . . . . . . . . . . . 266
5.10 Graphical illustration of the selection cases (II) . . . . . . . . . . . . . . . 267
5.11 Graphical illustration of the selection cases (III) . . . . . . . . . . . . . . 267
5.12 Economy without migration: Multiple Equilibria . . . . . . . . . . . . . 287
5.13 Technological progress using a Gompertz function . . . . . . . . . . . . 295
5.14 Equilibrium without migration . . . . . . . . . . . . . . . . . . . . . . . . 297
5.15 Brain drain or gain? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
5.16 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
5.17 Gompertz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
5.18 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
5.19 Computing π . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
5.20 Computing ĉ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
5.21 Computing ĉ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
5.22 Visual Basic Macro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
5.23 Representation of nested-CES structure. . . . . . . . . . . . . . . . . . . . 320
6.1 The model in a nutshell . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
6.2 The North-South wage gap . . . . . . . . . . . . . . . . . . . . . . . . . . 351

10 LIST OF FIGURES
List of Tables
3.1 Convergence in growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1 Inflows of permanent immigrants into selected OECD countries, 2007-

2013. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.2 Top 20 countries of origin of new immigrants to the OECD. . . . . . . . 231
5.3 Emigration stocks (in thousands) and emigration rates of high skill im-
migrants to OECD destinations. . . . . . . . . . . . . . . . . . . . . . . . . 275
11
12 LIST OF TABLES
Chapter 1
Introduction
The main goal of the course is studying the interaction between market integration
and economic development. How do international trade in goods and factors shape
the economic performance of countries at earlier development stages? The analyses
build upon the more common textbook approaches.
Many of the models and applications illustrated in the course may be familiar to
you from other courses. For instance, the Solow model of economic growth and the
canonical trade models have already been discussed in other modules. However,
we blend the models of national and international trade and migration with these
workhorse models. Hence, a sound knowledge of the contents of the previous mate-
rial is an advantage, but not a prerequisite. The course presents all the models from
scratch and includes detailed explanations of the relevant mathematical and econo-
metric tools. When appropriate, some recent stylized facts are used to motivate the
different frameworks.
The first part of the course summarizes a more recent debate about the middle income
trap. Researchers from the World Bank famously pointed out that many emerging
economies managed growing out of poverty into the middle-income region, but failed
to achieve high-income status. This observation is not new. Emerging economies
13
14 CHAPTER 1. INTRODUCTION
typically rely heavily on building up manufacturing activities with labor intensive

production of goods for the world market. Low labor cost is the main comparative
advantage at the beginning of the development process, but this advantage deterio-
rates when economies start growing as wages are increasing as well. Economic growth
rates slow down over time and sooner or later may come to a complete halt but there
are potential ways out of this trap. This trap is explained based upon the model of
Solow discussed in the third chapter but before turning to this model we take a closer
look at the role of regional integration.
The World Bank report on the middle-income trap favored urbanization and agglom-
eration as one potential strategy for sustainable development. Regional integration
allow firms to become more efficient due to economies of scale and sufficient supply
of labor for the labor intensive production in earlier stages of economic development.
We start our analysis by focusing on regional integration, which is a good starting
point for extending the analysis to the broader macroeconomic topics such as growth
and international trade. Regional integration is driven my internal migration. In a
first model, we allow for internal migration of workers and we broaden this analysis
by agglomeration of workers and firms in the last part of this chapter.
Chapter 3 provides a rigorous theoretical treatment of the Solow model and an illus-
tration of the related empirical growth studies. First, we tackle the question, "Why
do countries grow?" Capital formation is one potential source of economic growth,
but its growth potential is limited without technological change. Despite being quite
intuitive and straightforward, this concept is treated in-depth, following the illustra-
tion of the advanced textbook on economic growth by Acemoglu (2008). Equipped
with the insights derived from this model, we analyze the question "Why have some
nations failed to grow?" Indeed, while some countries have demonstrated sustained
economic growth through technological change, some others have remained stuck at
low levels of per-capita GDP and have not exhibited any growth. We dedicate one
section of the chapter to this puzzle, presenting the discussion as summarized in Ace-
15
moglu (2008). The prominent answer given in the literature relies on the existence of
institutions. Indeed, a country’s institutional setting can provide a safe environment
for entrepreneurs to invest. The absence of such an environment may render capital
formation inefficient, resulting in low rates of economic growth.
Another pillar of economic development is trade in goods and the movement of factor
inputs. The canonical trade models studied in Chapter 4 are able to rationalize inter-
national linkages between developed and developing countries based on technology
or endowment differences. Countries specialize in particular industries where they
produce with lower opportunity costs. The idea of a comparative cost advantage,
which determines international trade patterns, depends on country-specific differ-
ences in observable characteristics such as technology (Ricardo) or factor endowments
(Heckscher Ohlin). More recent models focus on intra-industry trade. This implies
that countries tend to export goods produced in sectors where they have a relative
cost advantage compared to the rest of the world, while they tend to import goods
that can be purchased cheaper on the world market rather than domestically. The
idea that comparative advantage matters appears to be plausible in the context of
developing economies. Indeed, when looking at trade between developing and devel-
oped countries, specialization in particular industries is evident in the data. However,
a drawback of these classical trade theories is that they are not able to explain why
similar countries import and export goods produced in the same industry.
At the end of this chapter, we are introducing migration into these canonical trade
models. This exercise is a good foundation for the more detailed discussion of mi-
gration studies in Chapter 5. We also look at a seminal paper in empirical migration
research. Motivated by the outcomes on migration in the canonical trade models,
Card (1990) analyzed the impact of an unforeseen migration shock on the US labor
market. The so called Mariel boatlift was a sudden inflow of migrants to Miami.
The canonical trade models would predict that the additional labor supply can be
absorbed by producers without having major impact on the natives’ labor market
outcomes. The empirical evidence on the labor market effects of the Mariel boatlift
supports this model outcome as the labor market condition for natives did not react
to the shock.
Indeed, migration movements represent a recurrent pattern from developing to de-

veloped countries. Moreover, the “loss” of individuals due to migration away from
developing countries has been a crucial topic both for academic and policy debates.
We devote another chapter of this course to migration, which is focusing on the more
recent developments in the research on the determinants of migration movements
and the interrelated Brain Drain phenomenon. Specifically, we answer the following
question: “Why do people migrate?" In doing so, we review the theoretical frame-
works that provide explanations behind the individual migration decision (that is, at
the micro-level). We then proceed by answering the same question from the aggregate
perspective, analyzing the determinants of the migration patterns at the macro-level
and presenting the associated empirical evidence. Not all individuals from a given
population have the same propensity to migrate. Thus, we inquire about who chooses
to migrate, stressing the importance of the issue of the immigrants’ “selectivity” for
the study of migration. Finally, we offer a brief overview of the economic effects of
immigration on the host country.
The final Chapter 6 discusses models that nest both trade and capital formation. These
extensions of the Solow growth model include versions with migration, foreign di-
rect investment or trade in goods. The canonical trade models are blended with the
workhorse model in the growth literature in order to understand their interactions.
Under autarky, the only way to build up a substantial capital stock is through invest-
ment. Households face a trade-off between consumption and savings that can be used
for capital formation. Moreover, due to diminishing returns to capital and labor, fac-
tor income depends solely upon factors’ marginal productivity. Once we open those
economic growth models to international trade in goods and factors, i.e. migration
and foreign direct investments, the pattern of economic growth is substantially differ-
17
ent. We have to take the evolution of world prices into consideration, which produces
outcomes that are not as straightforward as is the case in more ’basic’ frameworks.
For example, taking into account the effects of trade on economic growth in develop-
ing economies has important repercussions on some of the most important objectives
of government’s policy. The last part of the chapter is dedicated to the Brain Drain,
where we will show the most recent theoretical framework on the effects of the Brain
Drain for growth in the developing country.
Recommended literature. The foundation of this course is the textbook "Introduc-

tion to Modern Economic Growth" from Acemoglu (2008), which looks at different
aspects of economic growth from different angles. Many of the aspects covered in
this textbook are not touched upon in the lecture notes. We highly recommend this
book for a more involved study of the course subjects. Chapters on trade in goods
and factors build upon the textbooks from Feenstra and Taylor (2014) and Feenstra
(2004). The first gives a more intuitive treatment of international trade models and the
empirics of international trade, whereas the latter is more advanced. Another highly
recommended textbook for students interested in international economics is the book
from van Marrewijk (2012). It provides a highly recommended online library for Excel
programs that allow simulating the different models. All these books are highly rec-
ommended and can be used for other courses provided by the chair of "International
Economics" as well as Bachelor and Master theses on topics related to Globalization.
The treatment of the issues on international migration and on the brain drain is based
on different sources (notably Borjas, 1987; Dustmann and Görlach, 2016; Ottaviano
and Peri, 2012; Docquier and Rapoport, 2012).
Notice that the lecture notes cannot be seen as substitutes for the recommended books
as the descriptions in the textbooks are completely different. The textbooks describe
the models in a much broader context, whereas our courses focuses on more specific
issues.
Moreover, also notice that the basic models discussed in this lecture are covered by
many other textbooks as well but the structure may be different compared to the
structure of our lecture notes. Nevertheless, the questions covered by the final exam
will be based on the lecture notes and the tutorials provided online. The textbooks are
helpful auxiliary materials and not complements or substitutes for the lecture notes.
Chapter 2
The Middle Income Trap
Gill and Kharas (2007) were among the first using the term „Middle Income Trap“ to
describe the stylized facts captured in the following graph.
Per capita incomes are traced over the period 1900 to 2000 for countries in the follow-
ing three groups:
• Group 1: The eight largest Latin American countries (Argentina, Brazil, Chili,
Colombia, Mexico, Peru, Uruguay, and the República Bolivariana de Venezuela).
• Group 2: Five East Asian economies that have reached high-income levels (Hong
Kong [China], Japan, Korea, Singapore, and Taiwan [China]).
• Group 3: The five middle-income countries in East Asia (China, Indonesia,

Malaysia, the Philippines, and Thailand).
Obviously, countries included in the Latin America-8 group continuously grew until
the middle of the 1970s before per capita income began declining in the 1980s. There
is some recovery in the 1990s but overall the Latin America-8 block did not succeed
in reaching the high-income level above the cutoff represented by the solid black line
slightly above the 10.000 US-Dollar marker.
19
20 CHAPTER 2. THE MIDDLE INCOME TRAP
Figure 2.1: The Middle Income Trap

Source: Gill and Kharas (2007)
The Latin America-8 was left behind by the second block comprising the East Asia-5
high income countries mentioned above. All countries included in this group man-
aged to become high-income countries in the 1980s.
The third group caught up with the Latin America-8 in more recent years but they
had not reached high-income status back in 2000. The World Bank defined the cutoff
for high income status at a gross national income per capita above US-Dollar 12,695
(threshold in 2021).
It has been discussed if the developing East Asian-5, and in particular China, will
remain trapped at around the same average middle-income as the Latin America-8
group or if they can manage to escape the middle income trap by surpassing the cutoff
income represented by the solid line in Figure 2.1. China managed approaching this
goal in 2022 but it is still unclear if it will be successful in maintaining this trajectory.
Many hurdles have to be taken on the path to sustainable growth away from the lower
ceiling that separates the high-income from the middle-income group of countries.
21
But why do nations fail in growing out of the middle income trap? Potential reasons
are discussed in the world bank report published by Gill and Kharas (2007). We will
go through their most important findings in this chapter and concentrate on their
main argument: agglomeration.
2.0.1 Agglomeration: A way out of the middle income trap?
Table 2.2 compares past and future urban population in the ten years after 2000 for
countries in East Asia.
Figure 2.2: Urbanization in East Asia

Gill and Kharas (2007)
All East Asian countries’ urban population shares are expected to grow at remarkable
rates. Korea, a country with already more than 80 percent of its population living
in urban areas, is anticipating 86.3 percent in 2030. The share of urban population
in China is expected to grow from 40 percent in 2005 to 60 percent in 2030 at an
annual growth rate of 2.7, which amounts to a massive increase of 20 percentage
points. Comparing East Asia in total to the rest of the world on average in the last
two columns shows that East Asian countries urban population share grows by 0.6
percentage points more than the world average.
How is this development related to the middle income trap? Urbanization is one
potential way out of the middle income trap through agglomeration and external
economies of scale in manufacturing. Economic growth depends highly on manufac-
turing and manufacturing depends on the availability of both capital and labor. The
former can be achieved through investments. Without frictions in the financial mar-
ket or international restrictions on foreign direct investment, capital flows are guided
by the highest returns to regions where capital rentals are relatively high. Emerging
economies have little difficulties in attracting the relatively small amounts of foreign
capital necessary for building up the desired stock of capital. Production is usually
more labor intensive. Purchasing machines and other equipment needed in manu-
facturing of less sophisticated goods is therefore less capital intensive compared to
investments needed for producing more sophisticated goods. However, machines
must be operated by labor and the stock of capital is low in developing countries
when compared to their stock of labor. Both arguments highlight the importance of
the supply of labor at arm’s length to the manufacturing firms located at the economic
hot spots. Agglomeration is one channel through which firms in emerging markets
get access to an ample supply of workers and globalization may push agglomeration
further through magnified foreign demand.
Figure 2.3 presents some insights about trade integration in East Asia.
Both the share of exports and imports was increasing in the period from 1990 to 2003.
For instance, China’s trade share grew by 11 percentage points (exports and imports).
However, this export growth was driven by the performance of a few economic hot-
spots mostly located in close proximity to the major East Asian hubs. Moreover,
intensified international trade can be associated with soaring wages within these ur-
ban clusters, leaving behind the less developed rural areas in China. The search for
a job in one of the more prosperous cities fueled agglomeration and agglomeration
23
Figure 2.3: Regional trade integration in East Asia

Gill and Kharas (2007)
itself made those firms even more competitive according to the external economies
of scale argument discussed in Gill and Kharas (2007). Constant supply of workers
moving from rural to urban areas has a moderating effect on wage growth and pro-
vides enough labor for satisfying the additional demand. Especially the offshoring
boom at the end of the last millennium fueled East Asian countries’ integration into
the world economy and massive exchange of components among each other and with
other countries outside of East Asia.
The numbers in Figure 2.3 present the development of trade in components, which
may have been driven by the more recent rise in offshoring. Offshoring firms special-
ize production in certain tasks or activities and exchange these tasks through national
and international trade in intermediate goods. Instead of producing all production
stages within one firm, slicing the value chain into different chunks allows firms to
specialize. Put differently, instead of using labor and capital as sole input factors in
the production of final consumption goods, firms may add a third factor containing
intermediate goods. Those intermediate goods are produced by firms specializing
in particular processes, which is why these firms may easily realize positive external
scale effects. However, this specialization is also associated with more trade. Instead
of trading a good once when it is finished, the individual components used to produce
the final consumption good cross borders several times, which promotes regional in-
tegration of firms in East Asia and agglomeration. Agglomeration and economies of
scale allow firms to be internationally competitive by producing at the lower segments
of the cost curve.
The remainder of this chapter is focusing on the two-way (reverse) causality by focus-
ing on two very prominent explanations. The Krugman (1979b) model of international
trade is focusing on trade and external economies of scale. Trade liberalization fosters
firm selection through competition. Opening up the economy to international trade
allows firms to sell their product in additional markets, which allows them to become
more efficient by spreading the fixed costs on a larger amount of output. Average
fixed costs paid by a firm become lower when demand is increasing. However, the
amount of labor that can be used in the production process is limited. Thus, some
firms have to exit the market and the surviving firms become more efficient asso-
ciated with higher markups. This version of the Krugman model does not feature
agglomeration. Krugman’s seminal paper from 1991 augmented the trade model by
allowing for labor mobility between countries. Obviously, going from autarky to free
trade opens up another channel through migration in this paper. As argued before,
firms may become more efficient when serving additional markets but the bottleneck
in the production is usually the availability of labor. Allowing for migration between
regions mitigates this bottleneck to some extend as regions may extend their labor
force through migration. The source country will suffer from soaring labor shortage
due to emigration but the destination will be better off.
The model may result in regional agglomeration of manufacturing when trade costs
are sufficiently low and workers can choose the location of their permanent residency.
Firstly, we extend the original Krugman (1979b) model by internal migration and
agglomeration. Secondly, we will discuss a simplified version of the Krugman (1991)
framework as proposed in the textbook by van Marrewijk (2012). Several assumptions
in this textbook model made the original model more tractable without changing the
2.1. INTERNAL MIGRATION AND ECONOMIES OF SCALE EFFECTS 25
main insights of the model.
2.1 Internal Migration and economies of scale effects
We build our extension on the model by Krugman (1979b). See Feenstra (2004) for
more details.
There are two symmetric countries in the model, which are identical in every respect
except of the type of goods they produce. Goods are of the same sort but differ with
respect to some undefined properties. Suppose we are focusing on the production
of manufacturing goods. All goods under consideration sort into the class of man-
ufacturing goods but every entity in the model produces its own distinct variety of
manufacturing goods. None of the firms at Home or Foreign produces exactly the
same type of manufacturing good compared to the varieties produced by all other
firms at Home and Foreign. This product differentiation is the only source of het-
erogeneity across countries, everything else is exactly the same. Both countries are
equipped with the same amount of labor available for production. Workers receive a
wage w for supplying their time to the firm and this wage is spent for consumption
of the goods produced in the economy.
Moreover, all consumers purchase all goods available in the economy to maximize
utility, which is an important assumption when we talk about the role of transporta-
tion costs for agglomeration.
There are i = 1, ..., N varieties under autarky, which is the situation when we do not
allow for free trade. Index i identifies a particular variety within the total number of N
varieties available for consumption. The number of varieties N is determined endoge-
nously using the full employment condition discussed later on. Thus, the number of
available varieties is not constant. Exogenous is only the number of consumers, which
is equal to the number of workers L. The wage w is spent for consumption ci for va-
riety i. Notice that all consumers are homogeneous and therefore realize the same
individual consumption level ci . Consumers are homogeneous and they have identi-
cal preferences depending on the price charged by each producer. This assumption
allows aggregating the individual consumption decision over all consumers according
to
N
U= ∑ ν ( c i ). (2.1)
i =1
Equation 2.1 is aggregate utility obtained by summing individual utility obtained for
consumption of each individual variety over all varieties available in the market.
The function ν(ci ) is utility from consumption of individual variety ci and we are
assuming that total utility associated with the consumption of all varieties can be
computed by summing individual utility.
We are assuming that ν0 > 0, ν00 < 0. The first derivative of the preference function
with respect to ci is positive and the second derivative is negative. More consumption
of the same variety always raises utility obtained from the consumption of this variety
but the growth rates are declining in the level of consumption. Intuitively, this means
that there is some saturation in consumption. Consumers always prefer consuming
less of each individual variety in favor of more variety in total. Consumers benefit
from reducing the consumption levels for all individual varieties c already consumed
in exchange of adding new varieties to their consumption portfolio. A higher N im-
plies more variety as the number of varieties is increasing and all consumers consume
all varieties available. It is better to spread income on more varieties instead of in-
creasing ci for individual varieties.
The optimal amount of consumption ci can be solved by a maximization problem that

takes into account the budget constraint
N
w= ∑ pi ci . (2.2)
i =1
The budget constraint depends on the number of varieties and the wage, which are
both endogenous variables. Total expenditure must not exceed total income.
We solve the utility maximization problem by
N
L = U + λ ( w − ∑ pi ci ) (2.3)
i =1
Utility U shall be maximized by setting ci under restriction 2.2. The first order con-
dition can be solved for one distinct variety i. The solution for this one particular
variety represents the solutions for all other varieties. We obtain
ν0 (ci ) = λpi , i = 1, ..., N (2.4)
as solution for the first order condition for variety i, which represents the first order
conditions for all N varieties available in the market. The total differential of 2.4
together with the budget constrain is
dci λ
ν00 dci = dpi λ ⇒ = 00 < 0 . (2.5)
dpi ν
The left hand of 2.5 is the change in utility of i when prices change by dpi . We can
dci
rearrange to dpi , which gives us the change in utility for a change in the price of
variety i. Utility is decreasing in a positive change of its price (dpi > 0 ⇒ dci < 0, as
dci dci pi
dpi < 0).1 We can substitute by the elasticity of substitution µi = − dp i ci
.
Plugging the left hand side of 2.5 and 2.4 into the elasticity yields
ν0

dc p
µi = − i i = − >0 . (2.6)
dpi ci ci ν00
We are assuming that dµi /dci < 0. The elasticity of substitution is decreasing in con-
sumption, which is intuitive as consumers’ sensibility against price changes should
be lower when the consumption level is high but this result is not that obvious when
we look into the more common economic models. More common preference func-
1 λ is positive and ν00 is negative.

tions don’t feature this property as the elasticity in Constant Elasticity of Substitution
(CES) or Cobb-Douglas functions is constant. However, we need this property for
our selection results. Trade liberalization will have an effect on the elasticity through
changing the individual consumption levels and this changes the competition in the
market associated with different markups and firm exit due to economies of scale.
The intuition behind this result is straightforward. Opening up a country to free
trade extends the availability of varieties. Foreign suppliers provide their goods in
addition to the goods already supplied by domestic producers. Due to love of variety,
consumers will prefer reducing the individual demand for each variety in order to
spread their budget to more varieties, thereby affecting the elasticity of demand. The
elasticity itself is the crucial determinant of the firms’ markups, which is the wedge
firms can drive between unit costs and unit prices. The less sensitive consumers are
to changes in prices, the higher the markup firms can put through goods prices.
Production. Labor supply generates some income w for the workers who spend the
money for purchasing the goods produced in the economy. Production of good i is
denoted by yi . Labor is the sole input factor in the production. The amount of labor
used for producing good i is denoted by Li . We are using the parameter a as inverse
measure for labor productivity. To increase output by one unit (dyi = 1), producers
must increase the number of workers by a persons. Variable production costs are
therefore a × w. Firms must pay additional fixed costs, which are independent of
the amount of goods already produced. These costs are again paid in terms of wages.
Firms must hire α workers to pay the fixed costs. This assumption is a simplification to
keep the model as tractable as possible. Production of yi units of the variety requires
a total amount of labor
Li = α + ayi . (2.7)
Production costs amount to labor costs Li w. Figure 2.4 illustrates the economies of
scale in the model due to fixed costs. Firms become more efficient when production
is increasing as fixed costs are spread onto a higher number of output. Regardless
the level of production, firms always have to pay the α workers as compensation for
fixed costs. Thus, the average costs (DKi ) are decreasing in yi , whereas marginal costs
remain constant.
2.4: Average and marginal cost curve
𝐷𝐾𝑖 , 𝐺𝐾𝑖
𝐷𝐾𝑖
𝑎𝑤 𝐺𝐾𝑖
𝑦𝑖
Average costs become infinite when yi → 0 and approach the marginal costs when
yi → ∞. Average costs are calculated as DKi = (wLi )/yi so that
DKi = (wLi )/yi = (wα)/yi + wa . (2.8)
We can calculate the marginal costs as
GKi = wa, (2.9)
which is the first derivative of total costs Li w with respect to yi .

Firms price-setting behavior. Firms have some command over prices as competi-
tion is limited. Intuitively, this is because firms produce varieties of a good. For
consumers the different varieties are not perfect substitutes, which enables firms to
manipulate prices by changing outputs. A higher price will reduce the demand by
consumers but the demand effects by changes in prices are not as strong as it is the
case when goods are homogeneous. Producers can charge different prices through
adjusting production in a profit maximizing way, which takes the reaction in ci into
consideration.
Profits are revenue minus costs
πi = pi yi − wLi . (2.10)
we can setup the profit maximization problem by solving ∂πi /∂yi = 0. Firms continue
changing outputs and prices until this condition is fulfilled.
Producers should target a consumer demand where marginal revenue is equal to

marginal costs:
∂pi ∂L ∂p
pi + yi − w i = pi + yi i − wa = 0 . (2.11)
∂yi ∂yi ∂yi
Marginal revenue on the left of this equation tells us by how much revenue changes
for the next unit of output. As long as this additional revenue is more than the
additional costs incurred by this last unit, profits are increasing. The firm should stop
extending output when marginal costs exceed marginal revenues.
We obtain

1
MR = MC → pi 1− = wa (2.12)
µi

pi µi
= a . (2.13)
w µi − 1
dp ci dpi
The first equation stems from − pi µ1 = pi dc i = dyi yi . All workers consume variety
i i pi
i so that total consumption of variety i is equal to L × ci . All goods consumed in the

economy must be produced and there is only one particular firm that produces variety
i. This also implies that dyi = Li dci holds. Changes in the production of i match
∂p dpi
changes in the consumption of the respective variety, which translates into ∂y i ≈ dci .
i
We need to check the second derivative of the first order condition to see if we have a
maximum. With respect to yi it has to be negative.
However, there is another factor that limits the firms control over prices. New firms
can enter the market and they will do so as long as there are positive profits. New
entrants increase competition as consumers spread their income onto more varieties,
thereby mitigating economies of scale effects discussed earlier. In equilibrium, firms
continue entering the market until the free entry condition is fulfilled:
P = DK → pi = (wα)/yi + wa (2.14)

pi α
= +a (2.15)
w Lci
We can rearrange as shown in the last equation because yi = Lci must hold. The
average price of varieties becomes equal to the average costs. The only profit for the
firm is its markup. Firms will charge prices with constant markup over unit costs.
Based on these equilibrium conditions, we can solve the remaining variables of the
model using the labor market clearing condition. For a given labor supply we get an
endogenous solution for the number of firms N. This means that
N N
L= ∑ Li = ∑ (α + ayi ) = N (α + aLci ) (2.16)
i =1 i =1
must be fulfilled. All workers are used in the production of one of the N firms as
unemployment is zero.
The number of firms can be easily computed as we already know the consumption
behavior of workers in the model.
1
N= . (2.17)
[(α/L) + aci ]
General equilibrium. To solve the model, we need a solution for ci , which is the
equilibrium consumption of one consumer. This consumption level gives us the
amount of production needed to serve the consumers’ demand. Individual consump-
tion is obtained from interacting the free entry condition, which states that new firms
stop entering once prices are equal to average cost, and the profit maximization rule
we derived above applies:

α
ZZ: pi /w = +a , (2.18)
Lc
i
µi
PP: pi /w = a . (2.19)
µi − 1
We get a system of two equations with one unknown ci . Either we equate both and
solve for ci or we go for a graphical solution. The latter is the more elegant ap-
proach that allows analyzing the two equilibrium conditions separately. The solution
is unique when both curves intersect only once.
Figure 2.5 depicts the two equilibrium conditions.
The PP-condition is a graph that is increasing in ci . The ZZ-condition is decreasing in

ci .
To see that, we need to compute the first derivative with respect to ci . we can compute
∂{ pi /w}
∂µi = a (µ(iµ−−1)−
1)2
µi
< 0. At this point we need the assumption about the elasticity of
i
demand. µi is decreasing in ci by assumption.
Properties of the PP-condition

Increasing consumption of variety i is associated with more saturation. Con-
Figure 2.5: General Equilibrium in the Krugman Model
sumers become less sensitive against further changes in the goods prices. thus,
p/w is increasing in ci as ci ⇑ → µi ⇓ → pi /w ⇑ holds.
These properties of the two conditions guarantee the existence of an unique equilib-
rium. Figure 2.5 shows the equilibrium under autarky at cia . Our solution for cia pins
down the inverse of the real wage ( p/w) a on the Y-axis.
The endogenous number of firms under autarky is then determined by
1
Na = . (2.20)
[(α/L) + acia ]
2.1.1 Trade liberalization
The economy opens up from autarky to free trade. Thus, consumers have more variety
as domestic and foreign producers start competing for consumers’ demand. Without
further adjustments, the number of varieties available doubles and the demand for
each individual variety is decreasing. We are assuming symmetry between the two
countries: Foreign and Home are twins except of the type of goods both are produc-
ing. Labor endowments are the same, the production processes are the same and all
endogeneous variables are exactly the same in both countries.
Consumers benefit from trade liberalization through two channels. Producers become
more efficient due to the additional demand that translates into additional produc-
tion. Firms produce at lower average costs. Moreover, consumers benefit from the
additional variety. Individual consumption will be lower as consumers tend to spread
their income on more varieties. Remember that the second derivative of the preference
function is negative due to the saturation effect described before. The utility losses as-
sociated with decreasing ci are relatively low compared to the additional utility from
consumption of the newly added foreign varieties.
These two channels give rise to firm selection in the model. Firm i produces yi =
Lci + L∗ ci∗ = 2Lci following from the symmetry assumption. Spreading the fixed
costs on a larger amount of output decreases p/w as firms become more efficient.
We can show this result in our graphical solution.
The two conditions 2.18 and 2.19 are affected by trade liberalization in two ways.
Firstly, the ZZ-condition is affected directly. We substituted yi by Lci as the number of
consumers doubles when going from autarky to free trade: L + L∗ = 2L. This shifts
the ZZ-condition down. Suppose that ci remains constant at the autarky level cia .
Doubling the number of consumers is associated with a lower value on the right hand
of 2.18. At cia the inverse of the real wage p/w would have to jump down in order
to maintain the equilibrium again. The variables p/w and ci will adjust in several
Figure 2.6: Trade liberalization in the Krugman (1979b) model
𝑝𝑝
𝑤𝑤
𝑍𝑍𝑍𝑍 𝑃𝑃𝑃𝑃
𝑍𝑍𝑍𝑍´
(𝑝𝑝/𝑤𝑤)𝑎𝑎
(𝑝𝑝/𝑤𝑤)𝑤𝑤
𝑐𝑐𝑖𝑖𝑖𝑖 𝑐𝑐𝑖𝑖𝑖𝑖 𝑐𝑐𝑖𝑖
rounds until the equilibrium is satisfied again.
The PP-condition does not shift because its variables are not directly affected by trade
liberalization. These adjustments are illustrated in 2.6 where ci is decreasing due to
trade liberalization. There are adjustments in p/w and ci until the new equilibrium is
reached where the ZZ-curve intersects the old PP-curve. There are gains from trade as
consumers reach a higher level of welfare due to reduced ci and a higher total number
of varieties consumed. Moreover, the real wage is increasing as p/w is decreasing.
Do all individuals benefit from trade liberalization? No, we know that the number
of firms is determined by the conditions 2.17 and 2.20. Comparing Na (number of
firms in autarky) and Nw (number of firms under free trade) reveals that there is firm
selection in the model. From 2.20 we see that Na > Nw and Na∗ > Nw∗ must hold
because
1 1
Nw = < Na Nw∗ = ∗ ] < Na∗ . (2.21)
[(α/L) + a2ciw ] [(α/L∗ ) + a2ciw
Consumers benefit as markup prices go down (which is the same as an increase in the
real wage) but some firms will have to exit the market due to higher competition from
abroad. However, we are unable to identify the individual firms that have to exit the
market. One may assume that there is a random exit shock that destroys firms in each
period. Not replacing these exiting firms by newly entering firms would be enough
to get firm selection in this model. Notice that the model is static, which means that
the adjustment process (including the process of firm selection) is invisible. We are
comparing two long run equilibria without modeling the adjustments associated with
the transition from one to the other equilibrium. There are adjustments in between
the two long run equilibria but we put them into a black box instead of explicitly
modeling them. We address this issue in the next model based upon the Krugman
(1991) paper.
2.1.2 Agglomeration
Our first goal is extending the model by agglomeration. The World Bank argued that
urbanization is a potential way out of the middle income trap due to economies of
scale. we are introducing agglomeration by assuming that firms are located in the
core of each country, whereas workers are living in the rural area surrounding the
core region. Only the owners of the firms can afford the higher living costs in the city,
whereas workers have to commute from the rural to the urban area every day. Work-
ers accept commuting only when firms somehow cover the higher commuting costs.
We could model this as a wage premium. However, this would complicate the algebra
as consumption of transport would enter the utility maximization problem. Workers
should not be using the additional income for purchasing the consumption goods.
We circumvent this problem by assuming that firms directly pay for the commuting
costs, which does not affect the main outcomes of the model.
Figure 2.7 illustrate this assumption. Firms are located in the middle of each country
and workers are located in villages surrounding the firms.
Figure 2.7: Separation equilibrium
The assumption that workers are located on a circle with radius r is another sim-
plifying assumption that makes the extension more tractable. The distance between
each worker and the firm is the same for all workers employed by a firm. Thus,
the commuting costs are also identical across workers. We are assuming that these
commuting costs are paid by the firm in terms of wages.
The number of workers needed to produce yi becomes
Li
Li = α + ayi + ∑ d(ij) = α + ayi + Li d(r ) . (2.22)
j =1
The assumption that the distance between the workers’ residency and the location of
the firm is identical for all Li workers allows us to simplify the equation as shown on
the right of equation 2.22. The firm-worker specific commuting costs were substituted
by common commuting costs d(r ), where r is the radius of the circle around the firm.
We can rearrange this labor demand equation to
α + ayi
Li = (2.23)
1 − c (r )
The two equilibrium conditions become

α 1
ZZ: pi /w = +a , (2.24)
Lci 1 − c (r )

µi 1
PP: pi /w = a . (2.25)
µi − 1 1 − c (r )
At first glimpse, it seems that we are back in the standard model as the term 1 − c(r )
drops out when intersecting both equations. Indeed, the consumption level is exactly
the same as in the standard model. However, the real wage is different. We have
to insert the solution for ci into equation 2.24 or 2.25 for solving the inverse of the
real wage. The additional commuting costs are added to the prices charged by the
respective firm. These commuting frictions reduce the real wage of the workers as
they are passed on to consumers, which are the workers themselves.
The additional commuting costs are also relevant for the equilibrium number of firms
in the economy. The labor market clearing condition becomes
1 − c (r )
Na = . (2.26)
[(α/L) + acia ]
Each individual firm becomes larger as some additional workers are hired as compen-
sation for the commuting costs. However, the total number of consumers is the same
as in the standard model, individual consumption of one worker remains constant
and the real wage is declining. As the consumption level of each individual variety
is unaffected, the only possible adjustment is through the number of firms, which is
decreasing due to the additional costs. Thus, despite consuming the same amount of
each variety, consumers reach a lower level of welfare due to the reduced amount of
varieties. this result makes perfectly sense as real wages are lower with commuting
frictions. Notice that we would obtain the same result when modeling these costs as
additional fixed costs.
Now suppose we have the following scenario: Opening up the economy from autarky
2.2. AGGLOMERATION AND MIGRATION 39
to free trade is associated with higher real wages. This result is independent of the
newly introduced agglomeration frictions.
However, suppose that we have urbanization and trade liberalization at the same
time. It can be shown that the negative effects through firm selection could be mit-
igated when the government manages to foster urbanization simultaneous to trade
liberalization.
Also suppose that the government supports trade liberalization by fostering urban-
ization. For simplicity, we are assuming that all workers move at arm’s length to the
firms and change their place of residency to the core of the country. The commuting
costs become zero.
We obtain
1 1 − c (r )
Nwa = > Nwn = . (2.27)
[(α/L) + a2ciw ] [(α/L) + a2ciw ]
Notice that the equations that determine Nwa (number of firms under free trade with
agglomeration) and Nwn (number of firms under free trade without agglomeration)
are exactly the same except of their nominators. Thus, agglomeration spurs growth a
the extensive margin by reducing the negative growth effects associated with exit of
firms due to trade liberalization.
The individual consumption is the same for both scenarios but we have more firms
with agglomeration, so that total production is higher.
2.2 Agglomeration and migration
The model discussed in this chapter is based upon the Krugman (1991) model of ag-
glomeration. However, we are using the simplified version described in the textbook
by van Marrewijk (2012).
There are two regions producing agricultural and manufacturing goods by input of
labor. Labor supply in both regions together is exogenous and given by L. Krugman
(1991) assumes that the supply of labor is unity. Both regions share the total amount
of labor as workers can move freely between the two regions and between industries.
A fraction of δm workers are employed in the manufacturing sector and a fraction
1 − δm workers are employed in agricultural.
Distribution of workers. Workers in the agricultural sector are immobile across re-
gions. A constant share of φ1 workers lives in region 1 and a constant share of φ2
agricultural-workers is located in region 2.
Manufacturing workers are mobile across regions. A share λ1 is living in region

1 and a share λ2 is living in region 2. These two shares are flexible and depend
on the workers’ choice of residency, which is again driven by wages. Workers tend
to migrate to the location where wages are higher. Thus, there is a tendency that all
manufacturing workers move into one of the two regions leaving the other region with
agricultural production only. Corner solutions imply either λ1 = 0 or λ2 = 0. Both
regions can produce manufacturing goods only when there is factor price equalization
across regions. We start with the factor equalization scenario in which both regions
produce agricultural and manufacturing goods. The labor force in the four sectors
are:
• Labor force in the agricultural sector of region 1: L × (1 − δm ) × φ1
• Labor force in the agricultural sector of region 2: L × (1 − δm ) × φ2
• Labor force in the manufacturing sector of region 1: L × (δm ) × λ1
• Labor force in the manufacturing sector of region 2: L × (δm ) × λ2
Summing labor supply over all sectors gives us the total world supply of labor as
φ1 + φ2 = 1 and λ1 + λ2 = 1.
Income of agricultural workers. Worker productivity in the agricultural sector is

unity in both regions and prices of agricultural goods are the numeraire in this model.
These two assumptions imply that wages paid by agricultural firms are equal to unity
in both regions. The total income of agricultural workers in region 1 can be computed
as
Ia1 = L × (1 − δm ) × φ1 (2.28)
A similar equation can be derived for region 2.
Income of manufacturing workers. For the moment we take the manufacturing la-
bor shares λ as given. Notice that these parameters will change due to agglomeration.
Or, to be more precise, agglomeration is driven by changes in λ.
Workers in the manufacturing sector of region 1 receive a wage W1 and workers in

the manufacturing sector of region 2 get a wage W2 . Total income of workers in
manufacturing of region 1 is therefore
Im1 = L × δm × λ1 × W1 (2.29)
Total income of manufacturing workers in region 2 is
Im2 = L × δm × λ2 × W2 (2.30)
Computing W1 and W2 is a bit more complex. Workers get paid their marginal prod-
uct of labor but this depends on productivity and prices in manufacturing. Moreover,
we are assuming that those two goods are shipped by paying some iceberg transporta-
tion costs. We therefore have to care about demand and supply in order to derive the
equilibrium prices.
As a first step, we can solve the utility maximization problem to learn more about
the interaction between prices and demand. Consumers have CES preferences for
consuming varieties produced in both regions. This is similar to the assumptions

introduced in the previous section. Firms produce varieties of a differentiated manu-
facturing good. Again, we are focusing on one representative variety i and compute
the first order condition for this one particular good. This solution can be generalized
to all varieties available in the market.
The solution to the Lagrangian reads
c j = p− e e −1
j [P δm I ] (2.31)
Demand for variety j depends on the price charged for this particular variety. The
higher the price of variety j, the lower the demand for this one particular good as
consumers substitute it by other varieties available in the market. The reaction itself
depends on the demand elasticity e. However, demand also depends on the aggregate
variables P and I. The latter is straightforward. The higher the total income available
for purchasing all goods supplied in the market, the higher is the demand for variety
j. In fact, I raises demand for all goods including variety j.
The aggregate price P is the sum of all individual manufacturing prices
!1/(1−e)
N
P= ∑ p1i −e (2.32)
i =1
supplied in the respective region, where j = 1, ..., N.
Notice that N includes varieties produced by firms in both regions. Due to the sym-
metry assumption, we can split the sum into two different parts
!1/(1−e) !1/(1−e)
N1 N̄2
P= ∑ p1i −e + ∑ p1i −e (2.33)
i =1 i = N2l
where N2l stands for the first variety produced in region 2: We are ordering the N1
variety producers at the left and the N̄2 − N2l = N2 variety producers at the right. N2l
is the firm that separates region 1 and region 2 variety producers.
This is important as there is one difference between price setting behavior of region
1 and region 2 producers. Firms have to pay iceberg transportation costs and these
extra costs are passed on to the consumers by setting higher prices at the destination.
We can already simplify this equation by getting rid of the sums. Notice that all N1
firms charge the same price at region 1 and 2 and all N2 = N̄2 − N2l firms charge the
same price at region 1 and 2. Similar to what we have done in the previous section,
we can solve the number of firms N1 and N2 using the full employment condition for
manufacturing.
Firms need to hire

li = f + mxi (2.34)
workers for producing xi units of variety i: f workers are hired for paying the fixed
costs and one unit of the variety i requires m additional workers. The total number
of workers available for producing the N1 goods is Lδm λ1 . The full employment
condition therefore reads
N1
Lδm λ1 = ∑ Li = N1 Li (2.35)
i =1
We can solve this for

Lδm λ1
N1 = . (2.36)
f + mxi
N2 can be derived similarly. The next step is solving the profit maximization problem
you are familiar with from the last chapter. However, this time we are using the
CES-preferences instead of the more general formulation, which has some important
implications.
In general, the profit of variety producer i reads
π = px − W ( f + mx ) (2.37)
Revenue is px and costs are wage times the number of workers employed by the firm
W ( f + mx ). The first derivative with respect to x reads
∂π
= p(1 − 1/e) − mW , (2.38)
∂x
where e is the elasticity of demand. Profit maximization implies that the first order
condition is equal to 0, which allows us solving
p(1 − 1/e) = mW (2.39)
This equation is the pricing rule of firms under monopolistic competition. We can use
this pricing rule and derive the equilibrium production by
px − W ( f + mx ) = 0 (2.40)
This equation is essentially the free entry condition we have discussed earlier in this
lecture. We are again interacting the PP-condition and the ZZ-condition by inserting
the pricing rule into the free entry condition

e
mW x = W ( f + mx ) . (2.41)
e−1
Summing terms gives us

e
− 1 mWx = W f (2.42)
e−1
where W drops out and x can be solved as
f ( e − 1)
x= (2.43)
m
We can use this equation to simplify the demand for labor of firm i as
f ( e − 1)
li = f + mx = f + m = fe (2.44)
m
Let us go back to the number of varieties produced in region 1. We stopped at equa-

tion 2.36, where we can use the knowledge about the xi in order to derive
N1 = λ1 δm / f e (2.45)
Similarly, we get the number of firms at region 2 by solving
N2 = λ2 δm / f e (2.46)
Recall that both regions are symmetric, which explains why most parameters are the
same in both regions. The only difference is the distribution of workers and that
explains why the number of firms may be different as well. Again, the λ parameters
are the crucial parameters that determine agglomeration in the model, we come to
that later on.
We can use this knowledge to shed some more light on the aggregate price level.
Recall that the aggregate price is
!1/(1−e) !1/(1−e)
N1 N̄2
= ∑ p1i −e + ∑ p1i −e (2.47)
i =1 i = N2l
Our knowledge about the number of firms allows us to get rid of the two sums as
1/(1−e) 1/(1−e)
P1 = N1 p111−e + N2 p112−e , (2.48)
where price p11 is the price charged by firms from region 1 in region 1 and price p12
is the price charged by firms from region 2 in region 1. We can substitute for N1
and N2 by the equations derived above. Moreover, we now know the pricing rule of
firms. Firms in region s charge prices with markup over wages prs = mWs Trs /ρ in
region r. Variable T denotes iceberg transportation costs, which are 1 when indexes
r = s. Selling the good within the respective firm’s home market incurs zero iceberg
transportation costs.
We therefore get the following solution for the aggregate price level in region 1
1/(1−e) 1/(1−e)
P1 = (λ1 δm / f e) p111−e + (λ2 δm / f e) p112−e , (2.49)
Using the pricing rule gives us
1/(1−e) 1/(1−e)
P1 = (λ1 δm / f e) (m(Ws Trs /ρ))1−e + (λ2 δm / f e) (m(Ws Trs /ρ))1−e .
(2.50)
This equation can be simplified to
1/(1−e) 1/(1−e) 1/(1−e)

m δm 1− e

1− e
P1 = (λ1 ) (W1 T11 ) + (λ2 ) (W2 T12 ) .
ρ fe
(2.51)
Notice that we are summing prices charged by firms from region s (second index) in
region r (first index) and that wages paid in the source country (s) are relevant for
the prices in the destination (r). A further simplification is using the sum notation as
follows
1/(1−e) " 2 1/(1−e)
#
m δm
P1 =
ρ fe ∑ (λs ) (Ws T1s )1−e . (2.52)
s =1
The respective price level in region 2 can be derived similarly as
1/(1−e) " 2 1/(1−e)

#
m δm
P2 =
ρ fe ∑ (λs ) (Ws T2s )1−e . (2.53)
s =1
We still need to solve for the wage rate in both regions.
Wages and prices are closely related through the profit maximization problem. Firms
will set production costs as the lower bound for the price. Prices itself determine
demand. Thus, wages must be such that supply equals demand under consideration
of the pricing rule. Total supply is already known as (e − 1) f /m. Total demand can
be computed as the sum of demand from region 1 and region 2:
− e e −1 − e e −1
c1si + c2si = p1si [ P1 δm I1 ] + p2si [ P2 δm I2 ] (2.54)
Be careful with the subscripts. We are looking at production in region s, which is

consumed in the region 1 and 2. The relevant wages for consumers in region 1 and
2 are wages in region s as these wages are the basis for producers to calculate goods
prices charged in the destination country.
The prices charged by firms from region s in location 1 and 2 are
p1si = mWs T1s /ρ (2.55)
p2si = mWs T2s /ρ (2.56)
We can plug these prices into our aggregate demand functions, equate aggregate
demand with aggregate supply for variety i and solve the solution of this equation
for " !1/e
1/e # R
δm
Ws = ρm−ρ
( e − 1) f ∑ Ir Trs1−e Pre−1 (2.57)
r =1
We simplify the equation by assuming m = ρ and f = δ/e, which gives us
!1/e
R
Ws = ∑ Ir Tsr1−e Pre−1 (2.58)
r =1
Total income in country 1. Total income in one region is the sum of income received
by agricultural and manufacturing workers together
I1 = Ia1 + Im1 = L × (1 − δm ) × φ1 + L × δm × λ1 × W1 (2.59)

Accordingly, total income in region 2 is
I2 = Ia2 + Im2 = L × (1 − δm ) × φ2 + L × δm × λ2 × W2 (2.60)
Total income is spent for purchases of food and manufacturing goods.
Migration dynamics: Migration (agglomeration) is assumed to be
dλ1
= η (w1 − w̄) (2.61)
λ1
where w̄ is the weighted average wage in both regions. The more people migrate,
the higher the share of manufacturing workers in the destination country of migra-
tion. The parameter η is the adjustment speed. The higher this parameter, the faster
migrants move from one region to the other and the faster the share λ are changing.
Notice that the adjustment speed also depends on the wage difference itself. Mi-
gration speed is declining when the wage rate in one region approaches the average
wage.
2.2.1 Equilibrium
The model can be solved by interacting the following equilibrium conditions:
1. Aggregate prices in both regions
h i1/(1−e)
P1 = λ1 W11−e + λ2 { T 1− e
}W21−e (2.62)
h i1/(1−e)
P2 = λ1 { T 1−e }W11−e + λ2 W21−e (2.63)
2. Income in both countries
I1 = λ1 δW1 + (1 − δ)/2 (2.64)
I2 = λ2 δW2 + (1 − δ)/2 (2.65)
3. Wages obtained from the aggregate demand = aggregate supply condition
h i1/e
W1 = I1 P1e−1 + I2 { T 1−e } P2e−1 (2.66)
h i1/e
W2 = I1 { T 1−e } P1e−1 + I2 P2e−1 (2.67)
Notice that we have a system of equation with six equations for six endogenous
variables, which is good news as stable equilibria should exist. However, it be-
comes problematic once we allow for migration as λ becomes endogenous as well.
The solution adopted from the textbook by van Marrewijk (2012) is modeling λ as
functions depending on the wage in the respective region in relation to the average
wage in both regions. The share of workers in the respective region is growing
when the wage is above average and decreasing when the wage is below average,
which is adding two more endogenous variables and two more equations. We
then have 8 equations and 8 variables to solve. However, the model is too complex
for solving it analytically. Instead, we are focusing on the stable equilibria.
The three most important model outcomes:
• Spreading Equilibrium
• Agglomeration in region 1
• Agglomeration in region 2
Spreading is the scenario when both manufacturing labor shares are greater than
zero. This scenario is feasible when when both regions pay the same wage rate.
Whenever wages are different in both regions, workers tend to migrate from the
low-wage region to the high-wage region and all consumers purchase the goods
produced in the low-wage region.
Agglomeration in region 1 occurs when real wages in region 1 are higher than
real wages in region 2. All workers would have an incentive to move to region 1
and λ2 becomes zero.
Agglomeration in region 2 is when real wages in region 2 are higher than real
wages in region 1. Again, all workers would want to move to region 2 and λ1
becomes zero.
The following discussion builds on a simulation of the model with the following
parameters:
• The share of agricultural/manufacturing sector: δ = 0.4
• The markup: ρ = 0.8 (which implies that e = 5)
• Iceberg Transportation costs: T = 1.7
• The tolerance value for the solution: σ = 0.0001
• The labor force: L = 1

These parameters can be used to characterize the 6 equilibrium conditions. Take any
value for λ1 between 0 and 1 and notice that λ2 = 1 − λ1 . The 6 equations can be used
to identify the remaining 6 endogenous variables by trial and error, which means that
we have to change the endogenous variables until all 6 variables solve for (almost)
0. This rather inelegant way of solving the model implies that the equations will not
solve for 0 exactly. we would need infinite changes in the endogenous variables until
we find a solution that solves all equations exactly. This is why we need the tolerance
as stopping criteria for the alteration process. The first best solution that meets the
tolerance criteria will be taken as solution.
Let us start by the solution in point C. The share λ is equal to 0.5 at this point and
the wages are exactly the same in both countries. This is why spreading is possible in
this scenario. Can we verify this solution? Yes, we can by trying out W1 = W2 = 1,
which gives us the solution
• Price 1: P1 = [0.5 × 1(1−5) + 0.5 × 1.7(1−5) × 1(1−5) ](1/−4) = 1.156
• Price 2: P2 = [0.5 × 1.7(1−5) × 1(1−5) + 0.5 × 1(1−5) ](1/−4) = 1.156
• Income 1: I1 = 0.5 × 0.4 × 1 + (1 − 0.4)/2 = 0.5
• Income 2: I2 = 0.5 × 0.4 × 1 + (1 − 0.4)/2 = 0.5
The last two equations for the wage rate allow assessing the solutions. If the educated
guess was right, the wage equations should deliver the right values for our educated
guesses, the solutions should be equal to the educated guessed wage rates, which is
W1 = W2 = 1 in our case.
We obtain:
• W1 = [0.5 × 1.156(5−1) + 0.5 × 1.7(1−5) × 1.156(5−1) ](1/5) = 1
• W2 = [0.5 × 1.7(1−5) × 1.156(5−1) + 0.5 × 1.156(5−1) ](1/5) = 1

Our educated guess was right an the solution is exactly 0. Thus, we don’t need the
tolerance for this special case. Points A and E are also stable equilibria, which can be
computed easily as both λ-Parameters take extreme values of either 0 or 1. All other
λ-combinations are a bit more challenging. Instead of computing them, we discuss
these other points in the following graph.
Figure 2.8: Spreading and agglomeration

Source: van Marrewijk (2012)
There are five equilibria in this graph. Three of these equilibria are stable and we
discussed these three points (A,C,E) already. Points B and D are unstable equilibria.
If wages are exactly equal to 1 in both countries and λ takes exactly the value pinned
down in point B or C, we are in equilibrium but a minor deviation from these values
moves the economy away to one of the stable equilibria.
Let us take out point F. The relative real wage in region 1 would be higher, which gives
an incentive for workers from 2 to move into region 1. The share of manufacturing
workers in region 1 becomes bigger according to our law of motion defined in equa-
tion 2.61. This law of motion changes λ1 until the wage in country 1 is equal to the
average wage, which is the case when we are in the spreading scenario (W1 = W2 = 1).
This is why point C is stable and point F or B are not stable equilibria. We ca do the
same for all points between D and C. Here we find that the wage in region 1 is be-
low the average wage associated with declining λ1 until the spreading equilibrium is
reached again.
What happens if we are to the left of B or to the right of D? We would be moving

into the agglomeration equilibria, which is why points A and E are stable equilibria
as well.
Figure 2.9: Low transportation costs

Now, one may ask what the role of transportation costs is in this model. For very low
transportation costs, the agglomeration scenario becomes likely. The following graph
demonstrates this by setting T = 1.3.
Depending on the initial λ1 , the economy will move into agglomeration to country 1
or country 2. If λ1 is smaller 0.5, we will end up in the agglomeration to country 2
scenario. The manufacturing share in country 2 becomes zero as all workers migrate
to country 2. If the initial share is greater 0.5, all workers move into country 1 and the
share of of manufacturing workers in country 2 becomes 0. What happens if trans-
portation costs become large? In this scenario, spreading becomes the only stable
solution. Whenever the share of manufacturing workers deviates from 0.5, the dy-
namic forces will lead back to the spreading equilibrium as depicted in the following
graph.
Figure 2.10: Dynamics with high transport costs

Chapter 3
Economic Growth
Over the past two centuries some countries have exhibited sustained economic growth
resulting in high levels of GDP per capita. But many countries lag behind and con-
tinue to perform badly. Figure 3.1 illustrates the situation.1 Each line depicts the
cross-country distribution of per capita GDP in a different year. The blue line refers
to the earliest year, i.e. 1960.2 At this point in time, we observe a relatively low
degree of cross-country income inequality. Indeed, all sampled countries featured a
per capita GDP that lies close to the sample mean which is itself rather low. Most
countries were therefore relatively poor. However, during the most recent year cov-
ered by our data set - represented by the orange kernel density plot - we observe a
more dispersed distribution of per capita GDP with a much higher mean compared
to earlier periods. Hence, whereas many countries with low per capita GDP in 1960
grew through time this was not true for all countries. Rather a fraction of countries
remained at their low initial income level or even regressed. As a result, the mass
of countries at the bottom of the distribution decreased over time and the distribu-
tion shifted to the right. However, it is also evident that the shape of the distribution
1 The data used for this figure, as well as for the entire empirical analysis presented in this chapter are
taken from the "Penn World Table" version 8.1. This database is freely available online, and contains
many macroeconomic indicators which are harmonized to allow international comparison.
2 Notice that, although data exist for earlier years than 1960, their quality is questionable and it is also
not guaranteed that those data are comparable across countries.
55
56 CHAPTER 3. ECONOMIC GROWTH
changed significantly between 1960 and 1990 followed by modest changes over the
1990 to 2011 period. One potential reason for this pattern may be that the pace of eco-
nomic growth is faster at the beginning of the development phase, while development
subsequently slows down at later stages.
Figure 3.1: The per capita GDP distribution over time

.0002
Kernel Density Per Capita GDP
.00005 .0001
0 .00015
0 20000 40000 60000 80000

x
Per capita GDP (1960) Per capita GDP (1990)

Per capita GDP (2000) Per capita GDP (2011)
How can we explain these differences between less developed and more advanced
economies? And why do some economies remain in poverty without growing over
time? These questions are interrelated and can be answered using the results of one of
the earliest models of economic growth provided by Solow (1956). The Solow model
explains how investment in capital formation allows less developed countries to grow
relatively fast.
Growth rates then slow down as soaring depreciation rates associated with capital
formation kick in. A constant fraction of capital must be replaced each period. Capital
formation therefore slows down as more and more of the yearly investments go into
57
maintaining the capital stock at a given point in time.
The slowdown of economic growth in the data. The previous kernel density plots
hint towards a slowdown of economic growth over time. In order to verify this hy-
pothesis, we use the same data to estimate Barro-type growth regressions as described
in Barro (1996).3 The idea behind these regressions is rather simple. We compare the
initial level of a country’s GDP with its current level of GDP. One would expect that
low per capita GDP in 1960 is associated with much higher rate of economic growth
during subsequent years. In contrast, countries that reported a high level of liv-
ing standards in 1960 should have undergone a more moderate period of economic
growth between then and today. Using data from the Penn World Table we test this
relationship based upon the model
T
gi,t,t−1 = αyi,t−1 + Xi,t β + ei,t . (3.1)
The dependent variable is per capita GDP growth in country i between year t − 1 =
1960 and t = 2011. This is regressed upon the initial value of per capita GDP in
1960. Thus, the model postulates a relationship between the initial level of per capita
GDP and its growth rate. Based on the theoretical predictions we made earlier, one
would expect a negative relationship: the richer a country is in 1960, the slower it
will grow over time. However, economic growth may also depend on other factors,
such as the size of the population or human capital levels.4 The effect of those factors
is easily taken into account by including additional regressors in the hypothesized
model summarized in the vector X. We depart slightly from the established literature
by including the current population and the current level of human capital as control
variables instead of the levels from the year 1960. The coefficient α is reported in the
3 This section assumes some background knowledge of Introductory Econometrics and Statistics. Stu-
dents who are not already familiar with empirical economic methods can refer to the Appendix of
this Chapter, which provides the student with first working knowledge to understand the empirical
evidence discussed throughout the lecture notes.
4 By human capital we refer to the average years of schooling in the country combined with the average
rate of return to schooling.
first row of Table 3.1. Model (3.1) is estimated four times, each time using a different
set of control variables. The different outcomes are reported in columns (I) to (IV).
We are mostly interested in the coefficient α, which is reported in the first row of
Table 3.1 labeled as GDP (1960). The value −0.000025 is the coefficient α obtained
from a regression that does not include any further regressors. Notice that the sign of
this coefficient is negative, therefore confirming our a priori expectations but the high
standard errors raise doubts about the precision behind those point estimates.
The t-value is | − 0.00025/0.000062| = 0.403 < 2 indicating that the estimate is not
statistically significant at conventional levels. The overall picture changes once addi-
tional controls are added to the regression reported in column (II). Most importantly,
the coefficient α turns significant when we control for population and human capital.
Countries with higher initial per capita GDP tend to have lower growth rates and the
relationship is statistically significant. Adding those variables is sensible as we will
see that those variables are highly important drivers behind growth, which should
be controlled for in order to identify the true effect of initial per capita GDP on its
growth over time.5
This result must be seen in light of the low coefficient: A 1000 US Dollar increase in
initial per capita GDP is associated with a 0.469 to 0.717 percent lower per capita GDP
growth rate. Notice that a 1000 Dollar difference in per capita GDP is relatively large
for earlier periods with rather modest mean per capita GDP across countries.
3.1 The Solow (1956) model of economic growth
This chapter develops the well-known Solow growth model. You might have already
encountered this model in previous Bachelor courses but here the treatment is more
5 For example, failing to control for such determinants of growth performance such as population and
human capital risks that estimates of α are biased due to correlations between initial per capita income,
population and human capital.
3.1. THE SOLOW (1956) MODEL OF ECONOMIC GROWTH 59
Table 3.1: Convergence in growth
Estimated model: gi2011,1960 = αyi1960 + XiT2011 β + ei2011
I II III IV
GDP (1960) −0.000025 −0.000469∗∗ −0.000168 −0.000717∗∗

(0.000062) (0.000175) (0.000106) (0.000255)
Population (in millions) 0.000803
(0.000615) (0.000885)
Human capital index 4.395478∗∗∗ 6.216597∗∗∗
(1.136160) (1.615385)
Constant 2.274094∗∗∗ −6.240461∗∗∗ 4.187751∗∗∗ −9.502090∗∗∗
(0.546141) (1.791690) (0.917339) (2.787933)
R2 0.001 0.278 0.013 0.259

Observations 105 93 105 93
The Human capital index is computed using average years of schooling com-
bined with rate of returns to education. Standard errors in parentheses. *signif-
icant at 10%, ** significant at 5%, *** significant at 1%.
formal, and based on the textbook by Acemoglu (2008). In contrast to the textbook,
in these lecture notes we provide more detailed explanations for some basics that are
taken for granted in more advanced courses. Moreover, the discussion paves the way
for extensions that incorporate migration and trade.
3.1.1 Basic model
As a first step, we introduce the basic elements of the model. It is convenient to start
the description by specifying the production process as
Y (t) = O(K (t), L(t), ϕ(t)) . (3.2)
Notice that one implicitly assumes that the whole economy can be represented by
one production function without distinguishing sectoral or firm heterogeneity. Total
output is produced according to a function O that transforms the input factors capital
K at time t, and the input of labor L at time t into production Y at time t. The level
of technology ϕ at time t determines the efficiency of this transformation process.
Notice that technology in this model is represented by a simple shift parameter in the
production function. Improvements in technology simply shift the whole production
function upwards.
Each variable is characterized by an index that identifies the year of observation. The
model is dynamic in the sense that we analyze the response of output to changes in
labor and capital inputs over time.
Note that we have not specified the functional form of the production function in
equation (3.2) yet. This general formulation will be kept during major parts of this
chapter but we will also talk about the empirical relevance of the Cobb Douglas pro-
duction function. In fact, the Solow model and its results can be derived from a wide
range of production functions that fulfill some basic assumptions. Hence, we proceed
with stating these assumptions and explaining the intuition behind them.
Assumption 1 The production function O(K, L, ϕ) is continuous, twice differentiable, has

constant returns to scale, and diminishing marginal products.
The first two assumptions exclude less ordinary production functions with properties
that complicate the analysis. Functions that are continuous in their arguments are
more tractable in that they allow us to use differential calculus. We make use of this
particular property in order to check for diminishing marginal products in capital
or labor. Production functions have diminishing marginal products of their inputs if
the first derivative with respect to each input is positive but the second derivative is
negative. One can easily check if the functional form fulfill those properties using
differential calculus that must yield
∂O(K, L, ϕ) ∂O(K, L, ϕ)
OK (K, L, ϕ) = >0 , OL (K, L, ϕ) = >0 (3.3)
∂K ∂L
∂2O(K, L, ϕ) ∂2O(K, L, ϕ)
OKK (K, L, ϕ) = <0 , OLL (K, L, ϕ) = < 0 (3.4)
∂K2 ∂L2
The interpretation of the two equations in (3.3) are straightforward. The positive sign
of the first derivative of the production function with respect to capital or labor can
be interpreted as saying that increasing labor or capital inputs - holding everything
else constant - increases output. However, the second derivatives of the production
function with respect to both input factors are negative. Thus, output always grows
in L or K but it does so at diminishing rates. Hence, at the limit we find that a partial
change in labor or capital ends up having no more effect on output. That is, the pos-
itive effect becomes infinitesimally small. In contrast, a proportional increase of both
input arguments increases output by the same factor because of the constant returns
to scale assumption. Notice, that this result cannot be derived from the general for-
mulation of the functional form discussed above. It is a result that must be derivable
from the functional form of the production function. Suppose that inputs of both
labor and capital double, i.e. we perfectly replicate the factories which use labor and
capital in the economy. If the production function has constant returns to scale, then
this increase is allied to a twofold increase in output: two identical factories produce
twice as much output as one factory.
More formally, recall the definition of a homogeneous function. A function is homo-

geneous of degree m in the factors K and L if:
O(λK, λL, ϕ) = λm O(K, L, ϕ) for all λ ∈ R+ . (3.5)
that is, an increase in both input arguments by a factor λ increases output by a con-
stant factor λm . Notice that λ is restricted to take a positive real value. λ is a factor
that scales the amount of input factors up and down. There exists no plausible inter-
pretations for negative values of λ, although the algebraic proof of constant returns to
scale may still be valid. Production functions which are homogeneous of degree one,
i.e. m = 1, exhibit constant returns to scale, so that output changes proportionally to
the changes in the input arguments.
Factor supply and market clearing. Labor and capital used in the production pro-
cess must be employed at market prices. We assume perfect competition, meaning
that prices are taken as given by firms and consumers. A given amount of work-
ers can be recruited at wage w. Total labor supply at time t, L̄(t), equals total labor
demand, L(t). Households own the capital stock supplied to firms. Notice that we
normalize the price of the commodity good to 1. Thus, the final good is the numeraire
good. The final good can be used either as a consumption or as a capital good. Thus,
households face the following problem. They can consume goods at price 1 or rent
some of their capital to firms at an interest rate r (t) = R(t) − δ. We can think of
the interest rate as the compensation that the households receive for their forgone
consumption. Suppose one household decides to give up one unit of the commodity
good at time t − 1 to rent it to a firm. δ is the depreciation rate of capital. The capital
rent paid by the firm is equal to R(t) but a share δ of the capital stock depreciates in
every period. Thus, the interest rate is r (t) = R(t) − δ. At interest rate r (t) households
supply K̄ (t) units of capital through their savings and all capital is employed by firms
so that K̄ (t) = K (t).
But what drives firms’ demand for labor and capital? Firms’ factor demand is deter-
mined by profit maximization represented by the following maximization problem
max O(K, L, ϕ(t)) − R(t)K − w(t) L , (3.6)

K ≥0,L≥0
which means that firms choose labor and capital so as to maximize revenues minus
costs. Notice that the dependency of capital and labor on time is suppressed because
we are first looking at the long run equilibrium. Remember that we set the final goods
price to unity. Production, O(K, L, ϕ(t)), equals the revenue of the representative firm.
Production costs are equal to the sum of total interest paid on the capital stock and
the wages paid to workers. The first order conditions for problem (3.6) are
OK (K, L, ϕ(t)) = R(t) (3.7)
OL (K, L, ϕ(t)) = w(t) (3.8)
Labor and capital owners are paid according to their marginal productivity in the
production process. Taking the full employment conditions into consideration gives
a solution to the model in general equilibrium. Moreover, we know that the solution
implies that firms make zero profits in equilibrium. Notice that total production costs
are C = w · L + R · K. We assume constant returns to scale and hence we know that
the production function must satisfy
λO(K, L, ϕ) = O(λK, λL, ϕ) (3.9)
Differentiating this equation with respect to λ yields
1 · λ1−1O(K, L, ϕ) = OK (λK, λL, ϕ)K + OL (λK, λL, ϕ) L (3.10)
This relationship holds for any λ. In particular, setting λ = 1 in equation (3.10),

we can easily verify that equations (3.7) and (3.8) in C = w · L + R · K imply that
O(K, L, ϕ) = C. That is, firms make zero profits in equilibrium. The result we have
just derived is an important application of the Euler theorem, which we summarize
as Theorem I in the following box.
Theorem I
Consider a function g with Rk+2 → R. Variable k can take any integer from 0
to infinity and the output argument of the function is one real number R. Put
differently the function must have at least two arguments in order to be applicable
to Euler’s Theorem. Moreover, the function must be continuously differentiable

in two of its arguments x ∈ R and y ∈ R, with partial derivatives denoted by gx
and gy . Let g be homogeneous of degree m in x and y that means g(λx, λy, z) =
λm g( x, y, z) for all λ > 0. Then, the following results hold:
1. mg( x, y, z) = gx ( x, y, z) x + gy ( x, y, z)y for all x ∈ R, y ∈ R and z ∈ Rk
2. The partial derivatives gx and gy are themselves homogeneous of degree

m − 1 in x and y, i.e. :
gx (λx, λy, z) = λm−1 gx ( x, y, z), and
gy (λx, λy, z) = λm−1 gy ( x, y, z)
The full proof of the theorem is a straightforward application of calculus. An in-

terested reader can refer to the textbook by Acemoglu (2008). The properties of
the production function stated in Assumption 1 allow us to use the Euler theo-
rem for our analysis. Indeed, we are assuming a production function that is twice
continuously differentiable (notice that the Euler theorem requires the function to
be at least once continuously differentiable) with respect to capital and labor, and
homogeneous of degree m = 1 with respect to capital and labor.
We have already applied the Euler theorem when deriving equation (3.10). This
equation is an application of result 1, where m = 1, g( x, y, z) = O(K, L, ϕ),
gx ( x, y, z) = OK (K, L, ϕ(t)) = R(t) and gy ( x, y, z) = OL (K, L, ϕ(t)) = w(t).
Moreover, result 2 implies that the partial derivatives of the production function
with respect to L and K will be homogeneous of degree zero with respect to both
L and K. This means
OK (λK, λL, ϕ) = OK (K, L, ϕ), and
OL (λK, λL, ϕ) = OL (K, L, ϕ)

Intuitively, if we multiply K and L by the same factor λ, the partial derivative of

this modified production function with respect to λK or λL will be equal to the
partial derivative of the original production function. We will be able to appreciate
the usefulness of this result in what follows.
Assumption 2 The production function also satisfies the so called Inada conditions:
lim OK (K, L, ϕ) = ∞ and lim OK (K, L, ϕ) = 0 (3.11)

K →0 K →∞
lim OL (K, L, ϕ) = ∞ and lim OL (K, L, ϕ) = 0 (3.12)

L →0 L→∞
The Inada conditions ensure that the model has an interior solution. Intuitively, the
Inada Conditions guarantee that (i) the firm does not accumulate a very large (infinite)
amount of capital or labor, and (ii) always uses a positive (non-zero) amount of capital
and labor for production. Let one of the input factors tend to zero or to infinity
without changing the other factor. For instance, if K tends towards zero, then the first
derivative of the production function goes to infinity. Notice that the first derivative of
the production function determines its slope. Thus, the slope is approaching infinity
as K approaches zero. Now let K tend towards infinity instead. In this case the
slope of the production function approaches zero. Thus, additional changes in K
no longer translate into additional output. This is in line with the assumption of
diminishing marginal returns. However, if K is low, small changes in K are associated
with large increases in output but the higher K grows, the lower is the additional
output generated by the same change in K. For very high values of K, the change in
output is almost zero. Notice that neither K nor the change in output (slope of the
production function) can be exactly equal to zero.
Figure 3.2 gives an example of a production function satisfying the Inada condition.
The left panel plots the production function for changes in K for a fixed L and the
Figure 3.2: Production functions in line with the Inada conditions
𝑌𝑌 𝑌𝑌
𝑂𝑂(𝐾𝐾, 𝐿𝐿, 𝜑𝜑)
𝑂𝑂(𝐾𝐾, 𝐿𝐿, 𝜑𝜑)
𝐾𝐾 𝐿𝐿
right panel depicts the production function for changes in L for given K. Holding K
constant, output Y is increasing in L but the growth rates are diminishing in the level
of L. The slope of the function O tends towards infinity as K (L in the right panel)
tends towards zero and it approaches zero as K (L in the right panel) tends to infinity.
3.1.2 Dynamics in the short run
So far we have described the basic structure and properties of the economy in the
Solow model. Having understood these properties and their implications, we can
proceed to characterize the dynamics of growth. We will start with a baseline scenario
in which investment is considered as being the only determinant of economic growth.
In this context, it is possible to study the relationship between investment and growth
in a discrete time setting. Other determinants of economic growth, such as population

and technological change, are considered later in the chapter. The analysis will be
slightly more complex with the result that we solve those extensions in the more
convenient continuous time setting. At this point of the course it is important to notice
that we are dealing with a dynamic model. In contrast to a static model, equilibrium
in a dynamic model is computed for each period. Thus, there is a sequence of values,
each of which represents the different levels of the endogenous variable at any given
point in time. This sequence allows to describe the adjustment path of the endogenous
variables without gaps. This kind of adjustments also exists in static models but the
changes between different long run equilibriums are usually treated as black boxes.
Another important distinction is between stationary and non-stationary models. The

standard model is dynamic but stationary. The economy converges to its equilib-
rium and remains stable (stationary) afterwards. Later on we will add movements in
exogenous variables in order to get a non-stationary dynamic model. The dynamic
adjustment process towards the steady state equilibriums are characterized and due
to changes in technology, those equilibriums are changing over time (the equilibrium
is non-stationary).
The fundamental law of motion. Remember that a fraction δ of the capital stock is
lost every period due to depreciation. Thus, the capital stock one period in the future
will be
K ( t + 1) = (1 − δ ) K ( t ) + I ( t ) . (3.13)
Where the capital stock at the beginning of time t is K (t), but it decreases by δK (t)
during period t with the result that only (1 − δ)K (t) will remain available for the pro-
duction process at time t + 1. Investments made by the firms at time t will counteract
depreciation by contributing to the stock of capital at time t + 1. Total output Y (t) can
either be consumed or end up being invested such that
Y (t) = C (t) + I (t) (3.14)
We can easily solve equation (3.14) for I (t) in order to substitute I (t) by equation
(3.13). As a result we obtain equation (3.15). Recall that Y (t) = O(K, L, ϕ), meaning
that the equation for next period’s capital stock becomes
K (t + 1) = (1 − δ)K (t) + O(K (t), L(t), ϕ(t)) − C (t) . (3.15)
Where does investment is represented? Households’ behavior in the Solow model

is rather simplistic. We abstract from households’ preferences and assume that total
output is either consumed or saved. Households simply save a constant fraction of
their income and everything else is consumed within the same period:
S(t) = I (t) = Y (t) − C (t) = sY (t) (3.16)
Put differently, we can express consumption using the saving rate s
C ( t ) = ( 1 − s )Y ( t ) (3.17)
We can combine equations (3.16) and (3.13) in order to obtain the fundamental law of
motion in the Solow model as:
K (t + 1) = sO(K (t), L(t), ϕ(t)) + (1 − δ)K (t) . (3.18)
The benchmark model abstracts from changes in population or technology. Thus, the
labor endowment L and total factor productivity, ϕ, are assumed to be constant over
time. We will relax both of these assumptions in the next subsection. Furthermore,
let us define the per capita capital stock as:
K (t)
k(t) = . (3.19)
L
Given the constant returns to scale assumption, we can multiply the production func-
tion by 1/L to obtain O(K, L, ϕ)/L = O(K/L, 1, ϕ) = o (k (t)). It is straightforward to
verify that the per capita production function o (k (t)) has all the properties of the orig-
inal production function O(.), including homogeneity of degree 1 (constant returns
to scale). Indeed: o (λk ) = O(λK/L, λL/L, ϕ) = λO(K/L, 1, ϕ) = λo (k ). This means
that Theorem I can be applied to this production function. Setting λ = 1/L we obtain
OK (K, L, ϕ) = o 0 (K/L, 1, ϕ) . (3.20)
Hence, the marginal productivity of capital can be expressed as a function of the

capital-labor ratio only. We already know that in equilibrium the marginal productiv-
ity of capital is equal to the interest rate R(t). Hence we obtain
R(t) = o 0 (k (t)) > 0 (3.21)
Similarly, we can apply result 1 of Theorem I and write
O(K, L, ϕ) = OL (K, L, ϕ) L + OK (K, L, ϕ)K
Dividing both sides by L and recalling that OK = o 0 (k ) we get
o (k (t)) = OL (K, L, ϕ)1 + o 0 (k(t))k (t)
OL is the marginal productivity of labor, which is equal to the wage w(t) in equilib-
rium. This expression can be used in order to solve for
o (k(t)) = w(t) + o 0 (k (t))k (t)

which can be rearranged to
w(t) = o (k (t)) − k (t)o 0 (k (t)) > 0 . (3.22)
The law of motion in per capita representation reads as
k (t + 1) = s · o (k (t)) + (1 − δ)k (t) (3.23)
Notice that we have omitted the variable ϕ for the sake of simplicity. For now we ex-
clude the possibility of technological change. This has the effect that ϕ is meaningless
and can be normalized to 1. Equation (3.23) therefore describes the behavior of the
economy over time through the behavior of a single variable, that is, the capital-labor
ratio. This procedure simplifies the analysis of the equilibrium.
Graphical representation of the steady state equilibrium. Before providing a for-

mal definition and analysis of the dynamics in the model, we discuss an intuitive
graphical representation of the equilibrium. Figure 3.3 depicts the two elements that
contribute to the change of the capital stock: investment and capital depreciation.
Capital depreciates at a constant rate δ meaning that the capital depreciation is just a
straight line starting from the origin in the o (k ) − k space. Investments in every period
are a constant fraction s of output, hence the investment function has the same shape
as the production function, rescaled by the factor s. The vertical difference between
the production and saving functions corresponds to the level of consumption in the
economy at any point in time. Let us start with a low initial capital stock in period
0 and suppose that a fraction s of output ends up being saved. Thus, investments
during this initial period are equal to s · o (k) and the capital stock rises by exactly
this amount until the beginning of the next period. Although, as we know, some
of the investments must be used to compensate for capital depreciation, the capital
Figure 3.3: Fundamental law of motion
Output
𝛿𝛿𝛿𝛿(𝑡𝑡)
𝑜𝑜(𝑘𝑘 ∗ ) 𝑜𝑜(𝑘𝑘(𝑡𝑡))
Consumption
𝑠𝑠×𝑜𝑜(𝑘𝑘 ∗ ) 𝑠𝑠×𝑜𝑜(𝑘𝑘(𝑡𝑡))
Investment
𝑘𝑘(𝑡𝑡)
𝑘𝑘 ∗
stock is still small and the amount of depreciated capital is low. Depreciation is a
linear function of k, whereas investments are increasing relatively fast: this is one of
the implications of the previously discussed Inada conditions, which directly affect
the shape of the production and investment functions. At low levels of the capital
stock, the rate of change of investment tends towards infinity, while at a very high
level of capital stock the change in investment approaches zero. The Inada conditions
ensure that the two curves of capital depreciation and investment intersect at some
equilibrium level of the capital labor ratio, k∗ ∈ (0, ∞).
Moreover, Assumption 1, which establishes that the production function has positive
and decreasing marginal productivities, ensures that the equilibrium value of the
capital-labor ratio is unique, i.e. that the δk(t) and the s · o (k ) curves intersect only
once. Acemoglu (2008) discusses three examples where either the Inada conditions or
Assumption 1 are not satisfied. Figure 3.4 reproduces those cases.
72
Figure 3.4: Non-existent or not-unique steady states
𝑘𝑘𝑡𝑡+1 𝑘𝑘𝑡𝑡+1 𝑘𝑘𝑡𝑡+1

𝑠𝑠×𝑜𝑜(𝑘𝑘𝑡𝑡 ) + (1 − 𝛿𝛿 )𝑘𝑘𝑡𝑡 𝑠𝑠×𝑜𝑜(𝑘𝑘𝑡𝑡 ) + (1 − 𝛿𝛿 )𝑘𝑘𝑡𝑡
45° 45°
45°
𝑠𝑠×𝑜𝑜(𝑘𝑘𝑡𝑡 ) + (1 − 𝛿𝛿 )𝑘𝑘𝑡𝑡
𝑘𝑘𝑡𝑡 𝑘𝑘𝑡𝑡 𝑘𝑘𝑡𝑡

(𝑎𝑎) (𝑏𝑏) (𝑐𝑐)
CHAPTER 3. ECONOMIC GROWTH
The production function in panel ( a) does not satisfy the first part of the Inada con-
dition. The curve that pins down next period’s capital stock is always below the 45◦
line without intersecting in k∗ > 0. In panel (b) of Figure 3.4 the second part of the
Inada condition is not fulfilled.
Again there is no point where both curves intersect at any k∗ > 0. Finally in panel
(c) Assumption 1 is not satisfied: The production function is non-monotonically in-
creasing and hence the equilibrium is not unique. The equilibrium level of the capital-
labor ratio k∗ is called a steady state equilibrium, because once this level is reached, the
capital-labor ratio will tend to be constant.
The capital stock can adjust through savings but, without population growth, the
labor endowment is fixed. The economy then automatically moves towards its steady
state where the capital stock is k∗ . Suppose that capital grows beyond k∗ . Depreciation
would be higher than the investment made each period. Thus, depreciation pushes
the capital stock back to the level k∗ . The rate at which the economy converges to
its steady state equilibrium will depend on how far the economy currently is away
from the equilibrium itself: if an economy starts with a small value of capital, it will
tend to grow much faster towards the equilibrium than an economy that is already
at a later stage of development, that is, one with a higher capital stock. Notice,
however, that in the Solow model with neither population growth nor technological
change, there is no growth after the steady state is reached. Growth takes place during
the transition towards the equilibrium, but it comes to a halt once the equilibrium
is reached. Output and capital remain constant afterwards. Keeping the intuition
discussed above in mind, we can proceed with a more formal definition and analysis
of the steady state equilibrium in the Solow model.
Definition of a steady-state equilibrium. A steady state equilibrium in the Solow

model without technological progress and population growth is one in which k (t) =
k∗ for all t. This will correspond to the value of the capital-labor ratio for which the
capital depreciation and investment curves intersect. Thus, at k∗ we have
o (k∗ ) δ
s · o (k∗ ) = δk∗ ⇒ ∗
= . (3.24)
k s
We have previously discussed how the shape of the production and investment func-
tions matter to assure that such an equilibrium exists and that it is unique. Based on
Assumption 1 and Assumption 2 one can easily prove the existence and the unique-
ness of an equilibrium analytically. Indeed, if the Inada conditions (Assumption 2)
o (k(t)) o (k(t))
hold, then limk→0 k(t)
= ∞ and limk→∞ k(t)
= 0. Hence, the Inada conditions
hold also for the function o (k )/k. Moreover, from Assumption 1 we know that o (k ) is
continuous. Hence, for the intermediate value theorem6 there must exist at least a k∗
o (k∗ )
such that k∗ = δs . To prove that this point is unique we must be able to show that
o (k(t))
the function k(t)
is monotonic. The Inada conditions already tell us that the function
is decreasing, however, those conditions describe what happens at the boundaries of
the function.
In order to learn about the behavior of the function within its boundaries, we differ-
o (k∗ )
entiate k∗ with respect to k in order to obtain
∂(o (k )/k ) o 0 (k)k − o (k) w

= 2
=− 2 (3.25)
∂k k k
o (k(t))
Since k(t)
is monotonically decreasing in k, the equilibrium at k∗ is unique. Notice
that this result holds if o 0 (k ) > 0 which is always true if Assumption 1 is satisfied.
Steady state capital stock. We can compute the equilibrium capital stock k∗ by solv-
ing equation (3.24) to obtain
s k∗ o (k∗ ) δ
s · o (k∗ ) = δk∗ ⇒ = ⇒ − =0 , (3.26)
δ o (k∗ ) k∗ s
6 Consider a continuous function o with boundaries o ( a) and o (b) and consider a point s between o ( a)
and o (b). The intermediate value theorem states that there must exist at least one point c between a
and b such that o (c) = s.
where k∗ is the per capita capital stock for which investment within one period is
exactly equal to the level of capital depreciation.
Some knowledge about the production function allows us to compute the level of k∗
provided that all other exogenous variables are known. Based on the exact value of
k∗ one could also compute the steady state level of per capita output
y∗ = o (k∗ ) (3.27)
and the exact level of per capita consumption
c ∗ = (1 − s ) o ( k ∗ ) . (3.28)
Recall that we had normalized total factor productivity to unity. This normalization
was justified by the fact that technology was constant and therefore meaningless in
our analysis so far. However, we now want to allow for changes in ϕ in order to grasp
a first understanding of the effects of technological change on growth. We follow
Acemoglu (2008) by introducing a Hicks neutral shift variable ϕ > 1. The greater
ϕ, the higher the level of output for a given level of capital and labor inputs. The
production function becomes
o (k) = ϕõ (k) (3.29)
We can easily compute the steady state capital-labor ratio for a given production
function and for given parameters ϕ, s and δ using equation (3.26). But can we already
predict how k∗ reacts on changes in those parameters even without knowing the
functional form of O? Indeed, we know the properties of the production function
and this allows us to apply differential calculus using the Implicit Function Theorem.
Suppose to have a function that is O( x, y) = 0. Applying the total differential gives us
dx
∂O
∂x dx + ∂O
∂y dy = 0, which can be solved for dy = − ∂O ∂O
∂y / ∂x . The following comparative
statics exercises will make extensive use of the implicit function theorem to derive
∂k∗ ( ϕ, s, δ)
>0 (3.30)
∂ϕ
∂k∗ ( ϕ, s, δ)
>0 (3.31)
∂s
∂k∗ ( ϕ, s, δ)
<0 (3.32)
∂δ
and
∂y∗ ( ϕ, s, δ)
>0 (3.33)
∂ϕ
∂y∗ ( ϕ, s, δ)
>0 (3.34)
∂s
∂y∗ ( ϕ, s, δ)
<0 (3.35)
∂δ
Capital and total factor productivity. The first comparative statics exercise is on the
impact of a change in total factor productivity, ϕ, on the steady state per capita capital
stock. Equation (3.30) reports a positive relationship. The higher the technology
parameter ϕ, the higher the steady state per capita capital stock with the result that
o (k∗ )
output increases according to equation (3.33). We know that G (k∗ ) = k∗ − δ
s =0
must hold in equilibrium. The implicit function theorem can be applied to this in
order to compute the first derivative of k∗ with respect to ϕ as
∂k∗ ( ϕ, s, δ) Gϕ õ (k )k∗ −1
=− =− (3.36)
∂ϕ Gk∗ [o 0 (k∗ )k∗ − o (k∗ )]k∗ −2
The numerator in this equation is positive whereas the denominator is negative. The
latter is due to the condition that pins down the equilibrium wage: a worker’s income
must be positive in order to secure a strictly positive supply of labor. Moreover,
we know that w = o (k ) − ko 0 (k )>0. Multiplying by −1 yields −o (k ) + o 0 (k )k < 0,
which is equal to the term in squared brackets. Thus, the denominator in equation
(3.36) is negative. The negative denominator must be multiplied by the negative
sign in front of the fraction, thereby turning the sign of the whole expression from
negative to positive. An increase in total factor productivity, ϕ, is associated with

an increase in the equilibrium value of capital per worker. The intuition behind this
result is that labor and capital become more productive due to the improvements
in production technology. Investment increasing the capital stock leading to more
productive workers.
Capital and the saving rate. The second comparative statics exercise is on the impact
of a change in the saving rate, s, on the steady state capital per work. Equation (3.31)
reports a positive relationship. The higher the saving rate, the higher the per capita
capital stock and the higher the steady state output level according to equation (3.34).
o (k∗ )
G (k∗ ) = k∗ − δ
s = 0 and the Implicit Function Theorem can be used in order to
derive the partial derivative of k∗ with respect to s as
∂k∗ ( ϕ, s, δ) Gs + s −2 δ
=− =− (3.37)
∂s Gk∗ [o 0 (k∗ )k∗ − o (k∗ )]k∗ −2
The numerator of this equation is positive. We already know that the denominator
is negative. Multiplying the negative denominator with the negative sign in front of
the fraction yields a positive sign. An increase in the saving rate s is associated with
an increase in the steady state capital-labor ratio. The intuition behind this result
is straightforward. A higher saving rate boosts investments. The capital stock can
rise above its initial value since higher investments allow the economy to replace
the additional depreciation associated with a higher capital stock. The capital stock
expands and production increases to its new steady state value. As before, the higher
capital per worker is associated with a higher level of output as stated in equation
(3.34).
The stock of capital and the depreciation rate. The third comparative statics ex-
ercise investigates the impact of a change in the depreciation rate, δ, on the steady
state level of capital per worker. Equation (3.32) indicates a negative relationship. The
higher the depreciation rate, the lower the per capita capital stock and the lower the
steady state output level according to equation (3.35). Again, we can use G (k∗ ) =
o (k∗ )
k∗ − δ
s = 0 and the Implicit Function Theorem to calculate the partial derivative of
k∗ with respect to δ as
∂k∗ ( ϕ, s, δ) G −1/s
=− δ =− (3.38)
∂δ Gk∗ [o 0 (k∗ )k∗ − o (k∗ )]k∗ −2
The denominator and the numerator are both negative. The negative sign in front of
the ratio together with the two negative signs of the nominator and the denominator
yield a negative sign for the partial derivative that can be interpreted as follows. An
increase in the depreciation rate, δ, is associated with a decrease in the equilibrium
capital labor ratio. More severe depreciation requires more savings in order to keep
the capital stock at its equilibrium value. If the saving rate does not change capital will
be lost in the economy because the yearly investments are not sufficient to maintain
the capital stock, which is decreasing. Production declines to its lower steady state
level. The lower capital-labor ratio is associated with a lower level of output, which
corresponds to the outcome of the first derivative noted in (3.35).
Optimal consumption. We have learned two important outcomes of the model: for
given parameters, the capital stock moves towards its steady state level. Moreover,
changes in the parameters - especially changes in the parameter s - affect the equi-
librium capital-labor ratio and the steady state per capita output level. Moreover, we
also know how the outcome variables react to changes in s, δ, and ϕ. What is the
optimal steady state? Is it optimal to aim at a saving rate that maximizes output?
Certainly not! The highest output would be at a point where the saving rate is 100%.
Consumption would be zero and all GDP would be invested into the capital stock,
which yields the highest possible output. Such a strategy would certainly be difficult
to justify in the public debate. What is a high GDP good for if it comes at the ex-
pense of zero or low consumption? A rational decision would be to target a saving
rate that maximizes per capita consumption. The opposite extreme scenario would be
maximizing consumption by setting s = 0. In this instance, the capital stock would
decline to zero over time due to capital depreciation without any capital formation
due to zero investment. Consumption would approach zero as well, which cannot be
a sustainable equilibrium.
This discussion illustrates that there must be a unique savings rate somewhere be-
tween those two extreme values that maximizes consumption and that is not associ-
ated with the highest level of output possible. Starting from a low saving rate, we
can study how an increase in s affects consumption. There are two opposing effects
that have to be taken into account. On the one hand, consumption decreases in the
saving rate as a smaller fraction of output ends up being consumed. On the other
hand, consumption increases in s due to its magnification effect on capital formation.
Higher savings spur investments with the result that output increases leading to a
higher level of consumption. Starting from a very low level of s, we can be sure that
the latter effect outweighs the former. The positive effects of an increase in capital on
output are extremely high when the stock of capital is close to zero.
Notice that this result stems from the Inada condition and the fact that the slope of
the production function goes to infinity when the capital stock approaches zero. The
increase in output must be higher than the reduction in consumption. Changes in
output are extremely strong when k is small but diminishing returns to scale imply
the the output growth rates are vanishing when k gets larger. If s approaches 1,
consumption goes to zero again. The reason is a different one compared to before in
the scenario where s = 0. It is now the case that all output in each period ends up
being invested. Output would be at the maximum level, even if consumption is zero.
We can conclude that the relationship between savings and consumption is inverse-U
shaped with a maximum at a level of saving that we call the golden rule saving rate.
We denote this particular level of the saving rate by s gold . For all s ∈ (0, s gold ), the
increase in production through an increase in the stock of capital is still higher than
the depreciation associated with this change. The same logic allows us to derive the
converse result for s ∈ (s gold , 1). Depreciation would be higher than the change in
output if s grows larger than the golden rule saving rate.
Figure 3.5 gives a graphical solution for the golden rule saving rate, s gold . The level
lies between the two extreme scenarios s = 0 and s = 1. Associated with s gold , we find
a maximum level of consumption at a level equal to (1 − s)o (k∗gold ). In what follows
we will develop the reasoning above in a more formal way.
We already know that consumption in the Solow economy is a fraction of income not
being saved. At the steady state capital to labor ratio k∗ we have
c∗ (s) = (1 − s)o (k∗ (s)) (3.39)
= o (k∗ (s)) − δk∗ (s) (3.40)
The fact that the condition s · o (k∗ ) = δk∗ must be applicable at the steady state level
of per capita capital allows us to solve for the associated level of s by maximizing
equation (3.40) with respect to s. The first order condition reads
∂c∗ (s)
=0 . (3.41)
∂s
Applying the chain rule we obtain
∂c∗ (s) ∂k∗

= [o 0 (k∗ (s)) − δ] =0 . (3.42)
∂s ∂s
∂k∗
We know that k∗ is increasing in s (i. e. ∂s > 0) from equation (3.31), hence expression
(3.42) is true only at a level of saving such that o 0 (k∗ (s)) = δ, i.e. at a capital-labor ratio
such that an additional unit of capital would produce just as much as to compensate
for its own depreciation.
We can check if the corresponding saving rate is a local maximum by evaluating

the second derivative of consumption with respect to s at k∗ (s). The value of the
second derivative must be negative in order to fulfill the sufficient conditions for a
local maximum. The proof is straightforward and is left as an exercise to the student.
The saving rate that maximizes consumption is called the golden rule saving rate and it
fulfills
o 0 (k∗ (s gold )) = δ (3.43)
In order to prove that o 0 (k∗ (s gold )) = δ is a global maximum, we should be able to

verify that the second derivative is negative for any value of s ∈ (0, 1). However, an
easier way to prove uniqueness is to follow the line or reasoning below (see Acemoglu,
2008)7
Consider again equation (3.42). We have already shown that if s = s gold , then
o 0 (k∗ (s gold )) − δ = 0 .
For a value of s ∈ (s gold , 1) we have that o 0 (k (s)) < δ and by equation (3.42) con-
sumption decreases as the saving rate increases. The mechanisms through which this
happens can be explained as follows: a higher saving rate will raise the capital to labor
ratio. Due to diminishing returns to capital (Assumption 1) the marginal productivity
of capital at this new steady state level will be lower than the fraction of capital that
is lost due to depreciation. The higher output generated by the higher saving rate
is not sufficient to replace the depreciated capital stock, hence any new steady state
consumption level corresponding to a saving rate s > s gold will be lower than at s gold .
The economy could improve its consumption level by decreasing its saving rate.
Now consider the opposite case: if s < s gold and o 0 (k (s)) − δ > 0, that is, consumption
could increase by higher savings. A higher saving rate generates additional invest-
ments and a larger capital stock, which would raise production by more than what
is needed to compensate for capital depreciation. The economy could increase con-
sumption by saving more.
7 The proof is an application of the derivative test, which is used when the sufficient conditions for a
global maximum or minimum cannot be verified)
82
Figure 3.5: Hump shaped relationship between s and c
Consumption
∗
(1 − 𝑠𝑠)𝑜𝑜(𝑘𝑘𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 )
Saving rate
0 𝑠𝑠𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔
Hence, we have just proved that consumption at any saving rate below or above the
value s gold must be lower than consumption associated with s gold . The above reasoning
proves that s gold is the only saving rate that maximizes consumption.
Summary of the results derived so far:
Economies can grow through investments into their capital stock until the steady state
level of output is reached:
• Output is produced by inputs of labor and capital
– The amount of labor is determined by the country’s population
– The capital stock is flexible ⇒ The saving rate determines each period’s
investments as a fraction of output
• A fraction δ of the capital stock depreciates every period.
⇒ In the steady state this fraction needs to be replaced through investments.
⇒ For a given saving rate, the capital stock moves towards its steady state equi-
librium
⇒ Economies can grow up to the steady state level of capital that determines
equilibrium output
The steady state equilibrium level of capital and output can be influenced by changes in
the saving rate, the depreciation rate, or total factor productivity:
• An increase in s is associated with higher investments and hence higher

steady state output and capital stock
• An increase in δ is associated with a lower equilibrium capital stock
• An increase in total factor productivity is associated with a higher equilib-

rium capital stock
Higher savings lead to higher equilibrium capital stocks and output, but not necessarily to
higher consumption
• The effect of s on consumption is inverse-U shaped ⇒ More output 6= more

consumption
• The golden rule saving rate is the saving rate at which consumption is max-
imized.
3.1.3 Model solution under Cobb Douglas
The model presented so far features a general formulation of the production pro-
cess without imposing an explicit functional form of the production process itself.
We imposed certain assumptions that are fulfilled by a large number of production
functions. The advantage is that the results are more general without depending on
certain restrictions on the production process. The drawback of this approach is that
it precludes computing particular values of the different outcome variables and we
cannot make any statements about the magnitude of the effects. Instead we were
always focusing on general statements about the sign of the effects.
Many applications of the Solow model rely on a Cobb Douglas production function
in order to model the production process. Imposing a Cobb Douglas production
function allows us to grasp the key ideas of the model and then derive close-form
solutions. Suppose a standard Cobb Douglas production function that reads
(1− α )
Yi = ϕLiα Ki . (3.44)
Output, Yi , is produced by input of labor, Li , capital, Ki , and it depends on technology,

ϕ. Notice that we assume a Hicks neutral technology. It is easy to verify that equa-
tion (3.44) is continuous, has diminishing marginal returns in labor and capital and
satisfies the Inada conditions. Moreover, it has constant returns to scale properties
because α + (1 − α) = 1. Hence, constant returns to scale is not an intrinsic property
of the Cobb-Douglas production function but depends on the coefficient α.
Is it realistic to assume constant returns to scale and Cobb-Douglas technology? Based

on the results of the empirical literature, there seems to exist some support for this
view. Among the many empirical studies on the topic, we choose a very prominent
example for the US.
Notice that the production function can be linearized by taking logs on both sides,
which yields
ln Yi = α ln Li + (1 − α) ln Ki , (3.45)
where ϕ is assumed to be equal 1. This assumption does not do any harm because
the technology shifts the whole production without affecting the capital to labor ratio
in the production process. Thus, we can normalize it without loss of generality. The
log linearized version of the production function can be estimated using firm-level
data on output and inputs. Observed output and inputs of different firms are used to
estimate the functional form between output, labor and capital. The estimated model
reads
ln Yit = β 1 ln Lit + β 2 ln Kit + eit (3.46)
where β 1 = α and β 2 = (1 − α). The model can be fitted to the data using a simple
regression approach.
Based on the outcome of the regression, we wish to test whether the estimated co-
efficients fulfill β 1 = 1 − β 2 . Otherwise, the assumption of constant returns to scale
must be rejected. A rejection may still be caused by the choice of the production
function. It may be the case that other production functions with constant returns to
scale properties yield better results. In that case we would have to resort to another
functional form to represent the production process in the economy within the Solow
(and many others) growth models.
Figure 3.6 summarizes the results from different model specifications. Each row of
the table reports the coefficient estimate for each regressor. The row “Labor” reports
estimates of β 1 and the row "Capital" reports the coefficient β 2 . The other rows report
the estimates of the coefficients of other control variables.8 Standard errors are re-
ported below the respective coefficients. The column "Estimation Procedure" in turn
indicates the econometric method that generated the results. Remember that each
column is associated with one distinct regression variant. The functional form of the
regression is always the same but there are different ways of estimating it. The most
parsimonious way of estimating a linear regression is "OLS - Ordinary Least Squares",
the results of which are reported in column (5).
Without going into the details of how the coefficients are obtained through OLS, we
focus on the interpretation of the results of this model. The coefficient of Labor is
β 1 = 0.628 and it has a standard error of 0.020. The coefficient of capital, β 2 , is 0.219
with a standard error equal to 0.018. How do we interpret these coefficients? Taking
the partial derivative of equation (3.46) with respect to ln L and ln K we obtain
∂ ln Y
= β1 (3.47)
∂ ln Lit
∂ ln Y
= β2 (3.48)
∂ ln Kit
If ln L increases by ∆ ln L = 1 then ln Y increases by ∆ ln Y = 0.628. Notice that

we are looking at logarithmic changes and recall that a change in the logarithm of a
variable can be interpreted as a percentage change. Hence, the estimated coefficients
can be interpreted as elasticities and the interpretation reads "Keeping all other factors
constant, an increase by 100 percent in L is associated with an increase of 63 percent
in Y.”
By the same token, conditional on the input of labor, an increase in capital by 100
8 See the Appendix for further details on how to interpret the coefficients in a multivariate regression.
Figure 3.6: Estimated coefficients of a production function

Source: Olley and Pakes (1992)
percent is associated with a 22 percent

increase in output. Taking both effects together
we find that if both capital and labor increase by 100 percent then output increases
by 85 percent. This does not yet confirm our constant returns to scale hypothesis (for
that we would need the estimated coefficients of Labor and Capital to sum to 1), but
it is already very close.9
9 However, standard hypothesis testing techniques can be used to test whether β 1 + β 2 = 1.

The equilibrium conditions under the assumption of a Cobb Douglas production

function. We have seen that the Cobb Douglas production function seems to be a
plausible formulation of the production process in firms. We can now use the Cobb
Douglas production function as functional form for the function O(.) in the Solow
model
(1− α )
L α K (1− α ) K
o (k) = ϕ = ϕL(α−1) K (1−α) = ϕ = ϕ ( k ) (1− α ) (3.49)
Li L
We can substitute this into the equilibrium condition
ϕ ( k ∗ ) (1− α ) δ
− =0 (3.50)
k∗ s
For given parameters one may easily calculate the steady state capital per worker.
ϕ can be normalized to unity, α can be taken from the empirical analysis discussed
above and values for s and δ can be easily found in empirical studies or publicly
available data sets. Plugging these parameters into the equilibrium condition allows
us to compute a value for k∗ that can be used to obtain the steady state output from
the production function.
As we have seen in this section, the basic Solow model predicts a steady state growth
rate of zero in the long run. Capital accumulation leads to growth exclusively during
the transition towards the steady state: for instance if the country is at an early stage
of its development, or as an effect of a change in exogenous factors such as the saving
rate, technology or the depreciation rate. Once the steady state level of capital and
output is reached, growth rates are zero. How then can growth be generated in the
long run? While an economy where the saving rate keeps increasing is not conceiv-
able, given the limits imposed by the trade off between consumption and savings,
there are other two potential factors that might change the results of the base model.
These are changes in the size of the labor force, i.e. population growth, and techno-
logical advances. The former will be the object of the analysis of the next section.
The latter should already be familiar. We know how capital and output reacts to "one
3.2. POPULATION GROWTH IN THE SOLOW MODEL 89
time" changes in ϕ. In subsection 2.4 we will extend the Solow model to account for
a constant rate of technological change.
3.2 Population growth in the Solow model
This subsection deals with the role of population growth in the Solow model. We
assume that time is continuous in order to simplify the analysis. Population is no
longer constant. We assume that the labor force grows according to
L(t) = exp(nt) L(0) . (3.51)
According to this formulation, population grows at a constant rate n. As you may

remember from previous courses in mathematics, the rate of change of a variable can
be approximated by the difference between its natural logarithm for small changes.
Hence, the population growth rate can be approximated by L̇/L = ln L(t + 1) −
ln L(t). We follow the notation in Acemoglu (2008) and denote changes over time by
a dot above the respective variable. The rate of population growth can be approxi-
mated by the difference between log population between two different periods. This
approximation works for small changes in time and becomes less and less precise
when changes become bigger. We can substitute using the log version of equation
(3.51) in order to get
L̇/L = n(t + 1) + log L(0) − nt − log L(0) = n , (3.52)
where L(0) is some initial value of labor at time 0. This difference can be computed at
any point in time. The growth rate is constant over the whole time span. Remember
that
K (t)
k(t) = (3.53)
L(t)
Can we compute the change of k(t) over time? In a model with population growth, k
changes because K and/or L are changing. The change in k(t) can be computed using
the total derivative with respect to t as
dk ∂k dK ∂k dL
= + (3.54)
dt ∂K dt ∂L dt
dk 1 dK K (t) dL
= − (3.55)
dt L(t) dt L(t)2 dt
dk
dt 1 L(t) dK K (t) L(t) dL
= − (3.56)
k(t) L(t) K (t) dt L(t)2 K (t) dt
k̇ K̇ L̇
= − (3.57)
k(t) K (t) L(t)
K̇
= −n (3.58)
K (t)
Recall that the capital stock is determined by
K̇ (t) = sO(K (t), L(t), A(t)) − δK (t) (3.59)
What is the steady state in the Solow model with population growth? As before we
search for a solution in which the per capita capital stock remains constant over time.
k̇
Technically speaking, we are searching for the situation where k(t)
= 0. We already
k̇ K̇
know that k(t)
= K (t)
− n.
K̇
=n (3.60)
K (t)
That makes sense. The population grows at rate n every period so the capital stock
would have to rise at the same rate in order to hold the per capita level of capital
stable over time. The question is how do we achieve a rise in the stock of capital?
As long as we are not yet in steady state, the higher labor supply through population
growth spurs production. Higher production is allied to another round of investment
into the capital stock. A higher capital stock leads to another increase in production.
Investment increases again and so does capital formation. This procedure is repeated
3.2. POPULATION GROWTH IN THE SOLOW MODEL 91
over and over again until the new steady state is reached. Suppose we go from a
situation where n = 0 to a situation where n > 0. Output grows due to the reasons
discussed previously. But what about the per capita capital stock in the steady state?
Is it increasing or decreasing? The unexpected answer is that the per capita capital
stock must jump to a lower level. The model is already in steady state so that there
is no room for investments into the capital stock. Investments within each period
are already equal to the depreciation of capital every period so that the change in
capital is zero. Now we add population growth to the model. The supply of labor
suddenly increases by n every period. Firms will use more labor in the production
process. Suppose that the capital stock remains constant. The per capita capital
stock would decrease automatically due to the higher denominator of k. Nevertheless,
we also observe an increase in output associated with higher investments. Thus,
the capital stock must increase due to the additional investment. However, because
of diminishing returns to scale it is not likely that the change in the capital stock
is sufficient to compensate for the increase in population. Nevertheless, there are
different effects at work, which makes an assessment without algebra impossible. We
are able to show that the per capita capital stock decreases as the population growth
rate increases.
Before turning to the analytic proof, we have to rearrange the equilibrium condition.
Substituting K̇ in equation (3.60) by equation (3.59) yields
k̇ sO(K (t), L(t), ϕ(t)) − δK (t)

= −n (3.61)
k(t) K (t)
k̇
= sO(1, L(t)/K (t), ϕ(t)) − δ − n (3.62)
k(t)
k̇ o (k (t))
= s −δ−n (3.63)
k(t) k(t)
The last equation follows from the fact that O(K, L, ϕ)/K = O(K/L, 1, ϕ) · L/K =
o (k )/k. The steady state per capita capital stock can be determined by setting equation
(3.63) equal to zero in order to obtain
o (k∗ ) δ+n
∗
= (3.64)
k s
Capital and population growth. The third comparative static exercise is on the im-
pact of a change in the population growth rate, n, on the steady state level of per
o (k∗ ) δ+n
capita capital. We use G (k∗ ) = k∗ − s = 0 and the Implicit Function Theorem to
calculate the partial derivative of k∗ with respect to n as
∂k∗ ( ϕ, s, δ, n) Gn −1/s
=− =− (3.65)
∂n Gk∗ [o 0 (k∗ )k∗ − o (k∗ )]k∗ −2
The denominator of this equation is negative as is the numerator. Thus, multiplying

the negative nominator, with the minus sign in front of the fraction and dividing the
result by something negative gives a negative sign: an increase in n is associated with
a decrease in the per capita capital stock.
Graphical solution to the steady state. Figure 3.7 gives a graphical solution for the
steady state. The old "depreciation line" for the scenario without population growth
is inserted for comparison. Notice that the old equilibrium per capita capital stock is
at the intersection between s · o (k∗ ) = (δ + n)k∗ . The production function is expressed
in per capita units. The change in labor supply does not affect the functional form of
o (k). The new equilibrium is at k∗ where k̇
k(t)
= 0 is fulfilled at a lower k∗ .
Capital changes until the per capita capital moves to its steady state position in k∗
and remains constant afterwards. Does that mean that the economy does not grow
afterwards? Per capita output is indeed constant due to the constant per capita cap-
ital stock but total output is growing. Firms have more labor inputs with which to
produce. Higher output is associated with higher investments. Thus, production
increases due to i ) the increase of labor supply and ii ) the increase of capital. Never-
theless, an increase in the population growth rates is associated with lower per capita
3.3. ECONOMIC GROWTH AND TECHNOLOGICAL CHANGE 93
GDP.
Figure 3.7: Steady state equilibrium with population growth
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 (𝛿𝛿 + 𝑛𝑛)𝑘𝑘(𝑡𝑡)

𝛿𝛿𝛿𝛿(𝑡𝑡)
Consumption
𝑠𝑠×𝑜𝑜(𝑘𝑘 ∗ ) 𝑠𝑠×𝑜𝑜(𝑘𝑘(𝑡𝑡))
Investment
𝑘𝑘(𝑡𝑡)
𝑘𝑘 ∗
3.3 Economic growth and technological change
Building on the understanding of the population growth chapter, we will proceed to

extend the model by incorporating technological change. We will see that technolog-
ical progress can explain why countries grow over time. Not only total output grows
in the presence of technological improvement, but even per capita output rises over
time without being bounded by the steady state variables. Notice that technology is
denoted by ϕ and that we did not specify how ϕ affects output. In fact, there are three
ways how productivity may enter the production function.
Hicks neutral technology changes. The easiest modeling approach is called Hicks
neutral technological progress
Õ(K (t), L(t), ϕ(t)) = ϕ(t)O(K (t), L(t)) (3.66)
Changes in productivity, ϕ, shift the production function without affecting its shape.
Both input factors become more productive in the production process and the change
in efficiency is proportional.
Capital-augmenting technology. A second possibility is to assume that productivity

only affects capital. A higher productivity is associated with a more efficient capital
stock. Changes in productivity shape the production function. A production function
with capital-augmenting productivity reads
Õ(K (t), L(t), ϕ(t)) = O( ϕ(t)K (t), L(t)) (3.67)
Higher productivity is equivalent to a higher capital stock.
Labor-augmenting technology. A third possibility is to assume that productivity

only affects labor. A higher productivity is associated with more efficient labor.
Changes in productivity also shape the production function. A production function
with labor-augmenting productivity reads
Õ(K (t), L(t), ϕ(t)) = O(K (t), ϕ(t) L(t)) (3.68)
A higher productivity is equivalent to a larger labor endowment.
Graphical illustration of the different technology modes. Figure 3.8 compares the
three different ways of introducing productivity into production functions. The left
panel shows the isoquant associated with a Hicks neutral production function. An
increase in productivity, ϕ, shifts the isoquant from Ȳ to Ȳ 0 . Both factors become more
efficient when ϕ increases. Moreover, the efficiency gains are proportional. Increasing
both factors’ efficiency is equivalent to an increase in factor endowments by a factor of
ϕ. Constant returns to scale ensure that the isoquant shifts proportionally towards the
origin as proportionately less capital and labor is needed to obtain the same output
level.
The middle panel shows an increase in capital augmenting technology. Technological

changes shift the isoquant towards the origin. However, the shift is not proportional.
The isoquant becomes flatter due to the increase in efficiency associated with a higher
ϕ. To illustrate the reasons for why ϕ affects the shape of the isoquant, we analyze
its slope in Figure 3.8. The slope of the isoquant is smaller in b than in a. Remember
that the slope equals ∆K/∆L. Suppose there is a decrease in labor equal to ∆L = −1.
How much additional capital is needed to keep production constant at Ȳ? After the
capital-augmenting productivity change, the increase in capital needed to compensate
for the loss in labor is smaller because capital becomes more productive. Thus, the
slope in b is lower than the slope in a. The length of the triangle represents ∆L = −1,
which is equal at a and b. However, the height of the triangle, which represents
∆K, is larger before the capital-augmenting productivity change due to the reasons
discussed above.
96
Figure 3.8: Shifts of the production function due to population

growth
𝐾𝐾 𝐾𝐾 𝐾𝐾
𝑌𝑌�
𝑌𝑌�
𝑌𝑌�
c
�
a d
�
𝑌𝑌′ 𝑌𝑌′
�
𝑌𝑌′
b
𝐿𝐿 𝐿𝐿 𝐿𝐿
The right panel shows an increase in labor augmenting technological change. The iso-
quant moves towards the origin and becomes steeper due to the increase in efficiency
associated with a higher ϕ. To illustrate the reasons for why ϕ affects the shape of the
isoquant, we analyze the slope of the isoquant leaving the input of capital constant.
Now, the slope of the isoquant is higher in d than in c. Suppose a decrease in capital
equal to ∆K = −1. After the labor-augmenting productivity change, the increase in
labor needed to compensate for the loss in capital is smaller when labor is more pro-
ductive and is associated with a smaller slope in d than in c. The height of the triangle
represents ∆K = −1, which is equal at c and d. However, the length of the triangle
representing ∆L is larger before the labor-augmenting productivity change occurs.
The new steady state. We follow the established literature by assuming a labor-
augmenting technology
Y (t) = O(K (t), ϕ(t) L(t)) (3.69)
Without loss of generality, we assume that time is continuous. Technological change

ϕ̇
at rate g = ϕ > 0 affects ϕ at a constant rate and effective labor endowments are
increasing every period. Moreover, population grows at a constant rate n and the
capital stock at rate
K̇ (t) = sO(K (t), ϕ(t) L(t)) − δK (t) (3.70)
The production function depends on labor-augmenting productivity ϕ(t). It is conve-

nient to calculate per capita variables in terms of "effective labor" instead of popula-
tion. Notice, that this is a trick that allows us to interpret the effects on the per capita
variables. We are interested in per capita output computed in terms of population but
we need to know what happens to steady state per capita output in terms of effective
labor units. The effective per capita capital stock is
K (t)
k(t) = (3.71)
ϕ(t) L(t)
and changes over time according to
k̇(t) K̇ (t)
= −g−n (3.72)
k(t) K (t)
Output per effective unit of labor is
Y (t)
ŷ(t) = (3.73)
ϕ(t) L(t)
K (t)

= O ,1 (3.74)
ϕ(t) L(t)
= o (k(t)) (3.75)
However, per capita income y(t) = Y (t)/L(t) depends on population, not on effective
labor endowments
y(t) = ϕ(t)ŷ(t) (3.76)
= ϕ(t)o (k(t)) (3.77)
It is straightforward to show that per capita output y(t) grows if ŷ(t) is constant.
Equation (3.76) depends on a variable that is constant in the steady state, ŷ, and ϕ(t),
which is growing at constant rate g. Per capita output, y(t), is therefore growing at a
constant rate through time.
Combining equations (3.70) and (3.72) yields
k̇ sO(K (t), ϕ(t) L(t)) − δK (t)

= −n−g (3.78)
k(t) K (t)
k̇ o (k (t))
= s −δ−n−g (3.79)
k(t) k(t)
Proposition 1 Consider a Solow model in which production can be characterized by assump-

tion 1 and 2, population grows at rate n, and labor-augmenting technology grows at rate g.
The economy moves to its steady state variable k∗ , which is constant once the economy reaches
the steady state. The steady state variables can be determined by
o (k∗ ) δ+g+n
∗
= . (3.80)
k s
Output and consumption grow at rate g.
Proof: Investments are used in order to replace depreciated capital. Moreover, la-
bor endowments are increasing due to i ) population growth and ii ) better technol-
ogy. A higher labor endowments opens room for additional capital due to higher
investments. At given Capital stock, more labor input increases output, and output
increases investments and therefore capital. Growing capital leads to another round
of higher output, thereby increasing investment again. This process iterates on until
investment is again exactly equal to the change in the capital stock. The effective
capital stock remains constant once this point is reached. Does that mean that capital
remains constant as it was in the benchmark scenario without population growth and
technological change? No, the capital stock keeps on growing in order to match the
increase in population and technological progress. Equation (3.77) implies that per
capita output increases at rate g. The value associated with o ( g) is constant but ϕ(t)
increases by rate g every period.
Figure 3.10 shows the steady state equilibrium with technological change. The econ-
omy moves towards its steady state where investments per effective unit of capital
are equal to depreciation, plus the increase in capital due to population growth and
technological progress.
3.3.1 Dynamics with technological change
Figure 3.10 shows the transition dynamics in the Solow model for two different saving
rates. Growth rates in per capita capital stocks are plotted against the level of per
capita capital stock. The capital stock k∗ is associated with the steady state equilibrium
Figure 3.9: Steady state equilibrium with technological change
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 (𝛿𝛿 + 𝑛𝑛 + 𝑔𝑔)𝑘𝑘(𝑡𝑡)
Consumption
𝑠𝑠𝑠𝑠(𝑘𝑘 ∗ ) 𝑠𝑠×𝑜𝑜(𝑘𝑘(𝑡𝑡))
Investment
𝑘𝑘 ∗ 𝑘𝑘(𝑡𝑡)
where k̇ (t)/k (t) = 0. For k(t) < k∗ we find k̇ (t)/k (t) > 0. The per capita capital
stock is too low, which leads to an increase in k associated with positive growth
rates k̇(t)/k (t). To the right of k∗ we are in a situation, where k∗ is too high. Thus,
the capital stock must decrease to the level k∗ associated with negative growth rates
k̇(t)/k (t) < 0. The blue lines and arrows belong to the initial savings rate s. Suppose
that s increases to s0 . A higher savings rate fosters investments associated with a
higher per capita capital stock k∗∗ .
Figure 3.10: Dynamics due to technological change
𝑘𝑘̇(𝑡𝑡)
𝑘𝑘(𝑡𝑡)
𝑘𝑘 ∗ 𝑘𝑘 ∗∗
𝑘𝑘(𝑡𝑡)
0
3.3. ECONOMIC GROWTH AND TECHNOLOGICAL CHANGE
𝑠𝑠 ′ ×𝑜𝑜�𝑘𝑘(𝑡𝑡)�
− (𝛿𝛿 + 𝑔𝑔 + 𝑛𝑛)
𝑘𝑘(𝑡𝑡)
𝑠𝑠×𝑜𝑜�𝑘𝑘(𝑡𝑡)�
− (𝛿𝛿 + 𝑔𝑔 + 𝑛𝑛)
101
𝑘𝑘(𝑡𝑡)
3.4 Growth and internal migration
We studied the implications from internal migration in the beginning of this course.
Allowing for workers to move from agricultural to urban areas removed a friction
that allowed economies to grow more. Acemoglu (2008) presents a similar extension
of the Solow model of growth based upon the seminal work by Lewis (1954).
The model yields similar outcomes as derived earlier in our Krugman (1991) exten-
sion.
Lewis argues that growth in developing countries is below its potential due to labor
shortages in the urban area where manufacturing firms are located. Rural areas have
an ample supply of workers and a fraction of these workers is even unemployed.
Without mobility frictions, unemployed workers should move from rural to urban.
When firms in urban areas pay higher wages than the farms in the rural areas, all
workers would like to move from rural to urban. This is not what we observe in
developing economies. There is urbanization as we have seen in the introduction but
this process is slow and incomplete.
The model extension captures this idea by setting up a two-region model of a dual
economy. The rural region employs rural workers for agricultural activities, whereas
firms in urban regions produce modern goods. Workers are perfect substitutes and
can either be employed in farms or manufacturing firms. The difference is their
worker productivity changes when switching from agricultural to manufacturing.
There are mobility frictions that allow only a certain amount of workers moving from
rural to urban.
Time is continuous with infinite horizon and total population is normalized to unity.
we start the analysis at time t = 0, when LU (0) workers are in the urban area and the
remaining L R = 1 − LU (0) workers are living in the rural area. Workers in the rural
3.4. GROWTH AND INTERNAL MIGRATION 103
area are all working form agricultural farms producing food according to
Y A (t) = B A L R (t) (3.81)
which is a linear function. This formulation implies that labor productivity is constant
at B A > 0.
In the urban region, workers are all employed in the manufacturing sector, which
produces modern goods according to
Y M (t) = F (K (t), LU (t)) (3.82)
The variable K (t) denotes the stock of capital in period t, accumulated since the initial
period 0 by investments of the workers savings. We have discussed the function F in
great detail earlier in this course so we skip further explanations. Just to recap the
capital formation channel in this model, recall that investments are associated with
increasing capital stock according to
K̇ (t) = sF (K (t), LU (t)) − δK (t) . (3.83)
All variables are exactly as defined as previously in the discussion of the Solow model.
Notice that savings are entirely generated in the urban area as it is a fraction of the
modern goods output.
Workers are paid according to the value of their marginal product, which reads
∂F (K (t), LU (t))
wU ( t ) = (3.84)
∂L
w R (t) = B A (3.85)
The only reason for which workers have an incentive to move from rural to urban
are higher wages at urban, which is the case when marginal productivity in urban is
higher than marginal productivity in rural:
∂F (K (0), 1)
wU ( t ) = > w R (t) = B A (3.86)
∂L
This implies that there is always an incentive to migrate from rural to urban as wages
paid by manufacturing firms are always higher.
Mobility is assumed to be as follows:

= −µL R (t) if wU (t) > w R (t)







R
L̇ (t) = ∈ [−µL R (t), 0] if wU (t) = w R (t) (3.87)




if wU (t) < w R (t)

= 0

Case 1 is the scenario when urban wages are greater than rural wages, which is the
standard case. A fraction µ workers will migrate from rural to urban in each period.
Case 2 is a special case that defines the stopping criteria. As soon as wages in both
areas become equal, mobility switches from µ to 0 if there is no capital accumulation.
Case 2 also allows for migration rates between the maximum speed µ and 0, which
is relevant when the stock of capital is still increasing. A higher capital stock would
be associated with higher wages, thereby attracting workers to move from rural to
urban. The migration rates are sufficient to keep the balance between urban and rural
wages. Migration keeps the per capita capital stock constant at levels in accordance
with wU (t) = w R (t).
Case 3 is the unusual case that applies when urban wages dip below the agricultural
wages. Mobility would be zero as well.
Let us focus on the first period t = 0. Equation 3.86 tells us that there is always
some migration from rural to urban as wages paid by urban firms are always higher
compared to wages paid by rural firms. The most extreme scenario would be that all
workers are employed at manufacturing firms in urban at a given initial capital stock
3.4. GROWTH AND INTERNAL MIGRATION 105
K (0). If wages for this extreme scenario are greater in urban, wages for all LU < 1
must be higher as well as the marginal product of labor is increasing for lower labor.10
There is an incentive to move from rural to urban in period 0 and thus, there is also an
incentive to move in the periods after t = 0. However, the larger the amount of labor,
the lower the wages paid by urban firms. Sooner or later, wages are equalized so we
get case 2: migration switches from µ to 0 and remains zero afterwards. However,
migration is associated with capital accumulation when K (0) is below the steady state
capital stock. The increase in capital mitigates the decline in wages. It may even be
the case that wages in urban are increasing instead of decreasing due to an over-
compensation effect of capital formation on the wage rate. We are assuming that
there is migration from t = 0 on due to 3.86 and that the capital stock is not yet in
steady state as k (0) = K (0)/LU (0) is below its respective steady state level.
As before, we are expressing everything in per capita variables in order to derive the
capital formatio equation
k̇(t) = s f (k(t)) − (δ + µν(t))k (t) (3.88)
This equation is standard except of the variable ν(t), which captures - similar to
population growth - migration from rural to urban.
Notice that
L R (t)
ν(t) = (3.89)
LU ( t )
is a measure for the rural to urban labor share, which is decreasing when workers
move from rural to urban. The migration rate µ is constant but ν(t) is decreasing
over time. The migration rates are therefore declining over time. There is more
migration in the periods closer to t = 0 and the migration rates become rather small
in later periods. Finally, the migration rates become zero sooner or later. When
f (k (t)) − k (t) f 0 (k (t)) ≤ B A applies, migration stops and we are in case 2 or case 3.
10 Remember that each individual worker becomes more productive when they have relatively more
capital to work with.
There may still be some capital accumulation if capital is not yet in its steady state but
the changes are already too small to raise the workers’ urban wage above the rural
wage rate. The remaining workers in the rural area prefer staying instead of moving
to the city.
Let k̄ be the per capita capital stock for which urban and rural wages are exactly the
same so that
f (k̄ ) − k̄ f 0 (k̄ ) = B A (3.90)
Migration stops once we are in this scenario and the only remaining dynamic effects
are
k̇ (t) = s f (k (t)) − δk (t) (3.91)
There is zero growth of the labor endowments in urban areas. Thus, the capital stock
will be approaching its long run steady state level at
s f (k̂)
=δ (3.92)
k̂
Once this steady state level of per capita capital is reached, the economy is not grow-
ing any further.
Most interesting scenario. Acemoglu (2008) describes one interesting scenario that
describes the internal migration scenario of developing countries very well. Suppose
that the per capita capital stock in the initial period 0 is sufficiently low. Thus, there
is capital accumulation even without migration as k (0) < k̂ and k̇ > 0. The per capita
capital stock is also higher than the critical threshold k̄, which means that urban wages
are higher than rural wages. thus, there is migration from Rural to Urban. Acemoglu
(2008) assumes that the per capita capital stock is declining in the periods following
t = 0: k̇ < 0. This is the case when the additional labor is more than the additional
capital. We said that migration is rapid in the beginning of the structural change
process, which is why this scenario is indeed very likely. Two things may happen:
3.5. COMMUNITY ENFORCEMENT, MIGRATION AND DEVELOPMENT 107
1) The per capita capital stock reaches the lower ceiling k̂. Migration rates switch to
the one described by case 2 or case 3. Suppose we are in case 2. The migration rates
decline to levels sufficient for maintaining k̄. However, migration slows down further
and likely the migration rates become too low to maintain k̄: The capital to labor ratio
starts increasing and wages are again rising above B A . If there is still enough workers
living in the rural area, migration will start again at full rates. However, sooner or
later the equilibrium k̂ is reached. Migration and capital accumulation stops and
wages remain constant.
Alternatively, the wage may be decreasing in the beginning and increasing in a later
period when capital formation becomes stronger than the rural to urban migration.
The effects are ambiguous as migration and capital formation work into opposite
directions when wages are concerned. To summarize, the sign of the change in the per
capita capital stock k determines the sign of the wage change. When k is increasing,
wages are increasing as well. When k is decreasing, wages are following.
3.5 Community enforcement, migration and development
The model presented in this section puts rural to urban migration into the Solow
model of economic growth. Wages are again the crucial driver behind the migration
decision as urban areas pay higher wages. However, there are additional opportunity
cost of migration workers must take into consideration: Living rural areas yields
additional amenities that yield welfare on top of the welfare associated with the rural
income. Personal relationships and family ties for instance are expected to be closer
in rural areas where everybody knows each other. Migrants from rural to urban
must establish new networks as the family ties have little effect in urban areas. This
problem is more severe shortly after arrival in the city. Most migrants likely succeed
in building up new networks over time.
More details about the model can be found in the textbook by Acemoglu (2008), which
builds on the ideas formulated in the two papers by Banerjee and Newman (1998) and
Acemoglu and Zilibotti (1999).
Suppose there are two regions in a developing country: urban and rural. The econ-
omy is equiped with a constant labor force normalized to unity and workers can
choose the place of residency without paying any migration costs. Workers are ho-
mogeneous but their productivity is region specific as firms in rural areas use a more
sophisticated technology to produce modern goods. Workers in the rural area are
employed in agricultural firms producing food with traditional low-tech production
methods. Thus, workers become more productive when moving from rural to urban.
There is no capital accumulation in the model but output in the modern sector can
grow through technology spill-overs following
Y M ( t ) = X ( t ) F ( LU ( t ), Z ) . (3.93)
The variable X (t) denotes total factor productivity in the modern sector, which is
increasing over time. Acemoglu (2008) assumes some learning-by-doing externality.
Another specific input factor Z is constant over time. Z can be understood as a
constant capital stock. A fixed input factor implies that there is diminishing returns
to labor as marginal productivity of labor is declining when firms increase their labor
demand. Without unemployment, migration raises the supply of labor and firms
absorb the additional labor force at lower wages. Given the usual assumptions about
F, worker productivity is declining in LU . One may critically ask what happens to the
returns generated by the capital owners? We neglect this income as its effect on the
main outcomes is irrelevant. Taking care of those effects would complicate the model
a lot.
Economic growth is driven by the following technological progress

Ẋ (t) = ηLU (t) X (t)ζ (3.94)
where ζ ∈ (0, 1).
The opportunity costs of working in a modern firm located at Urban is another flow
cost that has to be paid every period in form of γ. Thus, the equilibrium without
migration is determined by
w M (t) − γ = w A (t) . (3.95)
As long as wages net of opportunity costs associated with living far away from the
home network are greater than the wage earned at rural, people decide moving to
Urban. Migration continues until wages minus opportunity costs are exactly equal to
the rural wage rate. The rural wage rate is a constant parameter B A determined by
the labor productivity of workers in agricultural.
Labor markets are competitive so that wages are dictated by the marginal productivity
and the level of total factor productivity.
∂F ( LU (t), Z )
w M (t) = X (t) = X (t)φ̃( LU (t)) (3.96)
∂L
A higher total factor productivity raises manufacturing wages, thereby attracting ad-
ditional workers from Rural. The higher labor supply counteracts the change in pro-
ductivity through lowering the marginal productivity until wages are back in equi-
librium. Thus, wages remain constant over time. Sooner or later, all workers have
moved to Urban and the productivity stops growing.
Notice that the function φ̃ is a variable that summarizes terms included in the first
derivative of F. The function is strictly decreasing in LU as capital is constant and the
usual Inada conditions apply.
We can use the labor market clearing condition that implies that both sectors can be
active only if wages in both sectors are similar.
w M (t) = γ + B A (3.97)
X (t)φ̃( LU (t)) = γ + B A (3.98)

X (t)

U
L (t) = φ (3.99)
BA + γ
The second line substitutes w M by X (t)φ̃( LU (t)), which is the marginal product of
labor, i.e. technology times the first derivative of the production function with respect
to LU . We are solving the equation in the midlle for LU in the last line, which may be
a bit confusing as a new variable φ is used. Without knowing the explicit functional
form of the production function F, the inverse φ̃−1 must be used in order to solve for
LU . For the sake of simplicity in notation, the inverse φ̃−1 is denoted by φ. Notice that
φ is a function that depends on the term X (t)/( B A + γ) because the first derivative
of the production function must be equal to the rural wage plus rural amenities γ
relative to the productivity at time t. This relationship is captured by
γ + BA
φ̃ LU (t) = , (3.100)
X (t)
which is strictly decreasing in LU as φ̃ is decreasing in LU . The term φ = φ̃−1 must be

increasing in LU as it is defined as the inverse of the first derivative of the production
function.
Let me explain this relationship using an example. Suppose that the production func-
tion is Y = L0.3 Z0.7 . The first derivative that determines the marginal product of labor
−0.7 1/0.3
is φ̃ = 0.3L0.3−1 Z0.7 and the inverse is φ =
φ̃Z
0.3 . Notice that
φ̃ = w M /X = ( B A + γ)/X
which explains why the inverse is a function of ( B A + γ)/X. The labor market clear-
ing and the technical progress conditions can be combined to
X (t)

U
Ẋ (t) = ηL (t) X (t) = ηφ
ζ
X (t)ζ (3.101)
BA + γ
The shape of the technological progress translates directly into the shape of the evo-
lution of the urban population share LU /L = LU due to the reformulation using
φ = φ̃−1 .
This formulation gives the typical shape of the urban population share depicted in
Figure 3.11. The urban population share starts at very low levels when technologi-
cal progress is at the beginning. The reason is that wages are too low to cover the
opportunity costs workers are paying when moving from Rural to Urban. However,
accelerating technological progress is associated with rapid increases in the urban
population share. When enough people are moving to Urban, the population share is
declining again due to the slow-down of technological progress. The urban popula-
tion share approaches unity when t gets large.
Notice that there are no migration frictions. Nevertheless, migrants do not move at
once due to competitive labor markets. The wage in Urban would decline below
the level of the rural wage when too many workers migrate at rates not supported
by technological progress. At each level of technological progress, there is a limited
amount of workers from Rural that can be absorbed in Urban, which gives a step-by-
step migration progress.
Let us focus more on the parameter ξ, which is the novel feature in this model. The
parameter captures an externality that benefits those living in rural areas and which
gets lost when workers move to the city.
The parameter itself enters the technological progress that determines X (t) through
the differential equation Ẋ (t). These rural amenities, Acemoglu calls them commu-
nity enforcement externalities, slow down technological progress by preventing rural
to urban migration. Migration is slower with amenities compared to the scenario
Urban population share
Time
Figure 3.11: Migration dynamics
without them. As a result, these amenities hinder economic development and reduce
welfare in the developing economy. However, Acemoglu argues that there is a trade
off between the dynamic gains through technological progress and wage growth in
both rural and urban areas and the static gains in the beginning of the economic de-
velopment. In period t = 0 the wage without the amenity would be lower as urban
manufacturers could pay lower wages. Wages in agricultural would be the same but
the amenity is another welfare effect that magnifies welfare in agricultural regions to
the level w A + ξ.
3.5.1 Empirical relevancy
The model extension neglects potential amenities in urban areas. One reason why
people migrate from rural to urban areas are facilities available only in bigger cites.
Figure 3.12 compares the livability of cities in East Asia to the livability of big cities
Figure 3.12: Livability in East Asia and other Middle-Income

Regions
Source: Gill, I. and Kharas, H. (2007)
in other middle-income countries. Interestingly, there is some significant difference

in the livability of cities across countries and this heterogeneity may matter for the
results derived from our model. So far, we have introduced factors pulling people
back to the rural areas. Pull-factors that motivate people moving into the bigger cities
may exist as well.
One could easily augment the model by including ξ U and ξ R . The former are the
opportunity costs of living in urban areas including better infrastructure, restaurants
and other things available only in the urban areas of a developing country. The
parameter ξ R is the well-known measure of rural amenities capturing family ties and
networks in areas more remote to the urban cities. However, distinguishing between
both type of amenities is not relevant if both are constant over time. We could easily
introduce ξ = ξ R − ξ U in order to solve the model as shown by Acemoglu (2008). This
kind of extension becomes more interesting when both ξ R and ξ U are changing over
time. It is likely that ξ U is declining over time due to agglomeration. Cities become
bigger and more crowded, associated with more social stress and longer commuting
time in bigger cities. ξ R may be changing as well. It may very well be the case that
the decline in rural population is associated with deteriorating family ties as families
become smaller. It may also be the case that staying in rural areas becomes more
tempting as people enjoy the less stressful life in rural areas coupled to increasing
availability of amenities in smaller cities due to some catching up process of these
rural areas.
Depending on the direction of the bias, economic growth would slow down or accel-
erate.
The bias depends on the change in net amenities ξ, which is the crucial determinant
of migrants’ decision to leave the rural area.
Figures 3.13 and 3.14 present some stylized facts that may explain why rural areas
became more popular in recent years. As argued above, the rising urban population
share is challenging for people living in bigger cities, which become more and more
Figure 3.13: Urban Transport and Road Safety

Figure 3.14: Livability in China vs. G-7

chaotic due to the increasing mass of people. Commuting time becomes longer, crime
is increasing and more people get killed in accidents when the city development gets
out of control.
In line with the model outcomes and the stylized facts presented in Figure 3.14, stabi-
lizing the city development is one crucial factor that can make the difference between
growth in Latin America and developing countries in East Asia. Especially regions
that grew out of the middle-income trap have very high scores in livability (Seoul,
Singapore, Hong-Kong).
Figure 3.15: Livability in cities: East Asia vs. Latin America

3.6. HUMAN CAPITAL IN THE SOLOW MODEL 117
3.6 Human Capital in the Solow model
A large strand of literature has emphasized the importance of human capital for eco-
nomic growth. Human capital is defined as the set of skills that any individual can
acquire during his or her lifetime and which are supposed to increase labor productiv-
ity. Intuitively, for any given task, a more skilled or trained individual will need less
time to perform it than a less skilled one. In this section, we will augment the Solow
model to allow for human capital as an additional factor of production. Similarly to
physical capital, we assume that individuals invest a given share of their income in
human capital. There exist more sophisticated growth models that incorporate the
micro-foundation of the decision surrounding education investments, however we do
not consider those models in this chapter. For now we aim to just study whether
the Solow model can be extended by adding more realistic features, such as human
capital as an additional factor of production.
The production function including human capital is
Y = O(K, H, ϕL) (3.102)
where we continue to assume that ϕ > 0 is the level of (labor augmenting) technology
in the economy.
With an additional production factor, we need to modify Assumption 1 and 2 above

so as to account for twice continuous differentially and constant returns to scale in
the three arguments, as well as to state the Inada conditions with respect to human
capital.
Assumption 3 The production function O(K, H, ϕL) is continuous, twice differentiable in

K, H and L, has constant returns to scale in its three arguments, and diminishing marginal
products.
Moreover, besides the already stated Inada conditions about K and L, we assume that
the following holds
lim O H (K, H, ϕL) = ∞ and lim O H (K, H, ϕL) = 0 (3.103)

H →0 H →∞
As with physical capital, we assume that during every period a constant fraction sh is
invested in human capital, which also depreciates at a constant rate δh . Moreover, we
continue to assume a constant rate of population growth and technological progress
as follows
L(˙t) ϕ(˙ t)
= n and =g (3.104)
L(t) ϕ(t)
As we have already seen in the previous section, the constant returns to scale as-
sumption allows us to define the production function in terms of effective physical
and human capital, by dividing each production factor by ϕL. Hence
K (t) H (t)

ỹ(t) = O , ,1 = o (k(t), h(t)) (3.105)
ϕ(t) L(t) ϕ(t) L(t)
Where k (t) and h(t) are physical and human capital per effective units of labor, respec-
tively. With the production function featuring another factor that can be accumulated,
and may grow over time due to investments, a steady state equilibrium will be one
where not only physical capital but also human capital grows at constant rate, to keep
up with the constant technological progress and population growth. The variables,
which remain constant over time, will be the ones defined in terms of effective units
of labor.
Using the same algebraic steps from the previous section, applied to both human and
physical capital, we obtain
k̇ (t) o (k (t), h(t))

= sk − δk − n − g (3.106)
k(t) k(t)
and
ḣ(t) o (k (t), h(t))
= sh − δh − n − g (3.107)
h(t) h(t)
The two equations above can be represented graphically in the h − k space. The
considerations we have made in the previous sections will help us understand the dy-
namics of physical and human capital in this context. Consider first equation (3.106).
Keeping h(t) fixed at a constant level, say h̄, the dynamics of capital per effective unit
of labor can be described exactly as before. Indeed, we know by the Inada conditions
that actual investment in physical capital sk o (k (t), h̄) will be steeper than break-even
investment (δk + g + n) for low levels of capital, but flatter for high levels of capital.
Moreover, the assumption of decreasing marginal returns states that sk o (k (t), h̄) will
increase at a decreasing rate as k increases. This assures that, keeping h(t) fixed at
a given level, there will only be one level of k (t) at which sk o (k̄, h̄) = (δk + g + n)k̄
and hence k̇ = 0. By repeating this exercise for any possible level of human capital
we obtain a curve in the h − k space, which represents all values of k (t) and h(t) at
which capital per effective unit is constant over time. Intuitively, such a curve will
be upward sloping (we formally derive the slope of the curve below). Indeed, the
higher the level of physical capital per effective unit in the economy, the higher will
be o (k∗ , h∗ ). Hence, the level of human capital per effective unit also has to be higher
(for constant δk , n and g) in order to keep physical capital per effective unit of labor
constant over time.
Now, let us start again from the perspective of human capital h̄ and consider a value
of capital per effective unit that is higher than k̄, for instance, k̃. Since k̄ was the level
of capital at which actual and break-even investment are equal, values of k higher
than that level are values at which the actual investment function starts to become
flatter than the break-even investment function. Hence by equation (3.106), for any
0
k̃ > k̄, k̇ < 0. The opposite happens at any level of k̃ < k̄. In this case, sk o (k (t), h̄)
increases by more than (δk + n + g)k∗ and k̇ > 0. This line of reasoning justifies the
direction of the arrows in Figure 3.16.
Analogous considerations can be made for the dynamics of human capital per effec-
tive unit, which is represented in Figure 3.17. Similarly to what we have done with
physical capital, we can first of all individuate the points in the graph where, for any
given level of k (t), ḣ = 0. This curve will also be upward sloping. If human capital is
higher, output per effective unit is also higher and the corresponding level of physical
capital which keeps ḣ = 0 must increase. Let us fix physical capital per effective unit
at the level k̄. There is only one level of h for which human capital per effective unit
remains constant, and we denote it by h̄. Let us consider point k̄, h̃‘ in the graph.
Since h̄ was the level of human capital at which actual and break-even investments
were equal, values of h higher than that level are values at which the actual invest-
ment function starts to become flatter than the break-even investment function. Below
the ḣ = 0 locus the opposite happens and human capital per effective unit will tend
to increase. The dynamics of human capital per effective unit is represented in Figure
3.16 with arrows pointing to the direction of changes of human capital.
Figure 3.16: Dynamics of physical capital per effective unit of labor.
𝑘𝑘̇ = 0
ℎ
ℎ�
𝑘𝑘� ′ 𝑘𝑘� 𝑘𝑘�

𝑘𝑘
Figure 3.17: Dynamics of human capital per effective unit.
ℎ�′
ℎ̇ = 0
ℎ�
ℎ�
𝑘𝑘�
𝑘𝑘
Steady state equilibrium. In a steady state equilibrium, the conditions
k̇ (t)
= 0 =⇒ sk o (k∗ , h∗ ) = (δk + n + g)k∗ (3.108)
k(t)
ḣ(t)
= 0 =⇒ sh o (k∗ , h∗ ) = (δh + n + g)h∗ (3.109)
h(t)
must be satisfied simultaneously. The equilibrium can be represented graphically as

in Figure 3.18. As shown in the graph, both curves are upward sloping. Indeed, take
equation (3.109): for a higher level of human capital per effective unit of labor, the
level of k that satisfies that equation has to be greater. The same holds for equation
(3.108).
We now proceed with proving the existence and uniqueness of the steady state equi-
librium. Existence follows immediately from the Inada conditions which affect the
shapes of the k̇ = 0 and ḣ = 0 loci. Consider the lower region on the left in Figure
3.18. The level of physical and human capital per effective unit is low. If starting from
that level capital increases, output per effective unit will increase by proportionately
greater amount, and so must h in order to satisfy ḣ = 0. The same holds for human
capital. If h increases the necessary change in physical capital to keep k̇ = 0 is higher
at low levels of human capital than at high levels. This intuitive line of reasoning
shows that the graph of the curve for ḣ = 0 is above the one for k̇ = 0 at low levels
of k. For a high level of capital, an increase in capital per effective unit of labor does
not change o (k, h) much, and h does not have to increase by a large amount in order
to satisfy equation (3.109). Hence, equation (3.109) is below (3.108) for a high level of
capital per effective unit of labor. The shapes of the curves ensure that there must be
a point at which they intersect at least once.
In order to prove that the steady state equilibrium is unique, we need to show that, at
the point where the two curves (each one with positive slope) cross, one is steeper than
the other. Moreover, we know that both curves are monotonically increasing, which is
another important property that secures the existence of an unique equilibrium. First
we prove that both curves are positively sloped. Consider the ḣ = 0 locus.
By rewriting equation (3.109) as
sh o (k∗ , h∗ ) − (δh + n + g)h∗ = 0
and using the implicit function theorem we get
dh sh ok (k∗ , h∗ ) sh ok (k∗ , h∗ )
=− = (3.110)
dk sh oh (k∗ , h∗ ) − (δh + n + g) (δh + n + g) − sh oh (k∗ , h∗ )
The numerator of the above expression is always positive. To see that the denominator
is also positive when along the locus where (3.109) is satisfied, we can consider that
sh o (k∗ ,h∗ )
when ḣ = 0, we know from (3.109) that h∗ = (δh + n + g). Since o (k, h) is
concave in h we can also write the following: 11
o ( k ∗ , h ∗ ) > o h ( k ∗ , h ∗ ) h ∗ + o ( k ∗ , 0) > o h ( k ∗ , h ∗ ) h ∗ (3.111)
Multiplying both sides of the inequality by the positive constant sh and rearranging
we can write
s o (k∗ , h∗ )/h∗ > oh (k∗ , h∗ )sh
|h {z }
(δh +n+ g)
which proves that the denominator of equation (3.110) in the ḣ = 0 locus is positive.
dh
Using exactly the same steps applied to equation 3.108 we can prove that dk > 0 when
k̇ = 0.
We now have to show that the two curves have different slopes, otherwise they would
never cross. Figure 3.18 actually shows that k̇ = 0 is always steeper than the ḣ = 0
locus and we provide a formal proof below.
We have derived the expression for the slope of the ḣ = 0 curve above. Using the
implicit function theorem we can also derive the slope of k̇ = 0, which is
dh (δ + n + g) − sk ok (k∗ , h∗ )
= k (3.112)
dk sk oh (k∗ , h∗ )
By setting (3.112) > (3.110) and following some simple algebraic steps (remember
that (δk + n + g) = sk o (k∗ , h∗ )/k∗ and (δh + n + g) = sh o (k∗ , h∗ )/h∗ ) we arrive at the
following inequality
o (k∗ , h∗ ) > ok (k∗ , h∗ )k∗ + oh (k∗ , h∗ ) h∗
Which is always true given that o (k, h) is concave in both k and h.
11 This follows from Assumption 1, which establishes the signs of the first and second derivatives with
respect to h. Moreover, the proof makes use of the definition of a concave differentiable function.
Recall that if a differentiable function o ( x ) is concave in its argument x we have that o ( x ) − o ( x ∗ ) <
o 0 ( x ∗ )( x − x ∗ ), for every x and x ∗ in the domain of o. Rearranging the expression and substituting x ∗
with k∗ and x with 0 we obtain the inequality in (3.111).
Comparative statics in the augmented Solow model with human capital. Having
understood the basics of the model, we can now ask how steady state physical and
human capital (and hence steady state output) per effective unit of labor change if
one of the exogenous variables changes. Let us consider the saving rate in physical
capital. Using the implicit function theorem on equation (3.108), we see that, similarly
to the standard Solow model, capital per effective unit increases as the sk increases.
Indeed,
∂k∗ o (k∗ , h∗ ) o (k∗ , h∗ )

=− = >0
∂sk sk ok (k∗ , h∗ ) − (δk + g + n) (δk + g + n) − sk ok (k∗ , h∗ )
We have already proved that (δk + g + n) − sk ok (k∗ , h∗ ) in the denominator is positive.

The numerator is of course also positive because it is the production function. Hence,
an increase in the saving rate in physical capital in the economy increases the steady
state physical capital. However, there will also be an indirect effect on human capital
too. Indeed, since h∗ increases as k increases (see equation (3.110)), the steady state
level of human capital will also be higher. This result is intuitive: with a higher level
of physical capital in the economy, total output will increase. Given that the economy
invests a fixed fraction of output on human capital, the new level of human capital
will also be higher. Thus, a higher saving physical capital saving rate has the effect
of increasing the steady state levels of both physical and human capital, and hence
total output will increase. The same derivations can be applied to the case in which
sh changes.
What happens to the variables in steady state if the rate of growth of population
changes? By using again the implicit function theorem on the equilibrium conditions
(3.110) and (3.112) we can see that both k∗ and h∗ decrease as population grows at
a faster rate. This result can be considered as an extension of what we have already
derived in the basic Solow model with only population growth. Intuitively, other
things being equal, if population increases at a faster rate, the economy will end up
with a lower level of physical and human capital per worker, and hence a lower level
3.7. EMPIRICAL APPLICATION ON INSTITUTIONS AND GROWTH 125
of GDP per effective units of labor.
Finally, notice that Figure 3.18 also includes the arrows showing the dynamics of
physical and human capital per effective units. It is immediately apparent from the
graph that the steady state equilibrium is stable, meaning that starting from any point
outside the equilibrium the system will tend automatically to it.
Figure 3.18: Steady state equilibrium in the Solow model with

human capital
ℎ
𝑘𝑘̇ = 0
ℎ∗ ℎ̇ = 0
𝑘𝑘
𝑘𝑘 ∗
3.7 Empirical application on institutions and growth
The main insight from the Solow model is that economies have limited potential for
sustainable economic growth without technological progress. This is one possible
answer to the question of why some nations failed in the past. As discussed in the
introduction to this chapter, many countries started to grow during their industrial-
ization phase. But comparing the situation in 1960 with today, we observe that only a
fraction of countries improved in terms of per capita GDP. One may argue that many
countries failed to catch up with the newest developments in production technologies
but still one may ask: why? In principle, it would have been possible to acquire new
technologies from the world market. So, why did those country not do anything to
acquire them? In other words, we are looking for what Acemoglu (2008) defines as
the fundamental causes of growth, that is, the deeper reasons that put some countries
in the position to accumulate more physical and human capital or better technology
than others.
As we shall see in what follows, many scholars argue that economic institutions are
the fundamental causes of economic growth. Institutions frame the rules and reg-
ulations behind economic activities. They may secure certain rights of citizens and
guarantee that they receive payoffs from the investments they make. We are mainly
interested in the payoffs for the provision of new production technologies. The de-
velopment of technology requires considerable investments in human and physical
capital that has to be built up over a long period of time. The aim of economic in-
stitutions is to provide economic agents with incentives to make those investments,
for instance by protecting the rights of the owners of the technology or of the other
production factors. Suppose that some new inventions could be copied by other free
riders. What would be the incentive for one individual to invest in the development
of those technologies? It would make sense to wait until others invent them. The
same holds for capital: there would be little incentive to invest in physical capital
(e.g. buying or renting land to establish a plant, buying machines and so on) or any
type of entrepreneurial activity without guarantees against expropriation. This line
of reasoning already hints that good economic institutions might indeed lie at the
core of economic growth and development. In this section we provide some sounder
arguments in support of the institutional hypothesis of economic growth.
Figure 3.19: Institutions and economic growth
Source: Acemoglu (2008)
Figure 3.19 illustrates the link between institutions and economic development in a
cross sectional graph. The graph is taken from the textbook by Acemoglu (2008)
and plots average protection against expropriation on the horizontal axis against per
capita income (measured in logs). The relationship is positive. Countries with bet-
ter economic institutions, approximated by protection against expropriation, tend to
have higher per capita GDP. Can we conclude that causality runs from institutions
to per capita GDP? Put differently, are countries able to boost their per capita GDP
by improving the quality of their institutions? Not necessarily. Causality may run in
both directions. It is likely that institutions provide the basis for economic growth
and thus contributed to higher per capita GDP, but it is also likely that higher per
capita GDP led to better institutions. For example, judicial services, patent offices and
police forces are expensive to maintain. Ideally, we would like to isolate both effects
in order to assess the importance of both channels.
Moreover, institutions may not be the only source of growth. Just to give one example:
the latitude of a country is also positively correlated with per capita income. Figure
3.20 shows a scatter plot with latitude on the horizontal axis and per capita GDP
on the vertical axis. Latitude is an exogenous variable that measures the distance
between a country’s capital and the equator. There is no obvious reason for reverse
causality. Latitude is exogenous to per capita GDP and there is no reason to believe
that per capita GDP affected the latitude of a country. But does that mean that latitude
affects per capita GDP? There are some explanations for such a causality. Climate
changes depending on the location of a country relative to the equator. Humidity and
heat are potential sources of inefficiency. Workers may be hampered by bad climate
associated with lower worker productivity. Remember that we identified workers’
productivity as the main source of economic growth. Technology was supposed to
increase workers’ productivity in a labor-augmenting manner but bad climate could
have an opposing effect on productivity.
The above examples help us to understand that it is extremely difficult to isolate the
most important determinant of economic growth by looking at simple correlations
between growth and each proposed fundamental cause. Indeed, there is a wealth of
factors that may influence economic growth simultaneously, and simple correlation
are just not able to disentangle them. In order to assess whether institutions are the
main determinants of economic growth we need to find a source of variation in in-
stitutions which neither affected the other potential causes (like geography) nor had
any direct impact on economic growth. Such a source of variation is called a "natural
experiment" in empirical jargon. Looking back at history might help find natural ex-
periments, since there have been events in the past which changed institutions while
holding constant the other determinants of economic growth. Acemoglu (2008) pro-
vides an anecdotal piece of evidence: the war in Korea and subsequent separation into
North and South Korea. Northern and Southern Korea had comparable economic per-
formance up to the point when both parts where split into a communist North Korea
and a more capitalist South Korea. We know that the institutions in North Korea
deteriorated soon after the division. In contrast, South Korea performed remarkably
well and became one of "Asian Miracle" countries with extraordinary high economic
growth rates. However, causality is a big issue here as well. Economic growth slowed
down in recent years and one may ask if that is a sign that institutions deteriorated in
more recent years. Again, institutions may serve as a necessary basis for technological
progress but there are many other factors that can explain the economic performance
of a country.
Although appealing, the main problem with the Korean example is that it is a nat-
ural experiment in a single country, involving the comparison between two extreme
institutional cases: the communist and capitalist systems. In order to reach satisfac-
tory conclusions on the general effect of institutions on economic growth, we need
a natural experiment which changed institutions in several countries and during the
same point in time. Acemoglu (2008) discusses the case of colonialism. In the six-
teenth century many European powers established colonies in Africa and America. It
is a historical fact that colonialism had a massive impact on institutions in the con-
quered areas: not only colonists changed the existing institutional settings, but they
also changed them in different ways across their colonies. It is also well known that
nowadays economic performance of the former colonies greatly differs: Australia,
New Zealand, Canada and the United States are among the most developed countries
in the world, while in some African or Latin America former colonies poverty and
famine reign. Specifically, it seems that the richest areas before colonization era are
the poorest ones nowadays. This phenomenon, called "the reversal of fortune", cannot
be explained by geography: if latitude determined growth, how can the same country
be very rich and very poor at two different points in time? As we will see in the
following discussion, the institutional hypothesis provides a convincing explanation:
colonialism changed institutions and through this channel it changed the growth path
of the former colonies.
Acemoglu (2008) argues that the mechanisms through which colonialism shaped insti-
tutions, and ultimately contemporary economic performance of the former colonies,
was driven by the economic needs and the settlement patterns of colonists. Areas with
abundant labor and natural resources experienced a so-called extractive colonialism,
i.e. they were used just as reservoirs for the extraction of precious metals and for
the exploitation of the local population through forced labor, taxes and tributes. In
these areas, which were at that time the most prosperous ones, European colonization
led to the deterioration of the institutional environment, because the colonists were
interested in dismantling the former economic institutions and certainly not in estab-
lishing private property protection, given that their aim was exactly expropriation.
Owing to the persistence of economic institutions over time, those previously pros-
perous areas found themselves unprepared to reap the benefits from industrialization
in the ninetieth century.
On the contrary, the less-densely populated and less urbanized areas with few natural
resources, where chosen by Europeans as settlement areas.12 The institutions in these
countries were positively influenced by colonization, since settlers established a social,
political and economic environment which resembled those in their origin countries.
The evidence that Acemoglu (2008) presents seems to validate this line of reasoning.
Indeed, he shows that former colonies that were more densely populated or had a
higher degree of urbanization (both indicators of economic prosperity) are the ones
with worse institutional quality and lower economic growth nowadays.
Another interesting mechanism leading from colonization to institutional change was

the settlement choices of the colonists. As we have already seen, settlers typically
chose to reside in less-densely populated areas where there was little or nothing to
extract, but this was not the only characteristic that attracted the settlers. A crucial role
might have been played by the healthiness of the areas, i.e. the life expectancy that the
settlers had there. Europeans were not used to the tropical climate, and they knew that
they would have hardly survived in the presence of tropical diseases. Hence the areas
featuring the shorter life expectancy for the colonizers (which very often corresponded
to the tropical and sub-tropical areas) were less likely to be chosen for settlement, but
12 Acemoglu et al. (2001) also show that settlement patterns were a function of the disease climate in
colonies. Europeans had never been exposed to malaria or yellow fever. When they came into contact
with these diseases they tended to die in large numbers. Hence, they chose to settle in areas where the
climate and disease burden was similar to their homelands. This explains why Europeans settled in
North America, Australia, New Zealand and to a lesser extent South Africa.
rather devoted to extractive activities. In order to prove this argument, we can look at
the correlation between settler mortality at the beginning of the colonization period
and institutional quality and per capita GDP nowadays. As shown in Figures 3.21 and
3.22, there is a positive correlation between settler mortality, economic institutions
today (again approximated by the average risk of expropriation) and per capita GDP.
The link between settlers’ economic needs and residential choices, institutional set-
tings in former colonies and their economic success seems to convincingly explain the
"reversal of fortune" phenomenon in former colonies. Moreover, the settler mortality
argument might also solve the puzzle surrounding the spurious positive correlation
between latitude and economic growth: the geographical position of a country might
just reflect how attractive the area was settlement by colonialists during the sixteenth
century, rather than the influence of climate on labor productivity or on the incentives
to invest in physical and human capital.
Figure 3.20: Geography and Economic Growth

Figure 3.21: Settler mortality and average protection against the risk
of expropriation
Figure 3.22: Settler mortality and per capita GDP

A. AN INTRODUCTION TO LINEAR REGRESSION 133
A An introduction to linear regression
Linear regression is one of the most widely used empirical methods in Economics. In-
deed, answering many theoretical economic questions ultimately rests upon whether
there exists a correlation or causal relationship between two or many variables. For
instance, in our analysis of growth convergence at the beginning of this chapter we
asked whether a negative correlation exists between initial per capital GDP and eco-
nomic growth.
The main difference between theoretical and empirical Economics is that in empirical
Economics we are not only interested in the sign of the correlation, but also in its size.
For instance, we are interested in knowing not only that countries with lower initial
level of per capita GDP grow faster than the others, but also how much faster they
grow.
The aim of this appendix is to illustrate the most relevant issues in regression analysis.
For expositional purposes, while explaining each concept we will refer to the growth
convergence regressions estimated earlier in this chapter.
As already noted, the Solow model implies that economic growth is inversely propor-
tional to the initial level of per capita GDP. The simplest formulation of the conver-
gence equation would be:
gi = α 0 + α 1 y i + ei (A-1)
Where we have omitted the time indexes for simplicity, but recall from equation (3.1)
in the main text that each variable is measured at different points in time. An impor-
tant clarification is necessary at this point: the model specified in (A-1) represents our
beliefs about the true relationship between the economic variables in the population,
i.e. in the "real world". In practice, we are never dealing with a dataset containing all
the countries in the world and all possible years, not only because there are countries
where data collection is impossible, but most obviously because we would need data
from the beginning of human history to nowadays. Instead, if you remember our
discussion at the beginning of the chapter, we are estimating (A-1) with a sample of
countries from 1960 to 2011. In a nutshell, econometric methods aim to understand
something about economic relationships in the population using only a sample of
data.
A-1 Dependent Variable, Independent variable and Regressions co-
efficients
The variable on the left hand side of equation (A-1) is called the dependent variable,
and represents the phenomenon that we want to explain. The variable on the right
hand side is called the independent variable or regressor. In the simple example we are
examining we have only one dependent and one independent variable. Discarding
the error term for the moment, which will be the focus of the next section, we can
graphically represent the relationship between gi and yi as a straight line in the gi − yi
space (and that is the reason why we talk about “linear” regression) with slope α1
and intercept α0 , as in Figure A-1. α0 and α1 in equation (A-1) are called regression
coefficients. The coefficient α0 , i.e. the intercept of the regression line, is the constant
term and has the interpretation of the value of the dependent variable when all other
regressors are zero.13 The coefficient α1 , the slope of the regression line, represents
the magnitude of the change in the growth rate for a unit change in y. Indeed,
dgi
= α1 (A-2)
dyi
The main objective of regression analysis is to obtain estimates for the regression
coefficients using actual data. Intuitively and non-technically speaking, econometric
13 Sometimes the constant does not have a meaningful economic interpretation. For instance, in our case,
it is not sensible to assume that a country has a per capita GDP of zero. In cases such as these the
constant has a purely mathematical interpretation - the level of the regression line.
estimates are the sample estimates of the unobserved parameters in the population.
Figure A-1: Regression line

𝒈𝒈𝒊𝒊
𝛼𝛼0 𝑔𝑔𝑖𝑖 = 𝛼𝛼0 + 𝛼𝛼1 𝑦𝑦𝑖𝑖
𝑑𝑑𝑑𝑑𝑖𝑖
𝛼𝛼1 =
𝑑𝑑𝑦𝑦𝑖𝑖
𝑦𝑦𝑖𝑖
A-2 From the theoretical to the empirical model: the error term
The model in (A-1) is made up of two parts:
• A deterministic part: α0 + α1 yi
• The error term: ei .
We have discussed all the components and interpretation of the deterministic part
above. We call it “deterministic” because, if this equation is exactly verified by the
data, the value of the growth rate gi would be determined only by the value of the
initial GDP per capita yi .
We began our discussion with the important distinction between the population and
a sample. Even if we know that the population of interest will never be observed, let
us imagine that such an ideal data set is available to us, i.e. one featuring the growth
rates and and per capita GDP of all possible countries and years. Imagine further that
you can use excel or any other software to plot the data on a two dimensional graph.
If the relationship between growth and GDP per capita were purely deterministic, you
would see that all data points lie on the regression line. However, what you will most
likely realize is that the actual data scatter around the regression line as in Figure
A-2. Some countries will have higher growth rates and some others a lower one than
what is implied by the regression line. Why does this happen? The main reason is
that the economic and empirical models are only approximations for the behavior of
real-world phenomena. It will never be possible to predict economic relationships
exactly, because some factors which influence the dependent variable will always
remain unobserved and will be missing in our regression model. For instance, as we
have also seen in this chapter, economic growth can be influenced by institutional or
cultural factors. These concepts are very difficult to measure and might cause the real
data to deviate from the relationship predicted by a simple regression.
For the above reasons we always introduce an error term in the analysis: by adding
the error term ei we admit that there are unobserved factors that may influence the
dependent variable besides our chosen regressors. If our empirical model is suffi-
ciently rich, we can be confident that the error will be small, but it will be never equal
to zero in an empirical model. As a matter of fact, we will never know how much
the error term will be, because, by definition, it remains unobserved. We will be just
able to compute estimates of the error terms, which we call residuals. Residuals are
the vertical distances between the estimated regression line and the true data at hand.
Without going into the technical details of how to estimate the regression coefficients,
it is worth mentioning that the most popular linear regression technique is Ordinary
Least Squares (OLS). In an OLS regression the coefficients are chosen so as to be sure
that, once the regression line is estimated, the difference between the points lying on
the estimated line and the true data is as small as possible.14 Although OLS is the
simplest way to estimate a regression, it represents the starting point of almost any
empirical work: before starting to estimate sophisticated models, researchers typically
check whether OLS regressions already hint at the existence of the relationship of
interest.
Figure A-2: Regression line and real data
𝒈𝒈𝒊𝒊
𝑔𝑔𝑖𝑖 = 𝛼𝛼0 + 𝛼𝛼1 𝑦𝑦𝑖𝑖
𝛼𝛼0
𝜖𝜖𝑖𝑖
𝜖𝜖𝑖𝑖
𝜖𝜖𝑖𝑖
𝑑𝑑𝑔𝑔𝑖𝑖
𝛼𝛼1 =
𝑑𝑑𝑦𝑦𝑖𝑖
𝑦𝑦𝑖𝑖
A-3 Standard errors and statistical significance
Besides estimating the regression coefficients, it is important in empirical research to

evaluate the margin of error of the estimates. To this aim, along with any regression
14 To be more formal: OLS minimizes the sum of the squares of the residuals.
coefficient, we also compute its standard error. Going back to table 2.1. in the main
text, we could ask: with which margin of error we can state that an increase of GDP
by 1000 increases the growth rate by 46 percentage points? We can remember the
following rule:15 the margin of error of any estimated coefficient is twice as much its
standard error. Hence, we can state that a one thousand increase in income increases
growth rate by 46 percentage points “plus or minus 0.35” (i.e. 2 × 0.000175 × 1000).
Is this margin of error acceptable? We have to compare it with the coefficient itself
(its absolute value). Intuitively, a coefficient of 1 with a margin of error of 0.5 is more
precise that a coefficient of 1 with a margin of error of 4. The rule of thumb is that the
margin of error (i.e. 2× the standard error) has to be lower than the coefficient. Or,
Absolute value of the coefficient

>2
Standard error of the coefficient
The ratio between the coefficient and standard error is called the t-ratio or t-statistic.16
Most econometric and statistical software provide the estimated coefficients together
with their associated standard errors and t-statistics. If the t-statistic is greater than
2 we say that the estimated coefficient is statistically different from zero. This means
that it is very likely that there exists a true correlation between the variables of interest.
But “how” likely will this be? This will be revealed by the significance level of the
coefficient. The usual levels of significance are 10, 5 and 1 percent. If we say that a
coefficient is significant at 10 percent level, it means that the likelihood that the true
relationship between the economic variables in the real world is zero (instead of a
the one suggested by the coefficient we have estimated) is 10 percent. Hence, the
lower the significance level, the more confident we can be, in statistical terms, that the
estimated coefficient are a true reflection of the actual relationship in the data and are
not a random outcome of an estimation performed with one specific data set.
15 We skip the technical details on how this rule of thumb is derived, but the interested reader may refer
to any introductory Econometrics or Introductory Statistics textbook.
16 The name derives from the fact that the random variable obtained by dividing the coefficient and its
standard error follows a Student-t distribution. For a formal proof or more technical details you can
refer to a standard Introductory Statistics or Econometrics textbook.
A-4 Linear regression with many variables
For illustrative purposes we have so far focused on regressions with only one inde-
pendent variable. However, most of the time we want to add more than one variable
to our regression model, for different reasons. First of all, the theory itself may sug-
gest more complex interdependencies than a simple binary relationship. For instance,
drawing from the Solow model, GDP per capita in 1960 is highly influenced by in-
vestment and saving rates, population growth and technological progress in 1960.
Those variables might also affect the growth rate between 1960 and 2011, since they
might be responsible for the transition from one steady state to the other. Hence, in
order to be sure that the coefficient α1 captures the degree of convergence towards
the steady state, keeping all other determinants of growth constant, we need to insert
those variables in our regression model whenever possible.
Moreover, given the richness of data available nowadays we may also include in the
model other variables which might not be directly present in our theory, but might
affect growth in the long run. For instance, we might want to add the exchange
rate and other monetary variables, indicators for whether the country experienced a
banking crisis in a given year, and so on. By including those factors in the regression
we achieve two results: we purge the effect of initial GDP per capita from other
confounding factors that might be related to it, and we get more precise predictions
of growth rates. In general, the main independent variables in the model (i.e. the
ones whose effect we want to measure) are typically referred to as the “regressors of
interest”, while all other variables which are included to rule out the effects of other
confounding factors on the dependent variable are usually called “controls”.
We can easily extend the model in (A-1) to many variables as follows:
gi = α1 yi + XiT δ + ei (A-3)
Equation (A-3) is an example of multiple regression, where the vector X contains all
additional regressors besides yi .17 What is the interpretation of the coefficients of a

multiple regression? Taking the partial derivative of equation (A-3) with respect to yi
we obtain:
∂gi
= α1 (A-4)
∂yi
The difference between this equation and equation (A-2) is that here we are dealing
with a partial derivative. The coefficient α can be interpreted as the effect of per capita
GDP on economic growth holding all other factors constant. We can now understand
more clearly why adding meaningful controls in a regression helps isolate the true
effect we want to detect: it purges it from the effect of other confounding factors.
A typical regression table. The concepts developed so far should now enable us to
interpret any regression table. To continue with our convergence regressions example,
we can examine Table 2.1. A good presentation of empirical results should enable
the reader to understand the model that has been estimated (in terms of method and
variables used), and to interpret the estimates in terms of magnitude and significance.
As we can see, Table 2.1. fulfills all the above requirements. First, the caption of the
table reports the model that has been estimated. The results of the four different
estimations of the model are reported in each column. In case the dependent variable
changes from one column to the other, it is customary to name the column with
the name of the dependent variable. In Table 2.1. this is not necessary because the
dependent variable is always the growth rate. The name of each row indicates the
name of each regressor. Typically, a regression table also reports the standard errors of
the coefficients. Moreover, the statistical significance is also indicated with the number
of “*” next to the estimated coefficients. A regression table should also indicate how
the variables are measured, so that the reader can get a sense of the magnitude and
the economic meaning of the presented coefficients.
17 Notice that if one element of the vector X is equal to 1, the corresponding element in the vector of
coefficients δ is the constant.
Chapter 4
Canonical Trade Models
The second main pillar of this course discusses the issues related to international
trade. As outlined in the introduction, economic growth and international trade are
important components of economic development strategies. Usually, both issues are
discussed independently of one another. However, there are many reasons to believe
that trade and growth are not so independent as they are treated in most of the
standard courses you may have already covered. Firstly, the transition dynamics may
be distorted if countries receive capital through foreign direct investments. Foreign
direct investment itself depends on the capital rental rates that are determined on
the world market if economies are open to foreign trade. Moreover, workers are able
to change their location. There are no labor market frictions that prevent workers
switching sectors if wages are different. One of the aims of this course is to combine
models of growth and trade in order to learn more about their interactions in general
equilibrium. To reach this goal in the final part of the course, we will study some of the
most important and parsimonious trade models, that is, the Ricardo and Heckscher
Ohlin models, before turning to blended models.
141
142 CHAPTER 4. CANONICAL TRADE MODELS
1 The Ricardo Model of Comparative Advantage
Why do some countries trade? The Ricardo (1821) model of comparative advantage
is one of the earliest attempts to answer this question in a very parsimonious way.
Why parsimonious? Most importantly because the assumptions about the production
process. Ricardo (1821) proposed a theory in which workers produce final goods
in different sectors. Labor productivity in his theory is constant and determined by
the state of technology in the respective sector in which the worker is employed.
Workers are able to switch sectors without frictions that prevent them from being
hired by particular firms. As long as some firms offer a higher income than others,
all workers will want to move to the high wage firms as a consequence. Trade takes
place between two different countries that are assumed to produce two goods using
only labor inputs. For the sake of simplicity, other factors such as capital or land
are omitted from the analysis. Thus, we refer to this model as a 2 × 2 × 1 model: 2
countries (Home and Foreign) trade goods from 2 different industries (Good 1 and
Good 2) that are produced by input of 1 factor (labor). The chapter builds on Feenstra
(2004) and Van Marrewijk (2012).
1.1 The model basics
Workers are assumed to be homogeneous. This implies that they have similar char-
acteristics, such as age, skill or ability. The implications of this assumption is best
described by the idea of clones working in factories with different technologies. Dif-
ferences in worker productivity stem from technology differences rather than from
worker heterogeneity. The latter channel is switched off by assumption. Productiv-
ity differentials among the same workers employed in different sectors and/or dif-
ferent countries emerge through differences in technology across sectors and across
countries. Only within a country-specific sector, all workers are identical and pro-
duce goods with the same labor productivity. Moving from one to another sector
1. THE RICARDO MODEL OF COMPARATIVE ADVANTAGE 143
changes the respective worker’s productivity depending on the technology differ-

ences between the two sectors.
Goods production. The assumptions on technology and the absence of more than
one input factor of production give rise to production functions with constant la-
bor productivity. Worker productivity is independent from the level of production
within a particular sector. One additional worker always increases production by the
same constant factor and the factor itself depends on the technology parameter. Put
differently, the marginal productivity of labor is constant. The production functions
themselves can be characterized by sector-specific technology parameters, denoted
by a, measuring the number of workers necessary to produce one unit of the output
good. The higher this input coefficient, the less efficient the technology within a par-
ticular sector i. The productivity of one worker is equal to the marginal productivity
GPi = 1/ai where GPi denotes the marginal productivity of a worker employed in
sector i.
The sectors within each country can be distinguished by index i. One unit of good i
must be produced by input of ai workers. The production functions of both countries
can be summarized by the following linear production functions
y1 = (1/a1 ) L1 , y2 = (1/a2 ) L2 ,
y1∗ = (1/a1∗ ) L1∗ , y2∗ = (1/a2∗ ) L2∗ . (4.1)
Output in sector i is denoted by yi and it depends on the total number of workers

employed in that sector, that is, Li . Put differently, one worker can produce 1/ai
units of the output good. Total production in each sector is therefore equal to the
total number of workers employed times individual worker productivity. All foreign
functions and variables are labeled using an asterisk.
Notice that consumers are indifferent between consumption of good i produced by

foreign firms and consumption of good i produced by domestic firms. Good i is
homogeneous and therefore identical across countries. Import or export may still be
beneficial for both Home and Foreign because of technology differences.
The endogenous variable in this model is the sector-specific labor input, Li . However,
labor inputs are limited by the total endowment. For the time being we assume that
there is no migration between countries and each country’s level of unemployment is
zero. Thus, the total labor endowment governs the available amount of inputs in both
sectors. We account for this restriction by introducing the following full endowment
condition:
L1 + L2 = L̄ , L1∗ + L2∗ = L̄∗ . (4.2)
All workers are employed either in sector 1 or in sector 2 so that the sector-specific
inputs of labor add up to total endowment L̄ in Home and total endowment L̄∗ in
Foreign.
We reduce the number of equilibrium conditions to one equation with two unknowns
in each country by combining the production functions (4.1) and the full employment
conditions (4.2) as
y1 a1 + y2 a2 = L̄ , y1∗ a1∗ + y2∗ a2∗ = L̄∗ , (4.3)
which can be solved to give the production possibility frontiers by solving for y2 and
y2∗
L̄ y a L̄∗ y∗ a∗
PPF: y2 = − 1 1 , y2∗ = ∗ − 1 ∗ 1 (4.4)
a2 a2 a2 a2
Figure 4.1 shows the PPFs for Home and Foreign. It is worth mentioning that the
slope of the PPFs represent the opportunity costs of production in sector 1. Suppose
you want to produce one more unit of good 1. How many goods of sector 2 must be
given up for the additional output in sector 1? The answer is that a1 /a2 goods have
to be given up in order to get the extra amount of labor input required in sector 1,
which is ∆L1 = a1 . How many goods could have been produced in sector 2 using this
particular amount of labor? One worker can produce 1/a2 units of good 2. Therefore,
the a1 workers could have produced a1 /a2 goods in sector 2.
Figure 4.1: The production possibility frontier at Home and Foreign
𝑦𝑦2
��
𝐿𝐿∗ 𝑦𝑦1∗ 𝑎𝑎1∗
𝑦𝑦2∗ = − ∗
𝑎𝑎2∗ 𝑎𝑎2
��
𝐿𝐿 𝑦𝑦1 𝑎𝑎1
𝑦𝑦2 = −
𝑎𝑎2 𝑎𝑎2
𝑎𝑎1
𝑎𝑎1∗ 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = −
𝑎𝑎2
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = − ∗
𝑎𝑎2
𝑦𝑦1
Moreover, the shape of the two budget constraints is different at Home and at Foreign
a1∗ a1
by assuming without loss of generality that a2∗ > a2 . The foreign production possi-
bility frontier is steeper than the domestic production possibility frontier. The slopes
represent the opportunity costs of production. Thus, opportunity costs in the foreign
sector 1 are higher than the opportunity costs in the domestic sector 1. This result
solely stems from the assumptions imposed on technology.
Optimal production program. Production in the closed economy must equal con-
sumption because goods cannot be imported in order to augment production facilities
and all production must be sold to domestic consumers. As a consequence, the invisi-
ble hand guides producers and consumers to the point where no resources are wasted
and production equals consumption. Every worker receives a wage w, which can be
spent on the two different commodities produced in the two different sectors. First
we derive all equilibrium outcomes for the autarky scenario in the closed economy
setting of the model. Those results serve as benchmark for evaluating the effects of
trade liberalization on welfare. We are going to ask how wages and utility change
when markets open up to free trade. The answer can be given comparing the key
variables before and after the goods market liberalization.
Consumption is limited by budget E that can be spent on C1 and C2 evaluated at

goods prices P1 and P2 . Total expenditure can’t exceed the household’s budget E,
which translates into
E C P E∗ C1∗ P1∗
C2 = − 1 1 , C2∗ = − ∗ . (4.5)
P2 P2 P2∗ P2
The budget constraint can be used in order to solve for the optimal consumption
and production program but before doing so we have to characterize the consumers’
preferences. Every possible consumption program can be associated with a particular
utility level that allows the consumer to discriminate between the different outcomes.
By consumption program we refer to a particular consumption mix C1 , C2 . To keep
things as simple as possible we assume that all consumers are represented by one
consumer consuming the complete production output from sector 1 and sector 2.
Preferences enable consumers to evaluate total consumption by
U (C1 , C2 ) = C1α C21−α . (4.6)
Unless otherwise stated, we assume that preferences are Cobb Douglas or at least
have comparable properties. Notice that one particular level of utility Ū can be asso-
ciated with many different combinations of C1 and C2 for which the level of utility,
U, is identical. We say that the representative consumer is indifferent between all
combinations of consumption programs associated with the same level of utility, Ū.
The preferences formulated in (4.6) allow consumers to substitute consumption of one
good with consumption of the other good, leaving the total level of utility constant.
The consumption ratio between both goods that is necessary to keep utility constant
is governed by the elasticity of substitution α and by the level of consumption C1 and
C2 . Generally speaking, we know that more consumption always increases utility.
Shifting the indifference curve outwards is therefore associated with higher welfare.
Rational consumers target the highest indifference curve that can be reached condi-
tional on the available budget of the household. The respective indifference curve can
be found using a utility maximization problem
max U (C1 , C2 ) subject to E = C1 P1 + C2 P2 (4.7)

C1 ,C2 ,λ
that has to be solved in both countries through changing C1 and C2 until the maximum
level of utility is reached. The solution must not violate the budget constraint (4.5).
We can use the Lagrangian method in order to compute the equilibrium consumption
levels through
L = C1α C21−α + λ( E − C1 P1 − C2 P2 ) , (4.8)
where λ is the shadow value of income. The first order conditions of this problem can
be derived through the first derivative of the Lagrangian with respect to C1 and C2 ,
∂L ∂L
= αC1α−1 C21−α − λP1 = 0 , = (1 − α)C1α C2−α − λP2 = 0 . (4.9)
∂C1 ∂C2
Both can be solved for λ in order to obtain
P1 α C2 ∂U/∂C1
= = . (4.10)
P2 (1 − α) C1 ∂U/∂C2
The highest indifference curve is the one that is tangential to the budget constraint,
which is the indifference curve with a slope equal to the relative goods price. More-
over, the consumption point must fall together with a feasible production point lying
on the production possibility frontier. Consumption points above the production pos-
sibility frontier may yield a higher utility but they are not reachable at the given bud-
get level. Points below the frontier are feasible but resources would be wasted. Thus,
the optimal solution is the consumption point on the PPF that fulfills the properties
derived in (4.10).
Wages in the Ricardo model. Labor markets are assumed to be competitive. Firms
in both industries compete for workers who may instantaneously switch between
sectors. Both sectors have to pay the same wages in order to avoid giving incentives
to move to the sector that pays a higher wage. Moreover, workers receive a wage
that equals the value of their marginal product evaluated at equilibrium goods prices.
Suppose that workers receive a wage w1 in sector 1 and w2 in sector 2, both being
equal to the value of the workers marginal productivity within the respective sector.
One additional worker in sector i generates additional output, 1/ai , of the respective
good sold for price Pi . In equilibrium w1 = w2 ⇒ P1 a11 = P2 a12 . Otherwise only one
sector, the sector that pays higher wages, can be active. We use this condition to derive
1 1
P1 = P2 , (4.11)
a1 a2
which can be rearranged in order to obtain
P1 a
= 1 (4.12)
P2 a2
Wages in both sectors can be equal if and only if the slope of the budget constraint
( PP12 ) equals the slope of the production possibility frontier ( aa21 ).
Notice the difference between absolute and relative prices. Wages depend on the
absolute price in a given sector, whereas most of the equilibrium conditions derived
in the remainder of this course depend on relative prices denoted by a small p. We will
always distinguish between the autarky and world market levels of certain variables,
which are distinguished by index w (world market) and index a (autarky).
Putting the whole model together. The optimal consumption and production pro-
gram can be derived in Figure 4.2. We draw the production possibility frontiers, the
budget constraints and the utility indifference curves for Home and Foreign in sepa-
rate graphs. The PPF and the budget constraints fall together under autarky and the
slope of the budget constraint must be equal to the relative goods price. The relative
goods price itself must be equal to the ratio of the input coefficients and the consump-
tion/production points must lie on both the PPF and the budget constraint. The only
way that all conditions are fulfilled is if both PPF and budget constraint lie together.
The first order condition of the utility maximization problem is fulfilled in Y1 , Y2 and
Y1∗ , Y2∗ in Figure 4.2. Obviously, the slope of the indifference curve equals the relative
goods price, represented by the slope of the budget constraint. The consumption
point is feasible because it lies on the PPF. Notice that we are using capital letters to
denote absolute production levels: for example, Y1 is the level of production in sector
1.
Figure 4.2: Optimal production/consumption under autarky
Home Foreign
𝑦𝑦2 𝑦𝑦2∗
𝑌𝑌2∗
𝑈𝑈 ∗
𝑈𝑈
𝑌𝑌2
𝑌𝑌1 𝑦𝑦1 𝑌𝑌1∗ 𝑦𝑦1∗

The consumption points are identical to the production points under autarky. We
therefore neglect them on the graph.
1.2 Gains from trade
Suppose that the economies open up to free trade. Producers can export goods at
world market prices. Consumers can choose between purchasing foreign goods in
order to substitute consumption of domestically produced goods. Thus, the autarky
prices determined in the previous paragraph are no longer binding. That said, we will
show that both economies are better off under free trade due to positive gains from
trade. Producers will specialize according to comparative advantage: every country
exports the good it can produce with lower opportunity costs of production and it
imports the good where it has an opportunity cost disadvantage. Due to the linearity
of the production function both countries completely specialize in the production of
one of the two sectors. All workers move into the comparative advantage sector and
the other sector shuts down completely.
Sound knowledge of the mechanics behind the effects at work is the key to understand
the relationship between the relative goods prices under autarky. Suppose the two
countries in our model can be characterized by
P2 P∗
1/p a = > 1/p∗a = 2∗ . (4.13)
P1 P1
This relationship tells us that the assumptions of the model are such that the relative
autarky price of good 2 is higher in Home than in Foreign. But how do such differ-
ences emerge? Notice that the relative autarky prices reflect the opportunity costs of
production under autarky. Opportunity costs can be explained by international tech-
nology gaps. The opportunity costs of producing good 2 are lower in Foreign than
in Home if the foreign country has a comparative cost advantage due to a relatively
more advanced technology within this particular sector. It is the term "relative" that
is important for understanding the determinants of international trade. The absolute

advantage is important in determining the relative price ratio under autarky. Instead,
the relative technology advantage compares one sector’s productivity relative to the
other sector’s across countries. The price differences across countries under autarky
allow us to derive the pattern of trade intuitively.
Consumers in Home have an incentive to purchase good 2 from foreign suppliers as

the good is relatively cheaper there. The autarky price differential generates incen-
tives for domestic firms in sector 1 to produce more. Their output can be sold for
higher prices in the foreign market. Thus, domestic resources shift towards sector 1
and the supply gap of sector 2 goods can be bridged through imports from Foreign.
For the same reason we find the opposite pattern in the foreign country. From a for-
eign perspective the world market price must lie between the autarky prices as well.
Having said that, it is easy to verify that foreign consumers can purchase good 1 from
firms in Home and focus their resources on sector 2 instead. Moreover, goods prices
adjust until trade between the two countries is balanced.
The free trade scenario is depicted in Graph 4.3. The slope of the budget constraint
equals the relative world market price pw = P1w /P2w , which is the same in both Home
and Foreign. Thus, the slope of the home and foreign budget constraints must be
equal under free trade. Moreover, we already know that the relative world market
price lies between the relative autarky prices: p a < pw < p a∗ . The budget constraint
in Home becomes steeper and the budget constraint at Foreign becomes flatter when
going from autarky to free trade.
The initial consumption point is no longer optimal because the representative con-
sumer can reach a higher indifference curve. The budget constraints can be shifted
outwards until they intersect the PPFs at the specialization points Y1 in Home and
Y2∗ in Foreign. This scenario is depicted in Graph 4.3. The first order conditions for
the utility maximization problem are fulfilled at the new consumption point, where
the slope of the indifference curve is equal to the relative world market price. This
point is represented by the tangency point between the indifference curve and the
budget constraint located at C2 , C1 and C2∗ , C1∗ . We can easily verify that the respective
world market price is such that world trade is balanced. Production in Home is Y1 .
All workers move into sector 1. Consumption is at a lower level C1 than production,
which is at Y1 . The difference between production and consumption is exported to
Foreign. Home imports are determined by the consumption level of good 2. Sector 2
is inactive in Home so that all consumption of good 1 is imported from abroad. The
opposite pattern can be found in Foreign. It can be seen that home imports of good
2 equal foreign exports of good 2 and home exports of good 1 equal foreign imports
of good 1. The world market price is such that trade between the two countries is
balanced. Moreover, we can see that both countries are better off. Home and Foreign
can reach a higher indifference curve U w and U ∗w .
Figure 4.3: Optimal production/consumption under free trade
𝑦𝑦2 𝑦𝑦2∗
𝑌𝑌2∗
𝐸𝐸𝐸𝐸𝐸𝐸∗
𝐶𝐶2∗ 𝑈𝑈 ∗𝑤𝑤
𝐵𝐵𝑅𝑅∗
𝐵𝐵𝐵𝐵
𝑈𝑈 𝑤𝑤 𝑈𝑈 ∗𝑎𝑎
𝑃𝑃𝑃𝑃𝑃𝑃 𝑈𝑈 𝑎𝑎
𝐶𝐶2
𝑃𝑃𝑃𝑃𝐹𝐹 ∗
𝐼𝐼𝐼𝐼𝐼𝐼 𝐼𝐼𝐼𝐼𝐼𝐼∗
𝑌𝑌2 𝐸𝐸𝐸𝐸𝐸𝐸∗ 𝑦𝑦1 𝑦𝑦1∗
𝐶𝐶1 𝑌𝑌1 𝑌𝑌1∗ 𝐶𝐶1∗
1.3 Trade liberalization and income
The gains from trade are one source of welfare improvements. Consumers are better
off due to higher utility. However, the concept of utility is somewhat difficult to inter-
pret. It is easy to use utility in order to discriminate between different consumption
bundles but the value itself has little meaning. It is extremely difficult to find a proxy
for utility in the available data but the model also permits analysis of the effects of
trade liberalization on income. Nominal income is increasing in both countries due
to the fact that workers move into the sector of comparative advantage. These effects
have to be weighted against the effects of a higher world market price for the export
good, which can be associated with a loss of purchasing power for this particular
good. Trade liberalization may not only reduce the positive income effects that stem
from higher wages, the purchasing power of consumers may decline if goods’ prices
increase by more than nominal income. Such a scenario is fairly implausible. We al-
ready know that there are positive welfare effects in both countries. It can be verified
that real wages are either increasing or at least stagnant.
Home specializes in sector 1 meaning that the nominal wage of all workers is deter-
mined by w/P1 = GP1 · P1 /P1 = 1/a1 . Whether the real wage increases or decreases
depends on which price we use to calculate the real wage. Measured in good 1 prices
we find that the real wage remains constant when going from autarky to free trade.
But consumers purchase good 1 and good 2. Calculated in good 2 prices we find that
the real wage increases due to trade liberalization as w/P2 = GP1 · P1 /P2 applies. The
marginal productivity of workers remains constant but the relative goods price P1 /P2
increases when going from autarky to free trade. We know that the relative price of
good 1 is higher under free trade than under autarky. We also know that consumers
purchase both goods so that, on average, real income and the workers’ purchasing
power must increase after countries open up to international trade.
Can we predict how real wages develop in Foreign? Sector 2 is the active industry in
Foreign. Measured in prices of good 2, foreign real wages are constant as w∗ /P2∗ =
GP2∗ · P2∗ /P2∗ = 1/a2∗ . In prices of good 1, the real wage is going up as well. Again, we
can compute the real wage using w∗ /P1∗ = GP2∗ · P2∗ /P1∗ and we know that the relative
price of good 2, which is the export good from a foreign perspective, rises as well. We
find that depending on the price used to construct real wages, the foreign real wage
either remains constant or increases as well.
1.4 Equilibrium goods prices
The relative world market price p f must be such that relative demand is equal to
relative supply. Suppose that Home has a comparative advantage in the production
of good 1 so that a1 /a2 < a1∗ /a2∗ . Relative world demand for good 1 (relative to good
2) is equal to
(C1 + C1∗ )/(C2 + C2∗ ) .
Aggregate demand of good 1 in both countries is related to sector 2 demand in both

countries.
Relative supply if Home specializes in good 1 reads
( L̄/a1 )/( L̄∗ /a2∗ ).
All sector 1 goods are produced by Home and all sector 2 goods are produced by
Foreign. The relative supply curve becomes
( L̄/a2 )/( L̄∗ /a1∗ )
if Home specializes in good 2.
We know that the relative world market price must be located within Area II (see
Figure 4.4). Notice that the relative supply curve in that particular segment is a
vertical line. Both countries specialize according to the comparative advantage as
long as the world market price lies between the two autarky prices. Supply is perfectly
Figure 4.4: Optimal world market prices

(1,1)
p
p∗a Relative Supply
Area II
p
pa Relative Demand
_ _
(L/a1 )/(L∗ /a∗2 ) (y1 + y1∗ )/(y2 + y2∗ )
inelastic within the interval [ p a , p∗a ]. Supply is constant but demand reacts to changes
in prices. Relative demand is monotonically increasing in the relative price for good
2. The higher the price of good 2 relative to the price of good 1, the more consumers
substitute good 2 by good 1.
The relative supply curve is not defined for relative goods prices above the foreign
relative autarky price of good 1. If the world market price is higher than both autarky
prices, both countries would want to specialize in sector 1. Production in sector 2
would be zero in both countries and the denominator in the relative supply curve
would be zero, which is not defined.
Relative supply is zero for all relative world market prices below the autarky price
at home. Both countries would have an incentive to specialize on sector 2. The
nominator of relative supply would be zero.
The only feasible world market price is the price associated with the point of intersec-
tion of the relative supply and demand curve.
2 The Heckscher-Ohlin model
So far we have seen how technology differentials can serve as a source of comparative
advantage. Countries specialize in sectors with lower opportunity costs of produc-
tion. Country characteristics such as endowments or the size of an economy have
not yet been taken into consideration. Intuitively, factor endowments should be an
important source of comparative advantage. For instance, the US is capital abundant
and it tends to export goods that are produced with a higher capital-intensity. At
the other extreme, we observe many developing countries that are labor abundant
and disproportionately produce labor intensive goods. China is among the most im-
portant textile producers in the world and well-known for being a labor abundant
country.
The Heckscher-Ohlin model provides a potential explanation for why endowments

drive trade patterns. Countries differ with respect to endowments rather than technol-
ogy. Technology is still different across sectors but it does not differ across countries.
We follow established textbooks that base the analysis on a standard 2 × 2 × 2 model
with two countries producing two goods using inputs of the two factors, capital and
labor. One sector is equipped with a capital intensive technology and the other sector
uses a labor-intensive technology. Both countries have the same technology within a
particular sector so that any differences emerging between the two countries are solely
due to endowment differences. That allows us to discriminate between the mecha-
nisms highlighted by Ricardo and Heckscher-Ohlin. Notice that both channels may
have some explanatory power for trade between developed and developing countries.
Part of the explanations may be based on technology differences and part of the in-
2. THE HECKSCHER-OHLIN MODEL 157
ternational shipments may be due to endowment differences. Nevertheless, the aim

is to isolate both mechanisms in order to evaluate their relevancy.
The model description is based on chapter 4 and 5 in Marrewijk (2012), as well as the
description in Feenstra (2004).
2.1 The 2 × 2 × 2 model
Suppose that production takes place according to a standard Cobb-Douglas produc-

tion function
Y1 = K1α1 L11−α1 ; Y2 = K2α2 L12−α2 . (4.14)
which combines labor, L, and capital, K, in order to obtain output Y1 and Y2 . Index
1 stands for some capital intensive manufacturing goods and index 2 stands for an-
other labor intensive sector. Technology is identical in both countries but it differs
across industries within the respective economy. Home and Foreign manufacturing
firms use identical technologies but the production functions in sector 1 and sector
2 are different. Differences are emerging through the parameter α, which governs
the degree of substitution between labor and capital in the production process. The
parameter is called the elasticity of substitution in the remainder of this chapter. We
impose the assumption that 0 < α < 1 holds. The bigger α, the more important is
capital in the production process. The capital intensive manufacturing sector 1 has a
higher α compared to the labor intensive sector 2.
Why do we say then that both factors are substitutes? We know that there is a ten-
dency that sectors with higher α produce with higher capital intensity. Both sectors
would be able to produce with high capital or high labor intensity if the two input
factors are substitutes but taking factor prices into consideration we can conclude that
the capital to labor ratio in the production will be different. Two industries with dif-
ferent α still face identical factor prices due to the assumption of competitive factor
markets. If wages are higher than capital rental rates both industries would want to
substitute the more expensive factor with the less expensive one but the sector with
higher α will always substitute relatively more capital than labor at identical factor
prices. Thus, α gives us information about relative factor intensities across different
sectors. We say that sectors with higher α tend to produce more capital intensive
goods.
Cost minimization. For given production functions, firms have to decide about op-
timal input intensities. We have already discussed the role of factor prices, w and
r, briefly in the last paragraph. Graph 4.5 gives a graphical solution of the problem
of the firms’ cost optimization. Suppose that firms produce according to (4.14) with
linear production costs
C1 = wa L1 + raK1 , C2 = wa L2 + raK2 , (4.15)
where we distinguish between production costs in the two sectors 1 and 2. Both
sectors face identical factor prices but total production costs vary with the amount of
capital and labor used in each sector. Thus, total production costs are given by C1
in the manufacturing sector 1 and C2 in sector 2. The two equations in (4.15) also
represent the isocost curves. The name isocost curves stems from the fact that they
represent all the possible input combinations of Li and Ki , that yield identical cost at
given factor prices. With fixed levels of C1 in the capital intensive manufacturing and
C2 in the labor intensive sector, one can use the cost function in order to search for
all input combinations of labor and capital that yield exactly the level of production
costs that was chosen at the given factor prices. The isocost curves then summarize
all of those input combinations associated with the specific level of costs and input
prices.
Firms in each sector optimize their production process by choosing the optimal level
of K and L subject to the factor prices and technology in each sector. We solve this
problem for the two output levels Ȳ1 = 1 and Ȳ2 = 1. The choice of those two
output levels is arbitrary. One may choose different levels as well but it appears
convenient to set both at a level equal to unity. Once we have solved for the optimal
labor and capital input requirements, for the particular case that production equals
unity (respectively in each sector), we are able to generalize those solutions to all
possible scenarios in a subsequent step. Knowing the optimal factor intensity at given
factor prices is sufficient to solve the general equilibrium using the full employment
conditions independently. Figure 4.5 shows the cost minimization problem for the
capital intensive sector, where we set the elasticity of substitution at α = 0.6.
Figure 4.5: Cost minimization in sector 1
𝐾𝐾
𝑌𝑌�1 = 1
𝑎𝑎1
𝐼𝐼𝐼𝐼𝐼𝐼 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
𝑏𝑏1
𝛼𝛼1 = 0.6
𝑤𝑤
𝐾𝐾1 (𝑤𝑤, 𝑟𝑟) 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = −
𝑏𝑏1 = 𝑟𝑟
𝐿𝐿1 (𝑤𝑤, 𝑟𝑟)
𝐿𝐿
Sector 1 firms produce with minimal production costs at point b1 . The production
could also take place in point a1 but b1 represents a Pareto improvement compared to
a1 . The same amount of good 1 can be produced cheaper at point b1 compared to the
production costs that firms would have to pay if they were producing at point a1 . The
more we move the isocost line towards the origin, the lower the production costs are.
However, the isoquant remains fixed so that we look at the same output level that is
produced at different production cost levels.
We are able to conclude that the optimal factor intensity to produce Ȳ1 = 1 is repre-
sented by point b1 . However, the total input of capital and labor in each sector has yet
to be determined. More than one good will be produced in this sector so that the total
input L1 and K1 will differ in absolute terms but the ratio aK1 /a L1 must be equal to the
optimal factor intensity determined at point b1 . L1 and K1 will be determined using
the full employment conditions and the two levels will be according to the optimal
factor intensity determined in this optimization problem. Thus, the solution obtained
from the full employment conditions must lie on the optimal factor intensity rays that
connect the origin and the point b1 .
Notice that the optimal factor intensity has been determined for given factor prices.
K1 (w,r )
Thus, we can use the notation b1 = L1 (w,r )
to highlight the fact that factor prices
were taken as given in the optimization problem. The next chapter will shed light on
the interaction between goods prices and factor prices. Nevertheless, remember that
changes in factor prices affect the optimal factor intensity as well. A higher wage rate
in comparison to the level of capital rentals increases the optimal capital to labor ratio
in the production process. Firms that see the wages paid to their employees rising
would be better off by adjusting their relative labor demand. They could produce
more efficiently with a lower labor to capital ratio in the production process.
The cost minimization problem can be applied to sector 2 as well in order to obtain
the optimal factor intensities in both sectors. Combining Figure 4.5 with Figure 4.6
enables us to characterize the equilibrium in one economy. Notice that we are still
under autarky and that the isocost lines in Figure 4.5 and in Figure 4.6 have the same
position and the same shape in both sectors. That result stems from the fact that firms
in both sectors must pay identical wages to their workers and to the capital owners.
Identical slopes of the isocost lines in both sectors are consistent with equality of w/r
Figure 4.6: Cost minimization in sector 2
𝐾𝐾
𝐼𝐼𝐼𝐼𝐼𝐼 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐
𝑌𝑌�2 = 1, 𝛼𝛼2 = 0.4

𝑎𝑎2
𝑏𝑏2
𝑤𝑤
𝐾𝐾2 (𝑤𝑤, 𝑟𝑟) 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = −
𝑟𝑟
𝑏𝑏2 =
𝐿𝐿2 (𝑤𝑤, 𝑟𝑟)
𝐿𝐿
in both sectors.1 The optimal factor intensity is represented by the vector b2 , which is
much flatter than the respective vector in sector 1. The different slopes are due to the
different elasticities of substitution, α, which shape the isoquant’s location.
2.2 Production, consumption and the gains from trade
This subsection determines the production and consumption program under autarky
and free trade. Total production in sector i is denoted by Y1 and Y2 , respectively.
Outputs Y1 and Y2 depend on inputs of capital and labor in the respective sector.
Before continuing with the discussion of the microeconomic foundations of the 2 ×

2 × 2 model, we briefly recap the results derived so far. We studied the link between
inputs and output in order to characterize the production process. Based on this
knowledge we were able to solve the cost minimization problem that gave us the op-
1 Why do the slope of the isocost curve is equal to −w/r? We know that the cost level is equal on the
curve so that ∆C = w∆L + r∆K = 0. We can solve this equation for ∆K
∆L = − w/r.
timal factor intensity in the production of goods 1 and 2. However, we still do not
know how the available capital and labor is shared between the two sectors. We have
to know the boundaries of production in both sectors in order to solve for the absolute
production levels. This is not an easy task as both sectors are interdependent. The
maximum output in one sector depends on output in the other sector. This interde-
pendency can be taken into account by using the production possibility frontier (PPF)
which has a concave form as depicted in Figure 4.7. The PPF can be derived as already
shown for the Ricardo model but the functional forms of the production functions are
different now. The PPF is a summary of all feasible combinations of good 1 and 2 that
can be produced at given endowments and technologies. The most extreme scenarios
are the points where the PPF intersects the X-axis or the Y-axis. Both points repre-
sent specialization in either good 1 or good 2. The international comparison of these
specialization points reveals the logic behind the different shapes of the Home and
Foreign PPFs. The PPF in Home is tilted more towards the good 1-axis while the PPF
in Foreign is tilted towards the good 2-axis. The different shapes of the home and for-
eign PPFs can be explained by the aforementioned differences in endowments. Both
countries are equipped with different labor and capital endowments that translate
into different specialization patterns in both economies.
Figure 4.7: The Gains from Trade
𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻
𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹
𝑌𝑌1 𝑌𝑌1∗
𝑝𝑝𝑤𝑤
𝑝𝑝𝑎𝑎 𝑝𝑝𝑎𝑎∗
𝑝𝑝𝑤𝑤
2. THE HECKSCHER-OHLIN MODEL
𝐶𝐶1𝑤𝑤∗
𝑃𝑃𝑃𝑃𝑃𝑃
𝑈𝑈 𝑤𝑤∗
𝑌𝑌1𝑤𝑤 𝑌𝑌1∗𝑎𝑎 = 𝐶𝐶1∗𝑎𝑎
𝑌𝑌1𝑤𝑤∗
𝑈𝑈 ∗
𝑌𝑌1𝑎𝑎 = 𝐶𝐶1𝑎𝑎
𝐶𝐶1𝑤𝑤
𝑈𝑈 𝑤𝑤
𝑈𝑈
𝑃𝑃𝑃𝑃𝐹𝐹 ∗
𝑌𝑌2𝑤𝑤 𝐶𝐶2𝑤𝑤 𝑌𝑌2∗
𝑌𝑌2 𝐶𝐶2𝑤𝑤∗ 𝑌𝑌2𝑤𝑤∗
𝑌𝑌2𝑎𝑎 = 𝐶𝐶2𝑎𝑎 𝑌𝑌2∗𝑎𝑎 = 𝐶𝐶2∗𝑎𝑎
163
This property of the production possibility frontier is evident at the specialization

points where the PPF intersects the vertical or horizontal axis. Home is capital abun-
dant and sector 1 produces with a technology that is relatively more capital intensive.
Using all capital and all labor in sector 1 must be associated with a level of output that
is higher than the level of output the same economy could attain if it uses all capital
and all labor in sector 2. The opposite pattern can be derived for the labor abundant
foreign country, which could produce more through specializing in the labor intensive
good 2 compared to specialization in the capital intensive good 1.
Autarky prices of goods 1 and 2 must reflect the opportunity costs of production
in both sectors. Thus, relative goods’ prices under autarky are different in Home
and Foreign. The relation between both relative price levels depends on factor en-
dowments and consumers’ preferences. Prices adjust as long as the production level
is different from the level of consumption. The invisible hand guides producers and
consumers towards the optimal production/consumption bundle that yields the high-
est utility possible.
The optimization problem itself can be solved using the standard utility maximization
problem as previously discussed in the Ricardo chapter. Production and consumption
are determined jointly by the point of tangency between the indifference curve and the
respective production possibility frontier. The slope of the indifference curve and/or
the slope of the PPF pin down the relative autarky prices p a = P2a /P1a , which must be
different in both countries due to the different shape of the PPFs.
We obtain the optimal autarky production/consumption in Home at Y1a = C1a and

Y2a = C2a . Home has more capital than labor, which is evident from the higher relative
production of good 1. The higher relative price p a ensures that the relative consump-
tion of the capital intensive good 1 matches the relatively higher supply of good 1.
Opening up the economy to free trade facilitates importing and exporting from and
to Foreign. Thus, the autarky prices are no longer relevant. Goods prices are then
determined on the world market and the adjustments of prices continue until trade
between the two countries is balanced.
Home has a comparative advantage in sector 1. The world market price of good 1
relative to the world market price of good 2 must be higher than the relative autarky
prices of good 1. Put differently, Home’s budget constraint under free trade is flatter
than its autarky price line. The latter also represents the budget constraint under
autarky. Good 1 was more expensive in Foreign before commodity markets were
fully liberalized. Domestic producers are able to sell their goods at higher world
market prices, which is essentially the same logic as already discussed in the context
of the Ricardo model. The budget constraint reacts to the new commodity prices
and becomes flatter in Home. The slope of the budget constraints depends on the
relative commodity prices. Under free trade the only relevant commodity prices are
the world market prices that pin down the slope of the budget constraint. Another
consequence is the shift in production towards the sector of comparative advantage.
More resources are dedicated to the sector with a comparative advantage in order
to exploit the relative cost advantage. The full employment conditions were already
fulfilled before trade liberalization. Thus, higher production in 1 must be associated
with a contraction of the other sector, 2. Both capital and labor are shifted from the
contracting to the expanding sector. Output rises from Y1a to Y1w but shrinks from Y2a
to Y2w . At the same time we observe that consumption patterns change as well. Good
2 can be imported at lower import prices from Foreign. Thus, the price of good 2
decreases relative to the price of good 1. Consumption rises to C2w in sector 2 but
decreases to C1w in sector 1.
Consumption and production are no longer equal and the difference between both is
either imported or exported. Whether the respective good is imported or exported
depends on the comparative advantage and the opposite pattern can be found in
Foreign.
Both countries benefit from trade liberalization. The representative consumer reaches
a higher level of utility. Rational consumers shift their consumption to the points,
where the higher indifference curves U w and U w∗ can be reached.
2.3 Factor prices
So far, we have analyzed the cost minimization problem that determines the factor
intensity at given prices in the first subsection of this chapter and we have developed
a deeper understanding of the production process and the role of goods prices. We
learned how goods prices change when going from autarky to free trade but we don’t
yet know how factor prices are determined in general equilibrium.
A Lerner diagram enables us to use goods prices and their adjustments to trade liber-
alization in order to determine the factor prices endogenously. Figure 4.8 summarizes
both the autarky and the free trade scenarios. We assume that the autarky price of
good 2 is identical in both countries. Now we assume that the world market price
∗
of good 2 equals the autarky price of good 2. Ȳ2 and Ȳ2 represent the isoquants
associated with the level of output in sector 2 necessary to generate a level of revenue,
which is exactly equal to 1 at some given price level. Thus, the respective isoquants
summarize all input combinations of L and K associated with this one particular out-
put level that is needed to obtain the desired level of revenue in the respective sector.
Since autarky and world market prices in sector 2 are identical, the two isoquants that
∗
represent the "revenue=unity" scenario are identical, too. Ȳ2 and Ȳ2 have the same
shape and position in Home and in Foreign and the autarky and free trade isoquants
in sector 2 are also identical in both countries.
The autarky and free trade isoquants in sector 1 must be different because of relative
autarky price differentials between Home and Foreign. Equal prices in sector 2 imply
that the price of good 1 must be higher in Foreign compared to the price of good 1
in Home. Otherwise, the relative world market price cannot lie between the autarky
prices. How does this translate into the location of the isoquants? Owing to identical
technologies, the shape of the isoquants are identical in both countries. However, the
positions of both isoquants are still different due to price differentials. A lower price
P1a in Home translates into a higher isoquant Ȳ1 . Remember that we are searching
for output levels that yield a total revenue that is equal to unity. The output level
∗
Ȳ1 must be higher than the output level Ȳ1 to ensure that revenue in both countries’
manufacturing sector is equal to unity.
But why do we need to construct this one particular case? We have to make the two
industries comparable to each other and we can do this using the perfect competition
outcome that profits are zero. Both sectors are represented by isoquants that yield the
same revenue=1. Thus, production costs are also equal to unity in both industries.
Perfect competition leads to a scenario where both sectors produce with zero prof-
its and the only feasible factor price ratio, w/r, for which all of those requirements
are met can be found in the tangency points between the isocost line and the two
isoquants. The assumption that revenue is exactly equal to unity is again arbitrarily
chosen. One can do exactly the same exercise for any level of revenue, the only re-
striction is that revenues are identical in both sectors in order to make sure that both
sectors are represented by the same isocost line.
The slope of the isocost line pins down the relative factor prices in Home and Foreign:
S a = w/r and S∗a = w∗ /r ∗ . The two intersection points of the isocost line and the x-
and y-axis determine the factor prices 1/w and 1/r ( 1/w∗ and 1/r ∗ in Foreign). The
isocost line represents all points where both sectors produce with unity production
costs (C1 = C2 = 1). Setting either K or L equal to zero yields the points, where the
isocost line intersects the x- or y-axis at 1/w or 1/r.
2.4 Trade liberalization and factor prices
How do factor prices adjust to trade liberalization? We can easily extend the Lerner
diagram in order to study the reaction of factor prices to trade liberalization. We
already know that the autarky and world market prices in sector 2 are falling together.
Thus, nothing changes in sector 2. The isoquants remain at the same position due to
the assumed constant goods price.
The isoquants in sector 1 adjust according to changes in goods prices. We know

that the price of good 1 increases in Home but decreases in Foreign. This change
translates into an inward shift of the home isoquant Ȳ1 and an outward shift of the
∗
foreign isoquant Ȳ1 . The intuition behind this result is straightforward. Remember
that we have fixed the level of revenue to a certain level. Suppose that the price is
increasing. To keep revenue fixed, one has to reduce output. The opposite pattern
can be found if the price is decreasing. Notice that all isoquants have the same shape
and position in the open economy due to identical technology.
The isocost curves adjust accordingly. The home isocost curve rotates and shifts until
it becomes tangential to the two isoquants under free trade. This adjustment is asso-
ciated with a change in relative factor prices. The home isocost line becomes flatter
implying a lower relative price of labor: S a > Sw . The opposite pattern can be found
in Foreign, where the new isocost line is steeper and is associated with an increase in
the relative wage w∗ /r ∗ . Despite the change in factor prices in Home and Foreign we
find factor price equalization. The slope of the two isocost lines in Home and Foreign
are identical after trade liberalization. That implies that both countries pay the same
wages to workers and capital owners.
Figure 4.8: Factor prices and trade liberalization
Home Foreign
𝐾𝐾 𝐾𝐾 ∗
∗ 𝑤𝑤
𝑌𝑌�1 = 1/𝑃𝑃1𝑤𝑤
��
𝑌𝑌
1 = 1/𝑃𝑃1
2. THE HECKSCHER-OHLIN MODEL
𝑆𝑆 𝑤𝑤 𝑌𝑌�1 = 1/𝑃𝑃1𝑎𝑎
∗
1
��
𝑌𝑌
= 1/𝑃𝑃1𝑎𝑎∗
𝑆𝑆 ∗𝑤𝑤 ��∗ 𝑤𝑤
𝑌𝑌
2 = 1/𝑃𝑃2
𝑆𝑆 ∗𝑎𝑎
𝑆𝑆 𝑎𝑎
𝐿𝐿 𝐿𝐿∗
169
Is inequality increasing or decreasing due to trade liberalization? The answer depends

on the initial scenario. We have seen that the ratio w/r is decreasing in Home but
increasing in Foreign. However, that does not necessarily translate into a scenario
where inequality is decreasing in Home and increasing in Foreign. The evolution of
inequality depends on the initial level of wages and capital rental rates. Suppose that
wages and capital rentals are exactly equal. This scenario would be associated with
the lowest level of inequality possible. Both factors get paid the same compensation
leaving them both with the same purchasing power. Trade raises inequality in Home
if the relative autarky wage satisfies w/r < 1. We have already derived the result
that trade liberalization reduces the wage paid to capital rental ratio and that this
reduction is reflected in a movement of the relative factor price further away from
the level 1. If the initial ratio satisfies w/r > 1 trade may even reduce inequality
as the income gap between workers and capital owners would narrow due to trade
liberalization.
The same ambiguous result can be found in Foreign, where trade rises the wage
to capital rental ratio. Trade liberalization in this particular country would reduce
inequality in Home if w/r < 1. Inequality would be increasing if w/r > 1.
2.5 Stolper Samuelson Theorem
Which factor benefits and which factor loses due to trade liberalization? We have
seen that the relative wage decreases in Home but increases in Foreign. However,
this information is not sufficient if we want to asses the changes of consumers’ pur-
chasing power. Can we really say that this change in relative factor prices reflect a
loss of workers’ purchasing power at Home and do foreign workers benefit? Such a
statement seems plausible but a more convincing analysis of the evolution of workers’
and capital owners’ purchasing power is needed. We will see that the scarce factor
loses and the abundant factor benefits from trade liberalization. This result is also
referred to as the "Stolper Samuelson Theorem". Applied to our initial scenario, we
would expect that the real wage is decreasing in Home but increasing in Foreign. The
real capital rental is expected to go up in Home but it should be declining in Foreign.
In order to prove this result analytically we are going to use Jone’s algebra.
Notice that we follow the notation in Feenstra (2004), which means that we con-
sider two general industries instead of distinguishing between manufacturing and
agricultural industries. The proof can be applied to any industry along the lines
shown below.
Ronald Jones, the inventor of this proof, expressed all equilibrium conditions in hat-
notation. Recall that the unit-cost function obtained from the cost minimization prob-
lem discussed graphically in the last subsection can be solved through the following
cost minimization problem
ci (w, r ) = min {wLi + rKi |Yi ( Li , Ki ) ≥ 1} , (4.16)

Li ,Ki ≥0
which has the solution

ci (w, r ) = wbiL + rbiK . (4.17)
The coefficients biL and biK are endogenously determined and pin down the optimal
amount of labor and capital used to produce one unit of good i. Solving the cost
minimization problem for both sectors gives us two cost functions for each country.
One can solve for the optimal input of capital and labor in the production function by
setting up a Lagrangian. For given factor prices, w and r, we search for the optimal
amount of labor and capital that is associated with the most cost efficient way a
good can be produced. This analytic cost minimization problem corresponds to the
graphical solution discussed in the first part of the Heckscher-Ohlin model.
The zero profit conditions state that total revenue within a sector must be equal to
total production costs in the respective sector so that
P1 = c1 (w, r ) , (4.18)
P2 = c2 (w, r ) . (4.19)
Moreover, we use the full employment conditions
b1L y1 + b2L y2 = L̄ , (4.20)
b1K y1 + b2K y2 = K̄ . (4.21)
From a firm’s perspective, prices Pi and endowments, L̄ and K̄, are exogenously given.
Firms adjust according to factor prices choosing the optimal factor intensity. Taking all
equilibrium conditions together gives us four equations and four unknowns: w, r, y1
and y2 . Wages and capital rentals can be easily solved by intersection of the two zero
profit conditions. Wages and capital rentals are equal in both sectors. Thus, we can
intersect the two zero profit conditions in order to pin down the levels of w and r
thereby fulfilling both zero profit conditions. The graphical solution can either be
derived in the Lerner diagram or as it is shown in Figure 4.9. The shapes of the cost
functions in this graph are determined by biL and biK . The position of the zero profit
conditions is determined by the goods prices Pi . The higher the price of the respective
good, the more the zero profit condition shifts outwards. Notice that we plot both
conditions in the wage-capital rental space. This graph is easier to understand but
less powerful than the Lerner diagram, where we are able to study different scenarios
within the same graph.
Why are the cost functions in Figure 4.9 different from the cost functions discussed in
the Lerner diagram? Costs and factor prices were held constant at a certain level and
input of capital and labor were the two factors that have been altered in order to find
the optimum. Figure 4.9 solves for an optimum by setting commodity prices. The
optimum is determined by alternating factor prices at given commodity prices until
the perfect competition condition is fulfilled.
The production program can be determined using the production possibility frontiers
and consumers’ preferences.
Figure 4.9: The zero profit conditions
b1
b2
P1 1 ,
P2 2 ,
Writing the model in algebraic form has one advantage. We can use it to prove the
Stolper Samuelson theorem. Applying the total differential at the two zero profit
conditions gives us
dPi wbiL dw rbiK dr

dPi = biL dw + biK dr → = + , i = 1, 2 (4.22)
Pi ci w ci r
The left equation can be obtained by dividing both sides of the total differential by
Pi . This allows us to to express everything as percentage changes. Remember that the
perfect competition assumption can be used to substitute the price by unit costs at
the left equation of the total differential (P = c). We will express percentage changes
using the "hat-notation" ( dP
P = P̂): the change in the price of good i relative to the
price level can be interpreted as the percentage change in prices.
We define the following variables as
wbiL
θiL = (4.23)
ci
rbiK
θiK = (4.24)
ci
θiL is the fraction of labor costs in total costs of sector i. θiK is the fraction of capital
costs in sector i as a share of total costs in sector i. We can use these cost shares in
order to rewrite equation (4.22) as
P̂i = θiL ŵ + θiK r̂ (4.25)
This equation relates the percentage change of sector i goods’ price to the percentage
change in the wage rate and the percentage change of the capital rental using the
above defined cost shares θ. It is convenient to stack the equations corresponding to
(4.25) into a matrix that represents both sectors
    
 P̂1   θ1L θ1K   ŵ 
 =   (4.26)
P̂2 θ2L θ2K r̂
The Matrix notation is helpful in this situation as we can make use of some matrix
algebra in order to solve for the variables of interest, which are factor price changes,
ŵ and r̂. The box on the following page presents a brief refresher on matrix algebra.
We will then apply this method to the matrix (4.26).
Refresher course: Matrix algebra

Notice that matrices are often used to solve systems of linear equations. Suppose
you have the following system in matrix notation
A·x = b (4.27)
where A is a (n × n) matrix that contains the parameters of the endogenous vari-

ables summarized in the vector x ∈ Rn . The vector b ∈ Rn contains the outcome
variables n ∈ N. In order to solve for the endogenous variables summarized by
vector x one has to solve for x using the following rule:
A −1 · A · x = A −1 · b (4.28)
However, this rule is applicable only if the inverse of matrix A exists. If that is the
case one can easily solve for vector x as
x = A −1 · b (4.29)
We still need to know how to compute the inverse of a matrix. Suppose we have a
2 × 2 matrix containing the elements a, b, c and d:
 
 a b 
A=  (4.30)
c d
The inverse of this matrix can be computed by changing the elements a and d,
switching signs of the elements b and c, and dividing the resulting matrix by the
determinant:  
1  d −b 
A −1 =  , (4.31)
DetA

−c a
where DetA = a · d − b · c is the determinant of matrix A.

The advantage of the matrix-notation is that it allows us to solve for ŵ and r̂
    
 ŵ  1  θ2K −θ1K   P̂1 
= (4.32)
Det(θ )
   
r̂ −θ2L θ1L P̂2
To solve equation (4.32) for factor prices we have to compute the determinant of the
matrix as: Det(θ ) = θ1L θ2K − θ1K θ2L The determinant can be rearranged as
Det(θ ) = θ1L (1 − θ2L ) − (1 − θ1L )θ2L
= θ1L − θ2L = θ2K − θ1K (4.33)
Remember that θ is a share which has to sum up to 1 such that θiL + θiK = 1. We need
some assumptions about the technology in industry 1 and 2. Industry 1 is assumed
to be the capital intensive industry. Thus, the capital cost-share in industry 1 must be
higher than the capital cost share in industry 2 so that
θ2K − θ1K < 0 (4.34)
Using this information one can easily verify that the determinant is negative because
Det(θ ) < 0. Finally, we can analyze the effect of an increase in the price of good 1.
We assume that the endowments are such that Home has a comparative advantage in
sector 1 meaning that trade liberalization leads to a change in goods prices according
to P̂ = P̂1 − P̂2 > 0.
To see this we first solve the matrix for ŵ as
θ2K P̂1 − θ1K P̂2 (θ − θ1K ) P̂2 + θ2K ( P̂1 − P̂2 )

ŵ = = 2K < P̂2 (4.35)
Det(θ ) (θ1L − θ2L )
The inequality holds since P̂1 − P̂2 > 0.

We can do the same for r̂, which satisfies
θ1L P̂2 − θ2L P̂1 (θ − θ2L ) P̂1 + θ1L ( P̂2 − P̂1 )

r̂ = = 1L > P̂1 (4.36)
Det(θ ) (θ1L − θ2L )
Again, the inequality holds because P̂1 − P̂2 > 0.
It has been shown that the wage increases by less than the price of good 2. Remember
we are looking at percentage changes that are comparable without further conversion
or standardization. Workers are worse off in Home. Their real wage declines below
its autarky level and they can afford to purchase less of good 2. But what about good
1? Workers can also afford less of good 1 because the percentage increase of the price
of good 1 was more than the percentage increase of the price of good 2 due to P̂1 > P̂2 .
We can summarize the effects as
ŵ < P̂1 < P̂2 (4.37)
For capital owners we find the opposite pattern. The change in the capital rental rate
is larger than the increase in price P1 and therefore larger than the increase in the
price of good 2, P2 .
r̂ > P̂1 > P̂2 (4.38)
Thus, capital owners benefit from trade liberalization due to the higher price of good
1. One can combine (4.37) and (4.38) as
r̂ > P̂1 > P̂2 > ŵ (4.39)
Remember that we made the assumption that sector 1 produces capital-intensive

goods so that
θ1L − θ2L < 0 . (4.40)

The input factor capital gains from globalization because it is used intensively in the
sector where the goods price is increasing.
Home has a comparative advantage in the sector that uses the abundant factor inten-
sively. Prices rise in the comparative advantage sector following trade liberalization
and this change in goods prices benefits the abundant factor, due to increasing export
prices. Capital owners benefit in real terms and workers are worse off.
What happens in the Foreign country? We can redo exactly the same proof for Foreign
and we will see the opposite pattern. Capital owners will loose from trade but work-
ers will benefit. Labor is the abundant factor in Foreign and the Stolper Samuelson
Theorem predicts that the abundant factor benefits, whereas the scarce factor loses
when going from autarky to free trade exactly as proved by Jone’s algebra.
Factor Price Equalization: We have learned that factor prices are determined by
marginal productivity of workers evaluated at the respective sector’s goods price.
Marginal productivity is equalized across countries by assumption. Goods prices
are different under autarky and so are factor prices. However, prices adjust to the
world market prices once we open up the economy to free trade. The values of the
marginal product of labor and capital are equalized by trade liberalization. Thus,
wages must be identical in both countries and the capital owners become the same
rental for their capital in Home and Foreign. We find Factor Price Equalization
across countries.
3. FDI AND OUTSOURCING IN A MULTI-INDUSTRY FRAMEWORK 179
3 FDI and outsourcing in a multi-industry framework
The Stolper Samuelson theorem states that some factors benefit from trade at the
expense of some other factors. Does that mean that inequality rises due to trade
liberalization? We have seen that the answer is complex and depends on the initial
wage ratio. However, it is possible to show that with reasonable assumptions about
the elasticity of substitution the 2 × 2 × 2 model yields a scenario where inequality
is rising both in the developing (Home) and in the developed (Foreign) countries.
But up to this point of the course we have only been concerned about wage differen-
tials between workers and capital owners. A more realistic distinction between low-
and high-skilled workers is still missing in our analysis. We therefore continue by
discussing a more recent model that features both high- and low-skilled workers as
well as capital. One may argue that this model is too recent to be labelled as canon-
ical. However, it is still among the most prominent approaches in the entire trade
literature. It is tractable and it features both foreign direct investment and offshoring,
which allows us to analyze different aspects of international trade in one single model.
Stylized facts about inequality. A first glance at the data suggests that trade rises
inequality independently from the factor endowments or from the level of develop-
ment. Many studies argue that this development can be rationalized by relative labor
demand shifts from low- to high-skill workers. Inequality has been rising throughout
the world and improvements in technology and/or more intense outsourcing may be
potential explanations for this phenomenon, as well as the increase in demand for
high-skilled workers. Improvements in production technologies for instance have al-
lowed firms to substitute low-skilled with high-skilled workers. Outsourcing works
in the same direction as skill-biased technology change. It allows firms to benefit
from a cost saving effect of offshoring. More expensive low-skill production stages
can be shifted abroad to countries producing at lower costs. The latter channel is
highly relevant in the context of trade between developed and developing countries.
Interestingly, we will see that the effect of offshoring on labor demand for high-skilled
is identical in both developed and developing countries. The offshored low-skill in-
tensive jobs are mostly jobs that are relatively high-skill intensive in the receiving
developing countries. Put differently, jobs that are declared "low-skilled" in Germany
may still be relatively "high-skilled" from a developing country perspective.
Thus, one would expect rising wages at the top and stagnant or even declining wages
at the bottom of the income distribution in developed and emerging countries that
are open to globalization. The stylized facts for Germany presented in Figures 4.10
and 4.11 support this hypothesis. The analysis is conducted for males and females
separately. Dustmann, Ludsteck, and Schönberg (2009, QJE) present evidence based
on data from the period 1990 - 2004. This period is often referred to as the second
"Golden Age of Globalization" characterized by soaring exports and heightened out-
sourcing activities. Thus, it is not surprising that the depicted evidence on inequality
is associated with globalization, although globalization is certainly not the only driv-
ing force.
The authors use German worker-level data to trace real wage developments at differ-
ent percentiles of the income distribution. The low-income group is represented by
the 15th percentile, the medium income group is represented by the 50th percentile,
and the high-income group is represented by the 85th percentile. First, all workers
representing the different percentiles in the different years are sorted within each
period. Notice that the workers who represent the different percentiles across the dif-
ferent years may be different. Depending on changes in the income distribution and
changes in the workforce competition different workers may be drawn from different
percentiles in different periods. Once the workers at the different percentiles are iden-
tified in each year, one can use the wage information associated with those particular
workers to compare them over time. The question is how does the wage of the worker
associated with the respective percentile in one particular year compare to the wage
of the respective percentile worker’s wage in the period before.2 Take the median
2 The data is the universe of German workers subject to social security contributions. Notice that the
income worker as example. Every period there is exactly one worker earning a wage
that is equal to the median wage. Comparing this median worker’s wage in period
t with the median worker’s wage in period t + 1 allows us to derive the evolution of
the median income between t and t + 1. Figure 4.10 reveals an interesting evolution of
incomes among the male workers in the sample. The wage rate of the 85th percentile
worker is rising over the period. The wage of the worker at the 50th percentile is
rising as well but not so fast as at the 85th percentile. Most interestingly, the wage
of the 15th percentile workers was stagnant in the first half of the sample period and
even declining after 1997. Taken together, we can conclude that inequality was rising
during this time. The wage pattern for female workers looks similar but we are not
able to replicate declining wage pattern in the low-income group represented by the
15th percentile of the income distribution.
Offshoring and skill-biased technological change are able to explain the stylized facts
but only if labor-demand shifted from low- to high-skilled workers during the same
period of time. The authors present some additional evidence on this issue by com-
puting labor demand shifts based on exactly the same data source. Labor demand
shares for different skills by occupations are computed as a ratio between the de-
mand for workers from skill group k and in occupation j over the total demand of
labor in skill group j. Those shares are averaged over all occupations and presented
as growth rates between different periods in Figure 4.11. Medium- to low-skill ra-
tios, and high- to medium-skill ratios are graphed separately. We find evidence that
is consistent with high-skill labor demand shifts which are growing over time. This
finding is in line with the hypothesized reasons behind rising inequality. Demand
shifted away from medium- to high-skill labor and from low- to medium-skill labor.
The contribution of offshoring can be explained using a simple model that features
analysis does not trace certain workers over time. Each period you group all workers according to
their income. The 15th percentile is the one worker with income higher or equal than 15 percent of
all workers at the bottom of the income distribution. Thus, the worker at the 15th percentile is not the
same in each period. Although the worker who represents the 15th percentile in one period may still
be in the sample one year after, her or she may no longer be the worker with income higher or equal
than 15 percent of his or her coworkers’ wage.
1975 1977 1979 1981 1983 1985 1987 1989
Year
15th percentile 50th percentile

182 85th percentile
CHAPTER 4. CANONICAL TRADE MODELS
E GERMAN WAGE STRUCTURE 0.2 851
0.2
and 85th percentile
15th, 50th, and 85th percentile

0.050.25 0.10.3 0.15
0.05 0.1 0.15

percentile
50th,
0 0.2
85th
15th,
0.05 0.15
0.05 0
15th, 50th, and
0.05 0.1
1990 1992 1994 1996 1998 2000 2002 2004 1990

Year

0
85th percentile
987 1989 1975 1977 1979 1981 1983 1985 1987 1989
Year

FIGURE II Figure 4.10: Wage
85th
inequality among males
percentile
in Germany
(Source: Dustmann, Ludsteck, and Schönberg (2009, QJE)
Indexed Wage Growth of the 15th, 50th, and 85
the Postunification Pe
0.2
Source. 2% IABS Sample for full-time workers

15th, 50th, and 85th percentile
0.15
The figures show the indexed (log) real wage gr

percentiles of the wage distribution. Panels A an
0.05 0.1
period between 1975 and 1989, with 1975 as the

to the post-unification period between 1990 and 2
0.05 0
002 2004 1990 1992 1994 1996 1998 2000 2002 2004
Year
period, both the 50th and the 85th perc 15th percentile
85th percentile
50th percentile
compared to 16% and 27% for

FIGURE II (Source: Dustmann, Ludsteck, and Schönberg (2009, QJE)
Figure 4.11: Wage inequality among females in Germany
men (Pane
th, 50th,period,
and 85thin contrast,
Percentiles: The wages
Pre- versus at the 15th p
the 85th
ostunification Periodpercentile experienced the hi
ull-time Panel
workersD).between Unlike 21 and the 60 1980s,
years of age. in the 1990
3. FDI
868AND OUTSOURCING IN AJOURNAL
QUARTERLY MULTI-INDUSTRY FRAMEWORK
OF ECONOMICS 183
0.6
0.4
Demand shifts
0.2
0
1975 1979 1983 1987 1991 1995 1999 2003

Year
Medium versus low High versus medium
FIGURE VI
Between-Occupation Demand Shifts:
Figure 4.12: Labor demand Medium/Low
shiftsversus High/Medium (Men)
in Germany
Source. 2% IABS Sample for men between 21 and 60 years of age working
full-time. (Source: Dustmann, Ludsteck, and Schönberg (2009, QJE)
The figure plots between-occupation demand shifts for the medium-skilled rel-
ative to the low-skilled, and for the high-skilled relative to the medium-skilled,
bothwith
FDI 1975
and asoutsourcing. The
the base year. The model
demandallows
shift of us to address
group k relativepotential labor
to base year 1975market
is computed as Dk = j (E jk/Ek)(E j /E j ), where k indexes education, and j in-
effects in developed
dexes occupation,economies thatinput
E j is total labor are driven
measuredby the emergence
in efficiency ofinmore
units competitive
occupation
j, and E jk/Ek is group k’s employment share (in efficiency units) in occupation j
low-income countries.
in the base year. We It also allows
distinguish us to explain the general equilibrium effects in
82 occupations.
developing economies’ labor markets that are comparable to the effects observed in
which may be seriously compromised due to the high incidence
high-income
of wagecountries.
censoring Quite
amongto thethe
contrary of the Stolper
high-skilled, Samuelsonittheorem,
and because does we
notthat
will see require an estimate
inequality is rising infor
boththe elasticity
countries of heightened
due to substitution. It demand
relative is
important to stress, however, that this index does not account
for high-skill workers.ofHigh-skilled
for the impact workers
price changes on benefit in both
observed countries, whereas
employment shifts. low-
Thus,
skilled if positive
workers’ incomeskill supply shocks cause expansions of high-skill
is declining.
occupations, this demand index will overstate the “demand”
shock. Conversely, if skill premiums rise due to demand shifts,
occupational shifts will be smaller than the price-constant coun-
3.1 terfactual,
The Feenstrathusand
leading the index
Hanson modelto understate the “demand”
shift. Figure VI plots the between-occupation demand shifts
of the medium- versus low-skilled, and the high- versus the
medium-skilled.
Production The
takes place in figure indicates
a continuum a considerable
of industries that rely on demand shift cap-
inputs of factors
favoring the high-skilled relative to the medium-skilled during
ital, high-skilled
the 1980s andworkers, andThis
1990s. low-skill workers.
demand shiftHaving more than twolarger
was substantially industries is
than
the key that favoring
to bring the modelthe medium-skilled
to the relative to
data in a more convincing theHowever,
way. low-skilled.
the idea of
To put these numbers into perspective, Katz and Murphy (1992)
a continuum of industries implies that the number of industries is infinite. Put differ-
ently, individual industries are not identifiable. Thus, we have to compute the mass
of industries instead of their number. A mass of industries aggregates the continuum

bounded by two thresholds that fulfil certain criteria. It allows us to quantify certain
outcome variables for the economy. Factor markets are competitive so that wages of
high-skilled, low-skilled, and capital owners are equal across industries.
We will find that each country specializes in a certain range of industries according
to the comparative cost advantage determined by production costs. Countries will be
active in industries that produce more cost-efficiently compared to the same industry
in Foreign. A key difference with standard trade models discussed above is the idea of
trade in intermediate goods rather than trade in final goods. The whole range of inter-
mediate goods produced within the individual industries is required to assemble the
final good. The intermediates themselves are produced using labor and capital inputs
in the continuum of industries. Imported intermediate goods are in fact goods that
were outsourced by the importing country when the economy moves from autarky
to free trade. The traded good cannot be consumed directly without going through
another assembly process that must take place in each economy separately. Interme-
diates that are produced in one country are combined with the intermediate goods
that are purchased from the rest of the world through outsourcing in order to obtain
the final consumption goods. Under autarky, there is no possibility to offshore such
that the whole continuum of intermediates must be produced and assembled within
the respective country. For the sake of simplicity, we will focus on the two-country
scenario where a developed North and an undeveloped South coexist. Equilibria for
autarky and free trade (offshoring) will be determined separately in order to evaluate
inequality under both scenarios. Moreover, we will allow for FDI in the free trade
scenario. Cross country capital flows trigger changes in production costs through
their impact on capital rentals. Thus, changes in capital rentals trigger a shift in the
comparative advantage across different industries, which has an effect on inequality.
How does outsourcing affect labor demand shifts from low- to high-skilled workers?
We will see that the developed country has a comparative cost advantage in more
high-skill intensive industries. Low-skill intensive industries are shifted to the devel-
oping country. Starting from the free trade equilibrium, one may ask how changes
in the cost of offshoring (i.e. through FDI or lower offshoring costs) affect labor de-
mand for high-skilled relative low-skilled workers. The developed North offshores
additional industries to the developing South. The resources that were freed up by
offshoring are reallocated to the remaining industries. From the perspective of the
North, the offshored industries that shift to the South are low-skill intensive. As
a result, the North loses some of the more low-skill intensive industries. From the
receiving country’s perspective (South) those industries are the more high-skilled in-
tensive industries. Thus, we can conclude that offshoring stimulates a labor demand
shift from relatively low- to relatively high-skilled industries in both the developed
and the developing country.
The model discussed in this chapter is taken from Feenstra and Hanson (1995). The
continuum of industries are ordered according to their relative high-skill intensity
in ascending order using index z that identifies one particular industry within the
whole continuum. Production in industry z takes place according to the Hicks neutral
production function
θ
L(z) H (z)

x (z) = A min , K1−θ , z ∈ [0, 1] (4.41)
a L (z) a H (z)
Parameter A denotes productivity, which is normailzed to 1 for sake of simplicity.

We assume that the range of industries in each country is bounded by 0 at the lower
bound and by 1 at the upper bound. Notice that this is the maximum range of in-
dustries that can exist. All industries must be active under autarky but some may be
offshored under free trade. The mass of one individual industry is zero so we rely
on the mass of a range of industries. The total mass of industries can be computed
through the integral at the lower and upper bounds of all industries. However, this
exercise is trivial as the mass is exactly equal to unity. Let the variables L(z) and
H (z) denote inputs of high- and low-skill labor in industry z. We say that production
in industry z follows a Leontief production function nested in a Cobb Douglas pro-

duction function. Low- and high-skilled workers are teamed according to the input
coefficients a L and a H , which determine the input requirement for one good in indus-
try z. The relationship is fixed which explains why we call this part of the production
function Leontief: high- and low-skilled workers cannot be substituted by each other.
Notice that these coefficients do not stem from any cost minimization problem. They
are given exogenously without any further micro- foundation. The labor-aggregate is
combined with capital in a Cobb Douglas production function. θ determines the Cobb
Douglas output elasticity. Thus, the model allows for some degree of substitution be-
tween labor and capital leaving the input relationship between low- and high-skilled
workers fixed. One may also think about substitution between capital and a labor
aggregate. The latter aggregates high- and low-skilled workers. Finally, K denotes the
input of capital in industry z.
We are able to solve the cost minimization problem using the standard solution
method the production function is Leontief in nature. Both types of workers can
be treated as combined aggregates so that we optimize production for labor (low-
and high-skilled bundled together according to the fixed relationship determined by
a L and a H ) and capital. The cost minimization problem can be solved as demon-
strated for the standard Cobb Douglas case with two input factors, the labor bundle
and capital. The solution yields the unit-cost function
c(w, q, r, z) = B[wa L (z) + qa H (z)]θ r1−θ (4.42)
Parameter B is a summary term that stems from the cost minimization problem and
solely depends on θ. Later in this chapter we will demonstrate how the endogenous
factor prices, w, q and r, can be determined. Once all output x in all industries z is
known, the number of final goods can be determined by the production function
Z 1
ln Y = α(z) ln x (z)dz . (4.43)
0
Final goods, Y, are assembled according to another Cobb Douglas production function
without incurring any additional costs. The production function in the assembly
process is different as we have to deal with a continuum of industries. Each industry
enters the assembling process with a Cobb Douglas share α that sums to unity over
all industries.
The comparative cost advantage. Conditional on wages and capital rentals in North
and South we are able to characterize the unit cost functions in both countries. The
graphical representation of both functions is shown in Figure 4.13. The slope of both
cost curves is an important determinant of comparative advantage. Recalling that
industries in both countries are ordered according to their high-skill intensity we find
that industries closer to the lower bound are the least skill-intensive industries. This
result applies for North and South. However, the steeper slope tells us that the skill-
intensity in the South is increasing faster than the skill-intensity in the North.
From the graph we can see that the undeveloped country has a comparative advantage
in all industries to the left of z∗ due to lower unit costs in these particular industries.
The developed country has a comparative advantage in all industries to the right of z∗ .
Industry z∗ is the one industry where both countries produce with identical costs so
that neither country has a comparative advantage. We pay attention to any particular
industry but rather focus on the range of industries between 0 to z∗ in South and z∗
to 1 in North.
What are the necessary conditions that secure a stable equilibrium? The assumption
q/w < q∗ /w∗ must hold. The high- to low-skilled workers’ wage ratio must be lower
in North than in South. This assumption is realistic for a North/South country setting.
We can apply Shephard’s lemma in order to solve for the demand for high- and low-
skilled workers in industry z as
∂c ∂c
ll = , lh = (4.44)
∂w ∂q
Figure 4.13: The comparative cost advantage
Total demand for low- and total demand for high-skilled workers can be computed
by integrating industry-level labor demand over the whole continuum of active in-
dustries. The solutions can be found in the numerator and denominator of North’s
relative demand function R1
z∗ ∂q x ( z ) dz
∂c
∗
D (z ) = R 1 ∂c (4.45)
z∗ ∂w x ( z ) dz
∂q x ( z ).
∂c
Total demand for highly skilled workers in industry z is Shephard’s Lemma
gives us the optimal labor demand associated with producing one unit of the good
∂c
as ∂q . Remember that the cost minimization problem was solved under the constraint
that one unit of the good can be produced. Total labor demand in industry z is thus
unit labor demand times the total output in industry z. However, labor demand in one
industry is meaningless in the continuous case. We aggregate by integrating over all
active industries using the integral from the lower to the upper bound of industries
where the respective country has a comparative advantage. Notice that the lower
bound in South is fixed to 0 and the upper bound in North is fixed to 1. The cutoff z∗
is the variable bound of active industries in both countries.
Labor market clearing. We are now able to solve the labor market clearing condi-
tion in relative terms. Notice that endowments are exogenously given and remain
constant throughout the analysis. In principle we only need to solve two equilibrium
conditions with two unknowns. The wage of high-skilled workers must be such that
total labor demand for high-skilled workers is equal to the exogenously given labor
supply of high-skilled workers, H. The wage of the low-skilled workers must be such
that total labor demand for low-skilled workers is equal to the exogenously given
labor supply of low-skilled workers, L. It is easier to reduce the problem to one equa-
tion and one unknown by expressing everything in relative terms. Relative demand
for low-skilled workers (related to high-skilled labor demand) must be equal to the
relative supply of low-skilled workers (related to high-skilled labor supply). To what
extent does labor demand depend on wages of high- and low-skilled? We assume
a Leontief production function so that the input relation is fixed. From Shephard’s
Lemma we know that the demand for labor depends on wages. However, wages still
affect labor demand through substitution between worker types and capital inputs.
The labor bundle becomes more expensive if q or/and w are rising so that demand
for both high- and low-skilled workers becomes lower. The input proportion between
low- and high-skilled is independent of changes in either wage rate. This can be easily
verified using the first derivatives of the unit cost function.
A graphical solution for z∗ and q/w can be found in Figure 4.14. The relative labor
supply is constant.3 Relative demand is a downward sloping function. The higher the
high-skilled wage relative to the low-skilled wage, the higher is the relative demand
for low-skilled workers in the economy. How can this result comply to the assumption
of a proportional input relation between high- and low-skilled within one industry?
The answer is complex and described in the proof to Lemma 1 b) discussed in the
Appendix of Feenstra and Hanson (1995). This course does not go into further detail.
A downward sloping labor demand curve with constant relative labor supply yields
3 This is different from the original version of the model where relative labor supply is a positive function
of the relative wage of high- and low-skilled workers. Increasing high-skilled wages induce positive
education decisions of low-skilled workers which increase the relative labor supply of high-skilled.
Figure 4.14: Labor market clearing
a unique solution for q/w at the intersection between the relative labor supply and
demand curves.
To fully characterize the equilibrium we have to solve the capital rental using Shep-
hard’s Lemma and the property
wL + qH
(1 − θ ) = rK (4.46)
θ
that stems from the assumption of a Cobb Douglas production function over the labor
aggregate and capital. Notice that K is fixed in the short-run without allowing for
foreign direct investment (FDI). Knowing the total labor costs and the total supply of
capital in the Cobb Douglas case implies that the capital cost share in total production
costs is proportional to the Cobb Douglas elasticity of substitution, 1 − θ. A share θ
of total production costs is spent on capital and a share θ on the labor bundle.
3.2 Capital market liberalization, offshoring and wage inequality
What happens if a government decides to liberalize the capital market? The situation
is depicted in Figures (4.15) and (4.16). Northern investors will have an incentive to
invest abroad providing the Southern interest rate is higher than that in the North.
Capital flows into the country in the form of FDI until the rates of return are equalized,
that is, r = r ∗ . The domestic capital stock decreases due to FDI with the result that
the capital rental rate increases. Capital becomes relatively more scarce, which is
reflected in higher capital rental rates. The opposite happens in the FDI receiving
country, where the capital rental rate is declining in order to restore capital market
equilibrium: increased capital supply must be met by higher capital demand induced
through lower capital rental rates.
The aforementioned adjustment process of capital rental rates in North and South
shift the unit-cost schedules, c(z) and c∗ (z), up or down. The direction of the shift
within each country depends on whether the change in r is positive or negative. Thus,
the two cost schedules shift in opposite directions. Suppose that wages and the cutoff
remain fixed. Firms in each industry have to pay a higher interest rate for capital
deployed in the production process. As a consequence, the equilibrium cutoff must
change, followed by increased offshoring in the form of additional industries being
shifted abroad. These industries survive the initial move from autarky to free trade
but they are no longer competitive due to the change in capital rental rates. The
country where capital flows out due to FDI looses its comparative cost advantage in
some of the industries located close to the former cutoff. These industries are not able
to compete with the FDI receiving country that benefits from lower capital rentals.
0
The cutoff shifts to the point z∗ at the right of the initial cutoff z∗ . All industries
between the initial cutoff and the new cutoff lose their comparative cost advantage to
South. Remember that the entire continuum of industries ranging from zero to one is
needed to produce one unit of the final output good. The shift in industries implies a
surge in offshoring of northern industries to the southern economy.
Figure 4.15: The effects of FDI on the comparative cost advantage
How do the northern and the southern labor markets react to this shock? We are
able to analyze the effects of FDI and offshoring within the same graph that we were
using beforehand to solve the labor market clearing condition. Relative labor supply
remains fixed at its initial level H/L but the relative labor demand curve shifts to the
right due to the change in the cutoff z∗ . This change implies a reduction of the range
of active industries in North. Only the most high-skill intensive industries can survive
in North and the higher relative demand for high-skilled workers must be met by a
higher relative wage q/w to stimulate low-skill labor demand. This counteracting
effect on relative demand is necessary to restore equilibrium. Notice that the total
supply of high-skilled and the total supply of low-skilled remains constant. Although
the input relation in one industry is fixed by the assumption of a Leontief production
function, changes in the relative wage rate affect the slope of the cost curve and
therefore total demand for different skills over the continuum of active industries.
Given the assumption that high-skilled wages are higher than the low-skilled wages
in both countries (q/w > 1 in both North and South), we are able to conclude that
inequality rises in North.
What happens in the southern country? Offshoring in this model increases high-skill
labor demand in both countries. The developing country becomes competitive in
industries that are more high-skill intensive from the developing country perspective.
Thus, the relative demand for high-skilled labor increases as well. Following the same
reasoning we can conclude that the relative wage of high-skilled must increase in the
developing country as well. Inequality increases in both countries.
Figure 4.16: Labor market clearing after FDI
3.3 The role of the continuum of industries
One may argue that the results derived in the last section hinge on the assumption
of a continuum of industries. To prove the generality of the results, we replicate the
main insights in a model with two countries, where firms produce in two industries
using inputs of high- and low-skill workers. The two intermediate goods are denoted
by y1 and y2 . Technologies are different in both industries such that optimal factor
intensities have to differ as well. Both intermediate goods are assembled to make a
final consumption good in the ultimate production step. Firms can assemble the final
good using intermediate inputs produced either in North or South. Corner solutions
where one or the other intermediate is entirely produced abroad are ruled out by
assumption and the two intermediate input goods are produced according to
y1 = f 1 ( L1 , H1 , K1 ) and y2 = f 2 ( L2 , H2 , K2 ) . (4.47)
We impose the assumption that y1 is the unskilled-labor intensive good, whereas good
y2 is the high-skill labor intensive good. Both intermediates are tradeable and the
variable xi is an import or export indicator variable that is positive if the intermediate
good is exported (xi > 0) and negative if the intermediate good i is imported (xi < 0).
Prices of intermediate good 1 and 2 are summarized by the vector p = ( p1 , p2 ). Under
free trade, prices are exogenously given at the world market.
Final goods in industry n are assembled using input of intermediates according to
y n = f n ( y1 − x1 , y2 − x2 ) (4.48)
Thus, total factor demand in country n must satisfy the endowment-constraints
L1 + L2 = Ln , H1 + H2 = Hn , K1 + K2 = Kn (4.49)
The value of output plus net-trade in intermediates can be determined by solving the
maximization problem
Gn = max pn f n (y1 − x1 , y2 − x2 ) + p1 x1 + p2 x2 s.t. (4.47) and (4.49)

xi ,Li ,Hi ,Ki
Countries choose the level of intermediate exports and imports, inputs of labor, inputs
of capital and the inputs of high-skilled workers in a way that maximizes the value of
final good output. Resource constraints (4.47), (4.49) are taken into account in order to
avoid the choice of non-feasible solutions obtained from the maximization problem.
Notice that pn is the price for the final good. Thus, the final good producer’s revenue
is pn yn , which depends negatively on the amount of imported intermediates and
positively on the amount of exported intermediates. Each country will specialize in
producing imports and exports of either intermediate 1 or intermediate 2. We focus
on one country. Prices are taken as given and trade is not necessarily balanced in this
chapter.
The final good assemblers maximize output by deciding about the amount of inter-
mediates produced in-house and the amount of intermediates imported from abroad.
Final good production is illustrated using isoquants that summarize input bundles
that give rise to the same level of final good production. Let yn denote the isoquant
in the initial scenario with only some imports of intermediate goods. The scenario we
have chosen is one with x1 < 0 and x2 > 0. Is it possible to prove this result? Pro-
duction is constrained by the full employment conditions that shape the Production
Possibility Frontier. The production point must lie somewhere on the concave graph
in Figure 4.17. It is the relative world market price for intermediates that determine
the production point A. Moreover, the relative intermediate price also determines the
maximum production at point B. Production is determined by point A. The differ-
ence must be imported so that x1 < 0. Next, let us turn to production in sector 2.
There, we find that point B is located below point A. Production of y2 exceeds the
level of inputs in the northern home country and the difference between production
and the quantity of this intermediate good used for the assembling process of the
final good is exported to the southern foreign country. The export volume is mea-
sured by x2 > 0. All input combinations are represented by the isoquant that is again
associated with a certain level of output of the final good. The output levels are held
constant at the levels we have chosen for the initial scenario with outsourcing (trade
in intermediates).
We are also able to determine the effects of an increase in the relative world market
price of intermediates. Suppose that world market conditions change in a way that
affects the terms of trade in favor of the Home country. Such a drop in the relative
price of the imported intermediate is depicted in Figure 4.17. Production shifts to-
wards the skill-intensive intermediate good represented by point A0 and trade shifts
to point B0 . Home produces more of the relatively more expensive good in order to
rise the level of exports above its initial level. The imported good becomes cheaper
so that less resources are used in the production of the good that can be purchased
even more cheaply on the world market. Resources shift towards the industry where
Home has a comparative advantage. As result, the final good producer can produce
more of the final good as indicated by the outward shift of the isoquant from yn to y0n .
The intuition is that imported intermediates become cheaper whereas exports become
more expensive due to the better terms of trade.
Perfect competition and prices. The assumption of perfect competition is main-

tained throughout the entire chapter. Firms produce intermediates without generat-
ing profits. Thus, a Zero Profit Condition can be used in order to determine the prices
of factors and intermediate goods. Can we conclude anything about the price of the
final good in industry n? We ignored the price Pn so far because it is not relevant for
the outcomes of the model. Intermediate good prices and factor prices are jointly de-
termined in a first step and those intermediate good prices pin down the final goods
price Pn . We can ignore Pn as long as we are not interested in discussing welfare
effects associated with trade liberalization.
More important than final goods prices are the prices of the intermediate goods. In-
termediate goods’ prices affect factor prices in our model as the Zero Profit Condition
pi = ci (w, q, r )
A‘
B‘
A ௡
Figure 4.17: Autarky and trade equilibrium
applies due to the assumption of perfect competition. The price of intermediate good
i must be equal to the unit production costs that depend on factor prices w (wage of
the low skilled), q (wage of the high skilled), and r (capital rental). Under autarky
prices are jointly determined with the price of the final goods. However, we focus
on the free-trade equilibrium where these intermediate goods prices are given by the
world market.
Figure 4.18 graphs the Zero Profit Conditions for the two intermediate goods p1 = c1
and p2 = c2 . Capital rental rates are implicitly fixed by graphing the iso-curves in
the w, q space. The intersection of both iso-curves determines equilibrium low- and
high-skill wages. More interestingly, we can analyze the effects of a decrease in the
price of intermediate good 1 (p1 decreases). Remember that good 1 is the low-skill
labor intensive intermediate good. The wage of the factor used intensively in the
respective sector decreases. High-skilled workers benefit from the decrease in the
price of intermediate good 1 and inequality rises in the developed country - similar to
the outcome in the model with a continuum of industries. That is not surprising. The
models are almost identical as we take one particular industry n out of the continuum
and look at this one particular slice of the economy in isolation from the remaining
industries. We ignore all general equilibrium effects that arise due to the existence of
other industries. Can we say something about the effects of offshoring in developing
countries? We have seen that the major advantage of the continuum of industries
was the result that more intense offshoring increased demand for high-skilled labor
in both the developed and the developing country. In fact, this result does not require
a general equilibrium framework with more than one industry. Countries that export
the low-skill intensive intermediate good would lose from the terms of trade changes
analyzed above. The price of their export good deteriorates, resulting in a shift to
more high-skill intermediate good production. Wages of the low-skilled decrease
relative to the wages of the high-skilled leading to higher inequality through the
assumption that q > w in both countries. However, the analysis lacks some clear
micro-foundation of the determinants of offshoring. The partial equilibrium analysis
is limited as we do not solve for world market prices in general equilibrium and we
also do not investigate the reasons for offshoring. FDI would be a good candidate.
One could analyze the effects of capital flows on the equilibrium capital rental rates
but for the sake of tractability we neglect r in our analysis. More information about
the production functions would be necessary to determine the effects of r on the Zero
Profit Conditions to learn about the associated price effects. The Feenstra and Hanson
model imposes very strong assumptions about the functional form of the production
process. These assumptions allow us to draw unique solutions and unambiguous
predictions about the effects of offshoring on inequality. The assumption of a Leontief
production function is especially restrictive, albeit one may conclude that a fixed input
relationship between high- and low-skilled is not as unrealistic as it appears on first
sight.
We close the discussion about theories of trade in intermediates by applying Jones

ଶ ଶ
ଵ ଵ
ଵ ଵ
Figure 4.18: Zero profit conditions and the effects on factor prices
algebra to the simplified model of offshoring. As a first step we compute the total
differential of the Zero Profit Condition as
dpi = dw aiL + dq aiH + dr aiK (4.50)
Dividing both sides by the price of the intermediate good and accounting for the Zero
Profit Condition on the left-hand side we obtain
dpi dw w aiL dq qaiH dr r aiK

= + + (4.51)
pi ci w ci q ci r
Accounting for the familiar cost shares θ allows us to rewrite the Zero Profit Condi-
tions as
p̂i = θiL ŵ + θiH q̂ + θiK r̂ (4.52)
Notice that θ denotes the share of costs falling on one particular input-factor in total
unit costs. The shares must sum up to unity as ∑ j θij = 1 where j identifies the
input-factor (low-skill labor, high-skill labor, or capital in our example). The system is
not fully specified since we only have two equations determining three endogenous
variables. We solve this problem by fixing capital. Capital has the same cost-share in
both intermediates, which implies that θ1K = θ2K .
We take the difference between the two intermediate-goods prices determined by

(4.52) as
p̂1 − p̂2 = (θ1L − θ2L )ŵ + (θ1H − θ2H )q̂ (4.53)
The cost shares of capital not only drop out in equation (4.53), but also imply that the
cost share for total labor (low- plus high-skill labor) must also be equal in both sectors.
If the share of capital costs in total production costs is equal in both sectors and the
total costs consist of labor and capital costs, then the labor cost share is also equal
in both sectors. However, that does not mean that the high- and low-skill cost share
is equalized across sectors as well. Suppose that the price of the intermediate input
increases such that p̂1 − p̂2 < 0 leads to a decrease in the relative price of unskilled
workers wage rate because
( p̂1 − p̂2 )
(ŵ − q̂) = <0 (4.54)
(θ1L − θ2L )
The model predicts that in high-skill abundant countries high-skill workers benefit
from outsourcing: high-skill workers’ wages are increasing due to offshoring.
3.4 Empirical tests of the Feenstra and Hanson model
The focus in this sub-section is on an empirical test of the predictions derived from
the two models of offshoring. First, a testable equation will be derived based on the
assumption of a translog cost function that is supposed to describe the functional form
of the unit cost function. Reverse engineering enables us to choose a cost function that
gives rise to an equation that can be taken to the data in order to quantify the effects
at work. It turns out that the translog cost function has the desired properties. A
detailed description of the function itself and the steps needed to derive the testable
equation can be found in the textbook by Feenstra (2004). We skip further discussions
and directly dip into the empirical model without spending too many words on the
details behind the translog function. The test of this model builds on the description
in Feenstra and Hanson (1995). The empirics shed light on the contribution of skill-
biased technological change and offshoring to the observable rise of the high-skill cost
share in production.
Suppose that the model discussed in this subsection can be characterized by a translog
function that gives rise to the first derivative
M K
∂ ln C
= αi + ∑ γij ln w j + ∑ φik ln xk , i = 1, ..., M. (4.55)
∂ ln wi j =1 k =1
∂ ln C ∂C wi
We know that ∂ ln wi ≈ ∂wi C . Thus, the first derivative of log-costs with respect
to the log-wage is approximately equal to the percentage change of total costs as a
∂ ln C
percentage change in the wage of input factor i. Put differently, ∂ ln wi = si is the
growth rate of the cost share of factor i. We can easily set up this equation for skilled-
workers as one particular input factor of interest instead of one general factor i. The
variable αi is a constant that determines the average demand for high-skilled workers.
Remember that the expression is part of the first derivative of unit costs with respect
to the wage of the high-skilled. According to Sheppard’s Lemma, this constant is
a measure for high-skill labor demand. The γ coefficients are associated with the
indirect effects of the other input factors. The last sum is a summary of all other
factors that influence unit costs indirectly through the high-skilled wages. Possible
factors include skill-biased technological change, offshoring or substitution of labor
by capital. Thus, the φ terms are the variable of interest in our setup. We want to
isolate the separate effects of technology and offshoring on high-skilled wages. The
aim is to back out estimates of the parameters γij and φik by fitting
∆snH = φ0 + φK ∆ ln Kn + φy ∆ ln Yn + φz0 ∆zn , n = 1, ..., N . (4.56)
The dependent variable is the cost share of high-skill workers and φ0 is a constant.
Moreover, we know from the model that we have to control for capital and output by
including proxies for Kn and Yn . The vector zn contains variables that shift unit costs,
which has an indirect effect on wages of the high skilled. We previously discussed
the problem that we are not able to say how offshoring or FDI affects wages of the
high- and low-skilled in Figure 4.17 without knowing the exact form of the production
functions. FDI and offshoring shifts the unit cost functions in Figure 4.17 but we don’t
know how q and w react without further knowledge about the production functions
and the form of the unit cost function associated with it. Thus, φz are the coefficients
that have to be estimated. The model can be translated into the question "If the high-
skill cost share increases, how much of the increase is due to an increase in capital,
output, and the structural variables technology and offshoring?".
The authors use the following data:
• Data source: NBER Productivity Database at the 4-digit Standard Industrial

Classification (SIC) level
• Coverage of the data: 447 industries within the U.S. manufacturing sector
over 1979-90
• Change of the non-production wage share over all manufacturing industries

from 35.4% to 42.4% used as proxy for low-skill labor wage share
• Regressors included:
– A proxy for output
– A proxy for the capital share in production

– Outsourcing measured as imported intermediate inputs as share of total

materials purchases
– The shares of computers and other high-tech capital in the total capital
stock
– The share of expenditures on computers in total investment
Table 4.19 reports some of the results reported in Feenstra and Hanson (1995). The
dependent variable is the change in non-production workers’ wage share in total
production costs. The authors look at changes between 1979 and 1990, which is a
relatively long period of time that featured both increases in offshoring and skill-
biased technological change. The change in the high-skill cost share is explained
using different regressors listed in the first column of Table 4.19. The second row
presents some summary statistics, rows 3 to 5 present the coefficients associated with
the different regressors and their standard errors in parentheses. The latter allows us
to evaluate statistical significance of the estimated coefficients.
Notice that the regressions differ with respect to the regressors included. Different
proxies for skill-biased technological change are included in order to prove robustness
of the reported results. The last column reports the contribution of each variable to
the change in the dependent variable. These figure allow us to asses the relative
importance of each variable of interest. Notice that the contribution is reported for
each variable except for the variable Other high-tech share.
204
Change in Nonproduction Wage Share, 1979-90, as Dependent Variable

Mean Regression 1 Regression 2 Regression 3 Contribution
∆ln(K/Y ) 0.71 0.05 0.04 0.04 7-9%
(0.01) (0.01) (0.009)
∆ln(Y ) 1.54 0.02 0.02 0.01 4-8%
(0.006) (0.006) (0.006)
Outsourcing 0.42 0.20 0.22 0.14 15-24%
(0.096) (0.10) (0.09)
Computer share 0.25 0.20 13%
(0.091)
Other high-tech share 0.14 -0.07 -
(0.14)
Computer share (alt. measure 1) 0.07 0.43 8%
(0.17)
Other high-tech share (alt. measure 1) 0.17 0.005 0.2%
(0.07)
Computer share (alt. measure 2) 6.56 0.02 31%
(0.01)
High-tech share (alt. measure 2) 0.40 0.03 3%
(0.05)
R² 0.16 0.16 0.19
N 447 447 447
Source: Feenstra and Hanson 2003, Table 3, as simplified from Feenstra and Hanson 1999, Table III, and empirical exercise 4.2.
Figure 4.19: Regression results to equation (4.56)

It is the only variable that is insignificant. This can be seen from a simple comparison
of the coefficient and the standard error reported below the coefficient.
The value associated with "other high-tech share" is -0.07. The coefficient is insignif-
icant indicating that the variable has no impact on the change in the high-skill cost
share. The range of contribution is reported for the variables that show up more than
once. To understand how the contribution is computed one must understand how to
interpret the coefficient itself. In our linear regression model the coefficient is equal to
the marginal effect of each variable. It tells us the reaction of the dependent variable
in a change of the respective regressors. The counterpart is the first derivative of equa-
tion (4.56) with respect to each variable. For instance, the coefficient of the variable
"Outsourcing" can be interpreted as the first derivative of equation (4.56) with respect
to ln K, which is ∂∆s/∂ ln K. Thus, the coefficient allows us to infer the effect of a
change in capital on the high-skilled wage share. The authors report that the mean
of the dependent variable is 0.389. This information together with the mean of the
respective variable of interest allows us to construct standardized coefficients, which
are directly comparable to each other. We are interested in the information about how
a one per cent increase in capital is associated with outsourcing. A change by some
certain value is meaningless if we are not able to relate that number to some bench-
mark value. Is the increase by one big or small? We don’t know without taking some
further information about the distribution of the variables into account. We normalize
the coefficients by multiplying the coefficient with its mean and dividing the result by
the mean of the dependent variable. The lower bound of the contribution of the first
variable in Table 4.19 is 0.04 · 0.71/0.389 = 0.07 = 7%.
Outsourcing turns out to be more important than skill-biased technological change in

most regressions. Only the regressions that use the second alternative way of measur-
ing the computer share skilled-bias technological change turns out to be more impor-
tant than outsourcing. The Other high-tech share variables are either insignificant or
quantitatively unimportant.
Standardized coefficients are plagued by many statistical problems. When the number
of observations is low one may argue that the mean values are not accurate, thereby
biasing the standardized coefficients. Moreover, the distribution of the variables must
be identical otherwise the means are not comparable.
The Table also reports information about the number of observations N and R2 which
is a measure of the overall fit of the model to the data.
Do we find the same results in developing countries? The model discussed at the
beginning of this section predicted similar effects of offshoring on high-skilled wages
through offshoring in developed and developing countries. The remainder of this
section deals with some empirical evidence for Mexico presented in Feenstra and
Hanson (1995). Mexico provides an ideal experiment for this question due to massive
FDI flows from the US to Mexico after the restrictions on capital flows were relaxed
by the Mexican government in the early 1980s. Feenstra and Hanson (1995) report
an increase in FDI from 478 million US Dollar to 3,635 million US Dollars. Most of
the capital flows went to so called Maquiladoras - mainly foreign owned enterprises
that are located close to the border in order to assemble products for their affiliated
firms in the US. The model would predict that these firms are the more skill-intensive
firms in Mexico. FDI flows trigger offshoring to the relatively more skill-intensive
industries in Mexico. This increases relative demand for high-skill workers who see
their wages rising relative to the wages of the low-skilled. Feenstra and Hanson (1995)
report interesting stylized facts about the change in wages during the transition from
capital restrictions to capital market liberalization.
The test discussed in this subsection exploits regional variation in the data. If the
changes in wages are mainly due to FDI and offshoring one would expect more pro-
nounced changes in the border regions where most Maquiladoras are located. The
authors use rich establishment-level data covering 2,354 manufacturing plants for the
period 1984 to 1990. The authors compare wage measures that relate the individual
Figure 4.20: Regional variation in the wage gap

Source: Feenstra and Hanson (1995)
plant wage to the industry benchmark and distinguish between annual and hourly
wages paid to the employees in the plant. The left panel reports the mean values and
the right panel reports changes over the period 1984 to 1990. The relative high- to low-
skill wage rate is approximated by the non-production to production wage ratio. The
data does not provide further information about different skills but the assumption is
that production workers are low-skilled.
High-skilled workers in coastal regions receive wages that are indeed much higher
than the standard wages paid elsewhere. Wages of the high-skilled relative to wages
of the low-skilled in more remote areas located in the South or Mexico City are below
average. One may conclude that FDI and outsourcing are natural candidates that help
explain higher wages paid by firms that are located closer to the US. The analysis of
the changes over time reinforce this explanation. Wages in regions that are more
closely connected to the US were rising after the ban on capital inflows was lifted due
to trade agreements. Others authors also report positive wage growth but the changes
are much smaller in magnitude.
Figure 4.21 provides a comparison of the Mexican and the US wage gap. Especially
after 1985, the time when both offshoring accelerated in both countries, the wage gap
between high- and low-skilled began to rise in both the developed US and the devel-
oping Mexico. This pattern further supports the theoretical considerations outlined
in the Feenstra and Hanson (1995) model.
Figure 4.21: The evolution of the wage gap in comparison to the US

Source: Feenstra and Hanson (1995)
4. MIGRATION IN THE 2 × 2 × 2 MODEL 209
Figure 4.22: FDI, remittances and aid

Source: The Economist
4 Migration in the 2 × 2 × 2 model
We have seen that FDI is a potential source of turmoil in the labor market in both de-
veloped and developing countries. Changes in comparative advantage due to changes
in capital costs can affect countries’ competitiveness and offshoring activities. Figure
4.22 compares the major capital sources in developing countries: FDI, remittances,
portfolio flows, and foreign aid. Not surprisingly, foreign direct investments are by
far the most important source of capital supply for developing-country firms. Remit-
tances, which are transfers from migrants to residents in their country of origin, are
also important. Not only is migration a source of additional workers in the destina-
tion countries, the origin countries also benefit from large transfer payments in form
of remittances. The total amount of remittances sum up to 50 percent of total FDI.
Foreign aid is remarkably low in comparison to the amounts of FDI or the amount
of remittances. Portfolio flows are another form of international capital flows that are
closely related to FDI.
Thus, migration is an important source of development for emerging economies. Yet

the public debate on migration usually neglects those transfer payments, which is
striking given their large volume.
The relevance of migration for both the destination and the origin country motivates
the remainder of this chapter, where we discuss different frameworks that permit
analysis of the effects of migration. We begin with the popular Rybzynski theorem.
The theorem builds on the 2 × 2 × 2 model augmented by international factor flows.
Migration or FDI affects either the capital or the labor endowment in both countries.
The model is silent on the determinants of both migration and FDI as one of the
main results in the 2 × 2 × 2 model was international factor price equalization. Trade
without frictions should equalizes capital rental rates and wages across countries.
One may argue that developing countries have an incentive to set high tariffs at the
beginning of their development, which prevents factor price equalization. As a con-
sequence of economic development, burdens of trade fall so that factor prices become
more equal to the factor prices in the developed countries. Figure 4.23 supports this
argument. There is a positive correlation between openness and barriers to trade. Tak-
ing China as anecdotal evidence shows exactly the pattern highlighted above. China
prevented import competition at early stages of its development but reduced tariffs in
most industries after its accession to the WTO. At the same time we observe soaring
wages in China as shown in Figure 4.24.
4.1 Endowments in the 2 × 2 × 2 model
In order to say something about the effects of endowment changes on wages and out-
put we start by describing the allocation of capital and labor between the two sectors.
We do not need a two country model to derive the Rybzynski theorem - it is sufficient
to derive the effects of a change in endowments in one country. Thus, we will focus
on the foreign economy without considering the home economy. We begin with a
discussion of how to extend the Lerner diagram in a way that allows us to analyze
the effects of migration. Additional endowment points are inserted in the diagram
as depicted in Figure 4.25. Home is assumed to be capital abundant and Foreign
40 100
Trade Restrictiveness Index
60
20 80
Trade Restrictions and per-capita GDP, 2010
6 8 10 12
Per-capita GDP (log)
Figure 4.23:95%
Economic
CI developmentTrade
and Restriction
infant industry
Index protection
Trade Restriction Index/Per-capita GDP (log)
Hongbin Li, Lei Li, Binzhen Wu, and Yanyan Xiong 59
Figure 1
Real Annual Wages of Chinese Urban Workers
(deflated to 2010 prices)
10,000 40,000
37,147 RMB
Left: Deflated by U.S. GDP deflator, U.S. dollars
9,000 35,000
Right: Deflated by China's PPI, RMB
8,000
30,000
7,000
5,487 25,000
6,000
U.S. dollars
5,000 20,000
4,000 15,000
3,000
10,000
2,000
1,004 U.S. dollars
5,000
1,000
2,607 RMB
0 0
1978 1982 1986 1990 1994 1998 2002 2006 2010
Source: China Statistical Yearbooks.

Note: PPI is producer price index.
Figure 4.24: Annual wage growth in China

In the first two decades of the reform period from 1978 up to the later part
Source: Li, Li, Wu, and Xiong (2012)
of the 1990s, the growth of workers’ wages in Chinese urban areas was relatively
low, as shown in Figure 1. According to the Statistical Yearbooks published by
China’s National Bureau of Statistics, the annual real wage of a Chinese urban
worker increased only slightly from $1,004 in 1978 to $1,026 in 1997, at an average
annual growth rate of only 0.1 percent (before tax, including pensions, and again
converted from yuan to U.S. dollars using the current exchange rate, and to the
2010 level using the U.S. GDP deflator). This growth rate of China’s urban wages is
significantly lower than China’s annual real growth rate of 4.0 percent (in real U.S.
1
is assumed to be the labor abundant country. So far we neglected the endowment

points because we were mainly interested in the effects of trade on wages. Using the
factor price insensitivity result we drew the conclusion that endowments are unim-
portant for determining factor prices. In the Heckscher-Ohlin model, endowments
were an important determinant of comparative advantage but wages depended only
on prices. Moreover, endowments were fixed in the absence of migration and foreign
direct investments. Neglecting the endowment points was permissible in the previous
subsections but we are now interested in the effects of changes in endowments due to
migration.
The receiving country’s labor endowment increases. One may look at migration be-
tween the two countries in Figure 4.25 but we opt for a partial equilibrium analysis.
This decision is justified by the factor price equalization theorem. Factor prices in
Home and Foreign are identical if both countries are allowed to trade goods with
each other. Thus, there is no incentive for migration once the equilibrium is reached.
Generally speaking, a higher wage in the foreign country (from the perspective of the
migrant) is the only reason why she may want to migrate in such a simple framework.
Thus, a rise in the labor endowments must be associated with a higher wage in the
destination country compared to the country of origin. The model is rather simple
and does not allow more complex migration scenarios. Does that mean that the re-
sults are unrealistic? Perhaps yes, but it offers a nice foundation for the later chapters
on more recent developments in the migration literature. The Rybzynski theorem
serves as a starting point for the more sophisticated models on migration discussed
in the next chapter.
Firstly, we ask if a positive migration shock results in declining wages of both the
native and migrant workers in the destination country. We start from the well known
free trade equilibrium depicted in Figure 4.25. We assume free trade, which is obvious
from the location of the isoquants at Home and Foreign. In both sectors, we find
that the two countries are represented by identical isoquants. Remember that we
assume identical technologies across countries so that the shape of the isoquants must
be identical by assumption. This result holds under both autarky and free trade.
The location of the isoquants are determined by output levels associated with unity
revenue of firms. Those quantities are determined by the goods’ prices, which are
identical in both countries. Thus, the locations of the isoquants must be identical as
well.
The slope of the isocost curve is determined by the relative factor prices denoted by
∗ = w/r. An equal slope of the isocost curve in Home and Foreign implies
Sw = Sw
factor price equalization, which is again the reason why migration between Foreign
and Home cannot occur in equilibrium. Factor prices and everything else is equal in
both countries. Thus, there is no mechanism that explains factor movements between
Home and Foreign.
The endowment points can be easily inserted at the locations L̄, K̄ and L̄∗ , K̄ ∗ , which
were chosen arbitrarily. The question is how are capital and labor distributed across
the two countries? The answer is that both factors move across sectors until the full
employment conditions are fulfilled and all sectors pay identical wages and capital
rental rates. Moreover, the capital-labor intensities must be set according to the op-
timal capital-labor intensity that stems from the optimization problem. We solve the
cost minimization problem for the case that revenue in both sectors is exactly equal
to one. This minimization problem gives a unique solution but the solution is not
the one we need to solve the full employment conditions. To solve both we draw fac-
tor intensity rays into Figure 4.25 that generalize the optimal solution for all possible
combinations of L and K in accordance with the optimal factor intensity determined
for the unit cost scenario.
214
Figure 4.25: Industry specific labor and capital demand
Home Foreign
𝐾𝐾 𝐾𝐾 ∗
�
𝐾𝐾 ��
𝑌𝑌1∗ = 1/𝑃𝑃1𝑤𝑤
𝐾𝐾1 𝑌𝑌�1 = 1/𝑃𝑃1𝑤𝑤
𝑆𝑆𝑤𝑤∗
𝑆𝑆𝑤𝑤
��
𝑌𝑌2∗ = 1/𝑃𝑃2𝑤𝑤
�∗
𝐾𝐾
𝐾𝐾2∗
𝐾𝐾1∗
𝐾𝐾2 𝑌𝑌�2 = 1/𝑃𝑃2𝑤𝑤
𝐿𝐿2 𝐿𝐿1 𝐿𝐿� L 𝐿𝐿∗1 𝐿𝐿∗2 𝐿𝐿�∗

𝐿𝐿∗
We are searching for sectoral labor and capital demand points lying on the two factor
intensity rays. To achieve this goal we begin with shifting the two factor intensity rays
in both countries until they intersect the factor endowment points. The shifts must
be parallel to the original factor intensity rays for that the factor intensities remain
unchanged. The original and the shifted factor intensity rays span a parallelogram.
Two corners of the parallelogram fall together with the origin and the endowment
point in each country. The other two corners, namely the corners where the original
and the shifted factor intensity rays intersect, determine the input of capital and labor
in compliance with the full employment conditions.
We can determine sectoral capital and labor inputs by tracing the two intersection
points to the horizontal and vertical axis. We find that the optimal input of capital is
at point K1 and K2 in the Home economy.
The optimal input of labor is at point L1 and L2 . Do those points fulfill the full
employment conditions? This can be verified easily by adding up capital and/or
labor demand in both sectors. We know that the distance between the origin and
point L2 plus the distance between the origin and point L1 must be equal to L̄. We
can see that the full employment condition for labor is identified. Replicating the
same procedure for capital also yields a solution where the capital demand in sector
1 plus the capital demand in sector 2 sum up to the capital endowment K̄. But is this
solution in line with wage equalization between the two sectors? Both solutions lie on
the optimal factor intensity rays. We know that the optimal factor intensities in both
sectors are such that the isocost curve represents both industries. Factor prices do not
change as long as the slope of the isocost curve itself is equal to the relative factor
price and as long as the factor intensity remains constant. The same logic applies to
the Foreign country, where we find different input points compared to Home.
4.2 The Migration Shock
Figure 4.26 simulates a positive migration shock in the foreign country. As mentioned
previously, we assume that workers flow into the Foreign country, where the labor
endowment point shifts to the right. Notice that the source country of the positive
migration shock cannot be Home. We were already under free trade meaning that
both countries pay the same wage rate. The positive migration shock must be origi-
nated in another country not modeled in our parsimonious two-country setup with
perfect free trade. That is, some third country.
How do wages react to this migration shock? Surprisingly, wages are not affected
at all. At odds with common sentiments against migration, the economy is able to
absorb the additional work force without adjustments in the labor market. How is
that possible? A closer look at Figure 4.26 reveals that industry outputs adjust, rather
than wages. The sector that uses the inflowing factor intensively expands and the
other sector contracts.
Suppose that the migration shock can be represented by a shift in the endowment
0
point in Foreign from L̄∗ , K̄ ∗ to L̄∗ , K̄ ∗ . Figure 4.26 depicts this situation for the des-
tination country without further analyzing the source country effects. Notice that we
abstract from any capital flows. Thus, the location of the endowment point on the
vertical axis is not affected by the migration shock. The dashed factor intensity rays
no longer intersect the endowment point, which shifts to the right. We therefore shift
the factor intensity rays until they intersect the new endowment point. These are rep-
resented by the red lines in the digital version of the lecture notes. Inputs of labor in
0
the labor intensive industry increase from L2∗ to L2∗ but do we find that the increase is
only due to the additional work force going into sector 2? We can see that the labor
inputs in the other sector 1, decrease. This implies that all migrant workers move to
the labor intensive sector 2, and additional workers from the capital intensive sector 1,
are drawn into the labor intensive sector 2. The sector that uses the migrating factor
Figure 4.26: Changes of capital and labor input due to migration
Foreign
𝐾𝐾 ∗
𝑆𝑆𝑤𝑤
��
𝐾𝐾 ∗
𝐾𝐾2∗′
𝐾𝐾2∗
𝐾𝐾1∗
𝐾𝐾1∗′
𝐿𝐿∗′ ∗
1 𝐿𝐿1
𝐿𝐿∗2 𝐿𝐿∗′
2 𝐿𝐿∗
𝐿𝐿�∗ ��
𝐿𝐿��
∗′
intensively expands and the other sector shrinks. This result can easily be verified
by looking at the movements of capital. Sector 1 not only looses labor, but capital
inputs decline as well. The capital freed-up in the capital intensive sector is absorbed
by the labor intensive industry to keep the factor intensities at the initial level before
migration. Thus, constant factor intensities in both industries imply that the wage
and the capital rental rate remain constant. The migration shock affects the output of
both industries but leaves factor prices unchanged.
As long as the endowment point shifts within the cone of diversification, solutions
that comply with the old factor prices are possible.
Figure 4.27: Cone of diversification
218
Foreign
𝐾𝐾 ∗
Cone of Diversification
𝑆𝑆 ∗
𝐿𝐿∗
Figure 4.27 illustrates this cone of diversification spanned by the two factor intensity
rays. Labor and capital can move between the two sectors in a way that allows them
to continue to produce with the initial factor intensities but only as long as the move-
ments are not strong enough to push the endowment points out of the cone. For
extreme migration scenarios we may find a situation where the Rybzynski theorem
does not hold anymore. The endowment point is less likely to move out of the cone
if migration is accompanied by heightened foreign direct investment which shifts the
endowment point diagonally.
Rybzynski Theorem:
Changes in factor endowments through migration or foreign direct investment have an
effect on output leaving factor prices constant.
We can summarize the effects on labor and capital input as follows:
• The sector that uses the migrating factor intensively expands:
– Point L2∗ moves to the right ⇒ More labor is used in the labor intensive
sector
0
– Point L1∗ moves to the left ⇒ Less labor is used in the capital intensive
sector
• The sector that uses the other factor intensively contracts:
– Point K2∗ moves up ⇒ More capital is used in the labor intensive sector
– Point K1∗ moves down ⇒ Less capital is used in the capital intensive
sector
4.3 Formal proof of the Rybzynski theorem
The Rybzynski theorem can be proved algebraically using Jone’s algebra on endow-
ment changes. We can construct the total differential of the factor endowments
a1L dY1 + a2L dY2 = d L̄ (4.57)
a1K dY1 + a2K dY2 = dK̄ (4.58)
aiL , aiK , the optimal factor intensities obtained form the optimization problem, do not
change since goods’ prices are constant and since the optimal factor intensities are
independent from endowments As before we obtain
Y1 a1L dY1 Y2 a2L dY2 d L̄

+ = (4.59)
L̄ Y1 L̄ Y2 L̄
Y1 a1K dY1 Y2 a2K dY2 dK̄
+ = (4.60)
K̄ Y1 K̄ Y2 K̄
This can be rewritten in percentage-changes
λ1L Ŷ1 + λ2L Ŷ2 = L̂ (4.61)
λ1K Ŷ1 + λ2K Ŷ2 = K̂ (4.62)
As argued previously for the analytical proof of the Stolper-Samuelson Theorem we

are able to simplify using the following definitions: The change in output can be
written as dY/Yi = Ŷi and likewise λiL = (Yi aiL / L̄) = Li / L̄ with λ1L + λ2L = 1 and
λiK = (Yi aiK /K̄ ) = Ki /K̄ with λ1K + λ2K = 1.
It is more convenient to write the system of equations in Matrix notation as
    
 λ1L λ2L   Ŷ1   L̂ 
  =  (4.63)
λ1K λ2K Ŷ2 K̂
5. THE MARIEL BOAT LIFT 221
The matrix can be rearranged as
    
 Ŷ1  1  λ2K −λ2L   L̂ 
= (4.64)
|λ| −λ
   
Ŷ2 1K λ1L K̂
The determinant of this matrix can be computed as |λ| = λ1L − λ1K = λ2K − λ2L < 0.
It is negative because we assume that industry 1 is the capital intensive industry.
Suppose that the labor stock increases due to migration so that L̂ > 0 and K̂ = 0. We
can solve the matrix from above for output in industry 1 as
λ2K
Ŷ1 = L̂ < 0 (4.65)
(λ2K − λ2L )
and output in industry 2 as
−λ1K
Ŷ2 = L̂ > L̂ > 0 (4.66)
|λ|
In line with the predictions of the Rybczynski Theorem we are able to find a contrac-
tion of output in industry 1, which is the capital intensive industry, and an expansion
of output in industry 2, which is the labor intensive industry. The latter result stems
−λ1K
from the fact that (λ1l −λ1K )
L̂ > L̂. We know that L̂ is positive due to the migration
shock. We also know that the denominator and the nominator are both negative.
+λ1K
Multiplying both by −1 yields (λ1K −λ1l )
. The whole term must be greater than 1.
5 The Mariel boat lift
A famous application of the difference in differences method that investigates the

labor market outcomes of a sudden migration shock caused by the Mariel botlift is
Card (1990). The underlying logic of this estimator can most easily be understood by
having a look at its graphical representation in Figure 4.28 adopted from the module
"Angewandte Ökonometrie".
𝛼1
𝛼2
𝑡
𝑡′
Figure 4.28: Difference in differences treatment effect
Here, α1 and α2 represent the averages of a given dependent variable within two
groups: The treatment group (α1 ) that is subject to an intervention in t0 and the con-
trol group (α2 ) that is not. The difference in differences estimate is then derived by
comparing the distance between both groups before and after t0 .
The most important assumption facilitating this identification strategy can also be
understood from Figure 4.28: By imposing "parallel trends", researchers stipulate that
other than treatment, no additional confounders drive the difference between both
groups. Or, put another way: If no treatment was assigned in period t0 , α1 would
follow the dotted line.
The study by Card (1990) has already been sketched out above, but shall be discussed
in more detail here. Since Cuban refugees are presumed to compete primarily with
black and particularly unskilled black workers, the authors initially compute the dif-
ferences displayed in Table 4.29 for those two sub-groups. It is within these groups
where a negative effect is expected to be identified in the data, specifically around the
1980 migration shock, i.e. comparing the years 1979 and 1980/81.
Figure 4.29: Estimation results from Card (1990)
Taken from Card (1990)
Overall, 8 differences are displayed in each row with standard errors added in paren-
theses below. While the first 4 columns depict all black workers, the latter 4 only
include those that are "low-educated". Focusing initially on the first part of the analy-
sis within the red box drawn in the digital "Studienbrief", result can be disaggregated
further into wage and employment effects. Looking at the former within the first
yellow box, additional clarifications are necessary: Card (1990) presents yearly differ-
ences both without and while adjusting for additional observed covariates. Looking
again at the former, i.e. (Difference in Log Wages, Miami - Comparison, actual), the rel-
evant numbers have been highlighted in yellow. Comparing both years, the wage
differential does not appear to have changed much. The average "treatment effect"
can be computed by taking the difference of these differences:
TE1980/79 = −0.16 − (−0.15) = −0.01 . (4.67)
If additional characteristics are controlled for (adjusted differences in Log Wages),

this difference is reduced to −0.12 − (−0.12) = 0. We can therefore conclude that
the shock has had no impact on the wage differential between both compared groups.
Technically, one would still need to test this difference for significance. However, since
the identified effect is extremely small, one can safely assume it is not.
The unadjusted difference in differences in 1982 compared to the base year is more
interesting. It can be computed as:
TE1982/79 = −0.24 − (−0.15) = −0.09 . (4.68)
Looking at this later period, the effect is indeed nine times higher than the originally
determined value. Card (1990) briefly discusses the possibility of the migration shock
having an impact on the labor market delayed by 2 years. The numbers support such
a conclusion. However, he also points out that the US was going through a recession
in 1982, which should have had repercussions for the labor market.
This criticism would be valid if the recession had affected regions differently. We
already underlined, that, in order to apply the method successfully, regions have
to be influenced equally by all changes. If this assumption cannot be upheld, the
differences might be driven by other characteristics except the treatment. If, however,
the recession affected all regions equally strongly, the difference would be unchanged.
In regression analysis the impact of the recession can be controlled for by including
yearly dummies.
A similar analysis is conducted for Cubans already in the US, results are displayed in
Table 4.30. Differences in column 10 are computed based on data from columns 2 and
9. Results likewise fail to identify a noteworthy effect of the migration shock on the
wage differential for the comparison of individuals in- and outside of Miami.
Figure 4.30: Further estimation results from Card (1990)
Taken from Card (1990)
Card (1990) discusses possible explanations for these surprising findings. One plausi-
ble reason is the migration shock’s negative impact on internal migration into Miami.
The influx of refugees could have diverted other groups originally headed there to
other regions. According to Card (1990) this logic is supported by the data as internal
migration to Miami has indeed decreased significantly in the years after the Mariel
boatlift.
Chapter 5
International migration
1 International migration
Besides trade and FDI, international migration from developing to developed coun-
tries, commonly labelled as "South-North" migration, is one of the key elements of
globalization. The most recent available data reveal that permanent immigration to
OECD countries has increased in 2014 with respect to the previous years: 4.3 millions
of people migrated to OECD countries in 2014. This has constituted an increase of
approximately 6 percent with respect to the previous year. Moreover, the absolute
number of individuals migrating to OECD destinations has exceeded 4.1 million -
the level of immigration before the 2007/08 global economic crisis. See Table 4.1.
(OECD, 2015). In relative terms, immigrants (i.e. foreign-born individuals) resident
in OECD countries accounted for, on average, 13 per cent of the total population of
such countries in 2013. See Table 4.2 (OECD, 2015).
The United States is the main destination country in terms of inflows of permanent
immigrants. In 2013, the US received almost 1 million permanent immigrants. Ger-
many and UK are the second and the third largest receiving countries, respectively.
See Table 4.3 (OECD, 2015) for further details. The most important country of origin
227
228 CHAPTER 5. INTERNATIONAL MIGRATION
for permanent immigrants is China. In 2013 outflows of immigrants from China to

OECD countries accounted for 10 per cent of the total flows to OECD destinations.
Next in line are Romania and Poland, respectively, which have become the most im-
portant sending countries from EU accession countries in 2007. Next in line is India,
whose outflow of immigrants in 2013 accounted for 240 thousand individuals, i.e. 4.4
per cent of the total flows to OECD countries. See Table 4.4 (OECD, 2015).
Figure 5.1: Permanent migration flows to OECD countries,

2006-2014.
Source of the data: OECD International Migration Outlook 2015.

1. INTERNATIONAL MIGRATION 229
Figure 5.2: Foreign-born individuals as percentage of the total

population. 2000 and 2013.
Source of the data: OECD International Migration Outlook 2015.
The aim of this chapter is to answer the following fundamental questions related to
the stylized facts presented above. Why do people migrate? We answer this question
from a theoretical perspective in sub-section 4.2. The objective is to understand the
main economic reasons behind the individual migration decision. We then look at the
empirical counterpart of the same question. Indeed, the chapter presents the empirical
evidence on the main determinants of bilateral migration flows among OECD coun-
tries. Not all individuals of a given population have the same propensity to migrate,
and immigrants tend to have particular characteristics (for example, they may be the
most skilled people from a given population). Therefore, we focus on the second
fundamental question, Who migrates? Most of the academic and policy debates have
been related to the effects of immigration on the destination country’s labor market.
In particular, the focus has been on the effect of immigration on the host country’s
wages and on the labor-market competition between natives and foreign-born work-
ers. Sub-section 4.3 therefore presents an overview of the empirical and theoretical
literature seeking to answer the question: what are the effects of immigration for the host
Table 5.1: Inflows of permanent immigrants into selected OECD

countries, 2007-2013.
Year % change
2007 2008 2009 2010 2011 2012 2013 2013/12 2013/07
USA 1052400 1107100 1130200 1041900 1061400 1031000 989000 -4 -6
DEU 232900 228300 210500 222500 290800 400200 468800 17 101
GBR 343300 317300 359200 394800 322600 286100 291000 2 -15
FRA 206500 214000 212100 224300 231500 251200 259800 3 26
CAN 236800 247200 252200 280700 248700 257900 258600 0 9
AUS 191900 205900 221000 208500 219500 245100 253500 3 32
ITA 571900 490400 390300 355700 317300 258400 245800 -5 -57
ESP 691900 409600 334100 300000 291000 209800 195300 -7 -72
CHE 122200 139100 114800 115000 124300 125600 136200 8 11
NLD 80600 90600 89500 95600 105600 96800 105500 9 31
SWE 74400 71000 71500 65600 71800 81700 86700 6 17
KOR 44200 39000 36700 51100 56900 55600 66700 20 51
AUT 47100 49500 45700 45900 58400 67100 65500 -3 38
NOR 43900 49300 48900 56800 61600 59900 60300 1 37
BEL 50300 51200 64200 64100 64300 65700 60300 -8 20
JPN 108500 97700 65500 55700 59100 66400 57300 -14 -47
MEX 6800 15100 23900 26400 21700 21000 54400 159 700
DNK 30300 45600 38400 42400 41300 43800 52400 20 73
NZL 51700 51200 47500 48500 44500 42700 44400 4 -14
IRL 120400 89700 50700 23900 33700 32100 40200 25 -67
CZE 100600 76200 38200 28000 20700 28600 27800 -3 -72
PRT 42800 71000 57300 43800 36900 30700 27000 -12 -37
Total 4451400 4156400 3893400 3791200 3783600 3750600 3864100 2 -14

Source of the data: OECD International Migration Outlook 2015
1. INTERNATIONAL MIGRATION 231
Table 5.2: Top 20 countries of origin of new immigrants to the

OECD.
Immigration % of total % of total Expatriation

to OECD OECD world rate
countries1 inflows population (per million
population)
2007 2009 2011 2012 2013 2013 2013 2013
CHN 520 463 531 507 557 10.3 19.1 410
ROU 557 274 310 294 300 5.5 0.3 15045
POL 339 221 277 284 290 5.3 0.5 7528
IND 213 229 243 228 240 4.4 17.6 192
MEX 164 180 162 166 152 2.8 1.7 1241
PHL 169 164 161 159 148 2.7 1.4 1505
USA 117 133 137 135 147 2.7 1.4 464
ITA 66 73 85 99 127 2.3 0.8 2130
GBR 149 129 108 111 108 2.0 0.9 1686
DEU 150 126 116 105 107 2.0 1.1 1323
FRA 82 93 96 97 105 1.9 0.9 1587
VNM 89 77 95 94 102 1.9 1.3 1139
HUN 37 43 68 87 96 1.8 0.1 9741
MAR 152 143 112 96 95 1.7 0.5 2865
BGR 87 67 98 101 93 1.7 0.1 12829
ESP 24 40 52 75 93 1.7 0.7 1988
RUS 68 68 71 77 86 1.6 2.0 597
PAK 75 77 106 86 75 1.4 2.6 412
COL 89 72 68 65 73 1.3 0.7 1513
KOR 72 79 71 70 72 1.3 0.7 1432

1 Data are in thousands. Source of the data: OECD International Migration Outlook 2015.
country?
2 Why do people migrate?
This section briefly reviews the main theories that seek to explain the reasons behind
individual migration decisions. It first focuses on the traditional framework proposed
by Sajaastad (1962) and then illustrates more recent theories, which take migration as
a temporary rather than a permanent decision. The section concludes by reviewing
empirical evidence on the determinants of international migration movements.
2.1 The determinants of the migration decision
Sajaastad’s theoretical framework. Economists often view the individual migration

decision as an investment in human capital. Indeed, income maximizing individu-
als choose a location which provides the highest net return to their human capital
(Chiswick and Miller, 2014a). The pioneering work of Sajaastad (1962) offered a gen-
eral framework to rationalize migration decisions, even if he did not give any mathe-
matical formalizations of the migration choice, he implicitly conceived the migration
decision as an inter-temporal optimization problem. In Sajaastad’s framework the
individual compares earnings in the origin labor market with earnings opportunities
in the alternative labor market when taking the migration decision.1 The individual
also considers the cost of living in the origin and in the alternative labor market, and
the migration costs. He then chooses the labor market which maximizes his inter-
temporal utility.
For a deeper understanding of Sajaastad’s insights on migratory behavior, it is useful

to illustrate a formalization of his framework. In particular, we show one provided by
1 Note that the alternative labor market may be interpreted either as a market in a different geographical
location or as a different sector or occupation. Therefore, from the individual maximization perspective,
there are no differences between international or internal/within-country migration between areas or
industries.
2. WHY DO PEOPLE MIGRATE? 233
Chiskwick and Miller (2014a).
Consider, in discrete time, a T periods model, where the individual retires at time T.
Suppose that there are two locations, O (the origin), and D (the alternative location
or the destination labor market). Assume that EarnO
t are the earnings in location O
at time t, and EarntD are the earnings in location D at time t. Similarly, the costs of
living in locations O and D at time t can be measured by CostLivO D
t and CostLivt ,
respectively. Migration involves both non-monetary and monetary costs, which are
indicated by C (dist, X ), where dist is the geographical distance between location O
and location D, and X are characteristics influencing the migration cost. Denoting the
discount rate with i, the individual inter-temporal utility from migration, φ, is given
by
T T
( EarntD − EarnOt ) (CostLivtD − CostLivO
t )
φ= ∑ (1 + i ) t
− ∑ (1 + i ) t
− C (dist, X ). (5.1)
t =1 t =1
The individual decides to move from O to D if the net present value from migration
is positive, i.e. φ > 0. In case of more than one alternative destination, the individual
computes φ for each of the alternatives and he then chooses the location yielding the
higher net present value.2
Despite being simplified, Sajaastad’s view of migration as an individual investment

decision in human capital has shaped most of the subsequent formalizations of the
migration choice. Moreover, the modern neoclassical analyses of the migration deci-
sion are rooted in Sajaastad’s basic framework (Chiswick and Miller, 2014a). How-
ever, it is important to point out the simplifications and limitation of his theoretical
setting. Firstly, his framework only allows for migration as a “once-and-for-all” de-
cision, namely the individual can migrate only once. The possibility of a temporary
movement is not considered and the return migration decision is ruled out. Similarly,
2 Sajaastad’s framework can be formulated also in continuous time, reported here for the sake of com-
pleteness. Following Chiswick and Miller’s notation, the net present value from migration in continu-
RT O −rt dt − C ( dist, X ).
EarntD − EarnO D

ous time becomes φ = t − CostLivt + CostLivt e
t =0
Sajaastad’s setting does not incorporate the possibility that the migration probability
varies over an individual’s life-cycle. For instance, the probability of migration may
be age dependent, a function of the individual’s career path, and could depend on his
relative preference of income over leisure time (Chiswick and Miller, 2014a).
The model also disregards remittances, i.e. the fact that migrants may send part of
their earnings from the destination back to the home country. Introducing remittances
would modify the individual net present value of migration, increasing the gains from
migration. However, if remittances are considered, the individual should also take
into account the exchange rate between the home and the destination country and the
functioning of the financial markets, thus requiring a richer and more complex model.
Implicit in Sajastaad’s framework is the fact that the individual is perfectly informed
about the earnings he would get in the destination. Uncertainty is excluded from his
model.
It is also implicit in the Sajaastad model that the migration decision only happens
at the individual level. However, the decision may also be taken at the family level.
Or, the presence of a family member may influence an individual’s migration and
location choice.
A more general and final observation on Sajaastad’s framework concerns the funda-
mental assumption of income maximization. It is well known that different factors
may drive migration decisions. For instance, migration can be motivated by the in-
tention to acquire human capital abroad (e.g. international migration by students) or
due to family reasons (e.g. family reunification). Individual also migrate when seek-
ing international protection or political asylum. Hence, the assumption of income
maximization does not include all the possible reasons for migration. The framework
of Sajastaad should be interpreted broadly, assuming that the individual chooses the
location which maximizes his utility, but not necessarily his income (Chiswick and
Miller, 2014a).
As anticipated, most of the subsequent theories seeking explanations for migration

constitute extensions of Sajastaad’s framework or develop the analysis of different

migration motives. Alternative migration theories have considered migration as a
consumption decision in which individuals migrate not only in search of higher net
returns to human capital, but also for different locational amenities, intended as non-
tradable goods or public goods (Chiswick and Miller, 2014a). Other strands of con-
tributions to economic migration theories have focused on the role of networks (e.g.
family, workers, and social networks) as factors which foster migration. Indeed, the
presence of an established network of migrants in the destination country lowers the
migration costs (e.g. the psychological or informational costs). Other extensions have
considered the family as the unit making the migration decision. One element of
uncertainty surrounding the migration decision has been introduced by Harris and
Todaro (1970). Their seminal paper modelled the phenomenon of migration from
the rural to the urban sector in developing countries. They argued that individuals
migrating from rural urban areas may not immediately find employment in the city.
Hence, to model this aspect of uncertainty, they introduce expected wages in the ur-
ban sector. Rural-urban migration will occur as long as the wage in the rural area
equates the expected urban wage, net of migration costs.
Beyond Sajaastad’s framework: modelling temporary migration. As previously

discussed, Sajaastad’s framework is very stylized and the individual migration deci-
sion is more complex than what it is suggested by the model discussed so far. One
of the main simplifications is that only permanent migration movements are modelled
in the standard theoretical framework and in its extensions. The possibility to tem-
porarily migrate is overlooked. Once migration has taken place, individuals remain in
the destination country permanently, and the possibility of returning to their home
country (return migration), or to move to another country (out-migration) is neglected.
The lack of attention to the temporary nature of immigration represents a gap in the
literature. Indeed, most migration movements tend to be temporary. If we look at the
individual migration intentions, we can infer that immigrants have a clear tendency to
remain only temporarily in the new country of residence. Interestingly, the fraction
of immigrants planning not to stay permanently in the host country is highest among
the most highly educated individuals. As suggested by Figure 5.3, which is based on
US and German survey data on immigrants’ intentions, there is an inverse U-shaped
relationship between the willingness to stay in the host country and the education
level, measured by years of schooling (see Dustmann and Görlach, 2016). Similarly,
if we look at realized migration movements, OECD estimates reveal that between one
fifth and half of migrants leave the destination country within the first five years fol-
lowing their arrival (Dustmann and Görlach, 2016).
Figure 5.3: The intention to stay abroad

Dustmann and Görlach: The Economics of Temporary Migrations 101
Source: Dustmann and Görlach (2016)
Panel A. US New Immigrant Survey Panel B. German Socio-economic Panel

0.45 0.65
0.4
0.6
Fraction
Fraction
0.35
0.55
0.3
0.25 0.5
0 5 10 15 20 5 10 15 20
Years of schooling Years of schooling
Figure 2
Notes: Fraction intending to stay permanently by level of schooling: (a) legal immigrants in the United States
aged 18–64 who arrived at age 16 or older; (b) immigrants in Germany aged 18–64 who arrived at age 16 or
older. Gray lines show the 95 percent confidence bounds of the fitted local polynomial regression line.
Sources: New Immigrant Survey (2003/2004); Socio-economic Panel, 2000–2012.
wages and savings. Since these choices are economic conditions in their home country
Despite this, the view of migration as an investment in human capital remains es-
determined by anticipated r emigration plans environment, as well. The key element that
(rather than e
x post realizations), we
sential and most recent theoretical studies have con- links economic
begun circumstances
to produce dynamic in the home
models
sider immigrants as permanent or temporary country and economic choices of migrants
depending
focusing on theontemporary
their intended migration
nature dura-
of migration inand
theonhost
thecountry is the intention
phenomenon of return of mi-
the
tions at any given point in time. migrant to remigrate at a later stage, and to
If the
gration. Theintended
model return itself by
proposed affects choices and
Dustmann spend part of(2016)
Görlach his or her
is alife cycle inexample.
notable the home
such as immigrants’ labor force participation country (or, in more complex settings, in a
3
andgeneral
Their human-capital
framework investment
allows one decisions, third
to analyze Obviously,
country).motives
different for researchers
return migra-have
then this introduces additional complexi- to make choices about which economic deci-
ties into the estimations of career profiles or sions to model and which sources of shocks
other types of behavior over and above those and heterogeneity to consider, depending on
instigated by selective out-migration. We the research question to be addressed. We
illustrate what we view as some of the core will discuss some of these below.
tion. Specifically, they consider preferences for consumption in the origin country
and differences in purchasing power parity between the origin and the destination as
possible factors influencing the return migration choice. Moreover, they analyze how
differences between the origin and the destination country in the returns to human
capital and skill accumulation influence the return migration decision. In their basic
framework, they do not account for the migration decision. Instead, they focus on
individual post-migration utilities, comparing the value of the option of staying in
the destination country with the value of the returning.
In their theoretical setting, Dustmann and Görlach (2016) assume that there are two
locations L = o, d where o is the origin and d is the destination. A is the individual’s
stock of assets and a is the individual’s age. Individual skills are denoted by S, and
they are assumed to accumulate according to Sa = Sa−1 + θSL . Notice that Sa−1 is
the level of skills attained in the year before the individual’s current age, i.e. a − 1.
Consider an individual aged a at the time of observation. The equation of skills’
accumulation indicates that the individual skills at a given age a are equal to the
skills accumulated since birth and the previous age, a − 1, plus a parameter θSL > 0.
Specifically θSL , i.e. the increase of skills within the year of observation, can differ
between locations and can change over time.
We distinguish between human capital and skills. In particular, we indicate the value
of attained skills as human capital. The same skills can be evaluated differently by the
two countries o and d (indeed, as mentioned previously, the parameter θSL can vary
across locations). This can be due to the fact that certain skills are not demanded
in a given location, but are required in the other one. Therefore, two individuals
with identical skills may have different human capital depending on the country they
are located in, i.e. human capital is country-specific. Hence, the individual’s human
capital in location L reflects both the stock of skills and their evaluation in the country
of residence. The country-specific level of human capital associated with a certain
L
level of skill, Sa , is given by H L = Sα1 , where α1L is a measure of preference for skills.
Figure 5.4: The wage profile for different α
238
𝑌𝑌 𝑌𝑌
𝑌𝑌(𝛼𝛼0𝐿𝐿′′′ , 𝛼𝛼1𝐿𝐿 )
𝑌𝑌(𝛼𝛼0𝐿𝐿 , 𝛼𝛼1𝐿𝐿′′′ )
𝑌𝑌(𝛼𝛼0𝐿𝐿′′ , 𝛼𝛼1𝐿𝐿 )
𝑌𝑌(𝛼𝛼0𝐿𝐿 , 𝛼𝛼1𝐿𝐿′′ ) 𝑌𝑌1′′

𝑌𝑌(𝛼𝛼0𝐿𝐿′ , 𝛼𝛼1𝐿𝐿 )
𝑌𝑌(𝛼𝛼0𝐿𝐿 , 𝛼𝛼1𝐿𝐿′ )
𝑌𝑌1′
𝑌𝑌1
𝑌𝑌1
𝑆𝑆 𝑆𝑆
CHAPTER 5. INTERNATIONAL MIGRATION
1 1
Thus, individuals can increase their level of human capital by changing location, i.e.
through migration3 .
It is further assumed that α0L is the log of the rental rate of human capital received
by the individual in location L. Earnings therefore differ between locations either
because the two countries reward human capital differently (e.g. due to different
rental rates α0L that shift the wage profile up or down), or because the returns to skills
α1L are different. The latter changes the shape of the wage profile without shifting it
up or down. Specifically, the earnings of an individual with skills S are wH L . Hence,
the log earnings in location L are equal to
lnY L (S) = y L (S) = α0L + α1L lnS, (5.2)
L
which follows from the fact that H L = Sα1 . Notice also that ln w = α0 . Figure 5.4
depicts the wage profile for different levels of α0L and α1L . In the left panel, α0L is
kept constant, while α1L changes. Suppose that the level of attained skills is equal to
one. The term α1L ln S becomes zero for all possible levels of α1 so that the wage is
Y (S = 1) = exp(α0L ).
Notice that α0L is in logs and ln 1 = 0. The three depicted wage curves in the left
panel of Figure 5.4 are based on the same α0L but different α1L . All the curves intersect
at S = 1 but the higher the parameter α1L , the steeper the wage profile. Notice that
000 00 0
α1L > α1L > α1L . A steeper wage profile is associated with a higher relevancy of
skills. Wages are increasing faster in the level of skill when α1L goes up. The right
panel of Figure 5.4 illustrates the role of α0L for constant α1L . The three wage profiles
000 00 0
are drawn for three different levels of α0L > α0L > α0L . A higher α0L shifts the wage
profile outwards, which has the same effect as the change in α depicted in the left
panel: the rewards for skill is increasing in α0L .
3 Notice that we are interested in the migration decision at the point of observation, when the individual
is aged a. Thus, all the events that occurred before the time of observation, implicitly summarized in
the variables indexed a − 1, are not relevant for the analysis.
Workers spend their wage on consumption c, which yields utility u. Individuals

can accumulate assets if consumption expenditure is less than income. Preferences
depend on the location of the individual. The immigrants’ individual preferences
for consumption are biased towards the destination country if π > 1. It is further
assumed that the individual utility in the destination d is given by ud (π, c). The
utility in location o is given by uo (1, c). The individual may also prefer consumption
in the origin country if π < 1. It is further assumed that consumption increases
∂u L
utility i.e. ∂c > 0, and that the relative preference for consumption in the destination
∂ud
increases the utility at destination, i.e. ∂π > 0. One may assume that preferences in
both countries are identical and π is a shifter that shifts the indifference curve in the
destination country relative to the indifference curve in the origin country. Moreover,
∂2 u L
it is assumed that ∂c2
< 0.
Dustmann and Görlach (2016) do not model the migration and the return migration
decisions jointly. Instead, they compare the utilities of an individual who has already
migrated to the destination country. Specifically, they model temporary migration as
an individual optimization problem with respect to consumption and locations. They
indicate with V L (Ω a ) the value of an individual with characteristics Ω a = ( a, A, S) for
being in location L.
The earnings of a migrant depend on the stocks of skills S, which in turn are deter-
mined by the individual’s age and by the country-specific parameter θSL . The higher
the wage, the higher consumption and utility. Moreover, the value at age a depends
on utility and also on the accumulated assets at that age. Dustmann and Görlach
(2016) assume that at the beginning of every period the immigrant decides whether
to remain in the destination country for one more period or to return to the origin
country according to
V (Ω a ) = max [V d (Ω a ), V o (Ω a )]. (5.3)
In each period, the migrant chooses the option associated with a higher utility and
stays if V d (Ω a ) > V o (Ω a ), thereby realizing the value V d (Ω a ). Conversely, the mi-
grant returns to the origin country if V d (Ω a ) < V o (Ω a ), ending up with a value

V o (Ω a ). Conditional on the location choice, the individual decides about the level of
consumption by solving the consumption maximization problem, which reads as
V d (Ω a ) = max ud (π, c) + βV d (Ω a+1 )

c
s.t. A a +1 ≤ (1 + r ) A a + Y d ( S a ) − c a
if the individual remains in the destination country. The maximization problem is

solved recursively by taking the discounted future value into consideration. This re-
flects a trade-off that the migrant faces. Indeed, the individual can increase consump-
tion today by using some of her assets, but this reduces the future value through lower
A a+1 . The trade-off depends on the individual’s time preference rate, β. The higher
the preference rate, the more the consumer prefers future consumption over today’s
consumption. Notice that r is the real interest rate and Y d (Sa ) is the real income in
country d. The individual maximizes utility over consumption subject to the budget
constraint. Individuals can spend their wage Y d (Sa ) on consumption. Moreover, they
can use their assets at time a for consumption. The assets in the following period de-
pend on the difference between expenditure at age a and the amount of income plus
assets. Suppose that the migrant decides to use all income in the respective period for
consumption without modifying their assets. Next period’s assets, which are given
by the assets at time a plus interest payments rA a , would increase due to interest pay-
ments amounting to rA a . The consumption maximization problem in case of return
to the origin country is given by
V o (Ω a ) = max uo (π, c) + βV o (Ω a+1 )

c
s.t. xA a+1 ≤ x (1 + r ) A a + Y o (Sa ) − c a (5.4)
where x is the real exchange rate and Y o (Sa ) is real income in country o. The migrant
has to convert her assets into the currency of her origin country. Indeed, those assets
were accumulated in the destination country (that is, in different currency to the
origin). Instead, wages are not transformed into the foreign currency since they are
obtained in the new country of residence (i.e. the origin), and paid in its currency.
It is further assumed that when migrating just for one period, a = a0 , A = A0 and
S = So hold. In the last period T, the final value function is set equal to V L ( T, A, S) =
0. Individuals use their entire assets for consumption smoothly over their life cycle
without leaving any assets that can be inherited by the next generation.
The migrant chooses at the beginning of each period whether to stay in d or to return
to o. He compares V d (Ω a ) and V o (Ω a ), given consumption in the previous period
which results in assets and skills accumulation. The migrant is indifferent between
staying in d and returning to o when V d (Ω a ) = V o (Ω a ), i.e.
ud (π, Y d (Sa ) + (1 + r ) A a − AdA+1 ) + βV d ( a + 1, Ada+1 , θSd Sa ) =
uo (π, Y o (Sa ) + (1 + r ) xA a − xAoA+1 ) + βV o ( a + 1, xAoa+1 , θSo Sa ). (5.5)
Again, we have to consider the utility obtained from consumption in period a plus
the value obtained from consumption in the following periods summarized in V d ( a +
1, Ada+1 , θSd Sa ) and V o ( a + 1, xAoa+1 , θSo Sa ), respectively. Taking the difference between
V o (Ω a ) and V d (Ω a ) and rearranging, it follows that
uo (π, Y o (Sa ) + (1 + r ) xA a − xAoa+1 ) − ud (π, Y d (Sa ) + (1 + r ) A a − Ada+1 ) =
β[V d ( a + 1, Ada+1 , θSd Sa ) − V o ( a + 1, xAoa+1 , θSo Sa )], (5.6)
where the left hand side of (5.6) is the current gain from returning to o in the current
period, while the right hand side is the future gain from remaining in d one period
more, discounted by the time preference rate, β. Depending on the values of the
parameters, return migration may occur for different reasons. Five possible outcomes
on return migration derived from the model are illustrated below.
The model is dynamic, which implies that the value associated with being in the des-
tination or origin country adjusts in each period. The model allows for three possible
scenarios (or solutions): the case of no migration and the one of permanent migration
(both are corner solutions), and temporary migration. In the first case, individuals may
find it optimal to remain permanently in their origin country, without migrating at
all. In the second case, individuals may decide to migrate from their place of origin
to the destination country permanently, i.e. without returning to their home country.
In the third scenario, individuals who migrate to the destination may find it optimal
to return to the origin country after a certain number of years. The evolution of the
two different value functions V d (Ω a ) and V o (Ω a ) is such that the choice remains the
same in every period.
I: Permanent migration case Dustmann and Görlach (2016) argue that some particu-
lar assumptions have to be imposed in order to guarantee the existence of the unique
corner solution. In particular, the preferences for consumption in both countries must
be sufficiently equal (for instance due to π = 1), there must be no differences in
purchasing power across countries (i.e. the real exchanger rate is x = 1), and accu-
mulation of skills does not occur (i.e. θSd = θSo = 0). Given these parameters’ values,
two sub-cases can be distinguished, i.e. yd (S) > yo (S) (permanent migration) and
yo (S) > yd (S) (no migration). When wages are higher in d, individuals will migrate
there and will always remain in that location. Identical consumption levels are eval-
uated equally in both countries and the wage is exactly the same at all ages due to
constant human capital. A higher wage in the destination implies that consumption
and/or assets are higher in all periods. There is little to no consumption smooth-
ing over time because the conditions are expected to remain constant due to the fact
that the state variables do not change at all. Age increases but it has no effect on
wages and consumption. Marginal utility is the same in all periods so there is little
to no incentive for asset accumulation over time. A concave utility function suggests
consuming exactly the same level in every period. The current value associated with
return migration is negative and the future value of being in the destination country
is always greater than the one from returning to the country of origin. This scenario
corresponds to the permanent migration corner solution. Observe that if wages are
higher in the origin country o, migration to d never occurs under the assumptions
discussed above. Indeed, the value of being in the origin V o (Ω a ) would be higher
than the one associated with migrating to d, V d (Ω a ).
II: Temporary migration due to preferences for consumption in the origin country
Corner solutions are likely to occur in the model without human capital accumula-
tion, biased preferences, or differences in the purchasing power. In what follows, we
will relax each of the above assumptions sequentially in order to analyze the chan-
nels through which the choice of residence changes over the life cycle of a migrant.
Intuitively, we need a channel which either affects the preferences in both countries,
or increases total income over the whole life time of an individual. Income can be
transferred inter-temporally through asset accumulation.
First, we focus on the role of preferences for return migration by disregarding all the
other potential channels. In this case, despite higher wages at destination (yd (Sa ) >
yo (Sa )), individuals may return to the origin after some number of years spent in
the destination. This result can be derived without human capital accumulation. A
higher wage may act as an incentive to accumulate assets that can be spent after re-
turning home, once the future value of returning home outweighs the higher current
utility of earning a higher wage in the destination. The intuition behind the result
is that an individual would prefer to work in the place where income is highest and
consume where consumption yields the highest utility. This path can be achieved
through temporary migration. Indeed, individuals work when young and keep con-
sumption at a minimum in order to spend everything once they are old and return to
their origin country. Notice that individuals must give up some consumption today
in order to build up assets that can be consumed in the future. Return migration is
more likely if there are some preferences for consumption in the country of origin
(i.e. their marginal utility from consumption is higher in o than in d, i.e. π < 1). As
in the previous case, it is assumed that there are no differences in purchasing power.
The dynamics in the model are simplified thanks to the assumption that skills do
not accumulate in both countries (i.e. θSd = θSo = 0). Despite these simplifications, dy-
namic behavior is influenced only by differences in preferences which induce workers
to save today in order to spend tomorrow. This “sequential” approach allows us to
disentangle the different drivers of return migration and to consider them separately,
thus simplifying the complex phenomenon at hand. However, notice the existence of
an interior solution with return migration highly depends on parametrization. With-
out simulating the model, we would not be able to show the existence of an interior
solution of return migration.
In Graph 5.5 below we summarize the results of the simulations by Dustmann and
Görlach (2016)4 .
The simulation of the model is based on the following parameters’ values:
• Preferences at destination and origin: uo = c1/2 and ud = πc1/2
• Interest rate r = 0.05
• Time preference rate β = 1/(1 + r )
• Year of Death after migrating T = 30
• Initial parameters A0 = 0, a0 = 0, and S0 = 2
• Skill accumulation θSd = θSo = 0
o
• Rental rate on human capital expα0 = 0.25
• Relevance of skill parameter α1d = α1o = 0.3
• Bias of preference parameter π = 0.75 (low preference asset profile)

and π = 0.8 (high preference profile)
4 For more details on the simulations, refer to Dustmann and Görlach (2016).
The indifference condition plots the left hand side of equation (5.6) against the right
hand side of (5.6). The left hand side, i.e. the current value of staying one more
period, is declining. The individual accumulates savings in every period by realizing
a consumption level below the one that could be realized at the wage earned in the
destination country. The marginal utility would be higher if those units would be
consumed in the country of origin. However, the immigrant remains in the destination
to accumulate assets for future consumption. A change of location would reduce life-
time income such that fewer units could be consumed in total over the remaining years
from switching to the destination (year 0) to death (30 years after migration in the
calibration of Dustmann and Görlach (2016)). The lower the total assets accumulated
after returning home, the lower the future value of return migration (right hand side
of equation (5.6)). The individual will switch once the future value exceeds the current
value associated with the intersection of the solid and the dashed line in the left panel
of Graph 5.5. Dustmann and Görlach (2016) distinguish two scenarios that differ in
the level of π, i.e. the variable of interest in this case. For π = 0.75 (low preference
asset profile) the migrant must decided whether to go back to the country of origin
at age 14. For a lower preference π = 0.8 the switch is a little bit later at age 21.
The parameter π is closer to unity, which implies that marginal utility is still higher
in the place of origin but the gap narrows compared to the situation with π = 0.75.
The right panel in Graph 5.5 illustrates the earning and consumption program for the
high preference profile π = 0.8, where the switch occurs at age 21. Consumption is
constant before and after the switch but it jumps up to a higher level between age
20 and 21, which is the year the migrant returns home. The opposite pattern can be
observed for earnings, which are higher in the destination country. Panel A in Figure
5.6 illustrates the savings profile for both scenarios. Individuals accumulate savings
before returning home and spend them after returning home. The return decision is
different depending on the preference parameter π.
III: Temporary migration due to higher purchasing power parity in the destination
Immigrants may temporarily migrate when the purchasing power of the destination
112 Journal of Economic Literature, Vol. LIV (March 2016)
Panel A. IndifferenceJournal
conditionof Economic Literature,
112 Figure 5.5: Return migration duePanel B. LIV
toVol.
homeEarnings and consumption
(March
bias 2016) II) paths
(scenario
Current gain from returning
Future gain from staying
Panel A. Indifference condition Panel B. Earnings and consumption paths
Current gain from returning

Future gain from staying
Earnings
Consumption
5 10 15 20 25 30 5Earnings 10 15 20 25 30
Years since immigration Years
Consumption since immigration
5 10 15 20 25 30 5 10 15 20 25 30
Figure 4. Preference for the Home Country
Years since immigration Years since immigration
Figure 4. Preference for the Home Country

Figure 5.6: Savings program with home bias or differences in
purchasing power (scenario II and III)
Panel A. Preference for the home country Panel B. Higher purchasing power of the
host country currency
under high host country utility

Panel A. PreferenceSavings
for the home country
Savings under low host country utility
Savingspower
Panel B. Higher purchasing under low
of purchasing
the power
Savings under high purchasing power
host country currency
Savings under low host country utility Savings under low purchasing power
Savings under high host country utility Savings under high purchasing power
5 10 15 20 25 30 5 10 15 20 25 30
5 10 15 20 25 30 5 10 15 20 25 30
Years since immigration Figure 5. Asset AccumulationYears since immigration
country’s currency is higher in the origin country. Immigrants may increase their life-
Figure 5. Asset Accumulation
time consumption with return migration by exploiting this difference in purchasing
power through asset accumulation. Indeed, when returning, the savings accumulated
abroad increases in real terms. This case is particularly relevant for price differentials
in tradable goods, i.e. the goods which should cost the same in all countries according
to the law of one price.
Temporary migration may even occur when nominal earnings are equal (yd (Sa ) =
yo (Sa )). The income in the destination would still generate a higher purchasing power
in the origin (i.e. x > 1). In order to benefit from those differences, the individual
must separate work and consumption. As before, it is assumed that θSd = θSo = 0 and
we also shut down the preference channel through setting π = 1. Finally, observe that
with equal earnings in the two countries the case of permanent migration cannot arise.
Indeed, the incentive to migrate in the first place is given by the fact that the currency
of the destination country has a higher purchasing power in the origin. Panel B in
Figure 5.6 depicts simulation outcomes for two different calibrations of parameter x.
The parameters used in the simulation of panel B in Graph 5.6 are
• Skill accumulation θSd = θSo = 0
o d
• Rental rate on human capital expα0 = expα0 = 0.5
• Relevance of skill parameter α1d = α1o = 0.3
• Bias of preference parameter π = 1
• Real exchang e rate x = 1.5 (high purchasing power asset profile)

and x = 1.1 (low purchasing power asset profile)
Migrants build up savings after migrating to the destination but before returning
home at age 12 (the low purchasing power scenario is represented by the solid line).
The value of savings jumps up during the transition from destination to origin, which
stems from the real exchange rate gap. Migrants use their savings in order to increase
consumption after returning home.
IV: Temporary migration due to human capital accumulation
Human capital accumulation may provide an incentive for individuals to migrate

temporarily. Suppose that there are no differences in purchasing power (x = 1), that
there are no locational preferences for consumption (π = 1), and that human capital
accumulates in the same way in both countries, i.e. θSd = θSo > 0. Assume further that
the (log) rental rate of human capital is lower at origin (i.e. α0o < α0d ), while the return
to skills is higher in the origin (α1o > α1d ). In this case, migration occurs due to higher
wages in d. However, when individuals accumulate skills, they prefer to return to the
origin country since the return to skills is higher in o. This could happen when the
lack of supply of skills in the origin country generates higher returns to those skills,
thus attracting immigrants to return. Observe that migration could be permanent
when the wage differential is relatively high or the differential in returns to skills is
relatively modest.
V: Temporary migration due to faster skill accumulation in the destination

Individuals may migrate only temporarily when the accumulation of skills is faster in
the destination country, i.e. θSd > θSo . Suppose that x = 1, π = 1, a1o > a1d , and a0o = a0d .
Despite earnings being higher in the origin country, individuals may temporarily
migrate if they can accumulate skills faster in d than in o. This could be the case
of student migration, where skills are accumulated faster in the destination country.
As with the previous arguments discussed for the case of a higher purchase power,
individuals have an incentive to migrate since they can accumulate those skills faster
in d. The rewards for those skills is then higher in o.
Dustmann and Görlach (2016) provide a simulation for both scenarios of human cap-
ital accumulation. In panel A of Figure 5.7, human capital accumulates at equal rates
in both countries. Wages are higher in the destination as long as the skill-level is
sufficiently low due to the assumption that α0o < α0d . However, the returns to skill are
higher in the country of origin due to α1o > α1d . Suppose the level of skill is equal to
unity. As discussed in Figure 5.4, wages would solely depend on α0 and the assump-
tion α0o < α0d would provide an incentive to migrate to country d but the higher α1
would lead to a situation in which the wage in the country of origin rises above the
wage in the new country of residence, d.
The left panel of Figure 5.7 illustrates this scenario based upon the following pa-
rameter values:
• Initial parameters A0 = 0, a0 = 0 and S0 = 2
• Skill accumulation θSd = θSo = 0.1
o d
• Rental rate on human capital expα0 = 0.25 and expα0 = 0.5
• Relevance of skill parameter α1d = 0.3 and α1o = 0.9
• Real exchange rate x = 1
The consumption level is equalized over the whole life cycle. Even after switching
back to the country or origin, individuals keep their initial consumption level. This
result is due to identical preferences in countries d and o. Although wages are chang-
ing over time, consumption smoothing allows individuals to maximize utility over
their whole life. Earnings are increasing due to human capital accumulation in both
countries. Despite the higher initial wage in country d, migrants choose to go back
in period 14 which is associated with an increase in the speed at which earnings are
increasing over time. That is entirely due to the higher α1 which indicates that human
capital is rewarded much better in the country of origin. Low-skilled workers mi-
grate to foreign countries in order to accumulate skills before returning to their home
country, which has a higher demand for those skills and thus offers a higher reward
to those skills.
Figure 5.7: Return migration and human capital accumulation

Dustmann and Görlach:(scenario
The Economics
IV) of Temporary Migrations 115
Panel A. Temporarily higher earnings Panel B. Faster skill accumulation

in the destination country in the destination country
Earnings Earnings
Consumption Consumption
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Figure 6. Earnings and Consumption Profiles
student migrations in which the knowledge migration will not occur because the sole
acquired abroad is more valuable in the purpose of migration is accumulation of an
home country or migrations in which skills asset (in this case, human capital), which
can be more easily accumulated in the work- is of higher value for income generation in
place in the destination country; for example, the origin than in the host country. As in the
through the higher skill levels of coworkers. previous case, the instantaneous utility gain
To illustrate this case, we again assume from returning increases with time in the
that x = 1and π = 1and that although the host country because the migrant returns
rate of return to skills is higher in the ori- with a larger stock of skills that are valued
Panel B in Figure 5.7 is based upon the following parameter values:
• Skill accumulation θSd = 0.2 and θSo = 0.1
o d
• Rental rate on human capital expα0 = expα0 = 0.5
• Relevance of skill parameter α1d = 0.3 and α1o = 0.9
• Real exchange rate x = 1
Wages are assumed to be higher in the country of origin than in the destination
country. Nevertheless, people have an incentive to migrate in order to accumulate
skills faster in the destination country, which would increase their earnings in the
country of origin after their return to o.
2.2 The determinants of international migration: empirical evidence
While the previous section has investigated the reasons for the migration decision
from a theoretical perspective, this section illustrates the main empirical findings on
the determinants of the international migration movements.
The gravity model of migration The empirical literature provides evidence of the
factors influencing migration movements using aggregate macroeconomic data of im-
migration stocks and flows.5 The main feature of these data is that they are dyadic (or
bilateral), i.e. they contain information both on the country of origin of the immigrant
(which can be either the country of birth or of citizenship) and on their country of
destination (residence). The approach commonly used to identify the factors influ-
encing the bilateral migration movements is the so called gravity model of migration,
originally developed to explain the determinants of the trade flows between importer
and exporter countries.
The main intuition behind the traditional “gravity approach” is that migration (or
trade) flows between two countries are inversely related to their distance and posi-
tively related to their mass, as in the Newtonian gravity law of physics. The distance
is usually measured with the geographical distance between the two countries, and
the mass as the product of the GDP of the origin and destination country (Bodvarsson
and Van den Berg, 2013).
Generalized gravity models of migration developed from the late nineties have been
specified as follows: (see Greenwod, 1997; Bodvarsson and Van den Berg, 2013).
ln( Migrod ) = lnβ 0 + β 1 ln( Distod ) + β 2 ln( Popo ) + β 3 ln( Popd ) + β 4 ln(Yo ) + β 5 ln(Yd ) +
Z Z
∑ βoi ln(Xoi ) + ∑ βdi ln(Xdi ) + eod
i =1 i =1
where the subscript o indicates the origin country and the subscript d the destination.
Specifically, Migrod is the migration (flow or rate) from the origin country o to the des-
tination d. Distod is the geographical distance between the origin and the destination
(usually measured as the distance between the capital of o and the capital of d). Popo
and Popd are the population sizes of the origin and the destination country, respec-
tively. Yo and Yd are the real per capita income of o and d, respectively. Xoi is a vector
5 For instance, both the OECD and the World Bank collect and provide these databases. A detailed list
of the macro data of migration and of their features will be illustrated in Chapter 5, Section 2 “The
Brain Drain”.
containing Z characteristics of the origin country, and Xdi for the destination country.
eod is the error term. When estimating this gravity equation, it is expected that the
distance is negatively associated with the migration from o to d (i.e. β 1 < 0), while the
other estimated coefficients (β 2 , β 3 , β 4 , and β 5 ) are expected to be positive. Indeed,
relatively high migration movements tend to occur between relatively big countries
(both in terms of population size and GDP per capita) (Bodvarsson and Van den Berg,
2013).
The most recent literature has tried to provide a “micro-foundation” of the gravity
equation (Beine et al., 2015). This means that the gravity equation used to estimate
the determinants of aggregate migration flows/rates (like the one presented above) is
derived from a model for the migration decision at the micro (individual) level. The
main intuition behind the micro-founded gravity models is that the individual utility
from migrating from a given origin to a given destination can be decomposed in two
parts. One part, is a deterministic component of the utility from migrating from an
origin to a destination. The other one, is a random component of the utility derived
from moving from the origin to the destination. Making some simplifying assump-
tions on the error term and on the components of utility, one can derive the share of
individuals who move from the origin to the destination. In turn, the gravity equa-
tion can be derived from this share. A detailed exposition of the micro-foundations
of gravity models and of the econometric issues arising when estimating the gravity
equations (both the traditional gravity equation and the micro-founded) goes beyond
the aim of this course6 .
Besides the econometric issues, the literature that estimates gravity equations has
provided the following evidence on the factors influencing bilateral migration move-
ments. Firstly, income in the destination country is commonly found as the most
important determinant of international migration movements. The higher the income
6 For a deeper analysis of these issues, please see Beine et. al. (2015).
in the destination country, the more attractive is such destination. Income is usually
measured as GDP per capita in the destination country. As a proxy for income, other
authors have calculated and used post-tax earnings or wages (Beine et al., 2015). Re-
gardless of the measure used, all studies have found a positive and robust correlation
between income in the destination and migration movements to that destination. Be-
sides the attractiveness of the destination country, the conditions of the origin country
also influence the migration movements. For instance, the lower is income in the ori-
gin country, the higher the credit constraints hindering migration. To capture the role
of the credit constraints (negatively correlated with income in the origin), some au-
thors have introduced higher-order terms of income at origin (see for instance Peder-
sen et al., 2008; Mayda, 2010 for an analysis of the determinants of bilateral migration
flows between OECD countries). The destination countries may introduce migration
policies which impose restrictions on access to potential immigrants from a given ori-
gin country. Identifying whether migration policies are correlated with bilateral mi-
gration movements is challenging due to the complexity involved in building reliable
and comparable measures of such policies. Using an index measuring the tightness
of entry into the destination country for OECD countries, Mayda (2010) and Ortega
and Peri (2013) find a negative correlation between the scale of migration inflows to
a given destination country and the strictness of its migration policies. The presence
of networks of previous immigrants from the same origin in the destination country
has also been extensively analyzed in the gravity literature (see for instance Peder-
sen et al., 2008, Beine et al., 2015; Bertoli and Fernandez-Huertas Moraga, 2015). A
common finding is that a network decreases migration costs, thus attracting inflows
of immigrants. Specifically, a ten percent increase of the network (i.e. the stock of
immigrants in the destination country) is associated with a four per cent increase in
bilateral migration flows (Beine et al., 2015).
3 Who migrates?
Immigration is a “selective” phenomenon. Indeed, not all individuals from a given

population have the same propensity to migrate. Among the population in the ori-
gin (or source) country, the more skilled and the more motivated individuals usually
choose to migrate. They could also represent a selective group from the income dis-
tribution of the population. Conversely, it could also be that the less skilled and edu-
cated individuals may decide to migrate. This means that immigrants are not a repre-
sentative sample of the population from the origin country, but they are a “selected”
sub-group of the population. The most popular work on selection in the field of mi-
gration economics are the models by Borjas (1987, 1991). His main contribution lies
in the recognition that immigrants self-select from the population of non-immigrants
according to different characteristics, either unobservable (like personal attitudes), or
observable (like schooling qualifications). This sub-section illustrates the 1987 Borjas
model which is based on the pioneering Roy model of individual self-selection (Roy,
1951)7 , and which is crucial to answer the question, “who migrate”?
3.1 The Borjas 1987 Model
Borjas’ framework assumes that there are two countries in the economy, i.e. the origin
(or sending) country denoted by 0 and the destination (or receiving) country denoted
by 1, which he interprets as the Unites States. Indeed, he intends to answer the
question of “who migrate to the Unites States?”. The origin and the destination countries
have different earning distributions; migration from the former to the latter is not only
due to average earnings differences, but also driven by the extent to which the two
countries reward immigrants’ human capital.
7 The seminal paper of Roy “Some thoughts on the distribution of earnings” (1951) is the first to recog-
nize the issue of individual self-selection. Using the example of the fishing and the hunting sectors,
Roy qualitatively describes the way in which workers self-select into these two different professions.
3. WHO MIGRATES? 257
Specifically, Borjas assumes that earnings in the origin country 0 are given by
ln w0 = µ0 + e0 , (5.7)
where µ0 is a constant term and e0 is the error term. e0 is assumed to be normally

distributed with mean zero and variance σ02 , namely e0 ∼ N (0, σ02 ). Equation 5.7 im-
plies that log wages in the origin country fluctuate around their mean value µ0 and
have variance equal to σ02 . Individuals in the origin country may have earnings either
below or above the country’s mean value µ0 and differences in earnings are due to
random factors captured by e0 . Hence, µ0 can be interpreted as the part of earnings
which is explained by observable socioeconomic characteristics of the origin country,
while e0 can be interpreted as the part of earnings influenced by individual charac-
teristics such as personal motivation, skill or ability, and other unobserved factors.
If individuals migrate to country 1, they get the earnings in the destination country
which are given by
ln w1 = µ1 + e1 , (5.8)
where µ1 is another constant term and e1 is also an error term. The error term is also
assumed to be normally distributed with zero mean and σ12 variance, i.e. e1 ∼ N (0, σ12 ).
This means that the wages in the destination country fluctuate around their mean
value µ1 and have variance equal to σ12 . As before, earnings can be seen as the sum
of two components µ1 and e1 : the former represents the part of the earnings due to
socioeconomic characteristics that are country-specific, while the latter is the part due
to unobserved characteristics8 . The variances σ02 and σ12 indicate the dispersion of the
earnings distributions around their means and they provide information on the in-
come inequality in the origin and destination. For example, in a country with highly
dispersed earnings, the high skill and the low skill individuals earn relatively high
and relatively low earnings, respectively. Instead, in a country where the earnings are
8 Borjas assumes that equation (5.8) also represents the earnings of the native population of country
1. Observe also that Borjas’ model ignores general equilibrium effects, namely the possibility that
migration affects the wage in the destination country 1 is ruled out. No matter how many migrants
flow into or out of a country, the means and variances of wages in both countries are not affected.
more concentrated around their mean, all individuals tend to earn the same. There-
fore, the variance or earnings inequality can be interpreted as the country’s returns to
skills.
Figure 5.8: Distribution of wages in both countries
𝜇𝜇0 𝜇𝜇1 log 𝑤𝑤
Figure 5.8 depicts the wage distribution in country 1 and country 2. Wages are nor-
mally distributed around the respective mean. The average wage is assumed to be
higher in country 1 compared to country 0, which gives an incentive for workers
from country 0 to migrate to country 1. However, earnings in the origin and in the
destination country may also vary in the same or in opposite directions. As a measure
for the correlation between the unobserved parts of the earnings, e0 and e1 , Borjas de-
σ01
fines ρ = σ0 σ1 , where σ01 = cov(e0 , e1 ) is the covariance between e0 and e1 . In the
case of a positive correlation, both countries reward the skills in similar way; individ-
uals who get a relatively high earning in a country tend to perform well in terms of
earnings also in the other country. Conversely, a negative correlation implies that the
skills valued in a given country tend to be undervalued in the other country.
It is then assumed that all individuals face the same migration cost which is propor-
tional to their earnings. The migration cost is equal to C = πw0 , where π ∈ (0, 1) is
a constant. The individual decides to migrate from 0 to 1 if the earnings differential
between the two countries, net of the migration cost, is positive. This corresponds to
!
w1
I = ln w1 − ln (w0 + C ) = ln ≈ (µ1 − µ0 − π ) + (e1 − e0 ) > 0, (5.9)
( w0 + C )
where I is the net earnings differential. Defining for simplicity ν = e1 − e0 9 and

( µ1 − µ0 − π )
z = − σν , the probability of migrating from country 0 to country 1 can be
expressed as 10
P = Pr [ I > 0]
= Pr [e1 − e0 > −(µ1 − µ0 − π )]

" #
ν −(µ1 − µ0 − π )
= Pr >
σν σν
" #
ν
= Pr >z
σν
= 1 − Φ ( z ).
The first line is the probability that the net earnings differential is positive (I > 0), thus
it corresponds to the probability of migration. The second line is obtained by substi-
tuting I with its definition given in (5.9). The third line is derived by using the above
definition of ν and by dividing both sides by the constant term σν . Notice that σν is
the standard deviation of ν. Since ν = e1 − e0 is a normal random variable, dividing
ν
by the standard deviation means that σν is a normal standard variable. This standard-
ization is introduced by Borjas for analytical simplicity. The fourth line is obtained
by plugging in the definition of z. The fourth line is the probability of the normal
ν
standard variable σν of being higher than a certain value z. The cumulative density
function (CDF) of a normal standard variable is defined as the probability that the ran-
9 Note that ν is a normal random variable since e0 and e1 are normal random variables.
10 Observe that the model does not allow for the possibility to migrate to a third country.
dom variable takes values equal or less than a given value z, i.e. Φ(z) = Pr ( Z ≤ z).
Hence, the fourth line is the complementary of the CDF, namely 1 minus the CDF,
from which the last line follows.
This equation allows us to identify the factors influencing the probability of migration.
The migration probability is inversely related to z. The higher the value z, the higher
ν
the value of the CDF (i.e. the probability that σν is less than or equal to z), the lower
is the probability of migration (1-CDF). In turn, z is decreasing with respect to µ1 ,
and it is increasing with respect to µ0 and π. Intuitively, this means that the higher
the average wage µ1 , the more attractive is the destination country. Therefore, the
probability to migrate to such a country increases. Conversely, the migration option
is less attractive when the average earnings in the origin country and the migration
cost are relatively high. Hence, the probability of migration is decreasing with respect
to µ0 and π.
Selection of immigrants. The issue of immigrants’ selection can be then analyzed

as follows. First, we calculate the expected earnings in the origin and in the destina-
tion countries, conditional on migration. We show that the conditional expectation of
earnings is the sum of two terms: the country’s average earnings plus another term,
called the “self-selection bias“. Secondly, we analyse the determinants of the “self-
selection bias“ terms and we show the possible cases of immigrants self-selection.
The expected earnings in the origin country conditional on migration are given by11 :
ν

E(ln w0 | I > 0) = µ0 + E e0 | > z
σν
σ0 σ1 σ0 φ(z)
= µ0 + ρ−
σν σ1 1 − Φ(z)
= µ0 + Q0 .
11 For the derivation of the conditional mean see the Appendix at the end of this sub-section.
Similarly, the expected value of the earnings in the destination country, conditional
on migration (see the Appendix for the details) is equal to
ν
E(ln w1 | I > 0) = µ1 + E e1 | > z
σν
σσ σ φ(z)
= µ1 + 0 1 1 − ρ
σν σ0 1 − Φ(z)
= µ1 + Q1 .
As anticipated, the conditional mean earnings are equal to the mean earnings (either

σ0 σ1
µ0 or µ1 ) plus a term indicating the self-selection bias, i.e. either Q0 ≡ σν ρ −

σ0 φ(z) σ0 σ1 σ1 φ(z)
σ1 1− Φ ( z )
or Q1 ≡ σν σ0 − ρ 1−Φ(z) . Importantly, the selection bias terms Q0
σ0 σ1
and Q1 depend on ρ and on the ratios σ1 and σ0 . As mentioned before, the correlation,
ρ, specifies whether the two countries tend to reward the same set of skills similarly.
The standard deviations, σ0 and σ1 , can be interpreted as the country’s return to
skills. Indeed, high skill individuals tend to earn more in a country with a relative
high income inequality (i.e. a relatively high variance and hence standard deviation
of the earnings).
Whether the conditional expected earnings are above or below mean earnings de-
pends on the sign of Q0 and Q1 , thus on the values of ρ, σ0 and σ1 . The sign and the
size of the bias terms determine how individual self-selection influences the migra-
tion.
In particular, Borjas distinguishes the following three cases:
• Q0 > 0 and Q1 > 0: case of positive selection;
• Q0 < 0 and Q1 < 0: case of negative selection;
• Q0 < 0 and Q1 > 0: refugee sorting.
Important notice: The parameter ρ is smaller than 1.

I: Positive Selection. In this case, individuals are self selected from the upper tail
of the earnings distribution of the origin country and they end up in the right tail
of the distribution in the destination country. This means that only the more skilled
individuals from the source country migrate. They move to a destination where they
earn a better return to skills.
Specifically, the positive selection appears when Q0 > 0 and Q1 > 0. From (5.11) and
(5.10), necessary and sufficient conditions for Q0 > 0 and Q1 > 0 are
σ1
1. σ0 >1

σ0 σ1
2. ρ > min σ1 , σ0 .
Condition 1 implies that the variance of the earnings of the destination country is
higher than in the origin country, namely the income distribution is more dispersed
in the destination than in the home country. As previously discussed, this condi-
tion implies that the destination has a higher returns to skill than the origin country.
Condition 2 states that the correlation between the returns to skill in the origin and
destination country is relatively high.
Interpreted together, the two conditions mean that in a country with relatively low
earnings dispersion skills are valued less than in a country with more unequal income
distribution. Skilled individuals will migrate from the country with lower returns
to skill (origin) to the country with higher ones (destination), if such returns are
sufficiently correlated between the two countries, namely if the skills valued in the
origin country are sufficiently valued also in the destination.
Hence, in the case of positive self selection, individuals are selected among the upper
tail of the income distribution of the origin country and they will migrate to destina-
tions with more dispersed earnings, where their skills are valued more. Only the high
skill individuals among the population of origin migrate. The most common example
of positive self-selection is the immigration of high skill individuals to the US, where
their skill earns relatively high returns.
II: Negative Selection. This case is the opposite to the previous case. With negative
self selection, only individuals from the left tail of the earnings distribution of the
origin country migrate. They move to a country with a more compressed earnings
distribution and they end up in the lower tail of the earnings distribution of the
destination country. However, they earn higher returns to skills since the destination
has more equal earnings than the origin. This type of selection holds when Q0 < 0
and Q1 < 0. From (5.11) and (5.10), necessary and sufficient conditions for Q0 < 0
and Q1 < 0 are
σ1
1. σ0 <1

σ0 σ1
2. ρ > min σ1 , σ0 .
Condition 1 states that the income distribution in country 1 (destination) is less dis-
persed than the one in country 0. Since the return to skills in country 1 is lower,
country 1 is less attractive to high skill individuals and more attractive for the low
skilled. As before, condition 2 states that the correlation between returns to skill in
the two countries is relatively high. Individuals are now selected from the lower tail
of the income distribution in the home country. The destination country, where the
returns to skill are relatively low but still sufficiently (positively) correlated with the
ones in the home country, attracts low skilled individuals.
III: Refugee Sorting. The last case occurs when individuals from the lower tail of the
earnings distribution of the origin country migrate and end up in the upper tail of
the distribution of the destination country. This happens when Q0 < 0 and Q1 > 0.
Necessary and sufficient condition for Q0 < 0 and Q1 > 0 is

σ0 σ1
1. ρ < min σ1 , σ0 .
Since Q0 < 0, immigrants are selected from the lower tail of the income distribution of
the home country, as in the case of negative selection. In contrast to before, condition
1 states that the correlation of the returns to skill is relatively low or it could even
be negative. This means that the skills valued in one country either receive a rela-
tively low return in the other country or they are not rewarded. Notice that the term
"refugee" is a bit misleading as the main incentive for refugees to migrate is most
likely not the higher rewards to skills in the destination country. However, the ex-
pected wages of refugees can be described by the "refugee sorting" scenario described
above. One may think about some skills that are redundant in a country at war but
of high value in the destination of asylum seekers.
Non-feasible scenario. Finally, in the case where Q0 > 0 and Q1 < 0 migration
is not rational since it would require that people move from the upper tail of the
distribution of the home country and end up in the lower tail of the distribution in
the destination. Moreover, it can be verified that the correlation ρ would be greater
than 1, which is impossible.
Summary. To summarize, the essential mechanism behind the self-selection of im-

migrants is based on the differences in the return to skills (or earnings dispersion)
between the source and destination countries. A destination with a more dispersed
distribution than in the origin (σ1 > σ0 ) tends to attract high skilled immigrants,
which enjoy higher returns to skill. Conversely, low skill immigrants tend to be at-
tracted to destinations with lower wage dispersion (σ1 < σ0 ), where they are insured
while the high skill are taxed (Borjas, 1987). Except for the case of refugee sorting,
selective migration occurs when the skills are rewarded in both countries (i.e. ρ is
positive and sufficiently high).
Cases of selection: a graphical illustration. The following graphical representation

of the Borjas model (see Bodvarsson and Van den Berg, 2013), provides a simple vi-
sualization of the three selection cases described previously (see Figure 5.9 below).
The idea behind this graphical representation is to capture the relationship between
the wages, the returns to skills and the skill level as predicted by the Borjas’ model.
Recall that the covariance of e1 and e2 were the main determinant of the three sorting
scenarios. The individual worker’s wage deviates from the mean but we were not
explicit about the determinants of those deviations. We left the particular reasons
unexplained when discussing the theoretical foundation of this chapter. One expla-
nation for a worker-specific wage rate is skills, which may be rewarded differently
across countries.
Suppose for simplicity that for any given country i, the following linear relationship
between the skills and wages holds12 :
log wi = αi + β i S (5.10)
where S is the country’s skill level and the slope β i is the rate of return to skills
in country i13 . The intercept αi indicates the wage paid to unskilled individuals in
country i, which is the counterpart to µ in the theory discussed above. The term β i S
captures the part of e that can be explained by skills. Suppose that β 0 and β 1 are both
positive. High-skilled workers earn more in both countries, which is associated with
a positive ρ. Thus, there will be a positive or negative selection of migrants. Whether
the selection is negative or positive depends on the interaction of the country-specific
parameters. Positive selection occurs when wages are less compressed in the destina-
tion country so that the rewards to skills are higher. Negative selection occurs from
countries with more dispersed wages to countries with less inequality. Using this
simplified linear relationship, Figures 5.9 to 5.11 show the possible cases of selection.
The first panel represents the case of positive selection: the return to skills (i.e. the
slope) of the country of destination (red line) is higher than that of the origin country,
12 Note that this illustration disregards the migration costs for simplicity.
13 Consistently with the model presented, in Figure 5.9 i = 0 indicates the country of origin and i = 1 the
destination.
which is represented by the blue line. The destination country is country 1 and we can
see that the wage profile is more dispersed as low and high skilled workers’ earnings
are much more unequal than the earnings of high and low-skilled workers in the
country of origin, which is country 0. In this case, only the high skilled migrate, i.e.
individuals whose skills are higher than s p .
Figure 5.9: Graphical illustration of the selection cases (I)
𝑷𝒐𝒔𝒊𝒕𝒊𝒗𝒆 𝑺𝒆𝒍𝒆𝒄𝒕𝒊𝒐𝒏
𝐿𝑜𝑔 𝑤𝑎𝑔𝑒
𝛼0
𝛼1
𝐷𝑜 𝑛𝑜𝑡 𝑚𝑖𝑔𝑟𝑎𝑡𝑒 𝑀𝑖𝑔𝑟𝑎𝑡𝑒

𝑠𝑝 𝑆𝑘𝑖𝑙𝑙𝑠
𝑵𝒆𝒈𝒂𝒕𝒊𝒗𝒆 𝑺𝒆𝒍𝒆𝒄𝒕𝒊𝒐𝒏
For any skill level above s p they earn a higher wage in the destination country. Con-
versely, in Figure 5.10, negative selection takes place. Migration is convenient only for
low skill individuals, specifically with skills lower than sn . Both figures differ with
respect to the parameters α and β. In the case of negative selection, the wage profile
in the destination country is less dispersed. That means that the average wage must
be higher in country 1 than in country 0. Indeed, this is the case due to α1 > α0 .
𝛼1
Furthermore, we have lower returns to skills. The returns are still positive so that
𝛼0 𝑀𝑖𝑔𝑟𝑎𝑡𝑒 𝐷𝑜 𝑛𝑜𝑡 𝑚𝑖𝑔𝑟𝑎𝑡𝑒 𝑆𝑘𝑖𝑙𝑙𝑠
more skilled workers earn more than low skilled workers but the wage gap between
𝑠𝑛
low and high skilled workers is much less pronounced compared to country 0.
In both cases of positive and negative self selection,

𝑹𝒆𝒇𝒖𝒈𝒆𝒆 𝑺𝒐𝒓𝒕𝒊𝒏𝒈the returns to skill of the ori-
𝐿𝑜𝑔 𝑤𝑎𝑔𝑒 𝛼0
𝛼0
𝛼1 𝐷𝑜 𝑛𝑜𝑡 𝑚𝑖𝑔𝑟𝑎𝑡𝑒 𝑀𝑖𝑔𝑟𝑎𝑡𝑒

𝛼0
3. WHO MIGRATES? 𝑠𝑝 𝑆𝑘𝑖𝑙𝑙𝑠 267
Figure 5.10: Graphical illustration

𝑠𝑝 of the selection cases (II) 𝑆𝑘𝑖𝑙𝑙𝑠
𝛼1
𝛼𝛼01
𝑀𝑖𝑔𝑟𝑎𝑡𝑒 𝐷𝑜 𝑛𝑜𝑡 𝑚𝑖𝑔𝑟𝑎𝑡𝑒
𝛼0 𝑀𝑖𝑔𝑟𝑎𝑡𝑒 𝐷𝑜 𝑛𝑜𝑡 𝑚𝑖𝑔𝑟𝑎𝑡𝑒 𝑆𝑘𝑖𝑙𝑙𝑠 𝑆𝑘𝑖𝑙𝑙𝑠
𝑠𝑛
𝑠𝑛
Figure 5.11: Graphical𝑹𝒆𝒇𝒖𝒈𝒆𝒆 𝑺𝒐𝒓𝒕𝒊𝒏𝒈

illustration of the selection cases (III)
𝑹𝒆𝒇𝒖𝒈𝒆𝒆 𝑺𝒐𝒓𝒕𝒊𝒏𝒈
𝐿𝑜𝑔 𝑤𝑎𝑔𝑒 𝛼0
𝛼0

𝑆𝑘𝑖𝑙𝑙𝑠
𝑠𝑟
𝛼1
𝐷𝑜 𝑛𝑜𝑡 𝑚𝑖𝑔𝑟𝑎𝑡𝑒 𝑀𝑖𝑔𝑟𝑎𝑡𝑒

𝑆𝑘𝑖𝑙𝑙𝑠
𝑠𝑟
gin and destination countries are positive. Instead, the case of refugee sorting occurs
when the origin and destination value individual skills differently, i.e. the correlation
ρ in the Borjas model is relatively low, or negative, as in the case depicted. Indeed, the
slope of the origin country is negative, while the other is positive. In the refugee sort-
ing case, individuals from the higher tail of the distribution in the origin countries,
i.e. high skill individuals, migrate and end up in the upper tail of the distribution in
the destination country. In the figure, all individuals with skills higher than sr migrate.
Empirical evidence on the Borjas’ model. Estimates of the Borjas model (1987) do
not provide strongly support for its theoretical predictions when analyzing immigra-
tion to the US. Using census data for 1970 and 1980 for immigrants from different
origins, Borjas finds evidence, though weak, that immigrants from Western European
countries tend to perform well in the US labor market and that they tend to have
higher skills than US natives. The opposite holds for immigrants from poorer coun-
tries. According to Borjas’ interpretation, his empirical results suggest that the former
tend to be positively selected, while the latter are negatively selected. However, as
noted by Bodvarsson and Van den Berg (2013), Borjas does not find evidence that a
higher income dispersion in the origin country relative to the US tends to lower the
immigrants’ skill level. Moreover, the credibility of Borjas’ empirical findings may be
challenged by the quality of the data he used in its analysis.
Both the theoretical work extending the 1987 Borjas’ model and the empirical litera-
ture testing its predictions are extensive and go beyond the exposition of these notes14 .
However, it can be concluded that the relevance of Borjas’ model lies in the fact that it
has provided a simple theoretical framework to analyze the issue of immigrants’ se-
lection and to acknowledge the fact that immigrants’ characteristics may substantially
differ from those of non-immigrants.
Notice that the following Appendix provides additional insights needed to solve
the equations presented in the main text. It is useful to go through the derivations
in order to develop a deeper understanding of the model.
14 See,for instance, Bodvarsson and Van den Berg (2013) for details and references on the theoretical
extensions and empirical studies based on the Borjas’ model.
APPENDIX: This appendix shows the derivation of the expected earnings in the ori-
gin and destination country conditional on migration, as reported in the text.
The expected earnings in the origin country 0, conditional on migration are given by
ν
E(ln w0 | I > 0) = E µ 0 + e0 | > z
σν
ν
= µ 0 + E e0 | > z
σ
e ν ν
0
= µ0 + σ0 E | >z
σ0 σν
ν ν
= µ0 + σ0 E ρ0ν + ξ | > z
σ σν
νν ν
= µ0 + σ0 ρ0ν E | >z
σν σν
σ0ν φ(z)
= µ0 +
σν 1 − Φ(z)
σ0 σ1 σ0 φ(z)
= µ0 + ρ−
σν σ1 1 − Φ(z)
= µ0 + Q0 .
σ0ν
where ρ0ν = σ0 σν . ρ0ν is the correlation between the unobserved part of the earnings
in the origin country, e0 , and the normal random variable ν = e1 − e0 . Note that the
covariance between e0 and ν is σ0ν = cov(e0 , ν) = σ01 − σ02 . Before explaining the
derivation of the formula, consider the following properties.
1. Given two random variables X and Y, the property of linearity of the condi-
tional expected value states that:
i) E( X + Y | Z ) = E( X | Z ) + E(Y | Z )
ii) E( aX | Z ) = aE( X | Z ).
2. The conditional expectation of two normal random variables X and Y is the

cov(Y,X )
regression coefficient E(Y | X ) = βX = var ( X )
X. Hence, Y can be written in
cov(Y,X )
the population regression function form, i.e. Y = βX + ξ = var ( X )
X + ξ, where
ξ is an error term with zero mean and independent from X.
3. Given the constants a, b, c, d, and the random variables X, Y the following

properties hold:
i) var ( a + bX ) = b2 var ( X )
ii) cov(cX, dY ) = cd cov( X, Y ).
4. Given a normal random variable X with mean µ and variance σ2 , the condi-
tional expectation of X of being higher than a certain value z is
z−µ
φ( σ ) (z)
E( X | X > z) = µ + σ z−µ = µ + σ 1−φΦ (z)
,
1− Φ ( σ )
where φ(z) is the probability density function and Φ(z) is the CDF. The last
φ(z)
term 1− Φ ( z )
is defined as the Inverse Mill Ratio, which gives the conditional
expectation of a standard normal variable of being truncated at point z.
The formula of the expected earnings in the origin country conditional on migration
is then derived as follows:
• First line: it follows from the definition of ln w0 in (5.7), and from the definition
of I and Pr ( I ).
• Second line: it follows from property 1.i above and from the fact that E(µ0 | I >
0) = µ0 .
σ0
• Third line: the term e0 is multiplied by 1 = σ0 . Then the property in 1.ii is
applied.
• Fourth line: it follows from the properties in 2 and in 3. Indeed, in this case
e0
since σ0 and eν
σν are normal standard variables, from 2 we have that E( σe00 | σνν ) =
e
cov( σ0 , σνν ) σ0ν σν2 ν
0
var ( σνν )
ν
σν = σ0 σν σν2 σν = ρ0ν σνν , where the variance and the covariance are calcu-
lated applying 3.i and 3.ii. Hence, the normal standard variables can be written
e0
in the population regression form as σ0 = ρ0ν σνν + ξ, where ξ is an error term
ν
with mean zero and it is independent from σν , and the third line of the equation
is then derived.
• Fifth line: it follows from property 1.i, which allows us to separate the two terms
in the expectation brackets. Notice that the expected mean of the error term is
zero. Thus, we can neglect the second term and focus on the first one. Applying
property 1.ii allows us to write ρ0ν in front of the expectation bracket.
• Sixth line: it follows from the definition of ρ0ν and from the property 4.
• Seventh line: it follows from the fact that cov(ν, e0 ) = σ0ν = σ01 − σ02 , which
is derived in the next box. Plugging this into the sixth line, we have that µ0 +
σ01 −σ02

φ(z) σ0 σ1 σ01 σ0 φ(z) σ0 σ1 σ0 φ(z)
σν 1− Φ ( z )
= µ 0 + σν σ0 σ1 − σ1 1− Φ ( z )
= µ 0 + σν ρ − σ1 1− Φ ( z )
.

σ0 σ1 φ(z)
• Eigth line: for simplicity, we define the selection bias as Q0 = σν ρ − σσ01 1− Φ ( z )
.
Finally, observe that the expected earnings in the destination country 1 conditional on
migration (i.e. E(ln w1 | I > 0), see (5.10) in the main text) can be derived in the same
way, applying the properties illustrated above.
Notice that we skipped the proof of cov(ν, e0 ) = σ0ν = σ01 − σ02 . You may simply
believe us that this equation is true or go through the following derivation. Notice
that this proof is not relevant for the final exam.
The equation σ0ν = cov(e0 , ν) = σ01 − σ02 can be derived as follows. Notice that
ν = e1 − e0 . e0 and e1 are normal random variables
h i
σ0ν = cov(e0 , ν) = cov(e0 , e1 − e0 ) = E e0 − E(e0 ) e1 − e0 − E(e1 − e0 ) =
h i
E e0 − E ( e0 ) e1 − e0 − E ( e1 ) + E ( e0 ) =
h i
E e0 − E(e0 ) (e1 − E(e1 )) − (e0 − E(e0 ))
For simplicity, let’s define A = e0 − E(e0 ) and B = e1 − E(e1 ). Plugging A and B

in the previous equation, it follows that
E[ A( B − A)] = E[ AB − A2 ]
Substituting now the equations for A and B we get that where the first term of
the last line follows from the definition of cov(e1 , e0 ), and the second term follows
from the definition of var (e0 ).
4 The brain drain
Intensified trade in goods is allied to international movements of workers. In par-

ticular, high-skilled workers are an important source of economic growth. However,
the ability of a country to invest in human capital, to retain high skilled individuals,
and to attract educated immigrants may be limited. Hence, the migration of the high
skilled is an extensively debated topic both at the academic and policy level. From the
destination country’s perspective the debates have focused on policies which attract
students, professionals and talented individuals. From the source country’s point of
view, the debate has concentrated on the possible detrimental effects of the emigra-
tion, and hence the loss, of highly educated individuals.
The intense emigration from developing to high-income countries (such us the US,
Europe and the Pacific region) of high skilled, professionals and highly educated in-
dividuals is labelled as the brain drain (Bodvarsson and Van den Berg, 2013). The
outflow of talented individuals from low-income countries has been considered by
development economists in the 1970s as a serious issue since it has represented a
loss of human capital in which developing countries had previously invested. Tak-
ing the perspective of the source country, this section investigates the phenomenon
of the brain drain, of its causes and measurement, and, most importantly, of its con-
sequences. Specifically, the focus will be not only on the possible detrimental effects
of the brain drain for the source economy, but also on the more recent view of its
beneficial consequences.
4. THE BRAIN DRAIN 273
Besides the opportunity to earn higher returns to human capital in high-income than
in low-income economies, other explanations exist for the migration of skilled individ-
uals from the latter to the former economies. First, the brain drain can be attributed
to the lack of demand for high skilled and educated individuals such as experts of
different professions and PhD graduates in developing countries. This could be the
result of economic policies inhibiting the demand for technical professionals and fail-
ure to foster innovation and adoption of new technologies. In addition, developing
countries tend to be capital-scarce. The lack of adequate technologies complementing
the work of technical professionals may induce them to move away from the country.
Labor market dysfunction, taxation of high incomes, corruption and appropriation of
the gains made by local enterprises might also be factors pushing high skilled indi-
viduals to move out from the country (Bodvarsson and Van den Berg, 2013).
Measurement of the brain drain: data and related issues. How is the brain drain
measured? To gain information either on the magnitude of the phenomenon, or on
the countries affected by it, different data sets based mainly on information provided
by the OECD have become available since the late 1990s. In particular, we refer to
the most commonly used database, i.e. the one of Docquier, Lowell and Marfouk
(2009). Their database is suitable to analyze the brain drain phenomenon since it
contains migration stocks by country of birth of immigrants and by their educational
attainment15 . As destinations, they use OECD countries. Their database contains
information for the years 1990 and 2000. Indeed, their main data source is made up
of national censuses, which have ten years frequency. Docquier et al. (2009) define
the high skilled immigrants as foreign-born individuals with tertiary education and
above living in an OECD country. Individuals are within the working-age (i.e. they
are aged 25 or more at the time of the census or of the survey)16 . Besides the above
15 Note that one of the main caveats of their database is that it does not contain information on the
country where the education is obtained (Docquier and Rapoport, 2012).
16 Besides the Docquier, Marfouk and Lowell database, which is based on Docquier and Marfouk (2006),
there exist other databases having different features and information. For instance, two examples are
mentioned data, Brücker et al. (2013) provides the most recent database of emigration
stocks and rates by skill level of education and by sex. Indeed, their data cover the
period 1980-2010, with five years frequency. Table 5.3 below is based on the Docquier
et al. (2009) and on Brücker et al (2013) databases. It reports the emigration stocks
and rates of high-skilled immigrants to OECD destination countries, for the years
1990, 2000, and 2010.
The bilateral emigration stock of high-skilled individuals (h) aged 25 and older, born
in the origin country o, living in the destination country d at time t is indicated as
h . Aggregating the stocks by destination country, we get the emigration stock of
Mot
D
h = h , where D is the
the high-skilled, which is reported in Table 5.3, i.e. Mot ∑ Modt
d =1
total number of destination countries. The Table also reports the emigration rate of
high skilled individuals (h) from the origin country (region) o at time t. Specifically,
the emigration rate (aggregated by destination countries) is defined as
h
Mot
h
mot = h + Mh
(5.11)
Not ot
h is the stock of emigrants as defined previously and N h is the stock of indi-

where Mot ot
viduals (25 years and older) who live in the origin o at time t. In Table 5.3, the origin
countries are grouped by income groups. Observe from Table 5.3 that the emigration
of high skilled individuals has increased from 1990 to 2010 both in terms of stocks and
rates. Lower-middle income countries display the highest emigration rates. Indeed,
high skill individuals born in such countries tend to have both the financial resources
to sustain the migration costs, and the incentive to move to earn better possibilities
and returns to skill abroad (Docquier and Rapoport, 2012). According to the Doc-
quier and Marfouk (2006) data, the regions displaying the highest emigration rates
are the Carribean, the Pacific Islands, and sub-Saharan Africa. In absolute numbers,
the Philippines and India are among the developing economies sending the highest
constituted by the database of Beine et al. (2007). They use the age of entry in the destination country.
They use this information to understand whether the education was obtained in the origin or in the
destination country.
numbers of high-skilled emigrants abroad, with 1.111 millions and 1.035 millions of
individuals in 2000, respectively (Docquier and Rapoport, 2012).
Table 5.3: Emigration stocks (in thousands) and emigration rates of

high skill immigrants to OECD destinations.
Emigration stocks Emigration rates

Year 1990 2000 2010 1990 2000 2010
By income group
High-income 5749 7911 9406 0.040 0.038 0.200

Upper-middle-income 2027 3729 8525 0.055 0.062 0.236
Lower-middle-income 3144 5691 7747 0.081 0.081 0.204
Low-income 1317 2918 1702 0.055 0.075 0.253
Notes: the source of the data for 1990 and 2000 is Docquier et al. (2009). The data for 2010 are own
elaborations based on the IAB Brain Drain dataset.
As previously mentioned, development economists from the 1970s focused on the po-
tential detrimental effects of the brain drain for the source countries. The brain drain
phenomenon was mainly associated with the “pessimistic” view according to which
the source country loses both the investment made to educate individuals and their
human capital when they emigrate permanently. Such a pessimistic view dominated
until the late 1990s. The recent “optimistic” view has shifted the attention to the pos-
sible beneficial effects of the brain drain. For instance, the emigration of the high
skilled turns out to be a form of brain circulation (rather than a brain waste) when
the individuals return to their source country, creating positive externalities through
the additional human capital they have acquired abroad or to the network they have
established. Moreover, the incentive to emigrate to earn better opportunities abroad,
may constitute an incentive for individuals to invest in education.
The next sub-sections are structured as follows. Sub-section 4.2.1. presents a theoreti-
cal framework exploring the conditions under which a “beneficial brain drain” arises
and Sub-section 4.2.2 provides a review on the recent empirical evidence on the brain
drain.
4.1 The brain drain: theoretical framework
This sub-section provides a theoretical framework, developed by Docquier and Rapoport

(2012), that enables us to analyze the so-called brain drain phenomenon. A brain
drain occurs when developing countries suffer from massive outflows of high-skilled
workers due to migration. The common perception is that high-skill emigration from
developing countries is associated with lower economic growth as the migrating high-
skilled workers are an urgently needed growth-stimulus in developing countries.
The brain drain is therefore often considered as detrimental but the prospect of fu-
ture migration possibilities may also induce more individuals to invest in education.
Emigration would still be bad for the economy because the amount of high-skilled
is decreasing but as long as relatively more workers are motivated to study, the total
number of high-skilled workers available for the production may be growing and so
does GDP according to the considerations discussed in the Solow model with hu-
man capital accumulation. This incentive to invest in human capital may therefore be
beneficial for the sending country as long as a large enough share of the individuals
acquiring education decide to later remain in the source country. The following theo-
retical framework investigates the effects of migration on the human capital formation
in the source country.
Docquier and Rapoport model. Docquier and Rapoport (2012) build a model of hu-
man capital accumulation with endogenous growth, which is illustrated below. The
model takes the perspective of the developing country characterized by relatively low
technology compared to the leading economy at the technology frontier. The rest of
the world (which is the advanced foreign country) is denoted by ∗ . It is assumed that
there are three periods and that firms and individuals live for all three periods. In
period 1, young individuals work as low-skilled and go to school if they decide to be-
come high-skilled in the second period. In period 2, some adults work as low-skilled
and some as high-skilled depending on their choice on education in the previous pe-
riod. In period 3 all workers retire. Adults give birth to young individuals associated
with some generation overlap in the model. The model first considers the benchmark
case of an economy without migration from the source country, migration is intro-
duced in the second part of this chapter.
Production: Firms produce a composite good Yt at time t employing physical capital

Kt and human capital (or efficiency units of labor) Ht at each time t. Human capital
is constituted by low-skilled and high-skilled individuals, which are assumed to be
perfect substitutes. The former contribute to production with 1 efficiency unit, while
the latter with 1 + θ units (where θ > 0). Assuming a Cobb-Douglas production
function, output per worker yt is given by
yt = At kαt h1t −α , (5.12)
where k t is capital per worker, ht is the average number of efficiency units, At ∈ [0, 1]
is a scale parameter (total factor productivity) and α is the elasticity of substitution
between physical and human capital. Each worker receives a wage per unit of labor
determined by margin productivity
∂yt
wt = = (1 − α) At kαt h− α
t . (5.13)
∂ht
It is assumed that physical capital depreciates after one period and that it is interna-
tionally mobile. Specifically, capital mobility equalizes the returns to capital across
economies all over the world. The return to capital is given by
∂yt
R∗t φt = = αAt kαt −1 h1t −α . (5.14)
∂k t
R∗t is defined as the risk-free international interest factor and φt is one plus the risk
premium.
Combining equation (5.12) and (5.14) and rearranging, the ratio of output per worker
between the domestic and the foreign country can be expressed as (for the derivation
see the Appendix at the end of this sub-section)
! 1−1 α ! 1−−αα !
yt At φt ht
= . (5.15)
y∗t A∗t φt∗ h∗t
Combining (5.12), (5.13), and (5.15) (see the Appendix for details) it follows that the
wage ratio between the domestic and the foreign economy is given by
! 1−1 α ! 1−−αα
wt At φt
= ≡ ωt . (5.16)
wt∗ A∗t φt∗
It can be further assumed that the scale parameter affecting productivity At is increas-
ing with respect to ht , namely that productivity is positively affected by the human
capital endowment of the economy. It is also assumed that the premium φt may de-
pend negatively on ht , i.e. a decrease in human capital increases the risk premium
on capital. These assumptions imply that the wage ratio in (5.16) is increasing with
respect to the domestic human capital. The wage setting equation can be defined as
ωt = W (ht , h∗t , Xt ). (5.17)
The wage setting equation depends on both the domestic and foreign human capital
0 00
and on the country characteristics Xt . It is finally assumed that Wh > 0, Wh ≶ 0 and
W (0, h∗t , Xt ) > 0. The first and the last assumptions are necessary conditions for a
stable equilibrium. However, we will discuss some limitations based on a graphical
illustration of the model. The second assumption can lead to a multiple equilibrium
as the shape of the wage curve switches between concave and convex depending the
level of h.
Human capital accumulation: All workers are treated as low-skilled workers in the
first period. Thus, their income is equal to wt and they have some basic consumption
µbt . Kids earn wages similar to the low-skilled adults but they have the chance to make
a binary decision of whether to invest in human capital denoted by xt (i.e. xt = 1 if
they invest in education and xt = 0 otherwise). If they invest in education, they incur
an individual monetary cost c, which is assumed to be uniformly distributed on [0, 1].
The remaining income cannot be saved but yields some additional utility in this first
period. In the second period, individuals work and earn wt+1 if they did not invest
in education in the previous period, while they earn (1 + θ )wt+1 if educated. Wages
are used either for consumption or for savings st+1 . In the last period, individuals
consume the returns yield by their savings from the previous period st+1 R∗t+2 . Their
logarithmic utility function is assumed to be
U ( xt , st+1 ) = ln(wt − µbt − xt cwt ) + (1 − λ)ln(wt+1 (1 + xt θ ) − st+1 ) + λln(st+1 R∗t+2 );

(5.18)
where λ is a parameter for the time preferences and the length for the retirement. Kids
receive a wage wt in the first period and at least the amount µ of this income must be
spent for consumption in the first period in order to avoid dis-utility. One may think
of some minimum consumption that is necessary to satisfy the basic needs. Utility
would turn negative if those basic needs cannot be satisfied. The remaining income
can be consumed in the same period or used for covering the education costs if indi-
viduals choose xt = 1. Young individuals are facing the following trade off: education
reduces income (and thus utility) in the first period but there are additional returns in
the second period that stem from a higher income paid to the high-skilled. Educated
workers receive a wage premium determined by their additional productivity θ.
Another trade off arises in the second period, when individuals have to choose the
amount of savings. Income can be consumed or saved for retirement. The savings
from the second period generate some additional income in the third period - the
period after retirement - depending on the capital rental, which is determined at the
international level. Thus, the model features some inter-temporal utility maximiza-
tion, which can be solved in two ways. The interrelation between consumption and
savings allows to express the model depending only on consumption or only on sav-
ings. In fact, one can be expressed in terms of the other, which allows to reduce the
number of endogenous variables. The more common approach is to express every-
thing in terms of consumption but the authors choose an expression that depends on
savings only. This approach is determined by the choice of the utility function, which
transforms expenditure directly into utility without modeling consumption explicitly.
As a consequence, utility must be maximized with respect to savings st+1 . Notice that
the inter-temporal utility maximization arises not till period 2 because individuals’
choice in period 1 is binary. The choice in period 1 is to invest or not to invest, which
can be easily modeled using a comparison of two different value functions derived
in the next step. The decision is straightforward as all individuals who can afford
to invest in education will do the investment if utility is increasing. The decision
between period 2 and retirement is a bit more complex as individuals smooth their
consumption over the remaining life time. Other than in period 1 and 2, there is no
income in the last period. Individuals must save in period 2 in order to prepare for
the time after their retirement.
The quasi-indirect utility can be derived following the steps explained in the Ap-
pendix at the end of this sub-section:
V ( xt ) = ln(wt − µbt − xt cwt ) + (1 − λ)ln(wt+1 (1 + xt θ )) + λln( R∗t+2 ) + F, (5.19)
where F is a constant. You may think of this indirect utility function as a function
that computes the total utility of an individual based on the variable x, which can
take only two values, the values 0 and 1. The decision on the optimal saving rate
is already incorporated in this indirect utility function, which can be seen from the
derivation in the Appendix.
The choice about investment into education also depends on the education costs
c, which are individual and not equalized across workers. This assumption is not
straightforward and must be defended. One may argue that this type of education
costs comprise common costs as fees for schooling and individual costs that are de-
pendent on the ability of the individual kid. The lower the learning-ability of the
person, the more effort he or she has to put into learning before receiving a high-
skill certificate. Individual costs can be paid for additional afternoon classes. The
higher the individual ability, the lower the amount of afternoon classes needed for
graduation.
But the choice not only depends on the trade-off between income and individual costs
in period 1, individuals also form some expectations about their life-time income and
make the choice by comparing life-time income with education costs. The investment
will be done if it is beneficial for the individual, where "beneficial" is evaluated using
utility functions. Kids choose to become high skilled whenever this choice results
in a higher (expected) life-time utility. The model is solved as following: Firstly, the
optimal saving decision in period 2 is derived generally. This allows us to setup two
different value functions describing the total value received by the kid if the choice is
low-skill education (x = 0) and the total utility if the choice is to become high-skilled
(x = 1). The only source of heterogeneity in this model is the cost parameter c, which
is individual to each worker. This heterogeneity gives some source of variation over
the entire workforce, associated with different choices for each individual. The com-
parison of V (1) and V (0) must be done for all workers, which leads to one critical
cost level for which the utility from investing in education (V (1)) is higher than the
one from not investing (V (0)). The value functions itself are derived in the Appendix
by backward induction. The optimal amount of savings can be solved generally as
a function of the choice of education (x). This general solution for s can be substi-
tuted into the utility function, which reduces the utility function to an equation that
depends on the common wage rate (w), the individual education cost parameter (c)
and other exogenous variables. From this comparison one can figure out a certain
value for c for which the respective individual is indifferent between investing into
education or not. Put differently, this one particular individual has the same lifetime
utility when the education costs are invested or when the education costs are not in-
vested. All individuals with costs lower than ĉ will invest and all individuals with
education costs higher than ĉ will refrain from investing into education. The value
can be computed as shown in the Appendix:
(wt − µbt )θ
ĉt ≡ > c. (5.20)
w t (1 + θ )
The critical cost value coincides with the fraction of individuals investing in education.
Let us first consider an economy without migration. In this case, the fraction of
individuals who invested in the first period (ĉt ) is equal to the fraction of educated
people in the next period πt+1 , namely πt+1 = cˆt , which is due to the fact that
education costs are equally distributed between 0 and 1. All individuals drawing
a cost parameter between 0 and ĉt will do the investment into education and all
individuals drawing education costs between ĉt and 1 refrain from education. The
cost level in (5.20) can be rearranged as
µ θ
ĉt = (1 − ) (5.21)
ωt 1 + θ
bt = µωt∗ . With the latter condition,

which follows from (5.16) and from assuming that µ
it is assumed that consumption in the first period in the source country is a fraction
of the wage in the foreign country.
One critical aspect in this model is the fact that all workers receive the same wage,
despite the fact that they have to pay individual education costs. The existence of het-
erogeneous education costs was explained by assuming that there is some individual
ability, but individual ability should be associated with differences in worker produc-
tivity as well. Put differently, worker heterogeneity should be reflected in individual
productivity parameters θ. However, θ is constant over the whole workforce and that
is why wages are equalized in the model. There is a premium paid to high-skilled but
wages among high-killed and low-skilled are equal. The individual ability is assumed
to affect the capability to achieve a certain skill. Graduation gives all high-skilled the
same ability and therefore the same labor productivity, reflected by the same income
among high-skilled coworkers. Indeed, this assumption is not realistic but it simpli-
fies the model and the conclusions drawn from the model are not affected by this
assumption.
In the economy without migration, human capital in the next period can be com-
puted by summing up all workers with low- and high-skills in period 1 using weights
for the high-skilled: One low-skilled worker is counted once and each high-skilled
worker is counted 1 + θ times in order to account for higher labor productivity. Re-
member that low- and high-skilled are perfect substitutes. The only difference is that
one high-skilled worker is 1 + θ times as productive as one low-skilled. However, a
high-skilled is also 1 + θ times as expensive as a low-skilled, which is due to perfect
substitutability. There is no incentive to pay additional income other then the pre-
mium that compensates high-skilled workers for their higher labor productivity. The
human capital stock in the beginning of period 2 is related to the total workforce by
dividing the amount of human capital by the total number of workers. The solution
is therefore given by
π t +1 θ cˆt θ
h t +1 = 1 + = 1+ , (5.22)
1+m 1+m
where m is the number of children of individuals in the first period. Notice that there
are L adults in the first period. A fraction of πt adults is high-skilled and a fraction
of (1 − πt ) are low-skilled. Their choice on education has been made in an earlier
period before time t. We take the number of adults in the first period as given by L.
Knowing πt+1 and assuming that π is constant over time, we can easily compute the
number of low- and high-skilled adults in period t, without even knowing the exact
number of workers available. Everything that happened before time t is treated as a
black box and you will see that the events in the past are not relevant for the model
at hand.
Each of the L adults has m kids. The total number of kids in the first period is therefore
L · m. One glimpse at the utility function informs us about the assumption that all
kids are low-skilled in the first period. Individuals spend their income wt at the time
when individuals decide about whether to invest in education or not. This decision
is made when individuals are young and implies that they are low-skilled workers
in the first period. This assumption is critical as one would expect that education
is financed by the adults of the respective kid. However, such a cross-generation
payment of education costs would complicate the model without bringing any benefit.
This simplification is not driving the main results in the model and therefore harmless
but one should keep in mind that this assumption is not too realistic.
Taking all these information together, we can compute the per capita human capital
stock as
L · (1 − π t +1 ) + L · π t +1 · (1 + θ ) + L · m
h t +1 = ,
L (1 + m )
which can be simplified to the first part of equation (5.22). The second part of this
equation can be obtained by substituting for πt+1 .
Combining (5.21) and (5.22), the skill setting equation is derived as
θ2 µ
h t +1 = 1 + 1− ≡ H ( ωt ) (5.23)
(1 + m)(1 + θ ) ωt
The skill setting equation expresses human capital in the second period as a function
of the wage ratio ωt in the previous period, (i.e. of the differential in the price of
0 00
labor). Finally, if ωt ≥ µ then H > 0 and H < 0 must hold.
Equilibrium in an economy without migration:
Using the setting above, it is possible to characterize the balanced growth equilibria,
i.e. the loci where the wage setting equation and the skill setting equation intersect
(indicated by (wss , hss ). The subscript ss refers to the steady state values of the wages
and human capital). Following Docquier and Rapoport (2012), the model can be
characterized by multiple equilibria but we stick to the more realistic scenario that
features one equilibrium only.
The graphs depicted in the following pictures have some special features that are not
as obvious in the model as one may wish. Firstly, both equilibrium conditions must
be solved for ω in order to draw them in the same diagram. We neglect the time
index for the endogenous steady state variables in order to simplify the notation. The
relative wage curve is already solved for ω and the skill setting equation can easily be
rearranged for ω as
µ
(1+m)·(1+θ )
=ω (5.24)
1 − ( h − 1) · θ2
Notice that Home is the economy that lacks behind. It is therefore reasonable to
restrict the model to parameters that yield solutions at some ω < 1. Moreover, we
know that ω must be greater than µ. The authors couple the minimum expenditure
that must be spent in period 1 to the wage in the foreign country. Suppose that this
wage is normalized to unity. This assumption would lead us to the conclusion that
w
the relative wage would be equal to the nominal wage at Home (ω = w∗ = w), which
yields a situation in which the relative wage must be greater than µ in order to get
utility greater than zero. This restriction is met if h > 1. For h = 1 we obtain µ = ω,
whereas h > 1 is associated with µ < ω. Notice that the denominator is increasing in
h but it becomes negative once a certain threshold is reached: the skill setting equation
is asymptotic in h. Put differently, ω goes to infinity when h approaches the critical
threshold h̄ and it jumps to negative values once h surpassed the critical level. The
threshold is determined by the level of h for which the denominator becomes zero
(1 + m ) · (1 + θ )
1 − (h̄ − 1) · =0 . (5.25)
θ2
The flowing graph ignores the implausible values for h below h̄. For values of h
between 1 and h̄ we can compute the first and second derivatives, which are both
positive. The first derivative with respect to h reads

(1+m)(1+θ )
∂ω −µ · − θ2
= 2 , (5.26)
∂h (1+m)·(1+θ )
1 − ( h − 1) · θ2
where both nominator and denominator are positive. Let us denote the nominator as
N and the denominator as D. The second derivative therefore reads
∂2 ω N · D0
= −
∂h2 D2 h i
(1+m)(1+θ ) (1+m)·(1+θ )
µ· θ 2 · 2 · 1 − ( h − 1 ) · θ 2 · −(1+mθ)·(
2
1+ θ )
= −
h D2 i
(1+m)(1+θ ) (1+m)·(1+θ )
µ· θ2
· 2 · 1 − ( h − 1) · θ2
· (1+mθ)·(
2
1+ θ )
= >0
D2
Notice that all terms in squared brackets are positive due to the restriction we put on
h. Thus, the function starts at h = 1 associated with ω = µ and goes to infinity when
h increases up to h̄. The shape of the function is convex as illustrated by the blue
graph in diagram 5.12.
The second equilibrium condition is represented by the red graph. Notice, that this
graph is assumed to be convex when human capital is relatively low but becomes
concave when human capital surpasses a certain threshold level. The authors do
not comment much on this special feature of the wage setting curve. The shape of
this curve is determined by the evolution of the technology, which is assumed to be
dependent on h. The higher the level of human capital, the closer the gap between
the lacking home economy and the leading foreign country. Suppose that the risk
premium is equal in both countries (φt /φt∗ = 1). The relative wage between Home and
Foreign would be solely determined by the technology gap A/A∗, which is smaller
1 in the beginning of the catching up process. A rising h closes the gap between A
and A∗ as A approaches A∗ . However, Home is imitating and not innovating. Home
technology will always be less or equal the foreign level of technology, which explains
why the wage ratio approaches 1 when h goes to infinity.
Figure 5.12: Economy without migration: Multiple Equilibria
𝜔𝜔
4. THE BRAIN DRAIN
𝟏𝟏 𝐻𝐻(𝜔𝜔𝑠𝑠𝑠𝑠 )
EQ1
𝑊𝑊(ℎ𝑠𝑠𝑠𝑠 )
𝑊𝑊 (0)
287
𝜇𝜇
1 ℎ𝑠𝑠𝑠𝑠
Home adopts foreign technology at increasing rates in the beginning of the catching
up but the catching up fades out when A approaches A∗ . The functional form is
not specified in the original paper. The next subsection provides a simple simula-
tion based on an approach that is very popular in the marketing literature, where
researchers tried to characterized the process of diffusion of innovations using a
Gompertz-Function. We will show that this functional form yields exactly the de-
sired form of technological progress needed to obtain the desired S-shape of the wage
setting curve.
The authors argue that this type of wage-setting curve explains the existence of mul-
tiple equilibria. Multiple equilibria can occur for very special parameter settings.
However, in most cases there exists exactly one equilibrium. We therefore draw the
more general case with one intersection in point EQ1 (Equilibrium 1) and leave the
more special case of two equilibria for the simulation in the next section.
Generally speaking, there is always one intersection of both equilibrium conditions if

W (0) (the level of W at h = 0) is greater µ and there can be two or even more solutions
depending on the shape of the two curves. There must be one solution as H goes to
infinity, whereas W converges to unity. Moreover, the restriction that W (0) (the level
of W at h = 0) is greater µ makes sure that the one intersection is not in the undefined
space at some h < 1. Notice that the S-shaped wage setting curve is not necessary
to get one equilibrium but it is a ”must have” for the two equilibria discussed in the
simulation. A concave wage setting curve would be sufficient for one equilibrium.
This stylized example is characterized by the absence of migration. This means that
all individuals (both the ones who invested in education - the high skilled- and the
ones who did not invest - the low skilled) remain in the origin country in the next
period. In the next paragraph migration from the source to the foreign country is
introduced.
Economy with migration: We consider now the case with migration and how the
movements from the source to the destination country affect human capital accumu-
lation in the source country. This allows us to investigate the brain drain and brain
gain phenomena. To model immigration, it is assumed that a fraction p of educated
individuals migrate in the second period. Hence, the proportion of high skill individ-
uals who does not migrate in the second period is given by
(1 − p)ĉt
π t +1 = . (5.27)
1 − pĉt
Suppose we have L adult workers in a period, where L is the total number of adults
who did not emigrate in the previous periods. A fraction ĉ of the L remaining adults
is high-skilled. The number of high-skilled workers is therefore given by Lĉ. How-
ever, there is a fraction of p high-skilled adults who successfully migrate. The number
of remaining high-skilled workers therefore becomes L(1 − p)ĉ. The denominator is
the total number of remaining workers at Home. Again, the total number of workers
in period 2 is L but we have to subtract the number of high-skilled who migrate to
foreign, which is L · ĉp. The term L cancels out and we obtain the solution stated
in equation 5.27. As mentioned at the beginning of the section, migration might be
harmful for the source country since it “drains” the high skilled out of the countries.
∂ĉt
This is indeed the case when ∂p = 0. The cutoff education cost must be independent
of the migration probability p. Is it plausible to assume that the emigration proba-
bility p does not affect the critical cost level for which individuals prefer to invest in
education than not to invest? One would expect that the human capital is is more
valuable at Foreign, where wages are higher in general. However, one may also ar-
gue that as previously illustrated, the critical cost level in (5.20) corresponds to the
proportion of individuals choosing education. In an economy where the emigration
of the high skilled (i.e. p) does not affect the proportion of individuals investing in
∂ĉt
education (i.e. ∂p = 0), the brain drain is harmful for the source country.
Indeed, it can be verified that
∂πt+1 −ĉt (1 − ĉt )

= < 0, (5.28)
∂p (1 − pĉt )2
meaning that the emigration probability of the high skilled decreases with the pro-
portion of educated individuals in the second period. Notice that this solution was
∂ĉt
derived using ∂p = 0. In turn, a decrease of πt+1 reduces human capital in the second
period ht+1 , thus shifting the skill setting curve H inward. After the inward shift, the
economy ends up at a new equilibrium (A) with a lower steady state value of human
capital.
Suppose now that the prospect of future migration by the high skilled increases the
incentive of individuals to invest in human capital. Indeed, individuals take into
account that in the second period a fraction p of high skilled individuals migrate. Im-
portantly, Docquier and Rapoport (2012) exclude the possibility that the low skilled,
i.e. individuals who do not invest in education in the first period, will also emigrate
in the second period. Specifically, they assume that in the first period, the individual
earns wt , consumes µbt , and decides whether to invest in education (bearing an edu-
cation cost proportional to its wage), as in the case without migration. In the second
period, a fraction p of individuals which previously invested in education (i.e. the
high skilled individuals for which xt = 1) migrate, thus earning the foreign wage
wt∗+1 . A fraction (1 − p) of high skilled individuals do not migrate and earns wt+1 .
The low skill individuals (i.e. individuals for whom xt = 0) earn wt+1 . In the second
period, individuals save st+1 . As before, in the last period, the individuals consume
their accumulated savings.
The expected utility in the case of education now is given by
bt − cwt ) + (1 − p)ln(wt+1 (1 + θ )) + pln(wt∗+1 (1 + θ )) + λln( R∗t+2 ) + F

V (1) = ln(wt − µ
(5.29)
As before, the expected utility for individuals who do not invest in education is
V (0) = ln(wt − µbt ) + λln( R∗t+2 ) + ln[wt+1 ] + F. (5.30)
The critical cost level making the individual better off by investing in education (than
not investing) is now given by
p
µ ω
cbt = 1 − 1 − t +1 , (5.31)
ωt 1+θ
which now depends on parameter p as well. Thus, we have to take the indirect effect
of p over ĉ on π into consideration when deriving the first derivative of π with respect
to p. Put differently, since individuals incorporate the prospect of future migration
(p), the critical cost ĉt for which investment in education is worthwhile is different
∂ĉt
from the case with no migration so that we now have that ∂p 6= 0. Indeed, for p < 1,
p
ωt+1 < 1 and cbt increases w.r.t. p17 . This can be shown using the first derivative of
equation 5.31 with respect to p:
∂cbt µ 1 p

= 1− · − ln(ωt+1 )ωt+1 > 0 (5.32)
∂p ωt 1+θ
which is positive. Notice that the derivative in the second term in brackets is derived
ω t +1 1 −1
p
using the chain rule. The derivative of the outer function is 1 − 1+ θ = 1. The
p
ω
∂ 1− 1t++θ1
p
derivative of the inner function is ∂p = − 1+1 θ ln(ωt+1 )ωt+1 .18 Therefore, the
first derivative of π with respect to p can be computed as
∂πt+1 (1 − p) ∂∂pĉt − ĉt (1 − ĉt )

= , (5.33)
∂p (1 − pĉt )2
which can be negative or positive depending on the numerator at the right hand
17 Notice that when p = 0, then ωt+1 = 1 and we come back to the case with no migration, where the
critical cost level is as the one in (5.20).
18 The first derivative of a x with respect to x is ln( a ) a x , which can be applied in order to derive the first
p
derivative of ωt+1 with respect to p.
of this equation. When the numerator of (5.33) is positive, the prospect of future
migration increases the fraction of high skilled individuals in the next period. Indeed,
more people have the incentive to invest in education and the economy benefits from
migration. We refer to this phenomenon as a brain gain. This happens for relatively
∂ĉt
low values of p when (1 − p) is larger than ∂p ĉt (1 − ĉt ). Docquier and Rapoport (2012)
further illustrate that a necessary condition for (5.33) to be positive is that ωss (i.e. the
differential in wages between the source and foreign economy) has to be higher than
a certain threshold level. Intuitively, the wage differential has to generate sufficient
incentives to invest in education, meaning that it must be relatively low. However, the
wage differential cannot be “too low", otherwise individuals would be constrained in
their capacity to invest in education. For instance, in relatively poor source countries,
individuals are credit constrained when investing in education. Instead, in relatively
rich countries, the wage differential is too high to create an incentive for individuals to
invest in education and then to migrate. Hence, the skill setting curve H might shift
inward (due to a change in p) for both relatively poor and rich countries. Instead,
for medium income source countries (where both the wage differential generates the
incentive to invest and individuals are not credit constrained), the effect of migration
might be beneficial, shifting the skill setting curve outward.
Equilibrium conditions with migration. The wage setting curve is not affected by
the possibility of migrating to Foreign. The functional form stays put at
! 1−1 α ! 1−−αα
wt At φt
= ≡ ωt . (5.34)
wt∗ A∗t φt∗
The human capital formation curve must be updated as π and ĉ are different when
migration is allowed for. Following the same steps as before we can derive
t (1− p)ĉ
π t +1 θ 1− pĉt θ
h t +1 = 1+ = 1+ (5.35)
1+m 1+m
where we can substitute ĉ in order to derive

p
ω
1− 1t++θ1
µ
(1− p ) 1− ω
t
p θ
ω
1− 1t++θ1
µ
1− p 1− ω
t
h t +1 = 1 + ≡ H ( ωt ) (5.36)
1+m
The solution looks messy but it is easy to verify that the solution with p = 0 is
equal to the scenario derived without migration. However, it is not possible to solve
this equation for ω. Does that mean that there is no equilibrium anymore? No, the
equilibrium can be computed using more sophisticated tools. Set all parameters as
given and choose any value for h in the respective wage setting and human capital
formation curve. ω can be directly computed using equation 5.34. The corresponding
ω from the human capital formation curve can be easily computed by searching for
an ω for which equation 5.36 is solved. The same step can be done for all possible
values of h in order to search for the one combination of h and ω for which both
equilibrium conditions yield the same (or almost the same) values for ω. This routine
can be easily programmed using programs as Matlab or R.
Even Excel can be used to back out the equilibrium. The model can be rearranged
to one equation with one unknown parameter that has to be solved for zero. Excel
features a solver that allows to solve for ω but some programming in Visual Basic is
needed. The next subchapter describes a way to compute the equilibrium using Excel.
4.2 Model simulation in Excel
This chapter demonstrates how easy it is to simulate a model in Excel. There are
more sophisticated programs to run simulations but Excel has the advantage that
it is installed on almost every computer and that most users are familiar with this
program. However, we will use some features in Excel that most users may not be
aware of. Firstly, we will use an equation solver and secondly, we will do some very
basic programming of Visual Basic macros.
We need to characterize some of the functions described by the authors before turning
to the simulation. Most importantly, we need to model the technological progress at
Home. From the functional form of the wage setting curve we get a rough idea that
that the technological progress at Home should be convex for values of h closer to
zero and concave for values of h closer to infinity, which translates into a slower
development in the beginning and the end of the technological catching up but a very
fast development at medium levels of h.
Moreover, we need to make sure that there is some minimum technology for very low
human capital stocks approaching zero. The wage setting curve must be positive at
h = 0 in order to secure that both equilibrium conditions intersect at a relative wage
ω that is high enough to cover µ. This can be done through assuming a constant tech-
nology, which is independent of h. The positive sloped s-shaped wage setting curve
can be characterized by a shifted Gompertz function, which is often used to describe
the diffusion process of innovations. There are some first-movers adopting new tech-
nologies at a very slow rate before the masses adopt the innovation in the mid-term.
After this fast adoption in the middle of the product-cycle, there are some late-comers
adopting the innovation in the end of the product cycle. The Gompertz function de-
scribes this process perfectly well and we are using this function for characterizing
the catching up process of technological progress at Home (developing country). One
of the main advantage is that we can assume a technology threshold the function is
approaching to. Notice that Home technology must approach the Foreign technology-
level without surpassing it. This can be easily done with the Gompertz function
−b·h
A(h) = ( A∗ − Amin ) · (1 − eb·h ) · e−ηe + Amin , (5.37)
which has the nice feature that it converges to the minimum technology level Amin
when h goes to Zero and it converges to the upper technology bound A∗ when h
goes to infinity. Moreover, the function is convex for low levels of h and becomes
concave when h is getting large. Thus, there is the possibility of multiple equilibria
as described by the authors of the original model this chapter is based upon. Our
characterization of the functional form yields
Figure 5.13: Technological progress using a Gompertz function
Notice that the technology at Foreign was set at A∗ = 3 and the minimum technology
at Home at a level Amin = 1.1. The values were set arbitrarily. The Gompertz function
uses the following parameters:
• Parameter b: 0.4.
• Parameter η: 10.
• Human capital goes from h = 0 to h = 20.
4.2.1 Autarky version of the model
Based on the technological progress at Home and the two equilibrium conditions
(5.16) and (5.23) we can characterize the following equilibrium. We need some as-
sumptions about the parameters used in both equations. The risk premium is sup-
posed to be equal in both countries. Therefore we can neglect the second term in
(5.16). The first term in brackets is determined by the Gompertz function and the pa-
rameters discussed in the last subsection. We can neglect the country-characteristics
due to the assumption of a lower technology bound at Home associated with a posi-
tive wage ratio at h = 0. The elasticity of substitution is assumed to be α = 0.4.
The human capital formation equation is assumed to be characterized by
• High-skilled labor productivity: θ = 20.
• Average number of kids: m = 1.1.
• Minimum expenditure: µ = 0.17.
The orange line represents the human capital formation curve and the blue line repre-
sents the wage setting curve. The relative wage converges to 1 due to the technology
progression described by the Gompertz diffusion process. The risk differential is 0
between both countries and the home technology approaches the foreign technology
associated with equalized wages across Foreign and Home.
The inverted human capital formation curve approaches 10 as threshold. Even when
the relative wage goes to infinity, the human capital stock will not exceed the amount
10.
The functional form assumed above has three equilibria. The first one is at a very low
level of human capital close to unity. The second equilibrium is at h ≈ 5 and the third
equilibrium is at h ≈ 8. The authors argue that developing countries may be trapped
at the low end of human capital formation associated with a relatively low GDP or
rise up to higher GDP levels due to capital formation. One of the drivers is human
capital formation, technological change and migration.
Figure 5.14: Equilibrium without migration
4.2.2 Model version with migration
The following graph illustrates the model with migration. The wage setting schedule
does not change as discussed in the model description. The orange line represents
the benchmark scenario with zero success rate of migration, which corresponds to the
no-migration scenario. Starting from this scenario, we simulate a modest change in
the migration probability going from 0 to 0.1. The new human capital formation line
is represented by the gray line in Figure 5.15. The authors argue that the brain gain is
more likely at low levels of p, which is indeed the case when we look at the middle
equilibrium point associated with modest levels of ĉ and a low level of p. In deed,
there is a brain gain highlighted by the red arrow pointing from the left to the right.
The increase of p is associated with a higher human capital stock in equilibrium as
discussed by the authors. However, this equilibrium, different from the other two, is
not dynamically stable, which makes it less likely to persist.
Figure 5.15: Brain drain or gain?

However, the same change of p can be associated with a brain drain if the economy is
in its more optimistic equilibrium at the right. As indicated by the red arrow pointing
from the left to the right, human capital is decreasing in a change from p = 0 to
p = 0.1. This can be seen from equations 5.32 and 5.33, which depend on p, ĉ and
ω. Put differently, a low p is a necessary but not sufficient condition to obtain a brain
gain in the model at hand.
The second migration scenario assumes a success probability equal to p = 0.6 de-
picted by the yellow line in Figure 5.15. The rather strong increase in p is associated
with a brain drain as the human capital stock is decreasing.
4.3 Summary
To sum up, the Docquier and Rapoport framework offers the following insights. The
brain drain does not necessarily represent a loss of highly educated individuals for the
source country. Indeed, the prospect of future migration opportunities increases the
incentive to invest in education, and hence the fraction of high skill individuals in the
source country. In particular, their model predicts a beneficial brain drain (or brain
gain) when the probability of migration of the high skilled individuals is relatively
low, when individuals are not credit constrained, and the wage differential generates
incentives to invest in education.
The illustrated framework is highly stylized. Indeed, the effect of high skilled emi-
gration on human capital formation in the source country depends on different inter-
acting channels (for some extensions of the model, see Docquier and Rapoport, 2012).
For instance, the temporary nature of the migration of the high skilled individuals
is another channel linked with the brain drain. Indeed, individuals who emigrate
may accumulate additional human capital abroad and then transfer it into the source
country in case of return migration. Finally, the presence of previously immigrated in-
dividuals could reduce transaction costs and stimulate trade between the source and
the destination countries, thus contributing to the development of the source country.
4.4 Appendix
AI: Derivation of the main equilibrium conditions
This Appendix shows the derivation of formula (5.15), i.e. the output ratio between
! 1−1 α ! 1−−αα !
yt At φt ht
the domestic and the foreign economy y∗t = A∗t φt∗ h∗t .
Firstly, note that from the equation of the risk-free international interest factor R∗t in
(5.14) it follows that
αAt kαt−−11 h1t−

−α
1
φt = . (5.38)
R∗t
where φt is the domestic risk premium. Similarly, the risk premium in the foreign
economy φt∗ can be derived as
αA∗t k∗t−α− 1 ∗1− α

1 h t −1
φt∗ = . (5.39)
R∗t
Recall from the main text that the output per worker in the domestic economy is
yt = At kαt h1t −α (5.40)
and the output per worker in the foreign economy is
y∗t = A∗t k∗α h∗t 1−α . (5.41)
Rearranging (5.40), the capital in the domestic economy can be expressed as
1
ytα
kt = 1 1− α
. (5.42)
At ht
α α
Similarly, in the foreign economy
∗1
yt α
k∗t = . (5.43)
∗ 1 ∗ 1− α
At α ht α
Taking the ratio between the output in the domestic economy and the one in the
foreign economy, it follows that
yt At kαt h1t −α α
=
y∗t A∗t k∗t α−1 h∗t 1−α α
φt k t
=
φt∗ k∗t
1− α
φ y α A∗ α h∗ α
1 1
t t t t
=
φt∗ y∗t At ht
where the first equality follows from the definition of output in the domestic and
foreign economy (i.e. from (5.40) and (5.41), respectively), and by multiplying the
right hand side by 1 = αα . The second equality follows by plugging in the first line the
definition of φt and φt∗ as in (5.38) and (5.39), respectively. The third line is derived by
substituting equation (5.42) and equation (5.43) in the previous line.
Finally, the output ratio as in (5.15) can be obtained with the following algebra
1− α
φt yt α A∗t α h∗t α
1 1
yt
= ∗ ∗
y∗t φt yt At ht
y 1− α1 ∗ 1 ∗ 1− α
t φt At α ht α

∗ = ∗
yt φt At ht
yt φ α−α 1 A∗ α−1 1 h∗ −1
t t t
∗ = ∗
yt φt At ht
yt φ 1−−αα A 1−1 α h
t t t
∗ = ∗ ∗ ∗ .
yt φt At ht
Let’s now show the derivation of equation (5.16), i.e. the he ratio of wages between
! 1−1 α ! 1−−αα
wt At φt
the domestic and the foreign economy wt∗ = A∗t φt∗ . Recall from equation
(5.13) that the wage in the domestic economy is
wt = (1 − α) At kαt h− α
t . (5.44)
ht
Multiplying the right hand side by 1 = ht
ht
wt = (1 − α) At kαt h−
t
α
ht
yt
= (1 − α )
ht
where the second line follows from the definition of yt . Similarly, the wage in the
foreign economy can be expressed as
y∗t
wt∗ = (1 − α) . (5.45)
h∗t
Thus, from the wage in the domestic economy and from (5.45) it follows that
wt yt h∗t 1
∗ = ( 1 − α ) ∗
wt h t y t (1 − α )
yt h ∗
= ∗ t
yt ht
φ 1−−αα A 1−1 α h h∗
t t t t
= ∗ ∗ ∗
φt At ht ht
φ 1−−αα A 1−1 α
t t
= ∗ ∗
φt At
where the third inequality follows from the definition of the output ratio.
The last part of the Appendix shows the derivation of the quasi-indirect utility func-
tion as in (5.19). From equation (5.18), utility is a function of education xt (which is a
binary variable) and of savings st+1 , i.e.
U ( xt , st+1 ) = ln(wt − µt − xt cwt ) + (1 − λ)ln(wt+1 (1 + xt θ ) − st+1 ) + λln(st+1 R∗t+2 ).

(5.46)
Since the education variable xt is binary (i.e. xt = 1 if the individual invests in
education and xt = 0 if not), we first maximize U ( xt , st+1 ) with respect to st+1 , which
is a continuous variable. Then, we substitute the optimal value of savings into the
utility function and we express utility solely as a function of education, i.e. V ( xt ).
The partial derivative of U ( xt , st+1 ) w.r.t. st+1 is equal to
∂U ( xt , st+1 )
= 0
∂st+1
(1 − λ ) λ
− + = 0
w t +1 (1 + x t θ ) − s t +1 s t +1
λ (1 − λ )
=
s t +1 w t +1 (1 + x t θ ) − s t +1
λ [ w t +1 (1 + x t θ ) − s t +1 ] = (1 − λ ) s t +1
st+1 = λ[wt+1 (1 + xt θ )].
st+1 ( xt ) is the value of savings which maximizes utility. Notice that it constitutes a
fraction λ of the income in the second period and it is a function of xt . Plugging the
maximum value of savings in the utility function, we get
V ( xt ) = ln(wt − µbt − xt cwt ) + λln( R∗t+2 ) + (1 − λ)ln[wt+1 (1 + xt θ ) −
λ(wt+1 (1 + xt θ ))] + λln[λ(wt+1 (1 + xt θ )]
Rearranging, we obtain
V ( xt ) = ln(wt − µbt − xt cwt ) + λln( R∗t+2 ) + ln[wt+1 (1 + xt θ )] + (1 − λ)ln(1 − λ) +
λln(λ)
Defining for simplicity F ≡ (1 − λ)ln(1 − λ) + λln(λ), we obtain
V ( xt ) = ln(wt − µbt − xt cwt ) + λln( R∗t+2 ) + ln[wt+1 (1 + xt θ )] + F. (5.47)
Since V ( xt ) is maximized w.r.t. savings and it is function of xt only, we can now

compare V (1) and V (0), namely the case when the individual invests in education
xt = 1 and when he does not invest xt = 0. Specifically, the utility from investing in
education is
V (1) = ln(wt − µ̂t − cwt ) + λln( R∗t+2 ) + ln[wt+1 (1 + θ )] + F, (5.48)
while the utility from not investing is
V (0) = ln(wt − µ̂t ) + λln( R∗t+2 ) + ln[wt+1 ] + F. (5.49)
Comparing (5.48) and (5.49), we can derive the critical cost level (5.20) such that the
individual gets a higher utility from investing in education than from not investing,
i.e.
V (1) > V (0)
ln(wt − µbt − cwt ) + ln[wt+1 (1 + θ )] > ln(wt − µ̂t ) + ln[wt+1 ]
ln(wt − µbt − cwt ) − ln(wt − µbt ) > ln[wt+1 ] − ln[wt+1 (1 + θ )]

h w − µb − cw i h w t +1 i
t t t
ln > ln
wt − µ̂t w t +1 (1 + θ )
h w − µb − cw i 1
t t t
>
wt − µ̂t 1+θ
wt − µbt
wt − cwt − µbt >
1+θ
wt − µbt
−cwt > − wt + µbt
1+θ
θ (µbt − wt )
−cwt >
(1 + θ )
θ wt − µbt
c <
1 + θ wt
which is the threshold cost in (5.20).

Finally, note that in the case with migration we can derive the critical cost level in
(5.31) in the same way, i.e. comparing the indirect utilities with and without invest-
ment in education, i.e (5.29) and (5.30), respectively.
AII: Description of the Excel simulation
The simulation can be replicated using an Excel file provided in Moodle. The file
contains one directory with three panels of variables (spread across columns 2 to
27) needed to simulate the autarky scenario without migration, a low migration-
probability scenario and a high migration-probability scenario.
Autarky scenario. Columns 2 to 12 contain the information shown in Figure 5.16.

The first column contains different values for the human capital stock. Those values
are set in a the first step to values going from 0.001 to almost 30. For each of those
values we get solutions for all endogenous variables based upon the common param-
eters set in columns 3 to 9. Technology at Foreign is set equal to the value 3. The
model assumes a catching up at Home until the foreign technology level is reached
in the long run. The catching up process is triggered by human capital formation at
Home following a modified Gompertz function. The catching up is characterized by
the two parameters b and η. The functional form of the Gompertz function enters
column 4, which transforms the level of h based upon the parameters b, η and the
Gompertz function into A reported in column 4.
Notice that we added an additional constant equal to 1.1 into the technological progress.
This term is a constant that ensures that the minimum level of Home technology is
not falling below the constant level 1.1. Put differently, the minimum technology ratio
at Home is slightly above 1/3, which is the constant for the wage setting schedule.
Figure 5.16: Overview

The intercept with the vertical axis must be large enough to ensure that the two
equilibrium conditions intersect at least once. We set the minimum value in a way
that this requirement is fulfilled.
It can be easily verified that the functional form for A entered in the Excel sheet
corresponds to the functional form discussed in the text by selecting any value of A
reported in column 4. The box highlighted in red displays the formula entered in the
respective cell.19 Moreover, the variables used will be highlighted in the data browser.
Notice that each cell illuminates in the color used to indicate the respective variable
in the formula box at the top of Figure 5.17.
Figure 5.17: Gompertz
The minimum productivity is assumed to be equal to 1.1 which is the reason why
h = 0.001 is associated with A ≈ 1.1. Home technology approaches the foreign
technology when h goes to infinity.
Based upon the technology gap between Home and Foreign, one can easily compute
the respective values forming the wage setting curve by computing the associated
outcomes for ω. The shape of the wage setting curve is mainly determined by the
shape of A. Notice that the risk premium at Home is assumed to be equal to the risk
premium at Foreign. Thus, the two parameters disappear in the wage setting curve.
The elasticity of substitution is set equal to 0.4 in the formula that computes ω in
column 10.
The formula is replicated across all rows for the different values of h, which is the
19 The formula can be uncovered by clicking a cell which contains a formula. The cell in column 4, row
5 was selected in order to uncover the formula used for computation of the result shown in this one
particular cell in the example in Figure 5.17.
changing parameter in our analysis.
Column 11 contains the formula which translates the remaining parameters (column
7 to 9) together with the respective value for h into the corresponding human capital
formation outcomes for ω. The human capital formation curve was rearranged for ω
in order to obtain the values of ω for which the respective level of h is an outcome.
Notice that the causality goes from ω to h but we rearrange this condition in order
to combine the two equilibrium conditions. All cells up to the cells associated with
h > 1 are empty due to the restriction in the human capital formation curve. Values
of h < 1 yield negative outcomes for π and ĉ, which can be ruled out in the set of
possible equilibrium outcomes.
Column 12 computes the distance between the outcomes for ω coming from the
wage setting schedule and the outcomes that stem from the human capital forma-
tion equation. Of all possible outcomes, the equilibrium is the one combination of
Omega_Wage and Omega_H associated with the lowest distance computed in col-
umn 12.
Migration scenario. Columns 14 to 20 contain the information needed to compute

the equilibrium with migration. Firstly, the probability of successful migration is set
as exogenous variable in column 14. We assume a rather modest probability equal to
0.1. The formulas that transform the input variables - the respective value of h, the
probability p and the exogenous variables set in the simulation of the autarky scenario
- into the outcome variables π and ĉ must be updated according to the formulations
discussed in the main text (equations 5.27 and 5.31). Based on the outcomes of ĉ and π
one can easily compute the equilibrium value of human capital using the human cap-
ital formation condition. However, the human capital formation curve with migration
is more complex and cannot be solved for ω. The solution is therefore more compli-
cated but we can compute all equations dependent on ω. The wage setting schedule
is not affected by the possibility of migration. We replicate the values derived in the
autarky scenario in column 20.
Figure 5.18: Overview
The value for ω must be set manually in column 19 in order to get solutions for π,
ĉ and H_mig based on the outcome for ω. The one solution for ω corresponding to
the equilibrium can be found using a trick. We rewrite the equilibrium condition in a
way that the equation must be equal to zero. The values computed in column 17 are
the values for h based on ω, where ω itself is set to any value in column 19.
The following two graphs (Figure 5.19 and 5.20) show how the formulas link the
different cells in each row. The formulas can be displayed in the formula-display by
clicking on the respective cell of interest.
The values in column 15 build upon the values in columns 14 and 16, whereas the
values in column 16 are computed based upon the values in column 19 (and some
other exogenous variables). The endogenous variables π and ĉ can be computed if
and only if a value for the endogenous variable ω is set in column 19.
Figure 5.19: Computing π
Figure 5.20: Computing ĉ
However, we are not searching for any value of ω, we are searching for the one value
of ω which yields an h equal to the value of h set in column 2. Thus, the difference
between the value set in column 2 and the outcome of h in the model must be zero
or at least very close to zero. We can compute the difference between column 2 and
column 17 in column 18. The value for ω in column 19 must be changed until the
difference in column 18 is as closest to zero as possible. This can be done manually by
trying out different values for ω but a more elegant way is using a solver that solves
for an ω in column 19 associated with the value zero in column 18. Excel provides two
different solvers that can do the job but there is no automation that allows to solve
all the solutions for the different levels of h at once. The solution to this problem
is a small program that can be written in Visual Basic. Excel contains an editor for
Visual Basic that allows us to write small programs in form of macros. Those macros
must be executed each time some of the other parameters are changing. Suppose for
instance that you are interested in the solutions for a different value of p. The values
in column 14 would have to be changed to the new value and the equilibrium values
of ω in column 19 would have to be recomputed using the respective macro.
The functioning of the macros will be discussed in the end of this sub chapter but
before we turn to the alternative migration scenario simulated using the information
provided in columns 22 to 27. The only difference to the initial migration scenario is
a higher probability of successful migration. More high-skilled workers will success-
fully migrate to Foreign. All variables are recomputed using the higher probability p
entered in column 22. The results in columns 23 to 27 are computed as before.
Figure 5.21: Computing ĉ
The wage setting schedule remains unchanged as in the initial migration scenario.
Some basic programming using Visual Basic in Excel. The solution can be obtained
by manually using the solver for each of the more than 5000 values of h but there is a
more efficient way using loops in a macro that can be easily programmed. The macro-
environment in Excel can be found in the user interface. There is the possibility to
record a macro. You can use this recording feature if you don’t know how to write
down the commands in Visual Basic. Simply start recording and solve for one of the
solutions for ω. The macro will record the command in the back. You can use this one
command and put it into a loop that replicates this command for each of the more
than 5000 values of h. The loop must be written by hand and some information must
be updated but the language is straightforward and easy to handle.
Figure 5.22: Visual Basic Macro

Figure 5.22 gives an overview over the two macros included in the provided Excel file.
You can reach them through ”Ansicht|Macros”. Simulation 1 computes the values for
the first migration scenario. The values for the second simulation with p = 0.6 can
be found by executing Simulation 2. You can open the code by pressing the button
Bearbeiten or execute the file using the button Ausführen, which will update all values
in column 19.
The code editor can be opened using the button Bearbeiten, which gives you the code
that had to be entered in order to replicate the solver command for all values. The
same procedure can be done with every command typed in by hand. One could easily
write down a program that generates all the information in the different columns and
even the formulas can be inserted using loops. However, the aim of this program was
to keep things as easy as possible.
The solutions for the endogenous variables were used to characterize the graphs dis-
cussed in the main text.
4.5 The brain drain: empirical evidence
Brain drain: evidence at the international level. This sub-section provides a brief
review of the empirical evidence on the brain drain. As illustrated in the theoret-
ical framework above, emigration from developing countries is expected to have a
potentially beneficial effect for human capital formation in such countries. The ex-
isting empirical evidence confirms this theoretical prediction. For instance, Beine et
al. (2001) find a positive correlation between the emigration rates from a sample of
37 developing countries and human capital formation. They use OECD data on gross
emigration rates as a measure of the brain drain. One of the limitations of these data is
that they do not contain information on the educational level of the emigrants. More-
over, the panel dimension is not exploited in this study due to the lack of harmonized
data of time series of human capital indicators.
When using migration data disaggregated by skill level (education), the theoretical
prediction is confirmed. Beine et al., (2007) use a cross section of emigration rates
of high skilled individuals for a sample of 127 developing countries and find that
the emigration rate in 1990 has a positive effect on the growth rate of human capital
between 1990 and 2000. They use migration data from the Docquier and Marfouk
(2006) database20 . As proxies for human capital formation, they include different
indicators such as literacy measures and school enrollment indicators. They estimate
the elasticity of human capital formation to high skilled migration to be equal to 5
percent. This means that a one per cent increase in the emigration rate of the high
skilled from developing countries increase the growth rate of human capital of such
countries by 0.05 percent.
Data constraints constitute the main limitation of the above mentioned studies. In-
deed, the use of cross sectional data does not allow researchers to tackle endogeneity
issues due to omitted variable bias and unobserved heterogeneity. An improvement
in this direction is the work of Beine et al. (2011), who exploit a panel data set of
emigration rates from 1975 to 2000, with five years frequency. They also confirm
the previous results. Specifically, they analyze the effect of emigration rates of high
skill individuals on the growth rate of human capital of natives, (i.e. residents and
emigrants). As an indicator for the human capital, they employ the proportion of
natives with a tertiary education. Using data on the stocks of migrants by skill levels
provided by Brücker and Defoort (2009), they find that emigration positively affects
human capital formation in low-income source countries. Instead, similar effects are
found neither for medium-income source countries, nor for high-income ones, con-
sistent with the predictions of Docquier and Rapoport (2012). In line with the same
theoretical framework, emigration is beneficial for the sending country when the per-
centage of high skilled individuals emigrating is relatively low, i.e. when it ranges
20 The Docquier and Marfouk database contains information on the stocks of immigrants (defined ac-
cording to their country of birth) by skill level of education, for the years 1990 and 2000. Information
are collected using information from censuses, registers and survey data. See Docquier and Marfouk
(2006) for further details on the data.
from 20 to 30 percent.
Case studies on the brain drain. Below we briefly present additional empirical ev-
idence on the brain drain of high skilled individuals. In particular, we consider the
two source regions most affected by the brain drain: Africa and India.
The case of Africa: The study of the brain drain of high skilled individuals from
the African continent is hampered by data limitations (Chiswick and Miller, 2014b).
The scarce existing literature is based either on survey data or on the Docquier et al.
(2009) data set. One study that uses this survey data is Mattes and Mniki (2007). In
2002, they conducted a survey on the intention to emigrate of high skilled individuals
from South Africa, specifically postgraduate and final-year students. They find that
the intention to emigrate (which has to be interpreted cautiously since it does not
imply realized emigration) is positively correlated with families’ financial situation.
Individuals whose family has better financial resources tend to report higher inten-
tions to emigrate. Interestingly, students who have to repay their scholarships to the
government have a higher tendency to emigrate than students’ whose scholarships do
not require repayment. Hence, according to the authors, the government attempt to
make emigration more difficult through the obligation of scholarship repayments has
the opposite effect of fostering the intentions to move abroad. Finally, better life ex-
pectations abroad than in South Africa positively influence the intention to emigrate.
Another example of the use of survey data are the studies on the brain drain of
medical doctors from Africa (see Docquier and Rapoport (2012) for a review on this
phenomenon). A general finding is that doctors tend to emigrate with the expecta-
tions of both better working conditions and higher wages abroad. Additionally, both
the risks related with the presence of HIV and with the care of HIV patients boost
the emigration of doctors. Whether the prospect of emigration positively influences

the choice of pursuing medical studies is also a subject of investigation (Docquier and
Rapoport, 2012). Overall, the existing literature finds that medical emigration does not
provoke doctors’ shortages in Africa. In addition, the prospects of future emigration
tend to increase the choice of medical training. However, the size of this phenomenon
is relatively small and it is not sufficient to generate a medical brain gain (Docquier
and Rapoport, 2009; 2012).
The analysis of the Docquier et al. (2009) data of bilateral stocks of immigrants by skill
level of education also provide some insights into the phenomenon of emigration by
high skilled individuals from Africa. For instance, Chiswick and Miller (2014b) use
such a data set to compute the emigration rates of high skill individuals from 53
African countries to 29 OECD destinations. They find that bilateral emigration rates
are positively correlated with the GDP per capita income gap between the origin and
the destination country. Moreover, higher GDP per capita in the source country is
associated with higher emigration rates. While geographical distance between the
origin and the destination country does not play a significant role in shaping the bi-
lateral emigration rates of the high skilled, coming from a landlocked country tends
to decrease the emigration. Finally, previous colonial ties between the origin and the
destination countries and the use of the same language in the two countries ease the
migration movements of the high skilled.
The case of India: India constitutes another example of the bran drain. Starting in
the 1990s the IT revolution has fostered international movements of highly educated
individuals. Engineers and IT specialist graduates in particular moved from India
to Western countries, mostly to the US and the UK. According to Census data, the
stock of high skilled Indian emigrants was above 1 million individuals in the 2000s,
representing 4.3 per cent of the total population (see Docquier and Rapoport (2012)).
5. THE EFFECTS OF IMMIGRATION ON THE HOST COUNTRY 317
Debates on the effects of Indian emigration have focused both on its cost due to the
loss of highly educated individuals and on its beneficial consequences for India. For
instance, Desai et al. (2009) estimate the net fiscal loss associated with emigration of
skilled individuals from India to the US. After estimating the counterfactual earnings
distribution of high skilled individuals if they had not emigrated and the expenditures
saved due to their emigration, they compute the net fiscal loss ranging from 0.25 per-
cent to 0.58 percent of Indian GDP in 2001. However, their estimates do not account
for the possible incentive effects of investing in education due to the prospect of fu-
ture migration. They also assume that all engineers emigrated would have worked as
engineers in case they would not have migrated, excluding the possibility that they
would have found better paid jobs in India (Docquier and Rapoport, 2012). Some of
the empirical studies investigate the positive channels of the brain drain for India.
Indeed, the emigration of Indian IT specialists served to convey information on the
quality of the sector in India, thus fostering exports of IT services. Moreover, the
movements of skilled emigrants has created opportunities for brain circulation. On
the one side, professionals in the IT sectors tended to return to India after the reces-
sion caused by the dot-com bubble (Docquier and Rapoport, 2012). On the other side,
Indian IT entrepreneurs who previously emigrated could benefit from networks and
professional experience gained abroad (Nanda and Khanna, 2010).
5 The effects of immigration on the host country
Recent political debates have illustrated how sensitive the topic immigration can be.
Often this debate arises because immigration has important ramifications on the host
country’s labor market and native workers’ employment and wage prospects.
One of the most influential contribution on the effects of immigration on the host
country labor market is Borjas (2003). He provides evidence of a negative impact
of immigration to the US on natives’ wages. Assuming that foreign-born and native

workers with similar education levels and different work experience are not perfect
substitutes, he finds a negative wage effect of immigration on wages. The negative
impact on wages differs according to the education level: the most adversely affected
by immigrants are the low educated (a ten percent increase in labor supply due to
immigrants reduces the wages of similarly lowly educated native people by eight
percent).
One of the most recent contributions on the wage effect of immigration to the US, is
by Ottaviano and Peri (2012). In contrast to Borjas, they find a relatively small but
positive effect of immigration on natives’ wages. The aim of this sub-section is to il-
lustrate the recent model of Ottaviano and Peri and its estimation in order to provide
a theoretical framework to analyse the wage effects of immigration.
The aim of the Ottaviano and Peri model (Ottaviano and Peri, 2012) is to examine the
effect of immigration on a host country’s wages. Immigration, the influx of foreign
workers, produces a labor-supply shock in the host labor market: how do wages of
native-workers respond to the inflow of foreign-workers? Specifically, what is the
effect of the increase of labor-supply of foreign-workers on wages of native-workers
with similar characteristics (such as education and experience)? Secondly, what is the
effect of immigration on all the other groups of native-workers (e.g. the ones with
different education and experience)? The former case is called the direct partial wage
effect of immigration, while the latter is the total wage effect of immigration. In general,
the effect of immigration on wages depends on the elasticity of substitution between
the two groups of workers, and on the size of the immigrants’ inflow. Intuitively, immi-
grants should substitute native-workers with similar characteristics. Hence, the rise
in labor-supply caused by immigration depresses the wages of natives. The opposite
happens when foreign-workers complement natives with different characteristics: im-
migration creates an upward pressure on natives’ wages.
Ottaviano and Peri (2012) adopt a nested Constant Elasticity of Substitution (CES)
framework to capture the effects of the labor-supply shock created by immigration on
natives’ wages. Their model focuses on the long-run, i.e. after capital has adjusted
to the increase in labor-supply. Moreover, their model does not take into account the
effects of the inflow of immigrants on labor productivity. Hence, it disregards possible
positive wage effects, for instance due to the increase in efficiency.
They assume a constant-returns-to-scale Cobb-Douglas production function, i.e.
Y = ALα K1−α , (5.50)
where Y is the total output, K is the stock of capital, A and α ∈ (0, 1) represent
the total factor productivity and the income share of labor, respectively. Workers are
assumed to be heterogeneous, i.e. they differ with respect to specific characteristics. In
particular, L is a nested-CES-aggregate of different worker types. Below we show how
L is defined for the general case where workers differ according to N characteristics.
Specifically, suppose that there are N + 1 characteristics labelled n = 0, 1, ..., N. It is
assumed that all individuals have in common characteristic 0, i.e. they all are workers.
We then assume that workers differ according to characteristics 1 in M1 ways. Hence,
we have i1 = 1, 2, ..., M1 distinctions in characteristic M1 . We then assume that workers
can differ also by characteristics category 2. As before, we distinguish M2 groups in
this category. We thus indicate with i2 = 1, 2, ...M2 the groups of workers differing by
characteristic 2. Proceeding in the same way for the next characteristics, we obtain a
“nested-structure” where each group of workers is nested into the previous one. To
ease the understanding of the nested structure, the sequential partitioning of the labor
force L is represented in Figure 5.23. Specifically, the figure shows workers differing
up to characteristic n. Using the same notation as before, this means that we have Mn
groups of workers in+1 (i.e. differing according to the first n characteristics), which
are nested into the in groups. We now focus the group of workers having the same
Figure 5.23: Representation of nested-CES structure.
Nested CES structure
Characteristics
0…………………………
1…𝑖1 ………………𝜎1 …………𝜎1 ……………𝜎1 …………………....
2…𝑖2 …………………………………………………………………...
N-1…𝑖𝑁−1 ………………………………………………………..…….
N…𝑖𝑁 ………………………………………………………………......
𝜎𝑁
characteristics up to n, i.e. i (n). Their labor supply is Li(n) and the CES aggregator is
given by
" # σ σn+−1 1
σn+1 −1 n +1
∑

Li (n) = θ i ( n +1) L i ( n +1) σn+1
, n = 0, ..., N. (5.51)
i (n+1)∈i (n)
θi(n) is the productivity of the group i (n)21 , and it depends on technological factors
which are exogenous.
Importantly, σn is the elasticity of substitution between workers of type i (n). It is as-

sumed that σn+1 > σn : this means that workers of groups i (n + 1) are more substi-
tutable than workers of type i (n). Indeed, the former have common characteristics up
to n + 1, while the latter have only n common characteristics, hence they are less sub-
stitutable. Finally, observe that the group i (0) is defined as the group of individuals
who have only characteristic 0 in common. Hence, i (0) corresponds to all workers,
i.e. the entire labor force. Thus, Li(0) is by definition equal to L used in (5.50), i.e. the
nested-CES aggregate.
Returning to the equation for aggregate production in (5.50), we can find the wage
of workers belonging to type i ( N ) which maximizes profits. In a competitive labor
market, the profit-maximizing wage wi( N ) is equal to the marginal product of labor of
group i ( N ). Hence, equating the wage wi( N ) to the marginal product of group i ( N )
and taking the natural logarithm, we have
!
N N −1
1 1 1
ln(wi( N ) ) = ln(αAk1−α ) + ln( L) + ∑ ln(θi(n) ) − ∑ − ln( Li(n) ) −
σ1 n =1 n =1
σn σn+1
1
ln( Li(n) ), (5.52)
σN
K
where k is the capital- labor ratio, i.e. k = L. Notice that the derivation of the
profit-maximizing wage of group i ( N ) is given in the Appendix at the end of this
21 Notice
that it is assumed that θi(n) is standardized, i.e. ∑i(n)∈i(n−1) θi(n) = 1 and that the factors
common to all groups are absorbed in the parameter of the total factor productivity A.
sub-section. We further consider two groups i ( N ) and j( N ) (having n characteristics

in common, i.e. having the same characteristics up to N − 1.). Observe that wi( N ) and
w j( N ) are the profit-maximizing wages for groups i ( N ) and j( N ), respectively, and
they can be expressed as in (5.52). σn is the elasticity of substitution between the two
groups. Taking the ratio of the logarithm of their wages, we get
! ! !
wi ( N ) θi (n) 1 Li ( N )
ln = ln − ln . (5.53)
w j( N ) θ j(n) σN L j( N )
Importantly, equation (5.53) provides a theoretical framework to estimate the param-

eter of interest, i.e. the elasticity of substitution between two groups of workers, σN .
The estimation simply requires the use of data on employment and wages for the two
θ
groups of workers. The factor ln θi(n) , i.e. the log ratio of the unobserved productivi-
j(n)
ties of the two groups, can be captured through the group-fixed effects. Additionally,
the Ottaviano and Peri framework allows computation of the effect of a percentage
change in the labor supply of workers i ( N ) on the percentage change in wages of the
other group of workers j( N ). The two groups have in common m characteristics. It
can be shown that the percentage change is
∆w0j( N ) /w0j( N ) s0i( N )

= > 0, m = 0 (5.54)
∆Li( N ) /Li( N ) σ1
and
∆wm /wm m −1 s n +1 − sin( N )
j( N ) j( N ) i( N )
∆Li( N ) /Li( N )
=− ∑ σn+1
< 0, m = 1, ..., N (5.55)
n =0
where sim(n) is the income share of labor of workers of type i ( N ). Equation (5.54)
indicates that when the labor supply of workers i ( N ) increases, then the wages of
workers j( N ) are increaseing as welll. This happens when the two groups of workers
do not have a common characteristic 1 (i.e. m = 0). Conversely, an increase in
labor supply of workers i ( N ) decreases the wage of the other workers when the two
groups have m = 1, ..., N characteristics in common. Intuitively, this result suggests
that the more similar the groups of workers are in term of characteristics, the higher
the downward pressure on the wages of the other group. Equations (5.54) and (5.55)
measure the effect of a change in labor supply of a group of workers on the wages of
workers with similar characteristics, keeping fixed the labor supply of other groups of
workers. As mentioned at the beginning of the sub-section, this is the so called direct
partial wage effect.
Ottaviano and Peri (2012) also compute the total wage effect of immigration on wages,
i.e. the effect which takes into account the change in labor supply of all groups of
workers induced by a change of labor supply of immigrant workers22 . Intuitively, the
total wage effect is obtained by first computing the effect of immigration on the wages
of different groups of workers using (5.54) and (5.55). Then, the percentage changes
in wages of all groups of workers are aggregated in order to obtain the average of the
wage effect for all groups.
Ottaviano and Peri use equation (5.53) to estimate σN , i.e. the elasticity of substitu-
tion between native and foreign-born workers with similar characteristics. In their
empirical analysis, they consider immigrants and natives as having common charac-
teristics such as education levels and experience. They use microdata from the US
Census (period 1990-2006) and consider native (US) and foreign-born workers, their
education level, and their years of work experience. This data source allows them
to build measures of both wages and employment for the different “education-years
of experience” groups of workers. When looking at native and foreign-born workers
within the same “education-experience” group, the estimated elasticity of substitution
is σN = 20. This suggests that immigrants and native workers with the same charac-
teristics are substitutable. When restricting the sample to specific groups of education
and experience, they obtain different results. Indeed, they estimate σN = 1.25 for
low educated workers; native and foreign-born low educated workers are therefore
imperfect substitutes. When looking at young workers (with relatively low years of
22 Forsimplicity, we do not provide here the computation of the total wage effects of immigration. For
further details please see Ottaviano and Peri (2012).
experience), the elasticity of substitution is σN = 6.6, thus providing evidence that

young foreign and native-born workers are imperfectly substitutable. Finally, when
computing the total wage effects of immigration in the US, Ottaviano and Peri find
that the effect on average native wages is positive and relatively small (+0.6 percent),
and it is also small for native workers holding a degree (the effect ranges from 0.6 to
1.7 percent).
APPENDIX:
We show the derivation of the profit-maximizing wage wi( N ) , in natural logarithm, as

given in (5.52). Using the notation of the nested-CES provided by Ottaviano and Peri
(2012) and described in the main-text, the labor-supply of workers of group i (n) (i.e.
workers having in common n characteristics) is given by
# σ σn+−1 1
σnσ+1 −1
"
n +1
Li (n) = ∑ θ i ( n +1) L i ( n +1) n +1
(5.56)
i (n+1)∈i (n)
Remember also that, by definition, L = Li(0) , i.e. individuals with only characteristic
0 in common, are all workers, hence they all belong to the labor force L.
First, to find the profit-maximizing wage of group i (n), we assume a competitive labor
market. Therefore, wi( N ) is equal to the marginal product of that type of labor. This
means that
! ! ! !
dY dLi(0) dLi(1) dLi( N −1)
wi ( N ) = ........................ (5.57)
dL dLi(1) dLi(2) dLi( N )
Secondly, notice that for a general group of workers with D common characteristics,
the CES-aggregator is
" # σ σD−1
D σD −1 D
L= ∑ θd ( Ld ) σD
(5.58)
d =1
The derivative of (5.58) with respect of Ld is
dL 1 − σ1
= θd L σD Ld D . (5.59)
dLd
Computing the derivative for each group Li (n) as in (5.59) and using the condition
(5.57) to find the profit maximizing wage in a competitive labour market, we get that
! ! ! !
1 − σ1 1
− σ1 1
− σ1
Aαk1−α
σ2 σN
wi ( N ) = θi(1) L Li(1)1
σ1
θi(2) Li(1) Li(2)2 ...... θi( N ) Li(n−1) Li( NN) . (5.60)
Taking the natural logarithm of (5.60) and rearranging, we have that
! !
N N −1 − σ1n + σ 1
1 1
ln wi( N ) = ln Aαk1−α + ln( L) − ln( Li( N ) ) + ∑ ln(θi(n) ) + ∑ ln( Li(n) n+1 ),
σ1 σN n =1 n =1
(5.61)
Further rearranging the last term, we derive can derive the equation for the natural
logarithm of wages as in (5.52).
Chapter 6
Trade and growth
In this last chapter we will present theoretical frameworks that combine models of
trade and growth. The aim of of combining trade and growth models is to look
at growth potentials for developing countries emerging through international trade.
Notice that the standard models of trade help explaining why trade liberalization is
beneficial for countries in terms of welfare or income effects. However, the standard
trade models discussed so far consider changes from one steady state to another
steady state without considering economic growth over time. The models presented
in this section go beyond comparing steady states.
The framework presented in the first sub-section helps explaining extraordinary high
growth rates of countries in the initial phase of development. Acemoglu (2008) moti-
vates the question at hand with the growth miracle of the East Asian tiger countries
in the 70s and 80s. Generally speaking, diminishing growth rates can be explained
by diminishing returns to capital. Intuitively, a growing capital stock comes at the
cost of higher depreciation every period. Without proportional growth in population,
output growth rates are plummeting over time. Countries can grow up to a certain
level by investing into their capital stock but sooner or later the additional investment
is just enough for renewing depreciating capital in each period. Without technologi-
cal progress, the economy gets stuck in its steady state. However, international trade,
327
328 CHAPTER 6. TRADE AND GROWTH
which is associated with higher growth rates in open economies, may attenuate this
process.
In the last sub-section of this chapter we will analyze the effects of innovation on eco-
nomic development and income. We will see that product innovation in a model of
North-South trade enables the more advanced North to prevent factor price equaliza-
tion, which was one of the more critical outcomes derived from the Heckscher Ohlin
model. Capital owners benefit from trade liberalization at the expense of a decline
in labor income. Workers in developing countries are supposed to benefit from trade
liberalization. The outcome of model studied in this last sub-section shows that eco-
nomic development can prevent the catching up of developing countries. Developing
countries imitate products innovated in the developed North, which motivates the
North to invest in research and development for innovating new products. The North
looses its advantage without continuous progress in research and development. The
developing country catches up to the advanced country, which is reflected in factor
price equalization. Workers in the North would end up earning the same wage as
workers in the South.
1 Economic growth in open economy models
The first model in this chapter can be understood as a Heckscher-Ohlin type of trade
model with economic growth due to capital formation. Labor endowments are simi-
lar across countries. Differences in endowments may be due to unequal investments
into the capital stock across countries. Moreover, labor and capital are indirectly
transformed into consumption goods through intermediates. Similar factor endow-
ments can be associated with unequal endowments of input factors in a two-stage
production process when labor and/or capital is used with different productivity in
the production of intermediates. The crucial determinants for the amount of final
good production are the amounts of available intermediates and not capital and labor
1. ECONOMIC GROWTH IN OPEN ECONOMY MODELS 329
endowments. Moreover, the availability of goods is independent of production in a

globalized world.
Without changes in endowments, output is constant over time but we will allow for
capital formation and population growth. This is the major contribution of the exten-
sion proposed by Ventura (1997). Adding these dynamics into the static Heckscher
Ohlin model extends the static framework to a dynamic model. A fraction of GDP can
be invested into capital endowments and growing endowments are always coupled
to economic growth as more intermediate input is always associated with more final
good output. But as discussed in the first chapter, these output gains may or may
not be enough to compensate for the depreciating capital. Sooner or later economies
stop growing no matter if capital endowments are changing or not. At diminishing
returns, the positive and negative effects sooner or later cancel out each other.
The production in this model takes place according to a standard production function
that is familiar to you from from the Solow model of economic growth. The functional
form is
Yj (t) = O( D Kj (t), D jL (t)) . (6.1)
The variable Yj (t) denotes the aggregate output produced using two intermediate
goods as inputs. The demand for intermediates by final goods producers is denoted
by D Kj (t) and D jL (t). The two intermediate goods, which are combined to the final
output good according to equation (6.1), are produced using the available capital and
labor endowments. Thus, capital and labor are not directly used in the production of
the final output good. Endowments are used in the production of the intermediate
goods instead, which gives rise to a two-stage production process. The subscript j
identifies one country. In particular, it is assumed that there are j = 1, ..., J countries in
the world. This framework therefore differs from the standard Heckscher Ohlin model
discussed previously. Nevertheless, the main implications derived from the 2 × 2 × 2
model can be replicated in this more general framework with J countries. Assuming
a two-stage production process with inputs used in the production of intermediates
and intermediates used in the production of final goods has only little effects on the
main insights discussed in context of the Heckscher Ohlin model. However, this two-
stage production process is more convenient in context of an economic growth model.
Countries have more control over the availability of endowments through imports of
foreign labor and foreign capital without allowing for migration or FDI. We will see
that this opens up a new channel through which emerging economies can prevent the
diminishing returns when growing through capital formation.
Intermediate goods are produced according to the production functions
YjL (t) = ϕ j L j (t) (6.2)
YjK (t) = K j (t), (6.3)
where superscript index K and L allow distinguishing between intermediates pro-

duced using labor or capital. Similar to the Heckscher Ohlin model studied earlier,
one sector is labor and one sector capital intensive. However, this time we assume
that each input factor is used solely in one of the two sectors. This is a more extreme
interpretation of capital and labor intensive production as inputs are sector specific.
Notice that we distinguish between Y and D in order to analyze the free trade equilib-
rium later on. The inputs demanded in a given country j are denoted by D jL and D Kj ,
whereas the level of production are denoted by YjL and YjL . Without trade in interme-
diates Y = D. Instead, they might differ when allowing for free trade in intermediate
goods.
Without international trade, both prices and factor intensities differ across countries1 .
Factor intensity itself is linked to factor prices.
But with free trade in intermediate goods, factor prices are determined at the world
market. We are assuming that some countries are small economies in terms of the
1 For more details, please see Chapter 3 on Canonical Trade Models. However notice that factor prices,
wages and capital rental refer to the prices, wages and rental paid for the labor- and the capital-
intensive intermediate goods.
endowments with intermediate goods. Labor and capital endowments are similar by
assumption but a low labor productivity translates into a very low production of labor
intermediates.
The world market prices, P L (t) and PK (t), are determined according to the balance of
trade conditions. For the sake of simplicity we are assuming that these world market
prices are unaffected by small changes in one country’s demand for those goods. This
is why small countries can change their strategy without having an impact on the
world market prices. Changes in the demand of a small country can be compensated
by the rest of the world without priced adjustments.
Final goods producers in all countries adjust the factor intensity in the final goods
production according to intermediate goods prices, which are exogenous from a small
country perspective. Firms can substitute labor intermediates by capital intermediates
and they keep on changing the input mix until the maximum output is reached.
Intermediate goods prices also determine the capital rental and the wage rate through
w j (t) = ϕ j P L (t) (6.4)
R j (t) = PK (t) (6.5)
Factor incomes are always determined by the value of their marginal product. The
value of marginal product of labor and capital itself depends on the production func-
tion for intermediates, which take a linear form. One unit of capital transforms into
one unit of capital intermediate and one worker can produce ϕ units of the labor inter-
mediate. Perfect competition with zero profits implies that the labor cost per worker
is exactly equal to the contribution of a worker to the firm’s revenue. One worker can
produce ϕ units of the intermediate good that sell for price P L at the world market.
Per worker revenue must therefore be equal to ϕ j P L (t).
Even though both factor prices are country-specific, equation (5.5) already indicates
that R is equated across countries due to common world market prices paid for the
capital-intensive intermediate good. Nominal labor income may differ across coun-
tries due to country-specific labor productivity. However, the wage equation hints at
factor price equalization conditional on the level of productivity, ϕ j . Wage differen-
tials across countries are proportional to country-specific labor productivity in this
setup. We say that there is conditional factor price equalization.
The capital rental equation also accounts for the rate of depreciation and reads as
r j (t) = R j (t) − δ = PK (t) − δ . (6.6)
Notice that we do not allow for inter-temporal lending and borrowing: trade must
therefore be balanced. Prices adjust until the following condition is fulfilled:
PK (t)[ D Kj (t) − YjK (t)] − P L (t)[ D jL (t) − YjL (t)] = 0 (6.7)
From a country’s perspective, the world market prices are given and its imports and
exports adjust accordingly. From a global perspective, the world market prices adjust
in a way that trade is balanced and the full employment condition for intermediates
J J
∑ D jL (t) = ∑ YjL (t) (6.8)
j =1 j =1
J J
∑ DKj (t) = ∑ YjK (t) (6.9)
j =1 j =1
is fulfilled. Moreover, the capital stock changes according to the following equation
K̇ j (t) = O( D Kj (t), D jL (t)) − Cj (t) − δK j (t) (6.10)
Each country generates GDP worth O( D Kj (t), D jL (t)). There is no trade in final goods.
Thus, GDP is either consumed by domestic households or saved. Savings are redis-
tributed to companies for investments. The model is simple by means of all savings
are invested into the capital stock. However, a fraction of the investment is used to
renew depreciating capital as indicated by subtracting δK j (t). To solve the model we

need to think about final goods consumption as the consumption decision pins down
the evolution of the capital stock.
1.1 Consumption of final goods
Notice that the final good is sold at numeraire price equal to unity. This is why
we can say that O( D Kj , D jL ) is equal to domestic GDP in perfect competition.2 The
change in the stock of capital is therefore equal to the part of total GDP which is
neither consumed, nor used for replacing the depreciated capital in the respective
period. This result is based on the strong assumption that one unit of output can
be consumed or invested without making any distinctions between consumption and
investment goods.
But how do individuals decide between consumption and investments? Firstly, the
consumption decision determines the investment in each period. It is therefore enough
to solve the consumers’ utility maximization problem. Consumers are assumed to
have constant relative risk aversion preferences
Z ∞
c j ( t )1− θ − 1
Uj = exp(−(ρ − n)t) dt (6.11)
0 1−θ
We assume that the level of consumption is chosen by one representative household

that adjusts the level of per-capita consumption c j = Cj /L j in a way that maximizes
lifetime utility. Countries differ with respect to their capital endowment but they
are equally populated so that L j = L in all countries. The time discount rate is
assumed to be identical across countries; the higher the time discount rate, ρ, the lower
the propensity to save. The assumption that ρ > n assures finite lifetime utilities.
2 Revenues for intermediate goods production must be accounted for but these revenues are costs for
the final goods producers. We can add intermediate goods producers’ revenues to total GDP and
subtract final goods’ producers costs for intermediate goods from total GDP in order to avoid double-
accounting problems. Both exactly cancel out each other, which is why we can ignore intermediate
goods production when calculating total GDP.
Notice that we have to take population growth into consideration. The number of
consumers is growing in each period so that total utility in one period is equal to
L(t)u j (c(t)) = exp(nt)u j (c(t)) if the labor force in the first period is given by L(0) = 1,
as suggested in Acemoglu (2008). Suppose that per-capita consumption is constant in
steady state but the economy keeps on growing at population growth rate n. Without
changes in the household’s preferences, utility per household member would be fixed
at the level u(c(t)). Instead, total utility is increasing by the factor of population
growth, which is determined by L(t) = exp(nt) L(0) = exp(nt). Moreover, utility at
each time is discounted back to period 0 at the time discount rate exp(−ρt). Hence,
exp(−ρt) exp(nt)u(c(t)) can be summarized as exp(−ρt + nt)u j (c(t)). The CRRA
c j ( t )1− θ −1
preferences are given by u j (t) = 1− θ when θ 6= 1.
The utility maximization problem can be solved using the following Hamiltonian
H = e−(ρ−n)t u(c(t)) + µ(t)[w(t) + (r (t) − n) a(t) − c(t)] , (6.12)
where a is the state variable that can be controlled through c and µ is the shadow
value associated with the constraint in squared brackets. It tells us by how much
utility changes if we relax the constraint by one unit. The shadow price translates
the change in the state variable (relaxation of the constraint) into changes in utility.
Notice that we are using the present-value Hamiltonian.
The following box shows how to solve the utility maximization problem based upon
the Hamiltonian approach for the more interested readers.
Recall that the interest rate, r (t), is pinned down at the world market through factor
price equalization. The world market price of capital intensive intermediates, PK (t),
determines the interest rate in all economies integrated through trade in intermedi-
ate goods. This issue is important for understanding the last step of the following
derivations.
The first order conditions can be easily computed as
Hc = e−(ρ−n)t u0 (c(t)) − µ(t) = 0 (6.13)
H a = µ(t)(r (t) − n) = −µ̇(t) (6.14)
The first FOC can be log-normalized in order to obtain
[−(ρ − n)t] + ln u0 (c(t)) = ln µ(t) (6.15)
∂ ln g(z)
We can use the chain rule and ∂z = 1
g(z)
g0 (z) in order to obtain the first derivative
of equation (6.15) with respect to time as
1 0 µ̇ 1 00 0 u00 (c(t)) 0
µ (t) = = [−(ρ − n)] + 0 u (c(t))c (t) = [−(ρ − n)] + 0 c (t) ,
µ(t) µ(t) u (c(t)) u (c(t))
(6.16)
where we substituted µ0 (t) with µ̇ and where c0 (t) is the change in consumption over
µ̇(t)
time dt = 1. We can solve equation (6.14) for µ(t)
in order to combine both FOCs as
u00 (c(t)) 0
−(r (t) − n) = [−(ρ − n)] + c (t) (6.17)
u0 (c(t))
u00 (c(t)) 0
−(r (t) − ρ) = 0 c (t) (6.18)
u (c(t))
u0 (c(t)) c0 (t)
−(r (t) − ρ) 00 = (6.19)
u (c(t))c(t) c(t)
We can substitute the first and the second derivative of the CRRA utility function:
1−θ
u0 (c) = c (t )−θ , (6.20)
1−θ
u00 (c) = −θc(t)−1−θ , (6.21)
in order to obtain
c0 (t) 1 1
= (r ( t ) − ρ ) = ( P K ( t ) − δ − ρ ) (6.22)
c(t) θ θ
One important consequence of the factor price equalization is that all countries end
up producing at identical factor intensity, where factor intensity refers to the input
ratio of capital and labor intensive intermediates. The input ratio can be defined as
D Kj (t)
D j (t) = . (6.23)
D jL (t)
Following the usual procedure, we solve the model using per-capita variables. How-
ever, differently from the standard Solow model discussed in Chapter 2, we are now
expressing output in terms of labor-intensive intermediates as
D Kj (t)
!
Yj (t) = O( D Kj (t), D jL (t)) = D jL (t)O ,1 (6.24)
D jL (t)
= D jL (t)o ( D j (t)) (6.25)
All countries make their decisions based upon the world market prices for capital
and labor intensive intermediate goods. Thus, the equilibrium will be determined
for the integrated world and not for the individual countries. The world consists
of different regions and the invisible hand guides world market prices to a level
where trade is balanced. Consumption, capital formation, and the capital to labor
intermediate goods input ratio adjust accordingly. The world equilibrium can be
derived as J country-specific outcomes for c j (t), k j (t), D j (t) based on the outcome of
the non country-specific world market prices PK (t), P L (t). A steady state equilibrium
is reached when all of those variables remain constant over time. Let us start with the
input ratio of capital and labor intensive intermediate goods. It can be shown that the
ratio is equalized across countries:
J
∑ j =1 k j ( t )
D j ( t ) = D j0 ( t ) = J
(6.26)
∑ j =1 ϕ j
Thus, the intermediate input ratio of country j is identical to the input ratio of any of
the other countries and it can be computed through the aggregated capital and labor
input ratios world wide.
Acemoglu (2008) provides an analytical proof for this result. Households optimize
input of intermediate goods according to the respective intermediate goods prices.
OK ( D Kj (t), D jL (t)) = PK (t) (6.27)
OL ( D Kj (t), D jL (t)) = P L (t) (6.28)
These equations state that intermediate goods are paid according to their marginal
productivity evaluated at the price of the final good, which is the numeraire. We can
compute the relative price ratio of intermediate goods as
OK ( D Kj (t), D jL (t)) PK (t)

= (6.29)
OL ( D Kj (t), D jL (t)) P L (t)
Constant Returns to Scale properties of the production function allow us to rewrite

this expression in terms of D j (t) as
o 0 ( D j (t)) PK (t)
= , (6.30)
o ( D j (t)) − D j (t)o 0 ( D j (t)) P L (t)
where the chain and product rules are applied as
∂D j (t) 1
OK ( D Kj (t), D jL (t)) = D jL (t)o 0 ( D j (t)) = D jL (t)o 0 ( D j (t))
∂D Kj (t) D jL (t)
and
∂D j (t)
OL ( D Kj (t), D jL (t)) = o ( D j (t)) + D jL (t)o 0 ( D j (t))
∂D jL (t)
= o ( D j (t)) + D jL (t)o 0 ( D j (t)){− D Kj (t)[ D jL (t)]−2 }
The right hand side can be treated as a constant in the open economy scenario,
whereas the left hand side is strictly decreasing in D j . All countries are equipped
with the same technology and intermediate goods prices are equalized across coun-
tries. D adjusts through trade in intermediates equally in all countries resulting in

one intermediate goods ratio, D (t), that applies to all economies in the integrated
world equilibrium. We therefore drop the country identifier j in the following steps
of the model description. Moreover, the intensity can be computed using aggregate
variables:
J J J
∑ j =1 K j ( t ) ∑ j=1 K j (t)/L(t) ∑ j =1 k j ( t )
D j (t) = J
= J
= J
(6.31)
∑ j =1 ϕ j L j ( t ) ∑ j =1 ϕ j ∑ j =1 ϕ j
Trade in intermediate goods allows countries producing final goods at identical labor
intermediates to capital intermediates despite different availability of labor and cap-
ital endowments. Thus, trade in intermediates is comparable to migration and FDI.
Factors can be used abroad without crossing borders. Moreover, all intermediates are
sold at identical prices in all countries.
The following Figure summarizes the model and presents a nummerical example with
particular focus on the pattern of trade in an integrated world equilibrium. We pick
out one particular period and show production and trade of intermediate goods based
upon the assumption that labor and capital endowments are identical across coun-
tries. Heckscher-Ohlin type of comparative advantage stems from differences in labor
productivity. All countries differ with respect to the amount of intermediates that can
be produced with the available labor endowment. The country equipped with the best
technology to transform labor into intermediates has the highest endowments of labor
intermediates. We can say that the country with the highest labor productivity is is
abundant in labor intermediates. We also compute the average labor productivity and
demonstrate how to compute the average labor productivity and world wide labor to
capital intermediate goods intensity in the final goods production. The graph also
demonstrates the inter-dependency between the two production stages. Technology
in the final goods production stage determines demand for intermediates produced
on the first production stage. The first derivative of the production function O with
respect to the respective factor price, which is the price of the intermediate good we
look at, determines demand for the respective intermediate good.
Production stage 1: Firms produce intermediate goods by input of labour and capital
Production stage 2: Firms produce final goods by input of intermediate goods produced in stage 1
Consumption: Households spend their labour or capital income earned by producing

intermediates for consuming goods produced on stage 2
Numerical example: Suppose we have three countries with equal labour endowments and equal
capital endowments of 100 each.
- Workers in country 1 are two times more productive than workers in country 2.
- Workers in country 2 are four times more productive than workers in country 3.
- One worker in country 3 can produce one unit of the labour intermediate good.
- Capital productivity is unity in all countries.
 Total world supply of labour intermediates is 100 + 400 + 800 =1300

 Total world supply of capital intermediates is 300
 Average labour productivity is 13/3
 World labour to capital intermediate input supply ratio is 13/3=4,33
 Individual labour to capital intermediate supply ratios are 1, 4 and 8
o Country 1 can export labour intermediates in exchange of capital intermediates

to reach the average ratio 4.33.
o Country 2 can export some capital intermediates in exchange of labour
intermediates to reach 4.33.
o Country 3 can export capital intermediates in exchange of labour intermediates
to reach 4.33.
The least productive country exports capital in exchange of labour through trade in
intermediates.
World market prices adjust such that ratios in all countries are 4.33 and trade is balanced.
Figure 6.1: The model in a nutshell
There is another inter-dependency between the second and the first stage. Consumers
have to decide about consumption of the final good and this consumption decision has
an impact on the production of intermediate goods through savings and investments.
The lower consumption, the higher savings and investments. Higher investments
spur capital endowments and intermediate goods production.
Steady State Equilibrium. We can summarize the results derived so far in two equa-
tions:
ċ(t) 0 k(t)

1
= o −δ−ρ (6.32)
c(t) θ ϕ
k(t)

k̇ (t) = ϕo − c(t) − (n + δ)k(t) (6.33)
ϕ
The first equation substitutes PK (t) = o 0 (k(t)) in equation (6.22). Per-capita consump-
tion converges towards its steady state level and remains constant afterwards. Using
this result in equation (6.33) implies that also the per-capita capital stock converges
to its long run equilibrium and remains constant after this point is reached. Equation
(6.32) can be set equal to zero in order to determine the per-capita capital stock at
which the consumption growth rate becomes zero, which yields the world market
price for capital intermediates
PK ∗ = δ + ρ (6.34)
Independently from the level of capital an economy is equipped with, it will converge
towards the same steady-state capital stock. The capital stock grows up to a certain
point and remains constant afterwards. In the long-run neither capital nor consump-
tion can grow beyond its steady state levels. However, the fact that input prices are
fixed at the world market open up a channel through which governments can spur
growth without technological progress.
Globalization and growth potentials. So how is trade related to a country’s growth

potential? Emerging countries may exploit the fact that globalization pins down the
intermediate goods prices to a certain level. All countries pay the same prices for
the inputs used in the production of the final goods: labor intermediates and capital
intermediates. World market prices for intermediates also determine factor costs in
the production of capital and labor as technologies are almost equal across countries.
The only difference between countries is labor productivity ϕ. Thus, there is fac-
tor price equalization in for capital and conditional factor price equalization in labor
income. The latter accounts for differences in labor productivity. Labor income is
proportional to labor productivity. We have seen that all countries behave equally by
producing with identical input ratios D j . Moreover, all countries converge to similar
capital endowments. Population growth is equal as well but the effective labor en-
dowments are different due to differences in labor productivity, which is the source
of Heckscher-Ohlin type of comparative advantage in this model. Countries produce
different amounts of the labor intermediate good but trade in intermediates allows all
countries to produce at similar labor to capital intermediate input ratios. Production
of labor intermediates (Y L ) differs across countries but the intensity at which coun-
tries use both intermediates in the final output goods production (D L over D K ) is still
similar in all countries due to the possibility of importing and exporting Y L and Y K .
Larger countries use more of both intermediates compared to smaller countries but
the ratios are identical as firms produce at maximum output.
Now suppose that one of the countries is able to change its consumers’ preference
for future consumption in order to spur savings. A lower discount rate for future
consumption would shift more income to the future. Savings and the capital stock
would increase. Such a strategy would have zero effect on growth without changes in
the endowments of labor intermediates. A one-sided increase in capital intermediates
would lead the economy out of the output maximizing labor to capital intermediate
input ratio. However, under free trade the economy can spur economic growth by
augmenting the stock of capital without departing from the optimal capital-labor ra-
tio. Depending on the direction to which the economy departs form the initial ratio,
economies can trade capital and labor intermediates at constant world market prices
in a way that the optimal input ratio can be maintained. Suppose that the country
is exporting capital intermediates. It could export even more at constant prices and
purchase even more labor intermediates for the export revenues. More final output
goods can be produced at the same input-intensity. If the respective country is ex-
porting labor intermediates to finance capital intermediates purchased at the world
market, additional capital intermediates can be used in domestic production. Less
capital intermediates can be exported and less labor intermediates must be imported
to maintain the desired input-intensity.
The necessary requirement for this strategy is being a small country. The changes in
imports and exports must be small enough so that the world market prices remain
stable. The rest of the world simply doesn’t care about the changes in the small
country that sets ρ0 < ρ in order to grow at a higher rate
ċ1 (t) 1
g1 = = (ρ − ρ0 ) > 0. (6.35)
c1 ( t ) θ

ċ1 (t) k(t) k(t)
Recalling that c1 ( t )
= 1
θ o 0 ϕ − δ − ρ with o 0 ϕ = PK ∗ and PK ∗ = δ + ρ yields
ċ1 (t)
c1 ( t )
= 0 when the respective country has the same δ and ρ as all other countries.
In fact, the world market price of PK was set according to these parameters. Zero
consumption growth can be escaped by altering these two parameters if and only if all

0 k(t)
other countries stick to the old parameters δ and ρ as o ϕ − δ − ρ0 at PK∗ = δ + ρ
cannot be equal to zero.
The country would be able to continue to grow even when all other countries already
reached their steady state. However, this does not mean that the country will continue
to grow forever (namely it will not be a small economy forever). Suppose that all
other countries react by lowering the discount rate from ρ to ρ0 . The developing
country’s strategy would be undermined by the rest of the world and world market
prices would have to react immediately to PK ∗ = δ + ρ0 . In this case, none of the
countries would be able to grow for a long time. Even if all other countries stick to
the initial discount rate ρ, world market prices would have to react sooner or later.
The optimal capital to labor intermediate input ratio was determined by world supply
of intermediates. Capital accumulation in the growing country affects world supply

of capital intermediates. We would expect that higher supply reduces the price of this
good in order to spur demand. However, the country of interest is small. That means
that whatever the country supplies at the world market will be demanded somewhere.
The country is sufficiently unimportant to affect the world market equilibrium. This
is good news for the capital accumulating country. It exports more at constant prices,
which generates export revenues. These additional revenues can be used to buy more
labor intermediates at the world market. Again, the additional demand for labor
intermediates is so small, all other countries don’t care and deliver whatever the
growing country demands.
But sooner or later the country would be too large to be neglected by the rest of the
world as it’s supply and demand start affecting the world market equilibrium. At this
point the world market prices start reacting too until a new equilibrium is reached at
which no country is growing anymore. The larger the escaping country or the higher
the number of countries that behave equally, the faster the readjustments in the world
market prices. Thus, this strategy can be applied by emerging countries in early stages
of their development. Sooner or later these countries must go over to technological
progress as described in the beginning of this course. However, this model helps
explaining why opening up to trade is an important strategy for emerging economies.
2 Trade and innovation in a North-South framework
The model in this last subsection in our study of international trade and economic de-
velopment focuses on the role of economic development in a model of trade between
advanced and developing countries. The world in this model consists of two coun-
tries that trade their products internationally. An advanced North innovates products
that are consumed in both the North and the developing South. South does not inno-
vate products but it replicates some of the innovations produced in the North. South
has a comparative cost advantage due to its relatively low labor costs. The reason
why South participates in international trade is because it is a low wage country. The
framework outlined in this chapter builds on the model by Krugman (1979a) and the
descriptions presented in the textbook by Acemoglu (2008).
Consumption. Consumers in the world have CES preferences for the varieties pro-
duced in the economy. As shown in Acemoglu (2008), total demand in country j at
time t can be described by
Z Ψ(t) ρ −1
ρ−ρ 1
D j (t) = d j (t, Ω) ρ dz (6.36)
0
Index j distinguishes between the two countries, North and South. The function d j
denotes the demand for product Ω produced by firm z. It can be shown that each firm
produces one variety of the good. We will not discriminate between Ω and z in the
remainder of this chapter. The total number of products is given by Ψ(t). All products
produced in either country are consumed in both countries. Total consumption can
be computed by aggregating consumption of each individual product over the whole
continuum of Ψ(t) products. Feenstra (2004) provides a detailed discussion of the
monopolistic competition framework in his "Advanced International Trade" textbook,
which provides more details on the CES preferences and the relationship between
prices and the level of consumption.
2. TRADE AND INNOVATION IN A NORTH-SOUTH FRAMEWORK 345
Perfect free trade allows consumers from both countries to purchase the whole range
of Ψ(t) products that are produced either in the North or the South.
The total number of products, Ψ(t), is treated as exogenously given in the first part
of this chapter but we will allow for innovation in the last subsection. Innovation
increases the number of available goods. Consumers can choose the optimal level of
consumption for each product based upon the elasticity of substitution, ρ > 1. A
constant elasticity of substitution implies that the level of individual consumption is
equal across products if they are sold at identical prices. Consumers spread their
available budget equally among all available products in order to maximize total util-
ity. One can easily prove that individual consumption for one product increases utility
at diminishing rates. More is always better but the changes in utility are diminishing
in the overall level of consumption.
Krugman (1979a) assumes that there is one difference between products from the
North and products from the South. New inventions are produced in the North only,
whereas Southern goods are imitations of older inventions from the North. Thus, the
model features a product cycle. A good is invented in the North and shipped to all
other countries until the South imitates its production in order to underbid the North-
ern competitors. South can imitate the products in a subsequent stage of the product
cycle. Being a low wage country, the South is able to produce the good at lower
costs. Krugman (1979a) and Acemoglu (2008) distinguish between a specialization
equilibrium and an equalization equilibrium.
Specialization. North will specialize its production on ”newly innovated goods” if

the number of new products is sufficiently high. We will see that wages in the North
are above the wages paid in the South. Competitive labor markets ensure that all
products are produced at identical wage costs. Higher wages in the North would
have to be matched with higher goods prices. Consumers are willing to pay those
higher prices for innovated goods that cannot be supplied by the low wage country
in the South. Selling goods that are also offered by the low wage country at a higher
price is not an option. Consumers would purchase all goods from Southern suppliers.
South will take over the whole production of ”old products” but Krugman (1979a) and
Acemoglu (2008) demonstrate that this is a stable equilibrium if the number of ”new
products” is sufficiently high.
Equalization. If the number of ”new products” is sufficiently low, North must main-
tain some of the production of out dated goods, which may also be produced at South.
The existence of this equilibrium depends on the wage at North, which cannot be
higher than the wage at South.
The North-South wage gap is the variable that allows to discriminate between the two
equilibriums. Suppose that the North specializes in the production of new goods and
the wage would adjust until the labor market clearing condition is fulfilled. The lower
the number of new goods, the more workers must be employed by each sector. The
low extensive margin must be compensated by higher labor demand at the intensive
margin. Lower wages stimulate labor demand at the intensive margin but sooner
or later the wage may also decline below the wages payed by firms in the South
producing the ”old products”. Specialization on "new products" is sensible if and
only if the Northern wage is above the Southern wage. The North should stimulate
labor demand at the extensive margin once the wage in the North drops below the
wage in the South. The available workforce would spread on a larger number of
products associated with lower demand for labor in the production of an individual
product and higher wages.
Notice that we still maintain the assumption of competitive labor markets. Thus, all
products are produced at similar labor costs. Suppose we are in the equalization
equilibrium. Both the new and the old products would be produced paying the same
wage to the workers employed in the North. Production of the old goods would be
shared between North and South until the wages in both countries are equalized.
However, it would be better to exploit the low wages in the South by importing the
old products at lower prices from the South. All workers in the North can be shifted
to the innovative sector and those goods can be exported to the South at export prices
above the prices paid for its imports. This patter is reflected in the international wage
gap if the North is able to remain in the ”specialization” equilibrium.
2.1 The specialization equilibrium
Suppose that North specializes on ”new products” and that this outcome is an equi-
librium. The total number of ”new products” at time t is Ψn (t), whereas the number
of ”old products” is denoted by Ψo (t). Both numbers sum up to the total number of
products Ψ(t) = Ψn (t) + Ψo (t). Recall that the existence of a separation equilibrium
depends on the size of the number of ”new products”. We can use the numbers of
products in order to compute the relative size of the ”new products” sector as the ratio
between Ψn (t) and Ψ(t). The description above was not precise enough. Not the ab-
solute number of ”new goods” but its ratio matters for existence of the specialization
equilibrium.
The production function is rather simple. Every worker can produce either one ”new
good” or one ”old good”. Prices across the ”new” and ”old” goods sector can be
different but within each sector prices must be equal and wages have to satisfy
Pn (t) = wn (t) (6.37)
Po (t) = ws (t) (6.38)
Workers are paid the value of their marginal productivity, which is one in both coun-
tries. Wage differentials must be due to price differentials across countries. North
produces ”new goods" by input of one worker per unit of output and South produces
”old goods” by input of one worker per unit of output as well.
We can solve the utility maximization problem associated with equation (6.36) in order
to obtain
−ρ
dn (t) Pn (t)

= (6.39)
do (t) Po (t)
Consumption of each individual good can be solved based upon the production func-
tions. Obviously, total consumption cannot exceed total production. Utility maxi-
mization implies that all consumers at North and South demand equal quantities of
the goods supplied. All individuals spread their income on all available goods ac-
cording to prices. Relative demand for ”new products” is lower or at least equal to
one in the specialization equilibrium because wages at North are at least as high as
Pn (t)
the wage rate in South, wn (t) ≥ ws (t). Thus, Po (t)
≥ 1 . The elasticity of substitution,
dn (t)
ρ, is greater one from which follows that do (t)
≤ 1.
Combining all arguments discussed so far, we can write down total production in
each sector as
Ln
dn (t) = (6.40)
Ψn (t)
Ls
do (t) = (6.41)
Ψo (t)
Notice that demand dn (t) comprises demand from Northern and Southern consumers,
whereas d j in equation (6.36) denotes country-specific demand for a single product.
How can we solve for the endogenous variables, the wages in the Northern and the
Southern country? All we need is to solve for the wage gap using the production
functions and the relative demand for ”new products”.
Recalling that
−1/ρ
Ln

dn (t) Pn (t ) −ρ Pn (t) dn (t) −1/ρ

Ψ n
 (s t) 
= ⇒ = = (6.42)
do (t) Po (t) Po (t) do (t) L
Ψo (t)
from where we obtain the relative wage
1/ρ
wn (t) LS Ψn (t)

= (6.43)
ws (t) L N Ψo (t)
allows us to solve the model. The equation postulates a negative relationship between
the ratio of ”new products” to ”old products” and factor endowments. For a given
relative factor endowment ratio we find a positive relationship between the ”new
product” ratio and the North-South wage gap.
2.2 Equalization equilibrium
Suppose that the ”new product” sector is too small to guarantee that the Northern
wage is equal or above the Southern wage rate. South would still specialize on the
”old products” but North would have to produce both ”old products” and ”new
products”. The number of workers competing for a job in the small ”new product”
sector is too large to guarantee a wage above the wage in South. However, North
can extend production to the ”old products” if wages are low enough to compete
with South. Put differently, the wage ratio must be smaller one in order to be able to
compete with the South.
The North would continue extending its production to ”old products" until wages
in both countries are equal. Extending the production at North drives up the wage
there. Such an extension is associated with a contraction in the South. Thus, the
North-South wage gap adjusts until it is equal to unity in equilibrium.
φLn
dn (t) = (6.44)
Ψn (t)
L S + (1 − φ ) L n
do (t) = (6.45)
Ψo (t)
A fraction of 0 < φ < 1 workers from North are used to produce all ”new products”.
The remaining 1 − φ workers at North are used in the production of ”old goods”.
However, the production of ”old goods” is shared with South, where all workers are
used in the production of ”old goods”. Notice that the world is integrated. It is not
important to determine the location of production of a particular good. All goods can
be transported to the respective consumer’s location at zero transport costs.
Figure (6.2) describes the equilibrium North-South wage ratio as a function of the ratio
between ”new products” and ”old products” conditional on factor endowments at
North and South. The curve tells us that the relative wage depends on the technology
gap. Let us compare two different situations, l1 and l2 . Both scenarios represent the
same North-South country pair. Thus, the endowments are identical in both l1 and l2 .
However, the North is relatively more advanced in scenario l1 compared to scenario
l2 because the ratio of ”new products” relative to the number of ”old products” is
higher in l1 than in l2 . Remember that the only country that can produce ”new goods”
is the North, which gives it an advantage in technology compared to the South. This
technology advantage is reflected in the higher relative wage at the North: E1 is higher
than E2 . The situation in l2 is special. The ratio of ”new products” to ”old products”
Ψn Ln
is exactly equal to the endowment ratio in l2 so that Ψo = Ls . One can easily verify
wn Ψn Ln
that this situation is associated with a wage ratio ws = 1 if Ψo = Ls in equation (6.43).
Figure 6.2: The North-South wage gap
𝑤𝑤 𝑛𝑛
𝑤𝑤 𝑠𝑠
𝐸𝐸1
𝐸𝐸2
1
2. TRADE AND INNOVATION IN A NORTH-SOUTH FRAMEWORK
351
0 𝑙𝑙1 𝑙𝑙2 𝑙𝑙
Suppose that the ratio of ”new products” to ”old products” decreases further to levels
Ψn Ln
Ψo < Ls . This situation is represented by the segment of the relative wage curve to
the right of A2 . The wage curve becomes flat. The wage would not fall below the
value one. The North starts producing ”old products” according to the parameter φ,
which is endogenously determined as explained above.
2.3 Endogenous innovation and imitation
The discussion of the model focused on a solution of the relative wage for given
numbers of ”new products” and ”old products” conditional on exogenous factor en-
dowments. All combinations of relative wages and relative number of ”new prod-
ucts” where summarized in the graphical solution of the model. Following Krugman
(1979a) and Acemoglu (2008), we solve the model in general equilibrium using a sim-
ple law of motion for the evolution of innovation and imitation at North and South.
At birth, the economy is equipped with a certain number of products, Ψ(0). All goods
in this initial period are innovations so that the number of imitations at South is zero.
South will start producing with a lag when some of the innovated products can be
imitated by the South. The law of motion for imitation is assumed to follow
∂Ψo (t)
= iΨn (t) , (6.46)
∂t
∂Ψo (t)
where ∂t is the change in goods that can be imitated, i is a parameter that takes a
value greater 0 and Ψn (t) is the number of ”new products” in period t. Every period
a fraction i of the ”new products” becomes available for imitation in the South. This
change reduces the number of ”new goods” but increases the number of goods that
can be produced by South. If specialization is a feasible equilibrium all ”new goods”
are produced by North and all ”old goods” are produced by South. Does that mean
that less and less goods are produced in the North but more and more goods are
produced in the South?
No, we assume that the North keeps on innovating new products. The total number
of products changes according to
∂Ψ(t)
= νΨ(t) (6.47)
∂t
Obviously, the change in the number of products is equal to the newly created ”new
products” in one period. The functional form is a differential equation. The number
of innovation is proportional to the total number of products (including both ”new
products” and ”old products”).
The total number of ”new products” is the difference between the total number of
products and the total number of imitations Ψn (t) = Ψ(t) − Ψo (t). Innovation and
imitation is unbounded. the number of total products may increase to infinity, which
makes it difficult to derive an equilibrium. Krugman (1979a) analyzes the evolution
Ψn
of σ = Ψ. He is able to show that the fraction of ”new products” converges towards
a steady sate value and remains constant afterwards.
Differentiation σ in order to derive its change gives us
∂Ψn (t) ∂Ψ(t) n

∂σ(t) ∂t Ψ ( t ) ∂t Ψ ( t )
= − (6.48)
∂t [Ψ(t)]2 [Ψ(t)]2
Thus, we obtain a differential equation in σ that can be simplified as
∂Ψn (t) ∂Ψ(t)

∂σ (t)
= ∂t − ∂t σ (6.49)
∂t Ψ(t) Ψ(t)
∂Ψn (t) ∂Ψ(t) ∂Ψo (t)

Substituting (6.46), (6.47) in ∂t = ∂t − ∂t allows us to rewrite
∂Ψ(t) o
∂σ (t) − ∂Ψ∂t(t) ∂Ψ(t)
= ∂t
− ∂t σ (6.50)
∂t Ψ(t) Ψ(t)
n
νΨ(t) − iΨ (t) νΨ(t)
= − σ (6.51)
Ψ(t) Ψ(t)
= ν − iσ − νσ (6.52)
We can compute an equilibrium value of σ by solving the differential equation for

∂σ
∂t = 0, which gives us
ν
σ= (6.53)
i+ν
Thus, the fraction of ”new products” converges to this steady state value and remains
Ψn (t)
constant afterwards. Notice that we need a solution for the ratio Ψo (t)
in order to solve
for the long run steady state wage ratio using equation (6.43).
We can rearrange as follows:
Ψn (t)
σ = (6.54)
Ψ(t)
Ψ(t) Ψ(t) Ψn (t) Ψn (t)
σ = = o , (6.55)
Ψo (t) Ψo (t) Ψ(t) Ψ (t)
where Ψ(t) = Ψn (t) + Ψo (t) can be used in order to obtain
Ψn (t) + Ψo (t) Ψn (t)

σ = (6.56)
Ψo (t) Ψo (t)
Ψo (t) Ψn (t)
σ = (1 − σ ) (6.57)
Ψo (t) Ψo (t)
We can solve this last expression for
Ψn (t) σ
= (6.58)
Ψ (t)
o (1 − σ )
Notice that we already solved for the steady state σ = ν

i +ν . We can plug this solution
into equation (6.58) in order to obtain
Ψn (t) ν
= (6.59)
Ψ (t)
o i
Ψn (t)
This tells us that the ratio Ψo (t)
converges to a constant steady state. Thus, the North-
South wage gap will converge to some constant steady state equilibrium as well. We
can solve for the relative wage by
1/ρ
wn (t) Ls ν

= (6.60)
ws (t) Ln i
All variables on the right side are exogenous parameters, which uniquely pins down
the wage gap between North and South.
2.4 Comparative statics
One interesting exercise is to study the role of innovation and imitation on the rela-
tive wage in North. Suppose that the parameter ν increases at constant i. The wage in
North would further increase relative to the wage in South. Innovations in North in-
crease relative to the imitations in the South, which increases the demand for workers
in the innovative goods production. Wages have to adjust accordingly.
3 Conclusion
The last chapter discussed in this course is a nice summary of many of the problems
studied in this course. We have seen that the implications drawn from a model of
economic growth are not as independent from the countries’ trade policy as most of
the applications suggest. Factor price equalization due to trade liberalization can be
a source of accelerated growth but this strategy is not sustainable.
The focus in the very last model presented in this lecture is on innovation and imita-
tion. Economic development takes place in the developed country but imitation from
the developing country puts pressure on the North to continue innovating new prod-
ucts. Without economic development in form of innovation, factor price equalization
would erode the wage premium in the North.
The mechanism in this model is highly relevant in context of interdependencies be-

tween developed and developing countries. On the one hand, firms suffer from the
lack of intellectual property rights in developing countries. The huge potential de-
mand at developing countries can be an important driver of economic growth for
firms in developed countries but the local competition with domestic firms imitating
their products at lower labor costs is a potential threat for highly innovative firms.
International trade carries this competition on to the domestic market, which drives
firms to differentiate by innovating new products. It usually takes some time until the
know how to imitate those new innovations.
For instance the pattern of trade between China and the West is described very well
by the model.
BIBLIOGRAPHY 357
Bibliography
Acemoglu, D. and F. Zilibotti (1999). Information Accumulation in Development. Jour-

nal of Economic Growth, 4(1), 5-38.
Acemoglu, D., Johnson, S. & Robinson, J. A. (2001). The Colonial Origins of Compar-
ative Development: An Empirical Investigation. American Economic Review, 91(5),
1369-1401.
Acemoglu, D. (2008). Introduction to modern economic growth. Princeton University

Press.
Banerjee, A. and A. F. Newman (1998). Information, the Dual Economy and Develop-
ment. Review of Economic Studies, 65(4), 631-653.
Barro, R. J. (1996). Determinants of Economic Growth: A Cross-Country Empirical

Study. NBER Working Paper, 5698.
Beine, M., Docquier, F., & Rapoport, H. (2001). Brain drain and economic growth:
theory and evidence. Journal of Development Economics, 64(1), 275-289.
Beine, M., Docquier, F., & Rapoport, H. (2007). Measuring international skilled migra-
tion: a new database controlling for age of entry. The World Bank Economic Review,
21(2), 249-254.
Beine, M., Docquier, F., & Rapoport, H. (2008). Brain drain and human capital for-
mation in developing countries: winners and losers. The Economic Journal, 118(528),
631-652.
Beine, M., Docquier, F., & Oden-Defoort, C. (2011). A panel data analysis of the brain
gain. World Development, 39(4), 523-532.
Beine, M., Docquier, F., & Özden, Ç. (2015). Dissecting network externalities in inter-
national migration. Journal of Demographic Economics, 81(04), 379-408.
358 BIBLIOGRAPHY
Bertoli, S., & Moraga, J. F. H. (2015). The size of the cliff at the border. Regional Science
and Urban Economics, 51, 1-6.
Bodvarsson, Ö. B. and Van den Berg, H. (2013). The Economics of Immigration: Theory
and Policy. Springer
Borjas, G. J. (1987). Self-selection and the earnings of immigrants. American Economic

Review, 77(4), 531-53.
Borjas, G. J. (1991). Immigration and self-selection. In Immigration, trade, and the labor
market ( 29-76). University of Chicago Press.
Borjas, G. J. (2003). The Labor Demand Curve Is Downward Sloping: Reexamining

the Impact of Immigration on the Labor Market. The Quarterly Journal of Economics,
118(4), 1335-1374.
Brücker, H. and Defoort, C. (2009). Inequality and the self-selection of international

migrants: theory and new evidence. International Journal of Manpower, 30(7), 742-764.
Brücker, H., Capuano, S., & Marfouk, A. (2013). Education, gender and international
migration: insights from a panel-dataset 1980-2010. IAB Methodology Report.
Card, D. (1990). The Impact of the Mariel Boatlift on the Miami Labor Market. Indus-
trial and Labor Relations Review, 43(2), 245-257.
Chiswick, B. and Miller, P. W. (Eds.). (2014a). The Immigrants. In Handbook of the

Economics of International Migration. Elsevier.
Chiswick, B. and Miller, P. W. (Eds.). (2014b). The Impact. In Handbook of the Economics
of International Migration. Elsevier.
Desai, M. A., Kapur, D., McHale, J., & Rogers, K. (2009). The fiscal impact of high-
skilled emigration: Flows of Indians to the US. Journal of Development Economics,
88(1), 32-44.
BIBLIOGRAPHY 359
Docquier, F. and Marfouk, A. (2006). International migration by educational attain-

ment, 1990-2000.
Docquier, F., Lowell, B. L., & Marfouk, A. (2009). A gendered assessment of highly
skilled emigration. Population and Development Review, 35(2), 297-321.
Docquier, F. and Rapoport, H. (2009). Documenting the Brain Drain of “La Creme de
la Creme”. Jahrbücher für Nationalökonomie und Statistik, 229(6), 679-705.
Docquier, F. and Rapoport, H. (2012). Globalization, brain drain, and development.

Journal of Economic Literature, 50(3), 681-730.
Dustmann, C., Ludsteck, J., & Schönberg, U. (2009). Revisiting the German wage
structure. Quarterly Journal of Economics, 124(2), 843-881.
Dustmann, C. and Görlach, J. S. (2016). The economics of temporary migrations. Jour-

nal of Economic Literature, 54(1), 98-136.
Feenstra, R. C. and Hanson, G. H. (1995). Foreign investment, outsourcing and relative

wages. NBER Working Paper 5121.
Feenstra, R. C. (2004). Advanced international trade: theory and evidence. Princeton Uni-
versity Press.
Feenstra, R. C. and Taylor, A. M. (2014). International Economics. Worth Publishers
Gill, I. and Kharas, H. (2007). An East Asian Renaissance: Ideas for Economic Growth. The
World Bank.
Harris, J. R. and Todaro, M. P. (1970). Migration, unemployment and development: a

two-sector analysis. The American Economic Review, 60(1), 126-142.
Krugman, P. R. (1979a). A Model of Innovation, Technology Transfer, and the World

Distribution of Income. The Journal of Political Economy, 87(2), 253-266.
Krugman, P. R. (1979b). Increasing returns, monopolistic competition, and interna-

tional trade. Journal of International Economics, 9(4), 469-479.
360 BIBLIOGRAPHY
Krugman, P. R. (1991). Increasing Returns and Economic Geography. Journal of Political

Economy, 99(3), 483-499.
Lewis, W. A. (1954). Economic Development with Unlimited Supplies of Labour. The

Manchester School, 22(2), 139-191.
Li, H., Li, L., Wu, B., & Xiong Y. (2012). The End of Cheap Chinese Labor. Journal of
Economic Perspectives, 26(4), 57–74.
van Marrewijk, C. (2012). International Economics. Oxford University Press.
Mattes, R. and Mniki, N. (2007). Restless minds: South African students and the brain
drain. Development Southern Africa, 24(1), 25-46.
Mayda, A. M. (2010). International migration: A panel data analysis of the determi-

nants of bilateral flows. Journal of Population Economics, 23(4), 1249-1274.
Nanda, R. and Khanna, T. (2010). Diasporas and domestic entrepreneurs: Evidence

from the Indian software industry. Journal of Economics & Management Strategy, 19(4),
991-1012.
OECD, (2015). International Migration Outlook 2015. OECD Publishing.
Olley, G. S. and Pakes, A. (1992). The dynamics of productivity in the telecommuni-

cations equipment industry. NBER Working Paper 3977.
Ortega, F. and Peri G. (2013). The effect of income and immigration policies on inter-
national migration. Migration Studies, 1(1), 47-74.
Ottaviano, G. and Peri G. (2012). Rethinking the Effect of Immigration on Wages.

Journal of the European Economic Association 10(1), 152-197.
Pedersen, P. J., Pytlikova, M., & Smith, N. (2008). Selection and network effects-
Migration flows into OECD countries 1990–2000. European Economic Review, 52(7),
1160-1186.
Ricardo, D. (1821). On the Principles of Political Economy and Taxation. John Murray.
BIBLIOGRAPHY 361
Roy, A. D. (1951). Some thoughts on the distribution of earnings. Oxford economic

papers, 3(2), 135-146.
Sajaastad L. A. (1962). The Costs and Returns of Human Migration. The Journal of
Political Economy, 70(5), 80-93.
Solow, R.M. (1956). A Contribution to the Theory of Economic Growth. The Quarterly
Journal of Economics, 70(1), 65-94.
Ventura, J. (1997). Growth and Interdependence. The Quarterly Journal of Economics,

112(1), 57-84.

3 - Lecture Notes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

3 - Lecture Notes

Uploaded by

Copyright:

Available Formats

Contents

2 The Middle Income Trap 19

2.0.1 Agglomeration: A way out of the middle income trap? . . . . . . 21

2.1 Internal Migration and economies of scale effects . . . . . . . . . . . . . 25

2.1.1 Trade liberalization . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.2 Agglomeration and migration . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.1 The Solow (1956) model of economic growth . . . . . . . . . . . . . . . . 58

3.1.1 Basic model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.1.2 Dynamics in the short run . . . . . . . . . . . . . . . . . . . . . . . 66

3.1.3 Model solution under Cobb Douglas . . . . . . . . . . . . . . . . 84

3.2 Population growth in the Solow model . . . . . . . . . . . . . . . . . . . 89

3.3 Economic growth and technological change . . . . . . . . . . . . . . . . . 93

3.3.1 Dynamics with technological change . . . . . . . . . . . . . . . . 99

3.4 Growth and internal migration . . . . . . . . . . . . . . . . . . . . . . . . 102

3.5 Community enforcement, migration and development . . . . . . . . . . 107

3.5.1 Empirical relevancy . . . . . . . . . . . . . . . . . . . . . . . . . . 112

3.6 Human Capital in the Solow model . . . . . . . . . . . . . . . . . . . . . 117

3.7 Empirical application on institutions and growth . . . . . . . . . . . . . 125

A An introduction to linear regression . . . . . . . . . . . . . . . . . . . . . 133

A-1 Dependent Variable, Independent variable and Regressions co-

A-3 Standard errors and statistical significance . . . . . . . . . . . . . 137

A-4 Linear regression with many variables . . . . . . . . . . . . . . . 139

4 Canonical Trade Models 141

1 The Ricardo Model of Comparative Advantage . . . . . . . . . . . . . . . 142

1.1 The model basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

1.2 Gains from trade . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

1.3 Trade liberalization and income . . . . . . . . . . . . . . . . . . . 153

1.4 Equilibrium goods prices . . . . . . . . . . . . . . . . . . . . . . . 154

2 The Heckscher-Ohlin model . . . . . . . . . . . . . . . . . . . . . . . . . . 156

2.1 The 2 × 2 × 2 model . . . . . . . . . . . . . . . . . . . . . . . . . . 157

2.2 Production, consumption and the gains from trade . . . . . . . . 161

2.3 Factor prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

2.4 Trade liberalization and factor prices . . . . . . . . . . . . . . . . 168

2.5 Stolper Samuelson Theorem . . . . . . . . . . . . . . . . . . . . . 170

3 FDI and outsourcing in a multi-industry framework . . . . . . . . . . . . 179

3.1 The Feenstra and Hanson model . . . . . . . . . . . . . . . . . . . 183

3.2 Capital market liberalization, offshoring and wage inequality . . 191

3.3 The role of the continuum of industries . . . . . . . . . . . . . . . 193

3.4 Empirical tests of the Feenstra and Hanson model . . . . . . . . 200

4 Migration in the 2 × 2 × 2 model . . . . . . . . . . . . . . . . . . . . . . . 209

4.1 Endowments in the 2 × 2 × 2 model . . . . . . . . . . . . . . . . . 210

4.2 The Migration Shock . . . . . . . . . . . . . . . . . . . . . . . . . . 216

4.3 Formal proof of the Rybzynski theorem . . . . . . . . . . . . . . . 220

5 The Mariel boat lift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

5 International migration 227

1 International migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

2 Why do people migrate? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

2.1 The determinants of the migration decision . . . . . . . . . . . . 232

2.2 The determinants of international migration: empirical evidence 252

3 Who migrates? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256

3.1 The Borjas 1987 Model . . . . . . . . . . . . . . . . . . . . . . . . . 256

4 The brain drain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272

4.1 The brain drain: theoretical framework . . . . . . . . . . . . . . . 276

4.2 Model simulation in Excel . . . . . . . . . . . . . . . . . . . . . . . 293

4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

4.4 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

4.5 The brain drain: empirical evidence . . . . . . . . . . . . . . . . . 313

5 The effects of immigration on the host country . . . . . . . . . . . . . . . 317

6 Trade and growth 327

1 Economic growth in open economy models . . . . . . . . . . . . . . . . . 328

1.1 Consumption of final goods . . . . . . . . . . . . . . . . . . . . . 333

2 Trade and innovation in a North-South framework . . . . . . . . . . . . 344

2.1 The specialization equilibrium . . . . . . . . . . . . . . . . . . . . 347