by trevor tombe - university of toronto t-space...penn state, simon fraser, toronto, wilfrid...

STRUCTURAL CHANGE AND INCOME DIFFERENCES

by

Trevor Tombe

A thesis submitted in conformity with the requirementsfor the degree of Doctor of Philosophy

Graduate Department of EconomicsUniversity of Toronto

Copyright c© 2011 by Trevor Tombe

Abstract

Structural Change and Income Differences

Trevor Tombe

Doctor of Philosophy

Graduate Department of Economics

University of Toronto

2011

Economic growth and development is intimately related to the decline of agriculture’s share of output

and employment. This process of structural change has important implications for income and pro-

ductivity differences between regions within a country or between countries themselves. Agriculture

typically has low productivity relative to other sectors and this is particularly true in poor areas. So, as

labour switches to nonagricultural activities or as agricultural productivity increases, poor agriculturally-

intensive areas will benefit the most. In this thesis, I contribute to a recent and growing line of research

and incorporate a separate role for agriculture, both into modeling frameworks and data analysis, to

examine income and productivity differences.

I first demonstrate that restrictions on trade in agricultural goods, which support inefficient domestic

producers, inhibit structural change and lower productivity in poor countries. To do this, I incorporate

multiple sectors, non-homothetic preferences, and labour mobility costs into an Eaton-Kortum trade

model. With the model, I estimate productivity from trade data (avoiding problematic data for poor

countries that typical estimates require) and perform a variety of counterfactual exercises. I find im-

port barriers and labour mobility costs account for one-third of the aggregate labour productivity gap

between rich and poor countries and for nearly half the gap in agriculture. Second, moving away from

international income differences, I use a general equilibrium model of structural transformation to show

a large labour migration cost between regions of the US magnifies the impact improved labour markets

have on regional convergence. Finally, I estimate the influence of structural change on convergence

between Canadian regions. I construct a unique dataset of census-division level wage and employment

levels in both agriculture and nonagriculture between 1901 and 1981. I find convergence is primarily

due to region-specific factors with structural change playing little role.

ii

Acknowledgements

I thank Xiaodong Zhu, Diego Restuccia, and Gueorgui Kambourov, for their extremely valuable guid-

ance and supervision over the past few years. I would also like to express thanks to those who con-

tributed with insightful comments, suggestions, and encouragements at various stages of this thesis:

Tasso Adamopoulos, Michelle Alexopoulos, Mickael Baker, Bernardo Blum, Branko Boskovic, Loren

Brandt, Jay Cao, Margarida Duarte, Andres Erosa, Berthold Herrendorf, Ignatius Horstmann, Sacha

Kapoor, Mara Lederman, Peter Morrow, Joanne Roberts, Andrés Rodríguez-Clare, Richard Roger-

son, Aloysius Siow, Kitty Wang, and Jennifer Winter. I also thank the various seminar participants at

Penn State, Simon Fraser, Toronto, Wilfrid Laurier, and York, along with conference participants at the

Canadian Economics Association 2009 and 2010 Meetings, Econometric Society North American 2009

Summer Meeting, Midwest Macroeconomics 2010 Meeting, and Tsinghua’s 2010 Macroeconomics

Workshop, for many helpful comments.

iii

Contents

1 The Missing Food Problem 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 A Model Consistent with Stylized Facts . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.2.1 Households’ Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.2.2 Production Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.2.3 International Prices and Trade Patterns . . . . . . . . . . . . . . . . . . . . . 10

1.2.4 Labour Market and Trade Balance Conditions . . . . . . . . . . . . . . . . . . 11

1.2.5 Equilibrium Definition and Solving the Model . . . . . . . . . . . . . . . . . 12

1.3 Calibrating the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.3.1 Productivity and Trade Costs . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.3.2 Subsistence, Service Sector Productivity, and Labour Market Distortions . . . 19

1.4 Results from the Baseline Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.4.1 Trade Cost Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.4.2 Sectoral Labour Productivity . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.5 Counterfactual Experiments: Trade, Productivity, and Income . . . . . . . . . . . . . . 23

1.5.1 International Food Trade Flows . . . . . . . . . . . . . . . . . . . . . . . . . 24

1.5.2 Cross-Country Productivity Gaps . . . . . . . . . . . . . . . . . . . . . . . . 25

1.5.3 Decomposition: Cross Country Aggregate Productivity and Income Variation . 26

1.6 Discussion and Robustness of Results . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.6.1 Alternative Values for θ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.6.2 Alternative Counterfactual Experiments . . . . . . . . . . . . . . . . . . . . . 28

1.6.3 Plausibility of Trade Cost Estimates . . . . . . . . . . . . . . . . . . . . . . . 28

1.6.4 Implications for Price Differentials . . . . . . . . . . . . . . . . . . . . . . . 30

1.6.5 OECD Agricultural Producer Support . . . . . . . . . . . . . . . . . . . . . . 31

1.6.6 Actual Development Experiences . . . . . . . . . . . . . . . . . . . . . . . . 32

1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 Regions, Frictions, and Migrations 472.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.2 Empirical Patterns, by Region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

iv

2.3 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.3.1 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.3.2 Households . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2.3.3 Market Clearing Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.4 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.5 Counterfactual Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

2.5.1 Labour Market Frictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

2.5.2 Goods Market Frictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

2.6.1 Effects of Transportation and Migration Costs . . . . . . . . . . . . . . . . . . 61

2.6.2 Calibration of Transportation Cost Parameter . . . . . . . . . . . . . . . . . . 61

2.6.3 Calibration of Peripheral Labour Market Frictions . . . . . . . . . . . . . . . 62

2.6.4 Calibration of Between-Region Migration Costs . . . . . . . . . . . . . . . . 63

2.6.5 Alternative Productivity Calibration . . . . . . . . . . . . . . . . . . . . . . . 65

2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3 Structural Change and Canadian Convergence 703.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.2 Convergence Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.1 Core vs. Peripheral Classification . . . . . . . . . . . . . . . . . . . . . . . . 73

3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

3.3.1 Variable Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3.5 Discussion and Sensitivity of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.5.1 Direct Comparison to US Experience . . . . . . . . . . . . . . . . . . . . . . 79

3.5.2 Exclusion of Western Provinces . . . . . . . . . . . . . . . . . . . . . . . . . 82

3.5.3 Alternative Regional Classifications of Census-Divisions . . . . . . . . . . . . 86

3.5.4 Agricultural Employment Definition . . . . . . . . . . . . . . . . . . . . . . . 86

3.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

Bibliography 99

v

List of Tables

1.1 Calibration of Model Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151.2 Main Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171.3 Selected Values from Stage-1 Calibration . . . . . . . . . . . . . . . . . . . . . . . . 181.4 Aggregate Productivity and Employment Shares, Model vs. Data . . . . . . . . . . . . 211.5 Baseline Model: Cross Country Productivity Differentials . . . . . . . . . . . . . . . . 221.6 Trade Between 1st and 4th Quartiles . . . . . . . . . . . . . . . . . . . . . . . . . . . 241.7 Results of Main Counterfactual Experiments . . . . . . . . . . . . . . . . . . . . . . . 251.8 Counterfactual Aggregate Productivity Gaps, with Fixed Labour Allocations . . . . . . 261.9 Contribution to Productivity Gaps, Various θ . . . . . . . . . . . . . . . . . . . . . . 271.10 Counterfactual Productivity Gaps, Various Experiments . . . . . . . . . . . . . . . . . 291.11 Counterfactual Productivity Gaps, Full Liberalizations . . . . . . . . . . . . . . . . . 301.12 Relative Productivity and Trade Estimates . . . . . . . . . . . . . . . . . . . . . . . . 34

2.1 Calibration of Model Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562.2 Calibration Performance vs. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 582.3 Isolating the Effect of Labour Market Improvements . . . . . . . . . . . . . . . . . . 592.4 Isolating the Effect of Transportation Cost Reductions . . . . . . . . . . . . . . . . . . 602.5 Model Performance under Various Migration Cost Assumptions . . . . . . . . . . . . 642.6 Average Annual Growth Rates of Key Variables, 1880-1990 . . . . . . . . . . . . . . 65

3.1 Key Statistics of the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.2 Convergence Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783.3 Results from Table 2 of Caselli and Coleman [2001] . . . . . . . . . . . . . . . . . . . 793.4 Convergence Decompositions (US Data from Caselli and Coleman [2001]) . . . . . . 803.5 Selected Comparison of US and Canadian Experience . . . . . . . . . . . . . . . . . . 813.6 Key Features of the Data - West Excluded . . . . . . . . . . . . . . . . . . . . . . . . 843.7 Convergence Decompositions - Excluding the West . . . . . . . . . . . . . . . . . . . 853.8 Labour Force Shares - With Farm Operators . . . . . . . . . . . . . . . . . . . . . . . 863.9 Classifications of All 1901-1981 Census Divisions (P=Peripheral, C=Core) . . . . . . 903.10 Agricultural Employment Share - With Operators . . . . . . . . . . . . . . . . . . . . 953.11 Annual Earnings, By Occupational Group, in Dollars . . . . . . . . . . . . . . . . . . 953.12 Convergence Decompositions - Including Farm Operators . . . . . . . . . . . . . . . . 963.13 Selected Earnings and Employment Share Data . . . . . . . . . . . . . . . . . . . . . 973.14 List of Key Variables for Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . 98

vi

List of Figures

1.1 The Food Problem in Poor Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Agricultural Import Share of GDP, by Country . . . . . . . . . . . . . . . . . . . . . . 41.3 Fit of the Stage-1 Calibrated Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 181.4 The Role of Subsistence Food Requirements . . . . . . . . . . . . . . . . . . . . . . . 201.5 Normalized Import Shares: No Import Barriers . . . . . . . . . . . . . . . . . . . . . 371.6 Normalized Import Shares: No Labour Mobility Costs . . . . . . . . . . . . . . . . . 381.7 Normalized Import Shares: No Import Barriers or Labour Mobility Costs . . . . . . . 391.8 Counterfactual Gains in GDP/Worker . . . . . . . . . . . . . . . . . . . . . . . . . . 391.9 Real Output-per-Worker in Agriculture Relative to Manufacturing . . . . . . . . . . . 401.10 Agricultural Labour Productivity, Model Estimates vs. Data . . . . . . . . . . . . . . 401.11 Trade Cost Estimates for Agricultural Goods . . . . . . . . . . . . . . . . . . . . . . . 411.12 Trade Cost Estimates for Manufactured Goods . . . . . . . . . . . . . . . . . . . . . . 421.13 Competitiveness Measure for Agriculture . . . . . . . . . . . . . . . . . . . . . . . . 431.14 Competitiveness Measure for Manufacturing . . . . . . . . . . . . . . . . . . . . . . . 431.15 Import Shares of Poorest Countries, by Source Country Percentile . . . . . . . . . . . 441.16 Import Shares of Richest Countries, by Source Country Percentile . . . . . . . . . . . 441.17 Increasing S-S Trade, Following Full Removal of Import Barriers . . . . . . . . . . . . 451.18 Increasing S-S Trade, Following Full Removal of Import Barriers . . . . . . . . . . . . 451.19 Increased Share of Agriculture in Total Exports Following Trade Liberalization . . . . 461.20 FAO Food Prices are Higher than ICP Prices, Especially for Poor Countries . . . . . . 46

2.1 US Census Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672.2 Agricultural Share of Employment, by Region . . . . . . . . . . . . . . . . . . . . . . 672.3 Relative Agricultural Wages, by Region . . . . . . . . . . . . . . . . . . . . . . . . . 682.4 Overall Average Wage Relative to Northeast, by Region . . . . . . . . . . . . . . . . . 682.5 Employment Relative to Northeast, by Region . . . . . . . . . . . . . . . . . . . . . . 69

3.1 Earnings and Convergence Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.2 Earnings and Convergence Plot (Using the US Data) . . . . . . . . . . . . . . . . . . 813.3 Foreign Born Population Share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.4 Earnings and Convergence Plot - West Excluded . . . . . . . . . . . . . . . . . . . . . 833.5 All North-South Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.6 Earnings and Convergence Plot, Including Farm Owner-Operators . . . . . . . . . . . 873.7 Earnings and Convergence Plot, Including Farm Owner-Operators - West Excluded . . 883.8 Illustration of Canada’s Census Divisions . . . . . . . . . . . . . . . . . . . . . . . . 94

vii

Chapter 1

The Missing Food Problem: How Low

Agricultural Imports Contributes to

International Income Differences

Abstract

This chapter finds an important relationship between the international food trade and cross-country in-come and productivity differences. Poor countries have low labour productivity in agriculture relativeto other sectors, yet predominantly consume domestically-produced food. To understand these observa-tions, I describe and exploit a general equilibrium model of international trade to: (1) measure sectoralproductivity and trade costs across countries; and (2) quantify the impact of low poor-country food im-ports on international income and productivity gaps. Specifically, I expand on Yi and Zhang [2010] andmodify an Eaton-Kortum trade model to incorporate multiple sectors, non-homothetic preferences, andlabour mobility costs. With this model, I estimate PPP-adjusted productivity from observed bilateraltrade data, avoiding problematic price and employment data in poor countries that direct output-per-worker estimates require. I find reasonable trade barriers and labour mobility costs account for thelow poor-country imports despite their low productivity. Through various counterfactual experiments, Iquantify how easing import barriers and labour mobility costs increases imports and within-agriculturespecialization, shuts down low productivity domestic food producers, and lowers the gap between richand poor countries. I also find an interaction between domestic labour-market distortions and tradebarriers not found in the existing dual-economy literature, which largely abstracts from open-economyconsiderations. Overall, I account for one-third of the aggregate labour productivity gap between richand poor countries and for nearly half the gap in agriculture.

1

CHAPTER 1. THE MISSING FOOD PROBLEM 2

1.1 Introduction

The difference in agricultural labour productivity between rich and poor countries is large and accounts

for most of the aggregate income and productivity gap.1 Average labour productivity in agriculture,

for example, differ nearly by a factor of 70 between the poorest- and richest-10% of countries, but

by only 6 in nonagriculture.2 Differences within agriculture drive differences in aggregate, since agri-

culture accounts for most employment and spending in poor countries (see Figures 1.1a and 1.1b);

Schultz [1953] calls this the Food Problem. Despite these productivity differences, Figure 1.2 shows

the food import share of GDP rises with income, implying poor countries do not generally substitute

imports for low productivity domestic producers.3 This motivates existing literature to abstract from

open-economy considerations and focus on domestic distortions within closed-economy frameworks to

understand agricultural productivity gaps. I depart from this approach and show limited food imports

itself inhibits structural change and lowers agricultural productivity in poor countries. There is also a

quantitatively important interaction between domestic distortions and trade barriers.4 Overall, I find lim-

ited food imports and labour misallocation account for nearly half the agricultural labour productivity

gap between rich and poor countries, and a third of aggregate income and productivity differences.

To demonstrate food imports have a first-order contribution to aggregate productivity gaps between

rich and poor countries, I present a trade model consistent with stylized facts of development that builds

upon Yi and Zhang [2010].5 Specifically, I embed an augmented Ricardian trade model into a dual-

economy (agriculture vs. nonagriculture) model of structural change. The model incorporates hori-

zontally differentiated and tradable agricultural and manufactured goods, individually structured as in

Eaton and Kortum [2002], and a nontradable service-sector. Product differentiation within each tradable

1See, for example, Hayami and Ruttan [1970], Kuznets [1971], Kawagoe et al. [1985], Hayami and Ruttan [1985],Rao [1993], Gollin et al. [2004], Cordoba and Ripoll [2006], Gollin et al. [2007], Adamopoulos [2010], Vollrath [2009],Adamopoulos and Restuccia [2010], Duarte and Restuccia [2010], Lagakos and Waugh [2010].

2These results are for 2000 and utilize PPP-adjusted agricultural value added data from the UN-FAO. The aggregatedifference in this sample of 173 countries is 35. Restuccia et al. [2008] find similar results: for 86 countries in 1985, thepoorest 10% have agricultural labour productivity 56 times lower than the richest 10%, but differ in nonagriculture by only5. Caselli [2005] finds that equalizing agricultural productivity across countries nearly eliminates all international incomedifferences. Specifically, he finds the 90/10 ratio of aggregate income falls from 19 to 1.9 in a sample of 80 countries.

3As a fraction of food spending, rich country imports are an order of magnitude larger.4Such interactions have been identified in other literature (see, for example, Kambourov [2009] and Artuc et al. [2010])

but studies accounting for cross-country productivity and income differences typically abstract from such considerations.5Yi and Zhang [2010] develop a multi-sector version of Eaton and Kortum [2002] and clearly link trade and structural

transformation. My contribution builds on their stylized treatment by quantitatively applying the framework to cross country.I also incorporate nonhomothetic preferences and labour-market distortions.


Figure 1.1: The Food Problem in Poor Countries

(a) Food Expenditure Shares, Selected Countries, 2005

ALB

ARG

ARM

AZE

BDI

BENBFA

BGD

BGR

BLRBOLBRA

CHL

CIVCMR

COL

CRI

DOMECU EST

ETH

GEO

GHAGIN

GMB

GTM

GUY HND

HTI

HUN

IDN

IND

JAM

JOR

KAZ

KENKGZ

KHMLAO

LKA

LTU

MARMDA

MDG

MEX

MKD

MLI

MRT NGA

NIC

NPL

PAK

PAN

PER

PHL

POL

PRY

ROMRUS

SEN

SLV

THATJK

TUR

TZA

UGA

UKR

URY

UZB

VEN

VNMYEM ZAF

.2.4

.6.8

Pop

ulat

ion

Ave

rage

Foo

d E

xpen

ditu

re S

hare

1000 3500 10000 35000 100000Real GDP/Worker, Log Scale

Source: Hoyos and Lessem (2008)

(b) Agricultural Employment Shares, by Country

ALB

ARE

ARG

ARM

AUSAUT

AZE

BGD

BGR

BHS

BLZ

BOL

BRA

BRB CANCHE

CHL

CHN

CMR

COL

CRI

CYPCZE

DOM

DZA

ECU

EGY

ESPEST

ETH

FINFRA

GBR

GEO

GER

GRC

GTM

GUY

HND

HRV

HUN

IDN

IRN

IRQ

ITA

JAM

JOR JPN

KAZ

KGZ

KHM

KOR

LKA

LTULVA

MAR

MDA

MDG

MEXMKD

MLI

MNG

MUS

MYS

NIC

NPL

NZLOMN

PAK

PAN

PER

PHL

PNG

POL

PRT

PRY

ROM

RUS

SAU

SEN

SLE

SLV

SURSVKSVN

SWE

SYR

THA

TJK

TTO

TUR

TZA

UGA

UKR

URY USA

VEN

VNM

ZAF

ZMB

0.2

.4.6

.8A

gric

ultu

ral E

mpl

oym

ent S

hare


Labour share source: World Bank for 2000 (or 95−05 average); GDP data source: PWT6.3 for 2000

sector avoids counterfactual specialization patterns found in homogeneous goods frameworks and, im-

portantly, allows for within-sector trade.6 I exploit the Eaton-Kortum structure to infer sectoral labour

productivity from observable trade data, avoiding data limitations that prevent more direct measures.7 I

also perform counterfactual experiments within the fully calibrated model to highlight the importance of

agricultural trade - or lack thereof - in accounting for cross country income and productivity differences.

I focus on two distortions that limit food imports: high international trade barriers and costly labour

mobility. Trade barriers, such as tariffs, quotas, regulations, or poor infrastructure, increase import

prices and lead consumers to opt for lower productivity domestic producers. Costly labour mobility,

such as regional migration restrictions or scarce rural education, makes switching to non-agricultural

activities difficult for farm labour, thereby increasing farm employment and decreasing farm wages.8 In

fact, wages in agriculture relative to non-agriculture increase strongly with a country’s level of develop-

ment, and differ by a factor of four to five in many poor countries. Low farm wages imply low output

prices and consumers - again - opt for lower productivity domestic producers over imports. Without

6Armington models, for instance, imply every country exports to every other country. The Eaton-Kortum structure allowsmany producers of a product variety but each supplying different parts of the world. Within-sector trade flows are necessaryto investigate specialization patterns within each sector, which relates sectoral productivity to import flows.

7Data limitations lead most studies of cross-country sectoral productivity to focus on developed economies. Rao [1993],Restuccia et al. [2008] are important exceptions, using FAO farm output prices to measure real agricultural productivity acrosscountries.

8Vollrath [2009] shows wage gaps do not reflect sectoral differences in physical or human capital endowments. See Caselliand Coleman [2001] for an exploration of the role learning costs play in structural change and, as a follow up, Tombe [2008]for how such costs may interact with transportation costs in a larger set of US states. Cordoba and Ripoll [2006] find schoolingor migration costs do not account for the sectoral labour productivity differences, but low quality of human capital in ruralareas.


Figure 1.2: Agricultural Import Share of GDP, by Country

ALB

ARE

ARG

ARM

AUS

AUT

AZE

BGD

BGR

BHS

BLZBOL

BRA

BRB

CAN

CHE

CHL

CHN

CMR

COL

CRI

CYP

CZE

DOM

DZA

ECU EGY

ESP

EST

ETH

FIN

FRAGBR

GEO

GERGRC

GTM

GUY

HND

HRV

HUN

IDN

IRNIRQ

ITA

JAM

JOR

JPN

KAZ

KGZKHM

KOR

LKA

LTU

LVA

MAR

MDA

MDG

MEX

MKD

MLI

MNG

MUS

MYS

NIC

NPL

NZL

OMN

PAK

PANPERPHL

PNG

POL

PRT

PRY

ROM RUS

SAUSEN

SLESLV

SUR

SVK

SVN

SWE

SYR

THATJK TTO

TURTZA

UGA

UKR

URY

USA

VEN

VNM ZAFZMB

0.0

05.0

1.0

15.0

2A

gric

ultu

ral I

mpo

rts

to G

DP

Rat

io


Trade data source: NBER−UN for 2000; GDP data source: PWT6.3 for 2000

these distortions, increased imports lead low productivity producers to shut down and labour to concen-

trate in fewer agricultural varieties or switch to nonagricultural activities.9 Admittedly, I incorporate

both distortions in a simple way and do not address complex political-economy issues to explain why

these distortions exist, such as balancing pressures between rural and urban residents.

While agnostic about the causes, I can quantify the costs of these distortions independently of imple-

mentation and transition issues governments would face. Specifically, I investigate: (1) lowering import

barriers everywhere to the average level of the richest countries; (2) eliminating labour mobility costs;

and (3) both together. The two distortions have important interaction effects, with trade liberalization

and improved labour mobility together driving the largest reductions in cross-country differences. The

aggregate productivity gap between the richest- and poorest-10% of countries shrinks by 33% when

both distortions are reduced, but only 23% and 7% when import barriers and labour mobility costs are

eased individually.10 These results are particularly important given the literature’s focus on domestic

distortions alone.

Key to these results are cross-country productivity and trade cost estimates, which I infer from bi-

9Existing trade literature consistently finds exporting is associated with a reallocation towards more productive plants,which is a quantitatively important component of overall productivity growth within the manufacturing sector. See Bernardand Jensen [2001], Pavcnik [2002], Bernard et al. [2006].

10Rich and poor countries are the top and bottom deciles of GDP/Capita in 2000.


lateral trade data within the Eaton-Kortum structure of the model. To see how this inference is made,

consider the following features of the global food trade between the 114 countries in my sample.11 For

the bottom quartile of countries, imports account for only 2% of food expenditures while imports are

approximately 40% of food spending in the top quartile. Poor countries source half of total food im-

ports from rich countries but very little from other poor countries.12 Similarly, rich countries source the

majority of their total food imports from other rich countries. While I present precise productivity and

trade cost estimates from thousands of importer-exporter flows in Section 1.3.1, the intuition is straight-

forward. I infer high labour productivity for a country if other countries import a disproportionately

high share from that country. I infer high import barriers if overall import shares are below the model’s

prediction, given a country’s productivity.13

For concreteness, consider trade between the United Kingdom, Cuba, and Canada. Both the UK and

Cuba allocate approximately 1.4% of their food spending to imports from Canada. On the other hand,

Canada allocates 1% of its spending to imports from the United Kingdom but only 0.2% to imports

from Cuba. The model infers low Cuban productivity from Canada’s low share. Given Cuba’s low

productivity, imports are an attractive alternative to domestic production. In the data, however, Cuba

imports a similar share to another - higher productivity - country, the United Kingdom. The model infers

high Cuban import costs from Cuba’s lower than predicted import share. Overall, I consider over 6,000

trade pairs in agriculture and over 9,000 in manufacturing to estimate that imports costs for the poorest-

10% of countries are approximately 43% more in agriculture than the average country, compared to 41%

less than average for rich countries.14 Labour productivity estimates suggest rich countries are over 100

11The precise data used and approach taken to estimate expenditure shares, both domestic and foreign (by source), issimilar to Bernard et al. [2003]. I use the NBER-UN Trade Database for the year 2000. I provide details in Section 1.3.1.

12The remaining imports are from the middle-income countries. For more on the low South-South trade levels, see Linder[1961], Markusen [1986], Feenstra [1988], Hunter [1991], Echevarria [2000], Fieler [2010]

13The procedure I employ to infer productivity and trade costs from observable trade flows, within the Eaton-Kortumframework, is broadly consistent with a large trade literature that I discuss briefly (see, for example, Eaton and Kortum [2001],Costinot et al. [2010], Levchenko and Zhang [2010], Waugh [2010]. For an alternative mapping of bilateral import sharepatterns to productivity and trade costs presented by Waugh [2010] will be investigated in Section 1.6.4

14To place the importer-specific fixed effects in context, note that Waugh [2010] finds a -62% fixed-effect for the UnitedStates’ manufacturing goods trade. My results suggest -76% for manufacturing and -55% in agriculture. Details will follow inSection 1.3.1. Overall, the trade weighted average import costs across all bilateral trading partners for the 114 countries in mysample is 263% in both agriculture and manufacturing. For rich countries, this number is 104% on average in agriculture and78% in manufacturing. For comparison, Anderson and van Wincoop [2004] report tariff-equivalent US trade costs of 170%.


times more productive in agriculture than poor countries, and over 70 times more in manufacturing.15

Non-tradable services productivity is calibrated so the baseline model aggregate matches data, resulting

in a factor of 10 difference between rich and poor.

Unlike direct estimates, labour productivity inferred from observed trade flows avoids using pro-

ducer price and labour input data, which are problematic for most developing countries. For cer-

tain years, industry-level producer prices among OECD countries are available through the Groningen

Growth and Development Centre [Inklaar and Timmer, 2008]. For developing countries, only expen-

diture prices, not producer prices, are available through the World Bank’s International Comparisons

Project (ICP). Productivity estimates from expenditure prices will be biased for two reasons: (1) dis-

tribution margins are systematically related to a country’s level of development [Adamopoulos, 2008];

and (2) expenditure prices capture many manufactured and service components of consumption, such as

packaging or preparation. With these limitations in mind, Duarte and Restuccia [2010] study structural

change and productivity growth over time in OECD economies with model-implied sectoral productivity

estimates; their approach, however, requires accurate employment data. For many developing countries,

standard surveys overestimate farm labour since rural residents and farm workers are treated synony-

mously [Gollin et al., 2004].16 For these reasons, I estimate productivity revealed through observed

bilateral trade patterns.17

I contribute to a large international macroeconomics literature on agriculture’s role in develop-

ment.18 This literature focuses on causes of low agricultural productivity and explanations vary from

inefficient farm sizes [Adamopoulos and Restuccia, 2010], poor domestic transportation infrastructure

[Adamopoulos, 2010, Gollin and Rogerson, 2010], comparatively low quality farm workers [Lagakos

15Agriculture’s relative labour productivity is low in poor countries, consistent cross country productivity comparisons inthe macro literature. This does not imply poor countries have comparative advantage in manufacturing, as in a pure Ricardianframework, since labour market distortions lower farm wages. Relative productivity to wages in agriculture is higher in poorcountries than rich. This is consistent with a Heckscher-Ohlin interpretation: poor countries are abundant in unskilled labour(or land) used intensively in farming.

16Brandt et al. [2008] and Brandt and Zhu [2010], for example, use household-level surveys to infer a 26% agriculturallabour share in 2007 rather than the official figure of 41%, when considering hours spent on farm work. Moreover, Gollinand Rogerson [2010] report that even in extremely poor rural areas of Uganda, over 40% of households are active in non-agricultural activities, mainly wholesale and retail trade and manufacturing.

17I recognize the trade data is not perfect. For example, shipments valued at less than $100,000, which represent approx-imately 1% of the transactions, were manually incorporated into the UN trade data by Feenstra et al. [2005]. To the extentthis adjustment is not exhaustive, I will not be able to capture small scale trade that is likely more important for trade betweendeveloping countries.

18Timmer [1988, 2002] provide an effective summary. Matsuyama [1992] and, more recently, Lucas [2009] highlightdynamic gains from labour reallocation, with learning-by-doing in manufacturing. This chapter focuses on static gains tostructural change.


and Waugh, 2010], barriers to labour or intermediate inputs [Restuccia et al., 2008], or just over-counting

farm workers in the data [Gollin et al., 2004]. Trade, however, substitutes for the lowest productivity

farms to meet subsistence food requirements, increasing average sectoral productivity as they shut-

down. My study is not the first to link trade to structural change. Stokey [2001], for example, finds food

imports account for the United Kingdom’s reduction in agricultural output between 1780 and 1850, and

for much of the increased manufacturing. More recently, Teignier [2010] demonstrates a similar pattern

for South Korea since the 1960s, although agricultural subsidies and tariff protection limited realloca-

tion and subsequent productivity growth. Rather than investigate time series growth patterns as in these

papers, I quantify to what extent the lack of food imports can account for the current cross-sectional

level differences.

The model’s dual-economy features closely follow recent structural change research, which ex-

amines the strong negative correlation between agriculture’s share of output and employment and the

overall level of economic activity. In particular, I model non-homothetic preferences as in Kongsamut

et al. [2001]. Consumers must satisfy a minimum food intake requirement before allocating income

across goods according to their preference weights.19 In addition, to capture large wage differences be-

tween agricultural and nonagricultural activities in poor countries observed, I incorporate labour market

frictions. This further increases agriculture’s share of employment in poor countries. Labour market

frictions are used by Caselli and Coleman [2001] in a dynamic model to explain the development ex-

perience of the Southern US and by Restuccia et al. [2008] to capture cross-country patterns. Given

the static nature of this chapter’s model, I adopt Restuccia et al. [2008]’s approach. Specifically, farm

workers face a cost to switch into nonagricultural activities, proportional to non-farm wages.

The model’s trade components follow a large literature based on Eaton and Kortum [2002].20 This

framework is particularly well suited to estimate productivity from trade flows. Of particular relevance

to this chapter, Costinot et al. [2010] and Levchenko and Zhang [2010] infer productivity and compara-

tive advantage using a similar framework, but only for manufacturing. Waugh [2010] studies trade flows

19This contrasts with Ngai and Pissarides [2007]’s approach, where differential productivity growth across sectors, coupledwith an elasticity of substitution different from one, generates structural change. Other approaches involve increasing consumergoods variety [Greenwood and Uysal, 2005, Foellmi and Zweilmueller, 2006] or capital deepening with sector-specific factorintensities [Acemoglu and Guerrieri, 2006]. Incorporating non-homothetic preferences is the most natural approach for thischapter.

20For recent studies utilizing a similar framework, see Bernard et al. [2003], Alvarez and Lucas [2007], Caliendo andParro [2009], Kerr [2009], Burstein and Vogel [2010], Chor [2010], Costinot et al. [2010], Donaldson [2010], Fieler [2010],Levchenko and Zhang [2010], Waugh [2010], Yi and Zhang [2010].


and the impact of trade on cross country income differentials, but - again - only for manufacturing.21

My model is distinct in two important ways. First, to capture declining food expenditure shares, con-

sumer preferences are non-homothetic. Fieler [2010] also employs non-homothetic preferences within

an Eaton-Kortum framework but my approach differs by linking low income elasticity to the good with

a high degree of international productivity variation - namely, agriculture. Fieler [2010] considers the

opposite case, which may be more relevant between manufactured goods than between agriculture and

non-agriculture broadly.22 The second distinct feature in my model, labour mobility costs, captures

large farm-nonfarm wage differences in poor countries. The model’s remaining features are standard:

perfectly competitive markets, trade arises through sectoral and international technology differences,

and labour as the only productive input.

1.2 A Model Consistent with Stylized Facts

This section presents the general equilibrium trade model that builds on Yi and Zhang [2010]. Overall,

the environment is composed of N countries each with three sectors: agriculture, manufacturing, and

services. I incorporate standard dual-economy features found in the macroeconomics literature: (1) non-

homothetic preferences and (2) labour market distortions between agriculture and nonagriculture. This

preference structure captures Engel’s law: food expenditure shares decline dramatically with income.

The labour market distortions capture large sectoral wage differences and high labour mobility costs in

poor countries. To incorporate open-economy considerations, agriculture and manufacturing are com-

posite goods composed of individually tradable and horizontally differentiated varieties. Each variety

is sourced from the lowest cost producer, whether at home or abroad, which introduces within sector

trade (exporting a subset of varieties to import others). Between sector trade is also available, where

a surplus of exports over imports in agriculture, for example, allows for a net import of manufactured

goods. Importantly, this structure does not imply perfect specialization and it links trade patterns to sec-

toral productivity and trade costs. I conclude this section by defining an equilibrium and by describing

21My findings are robust to alternative specifications of the bilateral trade-cost function. I reproduce my results using anexporter (as opposed to importer) specific trade costs specification. See, for example, Waugh [2010] on the role of export costswithin this class of models. I find evidence the type of trade cost asymmetry found by Waugh [2010] for manufactured goodstrade is also a feature of the agricultural goods trade.

22Additionally, she uses modeling features to generate variable budget shares in a fundamentally different manner from theStone-Geary preferences I use in this chapter.


a solution procedure for wages and labour allocations.

1.2.1 Households’ Problem

I index countries by i = 1, ...,N, and each is populated by Li agents endowed with an inelastically

supplied unit of labour, allocated between the three sectors. Within each country, households spread

consumption evenly across individual agents. I model non-homothetic preferences as subsistence food

requirements within a Stone-Geary type utility function. Households select consumption and labour

allocations to maximize

maxCik,Likk∈a,m,s

U(Cia,Cim,Cis) = εaln(Cia− a)+ εmln(Cim)+ εsln(Cis) (1.1)

s.t. ∑k∈a,m,s

LiPikCik = ∑k∈a,m,s

wikLik (1.2)

Preference weights εa,εm,εs determine the fraction of disposable income allocated to each type of

good. Consumer demands are standard: Cia = a+ εaMiP−1ia , Cim = εmMiP−1

im , Cis = εsMiP−1is , where

Mi is consumer income after subsistence spending,(

∑k∈a,m,swikLikLi

)− Piaa. As food subsistence

requirements increase - through a higher a - food’s share of total expenditures increases.

1.2.2 Production Technology

I model N-by-N bilateral trade flows with two differentiated tradable goods, agriculture and manufactur-

ing, similar to Eaton and Kortum [2002]. Goods, denoted k ∈ a,m, are composed of a continuum of

differentiated varieties. Firms produce individual product varieties, denoted z, with linear technology23

yik(z) = Aik(z)Lik(z).

Markets are perfectly competitive, which implies the producer price will equal marginal costs, wikAik(z)

.

More specifically, each variety is a contestable market with zero barriers to firm entry or exit; so, any

price deviation from marginal costs will result in a new entrant supplanting the incumbent firm. Pro-

ductive technologies for each firm/variety are independent random draws from a Frechet distribution

23Incorporating intermediate inputs into this production function is another standard formulation, which increases tradegains as input prices decline. To be conservative, I abstract from this consideration in the baseline results but report the casewith intermediates in the appendix.


specific to each country-i and sector-k, such that

Pr(Aik(z)≤ x) = Fik(x) = e−(x/Aik)−θk

,

where θk governs productivity dispersion and Aik the overall level of productivity, with Aik ∝ E [Aik(z)].24

Lower θ implies greater variability in productivity across firms and countries and higher A implies

greater average productivity. These productivity differences across producers provides the incentive to

trade: low productivity domestic producers may be shut down in favour of an import. For lower θ , the

incentive to trade, and the gain from doing so, increases.

Let yik(z) be the quantity of variety-z in country-i for good-k, either imported or produced domes-

tically. A domestic firm aggregates these into composite goods through a CES technology with an

elasticity of substitution of ρ ,

Yik =

[ˆ 1

0y1−1/ρ

ik(z) dz]ρ/(ρ−1)

.

Finally, nontradable services are produced with a similar linear production technology, Yis = AisLis.

1.2.3 International Prices and Trade Patterns

Firms producing the composite manufactured and agricultural good purchase individual varieties from

the lowest cost source - at home or abroad. As in Samuelson [1954], trade costs are iceberg: τi jk

sector-k goods are shipped per unit imported by country-i from country- j. To avoid shipments through

third-party countries, the triangle inequality holds: τi j < τihτh j, for any country h. Consequently, the

price of variety-z in country-i for good-k is the lowest price charged by producers, w jkA jk(z)

, adjusted for

transport costs, τi jk:

pik(z) = minj∈1,..,N

[τi jkw jk

A jk(z)

]. (1.3)

Substitute into a CES index to determine each country’s price for tradable good-k, Pik =[´ 1

0 pik(z)1−ρdz]1/(1−ρ)

.

Given the distribution of productivity across varieties, and assuming import costs and wages are not

24The constant of proportionality is Γ

[1− 1

θk

]−1. This relates to the scale parameter of a Frechet distribution. λik = Aθk

ik .


variety-specific, Eaton and Kortum [2002] demonstrate this index reduces simply to

Pik = γ

[N

∑j=1

(τi jkw jk

A jk

)−θk]−1/θk

, (1.4)

where γ = Γ

(1+ 1−ρ

θk

) 11−ρ

.25 Notice, Equation 1.4 is the price paid by consumers in country-i for the

aggregate good-k and no knowledge of individual variety sources is necessary.26

The share of country-i expenditures sourced from country- j capture trade patterns in the model.

This share, in turn, depends on the fraction of varieties produced in j that have the lowest price of all

producers in any other country, from the perspective of country-i consumers. As in Eaton and Kortum

[2002] the share of country-i spending sourced from country- j for good-k is

πi jk =ψi jk

∑Nj=1 ψi jk

, (1.5)

with ψi jk = τ−θki jk

(A jk/w jk

)θk as the product of trade costs and competitiveness of country- j from the

perspective of country-i consumers. A jk/w jk is a country’s competitiveness, which rises with techno-

logical productivity A jk and falls with labour costs w jk.

1.2.4 Labour Market and Trade Balance Conditions

Trade shares combine with household demand to determine each country’s total sales. Country- j spends

π jik fraction of its total consumer demand on output of good-k from country-i, which implies total

foreign demand is ∑ j 6=i L jPikCikπ jik. Since country-i spends πiik fraction on its own output, total demand

for country-i output of good-k is then ∑Nj=1 L jPikCikπ jik. With labour as the only productive input, total

sectoral revenue from all sources - foreign and domestic - equals labour income by sector for each

country:

wiaLia = PiaYia =N

∑j=1

[L j(Pjaa+ εaM j)π jia

], (1.6)

251+θk > ρ must hold, I set ρ such that γ = 1, which does not violate this restriction.26In order to perform proper PPP-adjustments to labour productivity across countries, individual variety price and quantity

information is required. To that end, I will numerically simulate the full model on 50,000 product types. I report details laterin the chapter. Unadjusted labour productivity from the model’s analytic solution is a good approximation to the PPP-adjustedvalues.


wimLim = PimYim =N

∑j=1

[L jεmM jπ jim

], (1.7)

wisLis = PisYis = εsMiLi. (1.8)

Labour demand by producers of each tradable variety, Lik(z), aggregate to sectoral labour Lik =[´ 10 Lik(z)dz

]. Also, sectoral labour allocations must total to national employment,

∑k∈a,m,s

Lik = Li ∀ i = 1, ..,N. (1.9)

I capture labour market distortions with a reduced-form wedge between sectoral wages.27 Specifically,

wia = ξiwi and wim = wis = wi, where ξi < 1 captures labour’s cost to move off the farm.

The sectoral revenue and labour earnings conditions of the previous section imply international

trade balances for each country. Specifically, combine Equations 1.6 to 1.8 with labour market clearing

Equation 1.9 and Mi +Piaa =(

∑k∈a,m,swikLikLi

)to yield

Li(Piaa+(εa + εm)Mi) =N

∑j=1

L j[(Pjaa+ εaM j)π jia + εmM jπ jim] ∀ i = 1, ..,N. (1.10)

Country-i appears on both the left and right side, so this equation is identical to imports equaling exports.

Alternatively, this condition states that total spending on tradable goods by country-i consumers will

equal total global spending on tradable goods produced by country-i firms.

1.2.5 Equilibrium Definition and Solving the Model

A competitive equilibrium in this framework is a set of prices Pia,Pim,PisNi=1, wages wiN

i=1, con-

sumption allocations Cia,Cim,CisNi=1 and labour allocations Lia,Lim,LisN

i=1 such that (1) given prices

and wages, households solve Equation 1.1; (2) given wages, price aggregates are consistent with Equa-

tion 1.4; (3) given wages, prices, and labour allocations, international trade balances through Equation

1.10; and (4) labour markets clear through Equation 1.9.

Given exogenous parameters (technology, Aik; trade costs, τi jk; preference weights, εk; subsistence

requirements, a; labour mobility costs, ξi; total employment, Li; and trade elasticities, θk), one can solve

27I abstract from how these differentials are supported in equilibrium. See Lagakos and Waugh [2010] for an excellenttreatment of the relationship between sectoral labour frictions and the food problem.


for wages and labour allocations as follows. First, combine prices from Equation 1.4 and trade shares

from Equation 1.5 with Equations 1.6 through 1.8. Wages and labour in agriculture, manufacturing, and

services is a set of 4N unknowns. Equations 1.6 through 1.8, with the labour market clearing Equation

1.9, is a set of 4N equations. Thus, equilibrium wages and labour allocations, given exogenous param-

eters, solves this system.2829 I solve counterfactual experiments in Section 1.5 using this procedure.

Technology and trade costs parameters, however, are not observable and I describe their calibration in

Section 1.3.

1.3 Calibrating the Model

To guide intuition through the calibration, I first describe the overall approach before moving into de-

tails. First, estimate competitiveness Aikwik

and trade costs τi jk from bilateral trade flows (details in Section

1.3.1). These estimates together imply prices from Equation 1.4 and trade shares from Equation 1.5.

Given prices and trade shares, determine international disposable income levels Mi to balance inter-

national trade from Equation 1.10. Given income and prices, consumer demands from the household

problem imply wages and labour allocations consistent with international demands and income levels

through Equations 1.6 and 1.8. The product of competitiveness Aikwik

and wages now implies sectoral tech-

nology parameters Aik. Importantly, I infer sectoral labour productivity from bilateral trade flows with

minimal structure. The Eaton-Kortum trade structure within agriculture and manufacturing generates

trade patterns independently of household preferences and trade balance conditions. Wage levels depend

on trade balance and household preferences but are independent of service-sector labour productivity.

Finally, I construct PPP-adjusted aggregate productivity in the model following similar procedures

as in the data. Given the three-sector structure of the model, PPP-adjusted GDP/Worker is total nom-

28Interestingly, wage levels and labour shares are independent of service-sector labour productivity. To see this, notethat if a = 0, ξi = 1, and there is only one tradable sector, the above system would collapse to wiLi =

ε

1−ε ∑Nj=1[L jw jπ ji

],

where ε is the tradable goods’ budget share. In this framework, the elasticity of substitution across goods is one (fromhousehold preferences) and, therefore, budget shares are constant. Thus, this system of equations determines wages acrosscountries, given technology and trade costs. These wage equations are similar to Equation 21 in Eaton and Kortum [2002],which corresponds to their special case of immobile labour. More general preferences, however, would imply εa,εm,εs arefunctions of an overall price index and, by extension, productivity in every sector, including services.

29Not allowing for trade imbalances will impact model wage estimates, since countries with large current account deficitswould have higher wages than in the balanced-trade case. In terms of productivity estimates, imposing trade balance willunderestimate productivity dispersion if rich countries typically have current account deficits. For poor countries, the impactwill be negligible, given their low import shares. Dekle et al. [2007] incorporated into an Eaton-Kortum framework and findmost wage estimates under imbalanced trade are within 10% of the balanced-trade case. They find imposing trade balanceresults in US wages of about 10% higher than with a current account deficit.


inal consumer expenditures deflated by a country-specific Geary-Khamis price index. This procedure

follows the World Bank’s International Comparisons Program and represents how Penn-World Table

measures of GDP/Worker comparable across countries would be constructed in a world with only three

goods [Heston et al., 2009]. To begin, find international prices of each good-k and purchasing power

parities for each country-i that solve the following system:

IPk =N

∑i=1

Pik

PPPiγik,

PPPi =∑k∈a,m,sLiPikCik

∑k∈a,m,s IPkCik.

PPPi is country-i’s purchasing power parity exchange rate and γik = LiCik

∑Nj=1 L jC jk

is a quantity (of total

consumption) weight for country-i and good-k. The common set of international prices to value con-

sumption is essentially a weighted-average of goods prices across countries. The model’s PPP-adjusted

GDP/Worker for country-i is then Yi/Li = PPP−1i ∑k∈a,m,sPikCik.

For the quantitative exercises, I use a set of 114 countries, listed in Table 1.12. A number of pa-

rameters can be set to generally accepted values in the literature; namely, the preference parameters and

the Frechet distribution’s dispersion parameter. In order: εa = 0.01, εm = 0.24, and εs = 0.75; and,

θa = θm = 7.30 Total employment is inferred from PWT6.3 as the ratio of total GDP to GDP/Worker. I

list the model parameters, their values, and calibration targets in Table 1.1. The following sub-sections

describe parameterizing productivity, trade costs, subsistence level of food consumption, and, finally,

labour market distortions. Given these, all other variables are endogenously determined. I proceed in

stages: (1) estimate competitiveness and trade costs to fit bilateral trade, independently of the structure

of the household sector supporting such flows in equilibrium; (2) select subsistence parameter to match

30Regressing ln(

πi jkπiik

)(see Section 1.3.1) on a measure of trade costs, τi j, from the CEPII trade database, along with

importer and exporter fixed effects, yields θ = 5.5 in agriculture and θ = 6.8 in manufacturing. For colonial India, Donaldson[2010] finds θ = 3.8 with the 17 agriculture varieties for which he has data, but θ = 5.2 with the entire sample of 85 com-modities. Lower theta in agriculture enhances my results by increasing the scope for comparative advantage within agriculture.To be conservative, I set θa = θm. For other estimates, Alvarez and Lucas [2007] set θ = 6.67, Eaton and Kortum [2002] setθ = 8.3, and Anderson and van Wincoop [2004] reviews the literature and finds anything between 5 and 10 reasonable. Finally,Waugh [2010] finds θ = 7.9 for OECD countries and θ = 5.5 for non-OECD countries, which is identical to my estimate foragricultural trade. All development accounting and trade flow counterfactual exercises are robust to alternative values (seeSection 1.6.1).


Table 1.1: Calibration of Model Parameters

Parameters Target Valueθa,θm Cost-Elasticity of Trade Flows 7,7εs,εm,εa Long-Run US Employment Shares 0.75,0.24,0.01

Ais Aggregate GDP/Worker Data Country-Specificξi Relative Wage Data Country-SpecificLi Total Employment Data Country-Specific

Aia,Aim,τi ja,τi jm

Bilateral Trade Data Country-Specifica US Sectoral Employment Data 0.0160

This table provides a list of model parameters that must be calibrated. All other variables in themodel are endogenously determined. The parameters in the bottom two rows are dealt with in detail asStage 1 and Stage 2, all other parameters either map to observable data or are generally accepted values.Long-run employment shares reflect the values to which US employment data appear to be converging. Ireport sensitivity of the model to various alternative values of θa,θm in Section 1.6.1.

US data; and (3) set service-sector labour productivity so the model’s aggregate labour productivity

matches data.

1.3.1 Productivity and Trade Costs

The empirical strategy relates variation in bilateral import and export flows, relative to each country’s

domestic purchases, to infer import barriers, export competitiveness, and bilateral trade costs. The share

of country-i expenditure imported from country- j, from Equation 1.5 can be expressed as

πi jk = Pθik

(A jk

τi jkw jk

)θ

= Pθik

(Tjk

τi jk

)θ

,

where Tjk = A jk/w jk is a country’s competitiveness, which rises with technological productivity A jk

and falls with labour costs w jk. Domestic spending shares are similar: πiik = Pθik (Aik/wik)

θ = PθikT θ

ik .

The ratio of πi jk to πiik is a normalized import share that depends only on competitiveness measures

(productivity per unit-input cost) and trade costs:

ln(

πi jkπiik

)= θ ln

(Tjk)−θ ln(Tik)︸︷︷︸ − θ ln

(τi jk)︸︷︷︸

Competitiveness Trade Costs

To estimate this expression, proxy trade costs with various bilateral characteristics and an importer-

specific trade barrier, Bik. The bilateral costs include distance between capitals and indicators for shared


border, common (ethnographic) language, and trade agreement status.3132 Importer-specific trade bar-

riers is a reduced-form approach to capture all import costs such as tariffs, non-tariff barriers, health

regulations, low quality local infrastructure, information asymmetries, among many others, in a single

number. Importantly, trade costs in this setup are asymmetric between countries: it is more expensive

to import goods from Canada into Cuba than from Cuba into Canada. Alternative frameworks, such

as in Anderson and van Wincoop [2003], employ symmetric trade costs between pairs.33 The precise

empirical specification I use, separately for each sector, is:

ln(

πi jk

πiik

)= β1kln

(Distancei jk

)+β2kBorderi jk +β3kLanguagei jk (1.11)

+β4kAgreementi jk +η jk +δik +νi jk,

where η jk is the exporter fixed-effect, δik the importer fixed-effect, and νi jk the random component. The

model parameter estimates are derived from coefficient estimates as: Tik = eηik/θ , Bik = e−(δik+ηik)/θ ,

and Pik = γ

[∑

Nj=1

(τ−1i jk Tjk

)θ]−1/θ

from Equation 1.4.

To fit trade shares πi jk to data, I construct trade share measures similar to Eaton and Kortum [2001],

Bernard et al. [2003]. Specifically, I take the ratio of country-i imports from country- j, reported in the

NBER-UN trade database, relative to country-i’s output less net exports

πi jk =Importi jk

SectoralOut putik−Exportsik + Importsik

I infer sectoral output from World Bank GDP shares.34 Bilateral trade data for 2000 is from the NBER-

UN Trade Database, which disaggregates by 4-digit SITC code.35 Agricultural trade flows are all bilat-

eral flows classified with an SITC 1-digit code of 0, such as 0573 (Bananas, Fresh or Dried). Finally,

31Data on pairwise characteristics and Capital coordinates are from CEPII.http://www.cepii.fr/anglaisgraph/bdd/distances.htm. Distance between importer-i and exporter-e:6378.7arccos(sin(late)sin(lati)+ cos(late)cos(lati)cos(longi− longe))

32I find the trade agreement variable particularly important for European bilateral pairs. Without this control, productivityinferences for these countries, given their high levels of trade, are extremely large. Data is from Fieler [2010].

33Asymmetric costs are not excluded from their framework, but it only identifies the average of any country-specific costsbetween members of a pair.

34Gross output measures are ideal but I lack internationally comparable measures. The FAO reports gross and net produc-tion values (PPP-adjusted, while trade flows are exchange-rate adjusted) and I find a gross-to-net ratio of approximately 5%among developing countries, compared to 15% for the rich. Net output inferred from GDP shares underestimates home-bias inpoor countries, so this approach is conservative. Consistent with my treatment in the manufacturing sector, I use the inferrednet output measure for agriculture as well.

35See Feenstra et al. [2005] for details regarding the construction of this data.


Table 1.2: Main Estimation Results

Agriculture Manufacturing(1) (2)

Ln(Distance) -1.037 -1.389(0.029)∗∗∗ (0.023)∗∗∗

Shared Border 0.574 0.472(0.105)∗∗∗ (0.098)∗∗∗

Shared Language 0.664 0.596(0.061)∗∗∗ (0.052)∗∗∗

Trade Agreement 0.354 -.323(0.124)∗∗∗ (0.122)∗∗∗

Exporter FEs Yes YesImporter FEs Yes YesObservations 6207 9014R2 0.971 0.971

The OLS esimates of Equation 1.11. The dependent variable is the normalized import share,for importer-exporter pairs from the NBER-UN trade database, for each traded sector. Data ondistance, borders, and language from CEPII; trade agreement indicator from Fieler (2010).

countries do not trade with every other country, leaving zeros in the data for those pairs. For my baseline

estimates, I estimate the above specification only on the pairs with positive trade with OLS.36

The basic gravity-specification implied by the theory captures the trade data well. The parameter

estimates are listed in Table 1.2, with 6,207 observed trade pairs in agriculture and 9,014 in manufac-

turing. To visualize the goodness of fit, I sum πi jkπiik

within countries for each sector, which represents the

relative importance of goods sourced from abroad relative to domestic purchases. The actual and fitted

values (summed in similar fashion) are found in Figure 1.3 and match extremely well.

My estimates of sectoral competitiveness, Tik, imply rich countries have a comparative advantage

in manufacturing and an absolute advantage in both sectors while poor countries have a comparative

advantage in agriculture. I report means for the richest and poorest countries in Table 1.3. On average,

rich country competitiveness in manufacturing is 2.4 times poor country competitiveness but only 1.6

times more in agriculture. This does not contradict earlier observations of larger labour productivity

differences in agriculture than manufacturing since farm wages are significantly lower than nonfarm

36As a robustness check, to handle this left-censoring of the data, I estimate a Tobit model with the minimum observedπi j for each country-i serving as the lower limit, below which statistical agencies do not observe the trade. This procedure issimilar, but not identical to, [Eaton and Kortum, 2001]. See Anderson and van Wincoop [2003] for more on estimating gravitymodels. The censoring threshold for πi j is selected as the maximum likelihood estimate πi = min j∈[1,..,N] πi j, such that if thetrue πi j < πi then I will not observe a trade flow in the data for that i, j pair. The geographic component of trade becomesmuch more important within the Tobit structure. I explore this alternative specification in the appendix.


Figure 1.3: Fit of the Stage-1 Calibrated Model

(a) Agricultural Trade

−7 −6 −5 −4 −3 −2 −1 0 1 2 3−7

−6

−5

−4

−3

−2

−1

0

1

2

3

Actual Log(Share Purchased Abroad / Domestic Share)

Mod

el L

og(S

hare

Pur

chas

ed A

broa

d / D

omes

tic S

hare

)(b) Manufacturing Trade

−4 −3 −2 −1 0 1 2 3−4

−3

−2

−1

0

1

2

3

Actual Log(Share Purchased Abroad / Domestic Share)

Mod

el L

og(S

hare

Pur

chas

ed A

broa

d / D

omes

tic S

hare

)

Displays the fit of the Stage-1 calibrated trade flows in the model to the data for each traded goods sector. The normalized trade flowmeasure is the share of consumer expenditures imported from abroad relative to the share sourced domestically. The vertical axis is the modelnormalized import rate and the horizontal axis is calculated from the NBER-UN trade database.

Table 1.3: Selected Values from Stage-1 Calibration

Mean for Competitiveness, ˆ(A jk/w jk

)Trade Cost, Importer Fixed-Effect, Bik

Countries in: Agriculture Manufacturing Agriculture Manufacturing

Top-10% 1.44 2.65 -41% -68%Bottom-10% 0.89 1.09 43% -1%

Competitiveness and importer-specific fixed effects implied by the bilateral pattern of sectoral trade. These results suggest rich countrieshave a comparative advantage in manufactured goods and an absolute advantage in both. Poor countries also face higher costs to import goods.Negative values for Bik imply imported goods cost less than in the average country.

wages in poor countries.37 For all countries, I plot the competitiveness estimates Tik in Figures 1.13 and

1.14.

When the model is solved, Frechet productivity parameters are Aik = Tikwik. Finally, I plot the

trade costs captured by the importer-specific fixed effect Bik in Figures 1.11 and 1.12. Note that these

fixed-effects can be below zero, which implies imports cost less than in the average country.38

37The price of agricultural goods relative to manufacturing, from ICP, is also consistent with this finding. The ratio Pa/Pmis 0.8 in the bottom quintile of countries and nearly 1 for the top. Overall, the relative price of food within the set of tradablegoods is U-shaped across countries, with middle income countries having the lowest Pa/Pm ratio.

38This interpretation was provided by Waugh [2010], who finds a -62% fixed-effect for the United States’ manufacturinggoods trade. My results suggest -76% for manufacturing and -55% in agriculture.


1.3.2 Subsistence, Service Sector Productivity, and Labour Market Distortions

An important driver of agriculture’s high employment and spending share in lower income countries

is the need to fulfill minimum food intake requirements. To capture this channel, without selecting

subsistence to target potentially suspect employment data in poor countries, I set a to match US data.

Specifically, Equations 1.6, 1.7, and 1.10 imply labour income in the tradable sectors must equal con-

sumer expenditures on tradables: PaCa +PmCm = ξ w(La

L

)+w

(LmL

). Normalize w = 1 for the US and

combine with household demands and labour market clearing conditions to yield

a = P−1a

[ξ la + lm−

(εa + εm

εs

)ls

].

To evaluate this expression, I use: prices from the previous section, which depend only on the trade cost

and competitiveness estimates; preference weightsεs,εm,εa; labour shares ls, lm, la= 0.743,0.232,0.026,

from the World Bank’s 2000 WDI; and US labour mobility costs ξUS = 0.8972, estimated from wage

data. I get a = 0.0160 and subsistence spending, Paa, of less than 1% aggregate GDP/Worker for the

United States - approximately $600 per year. For comparison, Restuccia et al. [2008] find a subsistence

value of approximately 2.2% of the US aggregate GDP/Worker in 1985.

Two parameters remain. I set service sector labour productivity, Ais, so the model implied PPP-

adjusted aggregate labour productivity matches data. Finally, each country’s labour market distortion,

ξi, can be matched to sectoral wage data from the International Labour Organization.39 Wage data is

unavailable for many countries. I use the observed relationship between relative agricultural wages and

GDP/Worker to fit ξi for each of the 114 countries. Details on this procedure are in the Data Appendix.

1.4 Results from the Baseline Calibration

I display the model-implied agricultural employment share in Figure 1.4a. For the richest and poorest

countries, I present each sector’s employment share and the relative GDP/Worker in Table 1.4. The

model corresponds well to sectoral data, despite explicitly targeting only aggregate productivity differ-

ences. If cross-country employment data are accurate, targeting US data accounts for approximately

half of the international variation in agriculture’s employment share. For many developing countries,

39http://laborsta.ilo.org/


Figure 1.4: The Role of Subsistence Food Requirements

(a) With Subsistence Food Requirement Set to Match USData

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Actual, World Bank WDI in 2000

Mod

el

(b) With Subsistence Food Requirements Set to Zero

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Actual, World Bank WDI in 2000

Mod

elDisplays the model implied share of employment in agriculture along the vertical axis relative to the share reported for 2000 in the

World Bank’s WDI database. The left panel is the model with subsistence set to match US data. The right panel is the model with subsistenceset to zero. Targeting US data allows the model to capture much of the cross country variation. All else equal, the sum of squared deviationsbetween the model and data is 8.18 with subsistence set to match US data and 17.1 with subsistence set to zero.

however, standard surveys overestimate farm labour since rural residents and farm workers are treated

synonymously. Brandt et al. [2008] and Brandt and Zhu [2010], for example, use household-level sur-

veys to infer a 26% agricultural labour share in 2007 rather than the official figure of 41%, when con-

sidering hours spent on farm work. Moreover, Gollin and Rogerson [2010] report that even in extremely

poor rural areas of Uganda, over 40% of households are active in non-agricultural activities, mainly

wholesale and retail trade and manufacturing. For these reasons, I proceed using the model-implied

employment shares.

1.4.1 Trade Cost Estimates

I decompose trade costs into bilateral components (distance between trading partners and separate in-

dicators for shared border, language, and trade agreement), importer-specific fixed effects, and an id-

iosyncratic component that is IID across products. These fixed effects capture many possible barriers:

tariffs, quotas, health regulations, or poor local road networks. Adjusting this variable downwards to the

typical level observed in rich countries will be a key counterfactual experiment I perform to determine

the impact of low food imports on international productivity and income differences.

Figures 1.11 and 1.12 display the trade cost estimates for both sectors. I separately plot the overall


Table 1.4: Aggregate Productivity and Employment Shares, Model vs. Data

Mean of Bottom-10% Mean of Top-10%

Data Model Data Model

GDP/Worker, Relative to Top * 0.04 0.04 1 1Agricultural Employment Share 0.67 0.45 0.04 0.02

Manufacturing Employment Share 0.08 0.13 0.29 0.23Services Employment Share 0.25 0.42 0.67 0.75

Agricultural GDP Share 0.36 0.27 0.02 0.01Manufacturing GDP Share 0.22 0.18 0.31 0.24

Services GDP Share 0.42 0.55 0.67 0.75

∗ denotes target. This table compares the model to data for various statistics.

measure (trade weighted across import sources) and the importer-specific fixed effect. In the figures, I

represent trade costs by their impact on prices: a trade cost of τ will increase prices by 100(eτ/θ − 1)

percent. The average import costs for the 114 countries in my sample is 263% in both agriculture

and manufacturing. For rich countries, this number is 104% on average in agriculture and 78% in

manufacturing. The importer-specific fixed component of trade costs average 43% for poor countries

and -41% for rich countries in agriculture. This implies that imports are typically 43% more expensive

in the poorest 10% of countries relative to the average country.40

1.4.2 Sectoral Labour Productivity

As noted by Costinot et al. [2010], Waugh [2010], Yi and Zhang [2010], A jk is the labour productivity

in autarky. With trade, labour productivity in sector-k is given by the conditional mean of operating

producer productivity, yik =wikPik

= E [Aik(z) | z ∈ yik(z)> 0], can be expressed as

yik = π−1/θkiik︸︷︷︸ · Aik︸︷︷︸

Trade Technology. (1.12)

In autarky, all producers operate and labour productivity is Aik. Imports, which lower πiik, leads average

productivity to grow as inefficient producers shut-down.41 Absent a very low πiik, measured labour

productivity will closely reflect the underlying technology parameter, Aik. This corresponds to recent

40See Section 1.6.3 for a discussion on the plausibility of these trade cost estimates.41See Costinot et al. [2010] for a detailed discussion of the extent to which selection impacts observed productivity.


Table 1.5: Baseline Model: Cross Country Productivity Differentials

Top-10% / Bottom-10% Variance of Logs

Sector Data Model Data Model

Aggregate 23.0 23.0 1.00 1.00Agriculture 79.5 103.9 1.73 2.14

Manufacturing - 77.4 - 1.99Services - 9.9 - 0.75

Nonagriculture 5.8 13.8 0.78 0.47

Presents the baseline estimates of sectoral productivity implied by bilateral trade flows andmodel wages. Services sector productivity is calibrated so the model aggregate productivity matchesthe dispersion of GDP/Worker in the PWT6.3. Agricultural labour productivity in the data is PPP-adjusted value-added per worker using net farm output data from FAO, valued at internationalprices, for 1999-2001. The precise procedure follows Restuccia, Yang, and Zhu (2008) and Caselli(2005).

literature finding low gains from trade in new-trade models (see, for example, Arkolakis et al. [2009]).

The counterfactual gains I find in Section 1.5 follow from poor-country πiik declining significantly. The

import-elasticity of labour productivity within a tradable sector is, from the above expression, 1/θk.

The value for θk influences the liberalization experiments I will perform in the next section. I will report

sensitivity of my results to various values for θk and I find the overall conclusions robust.

I plot relative productivity between agriculture and manufacturing in Figure 1.9; relative agricultural

labour productivity increases with a country’s level of development. In the figure, I separately show the

pure technology ratio AaiAmi

and the full observed productivity ratio, given trade selection.

Unfortunately, the model is unable to analytically produce PPP-adjusted estimates of sectoral labour

productivity since individual variety producer-prices and production quantities are unknown. To pro-

vide proper comparisons with data, I simulate the model on a set of 25,000 products for each sector

and country. Within the simulation, I track individual producer prices and quantities to construct PPP

adjustment-factors following the World Bank procedures. Overall, productivity estimates from Equation

1.12 match appropriate PPP-adjusted estimates very well. In fact, the agricultural and manufacturing

productivity gaps in the simulation are 102.4 and 73.2, respectively.42 Overall, the correlation between

the two measures are 0.997 in agriculture and 0.987 in manufacturing. Going forward, I report sec-

toral results from Equation 1.12 and leave further discussion of sectoral-level PPP-adjustments to the

appendix.

42Standard errors from 20 iterations are 0.8 and 7.


Table 1.5 displays the aggregate and sectoral productivity dispersion implied by the baseline model,

which match the data well for agriculture. I plot a complete comparison for all countries in Figure 1.10.

Nonagricultural productivity variation, however, is larger than data suggests, reconciled by the model’s

lower agricultural employment share. Similar comparisons within the manufacturing and service sectors

across a broad range of countries are difficult for lack of producer price data. Broadly speaking, however,

lower service-sector productivity variation than in tradable goods sectors is consistent with Herrendorf

and Valentinyi [2010].43

While direct international comparison is problematic, I examine a subset of countries for which real

labour productivity (per hour) exists in the GGDC Productivity Level Database [Inklaar and Timmer,

2008]. I find the variance of log manufacturing productivity is 1.42 in the model for these countries

and 0.19 in services, while GGDC figures are 0.53 and 0.16.44 Moreover, the ratio of the 75th to 25th

percentile is 7 in the model and 3 in the GGDC data. An alternative trade cost specification in Section

1.6.4 provides less variation in productivity estimates. The counterfactual results in the following section

are robust - indeed, strengthened - by this alternative specification, so I continue with the baseline model.

1.5 Counterfactual Experiments: Trade, Productivity, and Income

To account for the sources of productivity gaps between rich and poor countries, I perform a set of

counterfactual experiments within the model. Specifically, I investigate: (1) lowering importer fixed-

effects, Bik, to the average level of the richest-10% of countries for both sectors; (2) allowing full labour

mobility by setting ξi = 1 for all i; and (3), to capture interactions between the domestic and foreign

distortions, both (1) and (2) together. Following each of these experiments, poor countries increases

their level of food imports dramatically (see Figures 1.5, 1.6, and 1.7). Imports allow the lowest pro-

ductivity domestic producers to shut down and tradable-sector productivity increases, especially in poor

countries. I interpret the portion of the rich-poor gap that these counterfactual experiments eliminates

as the contribution of the two distortions. I present details in the following sections.

43For an interesting illustration of the difficulty of making direct cross-country productivity comparisons, especially withinthe service-sector, see Baily and Solow [2001]

44I use the GGDC figures corresponding to manufacturing less electrical equipment and services less postal and telecom-munications. Electrical equipment, postal, and telecommunications are aggregated into a single, separate category.


Table 1.6: Trade Between 1st and 4th Quartiles

(a) Baseline Model

Import Share from: Import SharePoor Rich of Spending

Poor 9% 47% 2%Rich 2% 74% 38%

(b) Counterfactual: Lower Importer Fixed-Effects

Import Share from: Import SharePoor Rich of Spending

Poor 25% 23% 67%Rich 10% 47% 58%

Displays the fraction of total imports by source-country income levels. Poor are the bottom quantile of countries in terms of GDP/Capitaand rich are the top. Large shares imported from Rich countries does not imply that rich countries export more food to poor countries thanvice-versa (in fact, the reverse is true). Prior to liberalization, poor countries bought very little from each other. Following liberalization ofimport and labour markets, food trade between poor countries rises dramatically. The fraction of varieties domestically produced also falls toone-third its original value. This Ricardian-selection is the source of the increased sectoral productivity. I show the shares across all incomepercentiles for both poor and rich importers in Figures 1.15 and 1.16.

1.5.1 International Food Trade Flows

I present the bilateral trade patterns for the trade liberalization experiment in Table 1.6. I show the

shares across all income percentiles for both poor and rich importers in Figures 1.15 and 1.16. Until

subsistence food requirements are met, poor country consumers allocate significant resources to agri-

culture since trade barriers inhibit their ability, and internal labour markets reduce their incentive, to

import food. Following liberalization, the fraction of varieties produced domestically falls below that

of rich countries. The fraction of food imports sourced from other poor countries more than triples and

the fraction from rich countries falls in half. Middle-income countries (see Figures 1.15 and 1.16) also

become an important source for poor-country food imports. I find some developing countries increase

their resource commitment to agriculture while others move labour into non-agricultural activities. In

essence, poor countries more efficiently allocate their food production among themselves. The counter-

factual volume of South-South trade grows by an order of magnitude to account for nearly one-fifth of

global agricultural trade, and together with North-South trade accounts for slightly more than half (see

Figures 1.17 and 1.18). These counterfactual trade patterns drive important changes in productivity and

income differences between rich and poor countries.

Importantly, despite low relative labour productivity, the counterfactual trade flows following re-

duction of import barriers confirm an earlier finding: poor countries have a comparative advantage in

agricultural goods. The fraction of total exports accounted for by agricultural goods rises in poor coun-

tries by more than in rich. I report each country’s percentage point change in agriculture’s share of total


Table 1.7: Results of Main Counterfactual Experiments

Liberalization Experiments

Liberalized MobileBaseline Trade Labour Both

Labour Productivity Top-10%/Bottom-10%

Aggregate 23.0 17.7 21.5 15.3Agriculture 103.2 88.5 98.3 57.5Manufacturing 77.4 47.0 78.4 52.8

Agriculture in the Poorest-10%

Employment Share 0.45 0.37 0.33 0.03Change in Labour Productivity - 17% 6% 60%Import Share of Expenditures 0.03 0.67 0.24 0.96

Displays the rich-poor productivity gaps and various statistics for the poorest-10% of countries. Specifically, the share of employmentin agriculture, the share of consumer spending on food, and the share of spending allocated to imports. The biggest reduction in cross countryproductivity differences results from liberalizing trade in the presence of costless labour mobility. Liberalized-trade involves lowering bothagricultural and manufacturing import barriers. Import barriers are lowered to the average for the richest ten-percent of countries, by sector.Mobile labour involves eliminating between-sector wage differences.

exports in Figure 1.19.

1.5.2 Cross-Country Productivity Gaps

Table 1.7 displays model-implied gaps in sectoral and aggregate productivity across countries, under

various measures and experiments. Reducing import barriers and allowing costless labour mobility

results in dramatic reductions in productivity gaps. The richest 10% of countries initially had aggregate

productivity 23 times the poorest but those same countries were only 15 times as productive after both

distortions were relaxed. The agricultural productivity gap for these countries is nearly cut in half, from

over 103 to 58. The reduction in log aggregate productivity variation is also significant. While not

displayed, nearly all the aggregate gains found in the broader liberalization experiments remain when

only agricultural import barriers are reduced. This is intuitive, given the importance of the agricultural

sector for poor country consumers resulting from subsistence food requirements.

Together, these results suggest that nearly one-third of the gap between rich and poor countries can

be accounted for by the lack of food imports. There are also interaction effects between domestic and

foreign (trade) distortions. Initially, the difference between the richest and poorest 10% of countries is

23. Lowering import barriers lowers the gap to 17.7, labour mobility barrier reductions lower the gap


Table 1.8: Counterfactual Aggregate Productivity Gaps, with Fixed Labour Allocations


Liberalize Liberalize TradeBaseline Trade with Fixed Labour

Aggregate 23.0 17.7 20.9

This shows aggregate productivity for the richest-10% of countries relative tothe poorest-10%. I restimate the trade liberalization counterfactual holding sectorallabour allocations fixed at their initial levels. This shows that approximately half(40%) the contribution of import barriers on aggregate productivity differences oper-ate through Ricardian trade selection.

to 21.5, combined the gap falls to 15.3. This implies 23% of the observed gap is from high import

barriers alone, 7% from costly labour mobility alone, but 33% from both distortions together. The re-

duction in cross-country income variation reveals a similar pattern. The variance in log GDP/Worker

across all countries in the sample falls by 7% following trade liberalization, 5% following labour mo-

bility improvements, and 17% following an improvement in both distortions. The contribution of both

distortions is greater than the sum of their individual contributions. This result is particularly important

given the literature’s focus on domestic distortions within closed-economy frameworks.

1.5.3 Decomposition: Cross Country Aggregate Productivity and Income Variation

Given technology levels, I decompose aggregate productivity changes into two broad channels: (1) Ri-

cardian trade selection and (2) structural change. Selection occurs because of low productivity domestic

producers shutting down with increased import levels. Recall Equation 1.12, yik = π−1/θkiik Aik, defines

sectoral labour productivity, which changes inversely with the domestic expenditure share. To determine

the contribution of trade selection in the reduction of aggregate productivity differences, I re-estimate

the trade liberalization counterfactual holding sectoral labour allocations fixed at their initial levels.45 I

display the results in Table 1.8. The aggregate productivity gap falls to 20.9 - instead of 17.7 - when

labour allocations are fixed. This indicates Ricardian trade selection accounts for nearly half (40%) the

contribution of import barriers to international productivity differences.

That productivity differences within sectors shrinks following liberalized trade and labour markets

45The labour mobility improvements are not relevant in this case, since ξi = 1 implies labour allocations are not relevantfor aggregate productivity.


Table 1.9: Contribution to Productivity Gaps, Various θ

θ = 7 θ = 5 θ = 10 θa = 5.5θm = 6.8

Aggregate 33% 60% 27% 46%Agriculture 44% 44% 46% 47%

Manufacturing 32% 42% 24% 32%

Displays the contribution to productivity gaps between rich and poor countries of both import and labourmobility distortions for various values of the productivity dispersion parameter θ . I report the baseline value of7 in the first column, followed by 5 and 10 to reflect the range suggested by Anderson and van Wincoop (2004).The final column reports for the values of θ implied by using CEPII trade costs to proxy τi j (see footnote 30).Excluding Niger results in lower contributions to aggregate productivity gaps, since Niger’s labour allocation ishighly sensisitive to trade costs. The alternative values are 23%, 27%, 21%, and 27%, respectively.

is an important point to emphasize. Trade models with horizontally differentiated goods and heteroge-

neous productivity across firms can account for Ricardian selection while homogeneous goods frame-

works cannot. It also suggests that many of the inefficient production technologies employed in low-

income countries - such as small farm sizes - may be abandoned if access to imports improves. To

reiterate, however, I am not advancing a specific policy recommending. Accounting for within-sector

changes is important to quantify the contribution of low imports and labour misallocation to observed

income and productivity differences across countries.

1.6 Discussion and Robustness of Results

1.6.1 Alternative Values for θ

The import-elasticity of labour productivity within a tradable goods sector is 1/θ . My baseline result

is that the two distortions contributes to one-third of the observed differences in aggregate labour pro-

ductivity between rich and poor countries. They further contribute to nearly half the difference within

agriculture and one-third in manufacturing. I repeat the experiments for various values of θ to en-

sure my results are robust. I report the alternative contributions in Table 1.9. Overall, the headline result

varies between 27% and 60% depending on the value of θ , with the baseline value yielding conservative

results.


1.6.2 Alternative Counterfactual Experiments

I investigate a more limited experiment involving reducing import-barriers in only the poorest-10% of

countries by fifty percentage points and improve poor-country labour markets only until ξi = 0.8. The

wage wedge implied by this value of ξ corresponds to an urban/nonagricultural unemployment rate of

20% and no rural/agricultural unemployment in the Harris and Todaro [1970] framework.46 I display the

results of this experiment in Table 1.10. The reduction in the gap between rich and poor is, as expected,

much less than previous experiments. The magnitudes, though, are still impressive given the limited

liberalization among only the poorest-10%. The gap between the richest and poorest falls by more than

18% (from 23 to 18.8) in aggregate and by nearly 20% within agriculture (from 103 to 84).

I perform three more counterfactual experiments to illustrate the behavior of the model. To isolate

food trade in particular, I liberalized trade within the agricultural and manufacturing sectors separately.

In these scenarios, the aggregate labour productivity gap falls to 15.8 when agricultural trade is lib-

eralized, in conjunction with fully mobile labour, but only to 21 in the case of manufacturing trade

liberalization. I report the results of these two experiments in Table 1.10. Finally, I examine the produc-

tivity and income response to a full liberalization of labour markets and international trade. Specifically,

I set ξi = 1 for all countries and τi j = 1 for all trading pairs (i, j) for both sectors. I report the results of

this experiment in Table 1.11. The average growth in aggregate real GDP/Worker amongst the poorest

countries of my sample is 120% under frictionless trade, and 160% when labour mobility costs are also

zero. For rich countries, the aggregate gains are approximately 30% under both scenarios. The resulting

gap in labour productivity between rich and poor countries falls nearly in half in aggregate and by two-

thirds in agriculture. Of course, this exercise merely illustrates the model’s behavior, since removing all

trade costs is not feasible.

1.6.3 Plausibility of Trade Cost Estimates

The country-specific import costs suggest nearly a one hundred percentage point difference between

rich and poor countries. These results are plausible, given the voluminous contributions to trade costs

beyond tariffs and transport costs that often cannot be directly measured (see Anderson and van Wincoop

46To see this, the agricultural wage wa will equal the expected nonagricultural wage wn. The unemployment rate reflectsthe probability of not securing employment at a given wage. So, wa = un0+(1−un)wn⇒ wa

wn= 1−un, which equals 0.8 if

un = 0.2.


Table 1.10: Counterfactual Productivity Gaps, Various Experiments

(a) Limited and Unilateral Liberalization in Poorest-10% of Countries





(b) Reduce Agricultural Import Barriers Only





(c) Reduce Manufacturing Import Barriers Only





This shows productivity for the richest-10% of countries relative to the poorest-10% for various alternative counterfactual experiments.The top panel: Liberalized-trade involves lowering both agricultural and manufacturing import barriers in the poorest-10% by fifty percent-

age points. Mobile labour involves setting ξ = 0.8 in poor countries. Even this limited and unilateral liberalization results in an 18% reductionin the aggregate labour productivity gap between rich and poor countries.

The middle and bottom panels: trade liberalization is for only agriculture (middle panel) and for manufacturing (bottom panel) separately.Mobile labour in both cases involves zero mobility costs (ξi = 1). The largest reductions come from liberalizing agricultural goods trade.


Table 1.11: Counterfactual Productivity Gaps, Full Liberalizations

Frictionless International Trade

Baseline Costly LabourMobility

Costless LabourMobility


Aggregate 23.0 14.3 12.6Agriculture 103.2 60.0 32.0Manufacturing 77.4 28.7 32.0

This shows productivity for the richest-10% of countries relative to the poorest-10%. I restimate the modelunder perfectly free trade (τi j = 1 for all (i, j) pairs) and, in addition, under zero labour mobility costs (ξi = 1).

[2004]). Traditional measure of trade costs can account for much of the estimate. First, observed

WTO average tariff rates are larger in poorer countries, on the order of 20% for agricultural imports

under MFN. Tariff costs go beyond average values, since variation across substitutable products matters

nearly as much. Kee et al. [2008], accounting for tariff variation across products and the different

product elasticities imply trade restrictiveness47 is 64% larger than average tariff rates imply. Large

distortions from product-line tariff variation is also found for the United States by Irwin [2010], with a

uniform tariff-equivalent estimate of 75%. Next, many studies find non-tariff barriers of roughly equal

importance (and often more important) for a country’s level of restrictiveness Kee et al. [2009]. A

host of other trading difficulties exist for poor countries that increase trade costs. Contracting costs

and insecurity, poor local distribution infrastructure, information gathering costs, currency controls,

local content regulations, or health regulations in the case of food. Distribution costs are no doubt a

significant driver of trade costs for poor countries, with such costs already on the order of 50% for rich

countries.

1.6.4 Implications for Price Differentials

Agricultural and manufacturing prices in the model decline with income. ICP data for 2005, however,

suggest price levels in these traded goods sectors are increasing with income. Moreover, the relative

price of agriculture to non-agricultural goods in the model is rising with income but strong declining

with income in the ICP data. Since model-prices capture the full price involved in purchasing goods,

47Trade restrictiveness is the uniform tariff rate that generate identical dead-weight loss as a particular tariff/NTB structure


including transport to the point of consumption, some of the differences may be illusory. Given low

infrastructure quality in poor countries, these concerns may be significant. However, one should not

even take ICP prices at face value. FAO farm-gate prices, on the other hand, which I display in Figure

1.20, suggests far higher agricultural prices in poor countries than ICP suggests. Putting this point aside,

and taking ICP price estimates as given, an alternative specification of trade costs allows the model to

more closely match ICP-implied price levels.

1.6.4.1 Exporter-Specific Trade Costs

I redo the analysis under the alternative form of trade cost asymmetry suggested by Waugh [2010];

specifically, country-specific export costs rather than import costs.48 The World Bank’s Doing Business

Index surveys the cost of exporting an identical shipment of goods from a variety of countries and

displays a clear decline in such costs with income. Poor country export costs are perhaps twice that

of rich. If exporter-costs are included only with the manufacturing sector, the baseline dispersion of

productivity across countries shrinks to 24 (from 77). Counterfactual experiments in this environment

yield even greater reductions in aggregate productivity since poor country manufacturing sectors - to

which farm labour will reallocate - have higher productivity. If exporter-costs are imposed on both

sectors, the productivity gap is 50 in agriculture and 30 in manufacturing. In this case, import barriers

and labour misallocation still account for nearly 20% of the aggregate gap between rich and poor. I

conclude that my results are largely robust to my choice of trade cost asymmetry and provide more

detailed results, with specific Tables and Figures, in the appendix.

1.6.5 OECD Agricultural Producer Support

Support programs for the agricultural sector in higher-income countries are large. The OECD estimates

producer support estimates as high as 60% of production in Korea and Japan, 31% in the European

Union, 22% in Canada, and 11% in the United States.49 My main productivity estimates, Aia, capture

producer supports. Previous counter-factual exercises apply if PSE levels remain unchanged. Removal

48I use the agricultural trade data and 2005 ICP prices to show asymmetric trade costs identified by Waugh [2010] formanufactured goods is also a feature of the agricultural goods trade. For example, it is generally most costly for the UnitedStates to import food from developed economies than for developing countries to import food from the US. Perhaps exportsubsidy programs may play a role here.

49Source: Agricultural Policies in OECD Countries 2009: Monitoring and Evaluation.


of support results in lower poor-country imports and higher rich-country imports. I present details of

this experiment in the appendix and I find all main results robust.

1.6.6 Actual Development Experiences

Finally, this chapter’s counterfactual experiments suggest food imports facilitate structural change and

development. Two notable historical experiences are broadly consistent with this claim. First, between

1780 and 1850, the United Kingdom experienced massive reallocation of labour off the farm at the same

time as food imports rose. Technological change in the manufacturing sector and the declining cost of

important inputs, such as power and transportation, increased manufacturing productivity. Were higher

trade volumes the result of higher manufacturing productivity or vice-versa? Stokey [2001] argues

increased food imports, independent of technical change, accounts fully for the reduction in domestic

food production and approximately half of real wage growth.

South Korea since the mid-1960s provides a second historical episode where increased food imports

may have facilitated structural change and increased aggregate productivity and income. The FAO Food

Balance sheets for South Korea show products accounting for over 75% of calorie consumption - cereals

and starchy-roots (potatoes, etc.) - were nearly all domestically produced in the early 1960s. By 2000,

imports were twenty-seven times their 1961 quantity (nearly 9% growth per year) and more than double

domestic production. Tariffs for some of the most important imported goods, such as Wheat, are as low

as 3% (applied, and 9% bound). Consequently, South Korea’s employment share in agriculture fell from

over 50% to less than 10%. The remaining domestic production has become increasingly concentrated

in fewer varieties, with rice alone accounting for more than 50% of cultivated land.50 Teignier [2010]

concludes food imports facilitated reallocation and productivity increases, though to a smaller extent

than possible given large support programs - among the highest in the world - for domestic farmers.

1.7 Conclusion

This chapter examines the relationship between the international food trade and differences in labour

productivity between rich and poor countries. A large literature finds labour productivity differences

within the agricultural sector accounts for nearly the entire aggregate productivity gap. To understand

50Source: South Korea Agricultural Policy Review, Vol. 5 No. 1. Agriculture and Agri-Food Canada.


these within-sector differences, existing studies focus on domestic distortions within closed-economy

frameworks. Instead, I describe and exploit a general equilibrium model of international trade to: (1)

measure sectoral productivity and trade costs across countries from observed import and export flows;

and (2) quantify the impact of low poor-country food imports on international income and productivity

gaps. Specifically, I expand on Yi and Zhang [2010] and modify an Eaton-Kortum trade model to

incorporate multiple sectors and standard features from the macroeconomics literature - namely, non-

homothetic preferences and labour mobility costs. With this model, I estimate PPP-adjusted productivity

without producer price or employment data. This is particularly important for developing countries,

where agricultural employment estimates are overstated [Brandt and Zhu, 2010, Brandt et al., 2008,

Gollin et al., 2004] and systematic non-agricultural producer price estimates are unavailable. I find

agricultural labour productivity differs by a factor of 100 between rich and poor countries, more than in

manufacturing and much more than in services.

Despite low agricultural productivity, poor countries import very little of their food. I focus on

two distortions to account for the low food imports: high international trade barriers and costly labour

mobility. Trade barriers increase import prices and labour mobility frictions increase farm employ-

ment and decrease farm wages. Both distortions lead consumers to opt for lower productivity domestic

producers. Counterfactual experiments within the calibrated model determine how much of the pro-

ductivity and income differences between rich and poor countries are due to limited food imports and

labour misallocation. Liberalization of domestic labour markets and lowering import barriers shut-down

low productivity domestic producers and facilitates labour reallocation out of unproductive agricultural

varieties. More specifically, both specialization within agriculture and trade between developing coun-

tries increase dramatically. Fewer farm workers are also required and the resulting labour reallocation

further increases aggregate productivity in poor countries. In addition, I find an interaction between

domestic labour-market distortions and import barriers, with both distortions together accounting for

more cross-country productivity differences than each separately. Overall, low food imports and labour

misallocation accounts for half the agricultural productivity differences between rich and poor countries

and a third of the aggregate differences.


Tables and Figures

Table 1.12: Relative Productivity and Trade Estimates

Country Real GDP/Worker Relative Ag. Home Bias Home Bias Ag Import Nonag Import(PWT 6.3) Productivity in Ag in Nonag Barrier Barrier

ALB 7590.271 .39115682 .96141678 .69874883 116.23558 35.421284ARE 66576.617 .44485646 .61258745 .42578775 -18.620369 -53.892677ARG 28930.211 .58822697 .92331129 .81713331 -31.249567 -54.02153ARM 9728.3887 .48998314 .98607594 .93895686 52.866177 38.422016AUS 60072.863 .57381749 .68545008 .54253531 -50.1661 -65.197693AUT 65356.531 .43096933 .4135825 .21421564 -20.688431 -58.447227AZE 8344.5225 .3909446 .95296329 .93601954 58.564125 11.980983BDI 1484.5964 .3851738 .9966318 .9570998 35.336205 47.586834BGD 4014.1082 .30144984 .99489081 .91947263 43.869297 -24.635656BGR 15948.272 .40791327 .95872879 .60820246 26.421473 -31.066128BHR 43883.574 .45361263 .37662947 .62962717 -14.125608 -41.283718BHS 48276.43 .57115817 .44156331 .27013373 -23.095102 -33.775581BIH 11547.183 .37477651 .87371927 .73112261 73.309517 15.562359BLR 25513.055 .44623926 .98642498 .9683097 91.117615 25.764761BLZ 21620.295 .73690593 .83211309 .79179543 -19.825785 -3.4059961BOL 7833.9316 .42441931 .95463169 .86706454 31.046803 -13.168035BRA 17660.801 .47839239 .94823343 .85341001 -24.6385 -58.092987BRB 40506.512 .62988985 .88369179 .81707412 25.189732 2.0579891CAN 60726.898 .52735758 .17978513 .10846359 -54.533997 -69.607185CHL 36284.004 .568097 .83947229 .77406633 -28.651699 -52.482285CHN 7559.1113 .27769059 .99373788 .90127778 5.4443336 -63.081043CMR 6600.2842 .40258005 .98312026 .90316433 32.569183 -7.6996498COL 13745.382 .52220678 .94951999 .85174996 .21882121 -36.463806CRI 22434.385 .65378511 .91839951 .74750924 -18.682486 -34.551842CUB 16589.055 .73064089 .90199858 .95611066 -15.214612 7.4105468CYP 43011.164 .54910803 .60781217 .30474305 -3.2015181 -28.692625CZE 31778.24 .40388408 .78479379 .45648283 -1.8401186 -46.059528DOM 18871.377 .57409 .94462961 .83033311 13.785944 -6.762392DZA 14551.485 .24858868 .8226999 .88474596 70.361885 -27.203905ECU 12178.112 .63464582 .80907476 .83972847 -28.973389 -25.910892EGY 15739.912 .42501581 .97270417 .89575201 42.46011 -22.299734ESP 55540.348 .47999349 .69178373 .52950901 -32.949368 -63.288963ETH 2001.5085 .34709328 .99612582 .9237144 43.246174 4.3161197FRA 61215.82 .42952317 .47984272 .2127251 -46.736046 -74.542259GAB 19198.613 .40116626 .85290688 .7041955 22.552519 -27.402355GBR 55386.16 .47294986 .19167638 .30696738 -51.370808 -70.356552GEO 8629.6055 .4205569 .97356004 .92317873 56.330154 11.483145GER 60376.449 .45014486 .1989603 .23394805 -52.131054 -74.166313GHA 3109.5029 .36462086 .98229206 .75973243 21.250496 -16.157288

Continued on next page...


... table 1.12 continued


GMB 2867.2598 .26624116 .90920943 .61966658 36.30434 -16.088514GRC 48963.172 .53542763 .79882061 .54401541 -8.650218 -47.199192GTM 19613.189 .8248201 .98411417 .84365767 4.647737 -9.9256105GUY 5755.3135 .470209 .9320249 .79902595 10.763121 -14.257613HND 8106.8535 .57012063 .97656822 .79237938 6.1110044 -17.688372HRV 21563.98 .47247696 .86566246 .589167 27.8923 -9.1185513HUN 32281.992 .45421082 .84972095 .25820506 -12.490937 -45.643238IDN 8827.8955 .31406319 .97858161 .91279435 -4.1200795 -56.2113IRN 24279.256 .47311798 .96224785 .90294707 13.080079 -31.706249ITA 65438.621 .43283814 .62625158 .45005322 -34.368046 -70.017349JAM 17163.977 .46379858 .90370435 .65703332 7.1718373 -29.846788JOR 16173.863 .4016206 .48802936 .6774441 14.357027 -19.893681JPN 53166.207 .31194833 .57334828 .61662436 -31.655426 -72.500595KAZ 15065.728 .41171774 .959952 .8687017 11.999783 -34.486782KGZ 7831.019 .41997975 .9946304 .93580574 90.92321 18.829765KHM 3811.2412 .24670576 .99335039 .7234664 124.21813 -15.165354KOR 39495.418 .28960797 .84103942 .45653099 -6.4119534 -71.588135LAO 3883.1799 .37166846 .99248093 .78021073 63.206375 6.1703076LBY 42501.613 .49303922 .77914327 .88688254 17.761953 -9.0768642LKA 10848.017 .49249989 .97575247 .83043873 -3.3311014 -28.865454LTU 17925.152 .42437679 .81858528 .51731342 15.331088 -16.681107LVA 17569.195 .36693966 .75201631 .28633946 18.558769 -29.939064MAR 13006.281 .45928419 .927104 .74079555 3.4475424 -32.662697MDA 5778.0376 .36736712 .97938186 .74590325 64.000237 9.0598278MDG 2116.3748 .30560085 .98440021 .7474438 16.96875 -22.301098MEX 26379.596 .393978 .81716233 .36352396 -14.011316 -59.917614MKD 15400.013 .43038976 .89876306 .66855162 54.501045 7.944171MLI 4272.3198 .30154476 .98487151 .85884619 67.92749 -7.9862909MNG 4949.4604 .32729781 .96017265 .60031152 37.980572 -11.134032MOZ 2643.6362 .36331731 .97207379 .86094666 32.287895 3.2383001MRT 5089.9858 .44948477 .91180551 .77724105 -10.426609 -17.935457MUS 34345.426 .62497592 .83736217 .79004896 -5.9959579 -19.806093MYS 33878.715 .39930263 .1155616 .09930795 -52.302139 -73.056992NER 2440.2144 .26194119 .98998141 .86838865 100.86871 18.319036NGA 4234.168 .23728889 .9738096 .82959747 43.586437 -41.20121NIC 5743.2651 .54447824 .9832809 .81121922 9.5045967 -6.6013904NPL 4679.5435 .37024894 .99880934 .9483645 104.91924 18.450016OMN 64533.863 .56666619 .48471743 .45963252 -20.699299 -41.747982PAK 9059.4248 .35754877 .99065626 .88874733 35.617798 -36.012218PAN 16515.746 .39707556 .87120342 .13788819 -20.300501 -55.828621PER 11026.101 .48054996 .91168296 .81760824 -17.149261 -42.004776PHL 9777.0186 .34894532 .95292133 .61925316 5.1450987 -49.129383PNG 5147.5142 .38255289 .97337526 .77683401 22.000086 -16.372852POL 23813.854 .45402408 .8617813 .65672165 -10.345575 -39.855751

Continued on next page...




PRT 38223.16 .4306064 .60384381 .42459369 -14.400739 -53.67141PRY 10000.54 .45730346 .9551084 .74019635 49.548306 -16.345976ROM 11892.55 .3478418 .95351768 .73561662 30.557854 -34.01915RUS 16792.43 .33892775 .90989524 .90612638 -6.7339177 -49.172535SAU 57897.566 .35190997 .76930201 .80598259 11.971082 -46.539055SDN 5175.5542 .36969495 .9942534 .91511351 61.522282 -8.2719154SLE 3026.7737 .2853317 .99199003 .89181894 95.976143 15.57388SLV 13053.375 .60670823 .96081477 .83843374 15.87324 -13.850939SUR 22195.484 .55933756 .90996349 .72771931 1.5714281 -22.388821SVK 24524.709 .39087147 .79760838 .44502318 17.925068 -33.120319SVN 38960.258 .3969709 .64999944 .27385515 11.000682 -41.536098SYR 8355.4863 .39964205 .94422925 .76356459 20.058931 -24.974756TGO 2544.9956 .27694082 .96148098 .61944824 20.037184 -23.952509THA 12530.275 .38763082 .92903525 .64243913 -32.958405 -63.073338TJK 6530.2212 .44627711 .97833043 .97203553 20.12665 12.346147

TKM 20911.957 .43660912 .99226308 .96496046 93.682884 20.936636TTO 33820.656 .57613021 .54945815 .75556767 -21.91465 -23.144445TUN 22505.564 .41643357 .9370814 .63800621 33.481415 -29.153522TUR 18381.109 .41552803 .97293139 .65911072 3.6455681 -51.358505TZA 1376.3237 .25590083 .99364263 .81406415 34.848835 -19.801512UGA 2519.8564 .42649615 .99762428 .95553136 48.702965 33.835415UKR 12087.757 .31721985 .98681301 .88269752 57.582664 -28.444921URY 24083.666 .64031476 .82175672 .71544683 -21.733906 -37.96516USA 77003.289 .49307323 .67464781 .58444643 -54.888073 -76.028572UZB 3857.9875 .28255773 .98547316 .84358966 40.940517 -23.165609VEN 25604.17 .42014474 .84088588 .82016724 .2384045 -41.836048VNM 4914.9805 .35769731 .98893237 .83897406 2.478817 -38.810883YEM 5050.4082 .32764158 .79635417 .84302258 15.492043 -19.490442ZAF 23749.947 .4343161 .89864075 .76840812 -23.157085 -54.141731ZMB 2729.6191 .26534802 .9891625 .66928852 51.460373 -19.407759ZWE 10903.287 .45615458 .99608648 .88828504 53.36005 -11.625922


Figure 1.5: Normalized Import Shares: No Import Barriers

7.5 8 8.5 9 9.5 10 10.5 11−8

−6

−4

−2

0

2

4

6

Log(GDP/Worker), PWT6.3 for Year 2000

Log(

Sha

re P

urch

ased

Abr

oad

/ Dom

estic

Sha

re)

CounterfactualBaseline

Display result of setting import barriers to the average level in rich-countries. Poor country normalizedimport shares increase slightly more than rich. The resulting normalized import share is unrelated to income.Dots represent countries with a quadradic best-fit line also illustrated.


Figure 1.6: Normalized Import Shares: No Labour Mobility Costs

7.5 8 8.5 9 9.5 10 10.5 11−7

−6

−5

−4

−3

−2

−1

0

1

2

3


Log(

Sha

re P

urch

ased

Abr

oad

/ Dom

estic

Sha

re)


Display result of removing labour mobility costs, ξi = 1, in all countries. Poor country normalized importshares increase as a result. Dots represent countries with a quadradic best-fit line also illustrated.


Figure 1.7: Normalized Import Shares: No Import Barriers or Labour Mobility Costs

7.5 8 8.5 9 9.5 10 10.5 11−8

−6

−4

−2

0

2

4

6

8


Log(

Sha

re P

urch

ased

Abr

oad

/ Dom

estic

Sha

re)


Display result of removing both labour mobility costs, ξi = 1, and setting import barriers to the average ofrich-country levels. Normalized import shares increase more in poor countries than rich. Dots represent countrieswith a quadradic best-fit line also illustrated.

Figure 1.8: Counterfactual Gains in GDP/Worker

7 7.5 8 8.5 9 9.5 10 10.5 11 11.510

−1

100

101

102

103

Change in GDP/Worker, Post−Liberalization


Per

cent

age

Poi

nt C

hang

e in

GD

P/W

orke

r

Trade Liberalization OnlyBoth Trade and Labour Liberalization


Figure 1.9: Real Output-per-Worker in Agriculture Relative to Manufacturing

7 7.5 8 8.5 9 9.5 10 10.5 11 11.50.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9Real Relative Labour Productivity

Log(GDP/Worker), PWT6.3

Technology Onlyw/ Trade Selection

Figure 1.10: Agricultural Labour Productivity, Model Estimates vs. Data

−5 −4 −3 −2 −1 0 1 2 3−5

−4

−3

−2

−1

0

1

2

3Agricultural Labour Productivity

Data, PPP Adjusted (Rescaled, Mean=0)

Mod

el (

Res

cale

d, M

ean=

0)


Figure 1.11: Trade Cost Estimates for Agricultural Goods

(a) Overall (Trade Weighted) Trade Barriers

AGO

ALB

ARE

ARG

ARM

AUSAUT

AZE

BDI

BENBFA BGD

BGR

BHR

BHS

BIH

BLR

BLZ

BOL

BRA

BRB

CAF

CAN

CHECHL

CHN

CIV

CMR

COG

COL

CRI

CUB CYP

CZE

DJIDOM

DZA

ECU

EGY

ESP

EST

ETH

FINFJI

FRA

GAB

GBR

GEO

GER

GHA

GINGMB

GNB

GNQ

GRC

GTM

GUY

HND HRV

HTI

HUN

IDN

IRN

IRQ

ISR

ITA

JAMJOR

JPN

KAZ

KEN

KGZ

KHM

KOR

KWT

LAO

LBN

LBR

LBYLKA

LTU

LVAMAR

MDA

MDG

MEX

MKD

MLI

MLT

MNG

MOZMRT

MUS

MWI

MYS

NER

NGA

NIC

NPL

OMN

PAK

PANPER

PHLPNG

POLPRT

PRY QATROM

RUS

RWA

SAU

SDN

SEN

SLE

SLV

SOM

SUR

SVKSVN

SWE

SYRTGO

THA

TJK

TKM

TTO

TUN

TURTWN

TZA

UGA

UKR

URY

USA

UZB

VEN

VNMWSM

YEM

ZAF

ZAR

ZMBZWE

020

040

060

0P

erce

nt


(b) Importer-Specific Fixed Effects

AGO

ALB

ARE

ARG

ARM

AUS

AUT

AZE

BDI

BENBFA

BGD

BGR

BHRBHS

BIH

BLR

BLZ

BOL

BRA

BRB

CAF

CAN

CHECHL

CHN

CIV

CMR

COG COL

CRICUB

CYPCZEDJI

DOM

DZA

ECU

EGY

ESP

EST

ETH

FINFJI

FRA

GAB

GBR

GEO

GER

GHA

GINGMB

GNB

GNQGRC

GTMGUY

HND

HRV

HTI

HUNIDN

IRN

IRQ

ISR

ITA

JAMJOR

JPN

KAZ

KEN

KGZ

KHM

KORKWT

LAO

LBN

LBR

LBY

LKA

LTULVA

MAR

MDA

MDG

MEX

MKD

MLI

MLT

MNGMOZ

MRTMUS

MWI

MYS

NER

NGA

NIC

NPL

OMN

PAK

PANPER

PHL

PNG

POL PRT

PRY

QAT

ROM

RUS

RWA

SAU

SDN

SEN

SLE

SLVSOM

SUR

SVKSVN

SWE

SYRTGO

THA

TJK

TKM

TTO

TUN

TUR

TWN

TZA

UGAUKR

URY

USA

UZB

VENVNM

WSMYEM

ZAF

ZARZMB ZWE

−50

050

100

150

Per

cent



Figure 1.12: Trade Cost Estimates for Manufactured Goods

(a) Overall (Trade Weighted) Trade Barriers

AGO

ALB

AREARG

ARM

AUS

AUT

AZE

BDI

BEN

BFA

BGD

BGRBHR

BHSBIH

BLRBLZ

BOL

BRA

BRBCAF

CAN

CHL

CHN

CIVCMRCOG

COLCRI

CUB

CYP

CZE

DJI

DNK

DOM

DZAECU

EGY

ESP

ETH

FJI

FRA

GAB

GBR

GEO

GER

GHA

GIN

GMB

GNB

GNQ

GRC

GTMGUYHND

HRV

HTI

HUNIDN

IRN

IRQ

ISL

ISR

ITA

JAMJOR

JPN

KAZ

KEN

KGZ

KHM

KOR

KWT

LAO LBN LBYLKA

LTU

LVA

MAR

MDAMDG

MEX

MKD

MLI

MNGMOZMRT MUS

MWI

MYS

NER

NGA

NIC

NOR

NPL

NZL

OMNPAK

PANPER

PHL

PNG

POLPRT

PRYQAT

ROM

RUS

RWA

SAU

SDNSEN

SLE

SLV

SOM

SUR

SVKSVN

SYC

SYR

TCD

TGO

THA

TJK

TKM

TTO

TUN

TUR

TZA

UGA

UKRURY

USA

UZB

VENVNM

WSM

YEM

ZAF

ZAR

ZMBZWE

020

040

060

080

0P

erce

nt


(b) Importer-Specific Fixed Effects

AGO

ALB

AREARG

ARM

AUSAUT

AZE

BDI

BEN

BFA

BGDBGR

BHRBHS

BIH

BLR

BLZ

BOL

BRA

BRB

CAF

CAN

CHL

CHN

CIVCMR

COG

COL CRI

CUB

CYP

CZE

DJI

DNK

DOM

DZAECUEGY

ESP

ETH

FJI

FRA

GAB

GBR

GEO

GER

GHA

GIN

GMBGNB

GNQ

GRC

GTMGUY

HND

HRV

HTI

HUN

IDN

IRN

IRQISL

ISR

ITA

JAM

JOR

JPN

KAZ

KEN

KGZ

KHM

KOR

KWT

LAO

LBN LBY

LKA

LTU

LVAMAR

MDA

MDG

MEX

MKD

MLIMNG

MOZ

MRT MUS

MWI

MYS

NER

NGA

NIC

NOR

NPL

NZL

OMNPAK

PAN

PERPHL

PNG

POL

PRT

PRY

QATROM

RUS

RWA

SAU

SDN

SEN

SLE

SLV

SOM

SUR

SVK

SVN

SYC

SYR

TCD

TGO

THA

TJK

TKM

TTOTUN

TUR

TZA

UGA

UKR

URY

USA

UZB

VENVNM

WSM

YEM

ZAF

ZAR

ZMB

ZWE

−10

0−

500

50P

erce

nt



Figure 1.13: Competitiveness Measure for Agriculture

AGO

ALB

ARE

ARG

ARM

AUS

AUT

AZE

BDI

BEN

BFA

BGD

BGR

BHR

BHS

BIH

BLR

BLZBOL

BRA

BRBCAF

CAN

CHE

CHL

CHN

CIV

CMR

COG

COLCRICUB

CYP

CZE

DJI

DOM

DZA

ECU

EGY

ESP

EST

ETH FINFJI

FRA

GAB

GBR

GEO

GER

GHA

GIN

GMB

GNB

GNQ

GRC

GTM

GUY

HND

HRV

HTI

HUN

IDN

IRN

IRQISR

ITA

JAM

JOR

JPN

KAZKEN

KGZ

KHM

KOR

KWT

LAO

LBN

LBR

LBY

LKA

LTU

LVA

MAR

MDA

MDG

MEX

MKDMLI MLTMNG

MOZ

MRT

MUSMWI

MYS

NER

NGA

NIC

NPL

OMN

PAK PAN

PER

PHL

PNG

POL

PRT

PRY

QAT

ROM

RUS

RWA SAUSDN

SEN

SLE

SLV

SOMSUR

SVKSVN

SWE

SYR

TGO

THA

TJK

TKM

TTOTUN

TUR

TWN

TZA UGAUKR

URY

USA

UZB

VEN

VNM

WSM

YEM

ZAF

ZARZMB

ZWE

.51

1.5

2C

ompe

titiv

enes

s M

easu

re


Figure 1.14: Competitiveness Measure for Manufacturing

AGO

ALB

ARE

ARG

ARM

AUS

AUT

AZE

BDI

BENBFA

BGDBGR BHR

BHS

BIH

BLR

BLZ

BOL

BRA

BRBCAF

CAN

CHL

CHN

CIVCMR COG

COL

CRI

CUB CYP

CZE

DJI

DNK

DOM

DZA

ECU

EGY

ESP

ETHFJI

FRA

GAB

GBR

GEO

GER

GHA GIN

GMBGNB

GNQ

GRC

GTMGUY

HND

HRV

HTI

HUN

IDN

IRNIRQ

ISL

ISR

ITA

JAMJOR

JPN

KAZ

KEN

KGZ

KHM

KOR

KWT

LAO LBN

LBYLKA

LTULVA

MAR

MDA

MDG

MEX

MKD

MLI

MNGMOZ

MRT MUS

MWI

MYS

NER

NGA

NIC

NOR

NPL

NZL

OMN

PAK

PAN

PER

PHL

PNG

POL PRT

PRY

QAT

ROM

RUS

RWA

SAU

SDN

SENSLE

SLV

SOM

SUR

SVK SVN

SYC

SYR

TCD

TGO

THA

TJKTKM

TTO

TUN

TUR

TZA

UGA

UKR

URY

USA

UZB

VENVNM

WSM

YEM

ZAF

ZAR

ZMB ZWE

11.

52

2.5

33.

5C

ompe

titiv

enes

s M

easu

re



Figure 1.15: Import Shares of Poorest Countries, by Source Country Percentile

0.2

.4.6

.8

1 2 3 4 5 6 7 8 9 10

Manufacting Import Shares, by Source, of Poorest−10%

Before After

Figure 1.16: Import Shares of Richest Countries, by Source Country Percentile

0.2

.4.6

.8

1 2 3 4 5 6 7 8 9 10

Manufacting Import Shares, by Source, of Richest−10%

Before After


Figure 1.17: Increasing S-S Trade, Following Full Removal of Import Barriers

01,

000

2,00

03,

000

4,00

05,

000

N−N N−S S−S

S is below median GDP/worker

% Increase in Trade Flows

Agriculture Manufacturing

Figure 1.18: Increasing S-S Trade, Following Full Removal of Import Barriers

S−S N−S

N−N

Baseline

S−S

N−S

N−N

After Trade Liberalization

Increasing S−S Trade


Figure 1.19: Increased Share of Agriculture in Total Exports Following Trade Liberalization

7 7.5 8 8.5 9 9.5 10 10.5 11 11.5−10

−5

0

5

10

15

20

ALB

ARE

ARG

ARMAUS

AUT

AZE

BDI BGD

BGR

BHR

BHSBIHBLR

BLZ

BOL

BRA BRB

CANCHLCHN

CMR

COL

CRICUB

CYP

CZE

DOM

DZA

ECU

EGY

ESP

ETH

FRA

GAB

GBR

GEO

GER

GHA

GMB GRCGTM

GUY

HND

HRV HUN

IDN

IRN

ITA

JAMJOR

JPN

KAZ

KGZ

KHM

KOR

LAO

LBY

LKA

LTULVA

MARMDA

MDG

MEX

MKDMLI

MNG

MOZ

MRT

MUS

MYS

NER

NGA

NIC

NPL

OMNPAK

PANPER

PHL

PNG

POLPRTPRY

ROMRUS

SAU

SDN

SLE

SLV

SUR

SVK SVN

SYR

TGO

THA

TJK

TKMTTO

TUNTUR

TZA

UGA

UKR

URY

USA

UZB

VEN

VNM

YEMZAF

ZMBZWE

Change in Agriculture‘s Share of Exports

Log(GDP/Worker), PWT6.3

Per

cent

age

Poi

nt C

hang

e

Figure 1.20: FAO Food Prices are Higher than ICP Prices, Especially for Poor Countries

ALB ARG

ARM

AUSAUT

AZE

BDI

BEL

BFA

BGD

BGR

BIH

BLR

BOL

BRA

BTN

CAN

CHE

CHL

CHN

CIV

CMR

COG

COLCYP

CZE

DNK

ECU

EGY

ESP

EST

ETH

FIN

FRA

GBR

GEO

GHA

GIN

GMB

GNQ

GRC

HRVHUN

IDN

IND

IRL

IRN

ISL

ISR

ITAJOR

JPN

KAZ

KEN

KGZKHM

KOR

LAO

LBNLKA

LTU

LUX

LVAMAR

MDA

MDG

MDV

MEXMKD

MLI

MLT

MNG

MOZ

MUS

MWI

MYSNAM

NER

NGA

NLD

NOR

NPL

NZL

PAK

PER

PHL

POL

PRT

PRY

QAT

ROM

RUS

RWA

SAU

SDN

SGP

SVK

SVN

SWE

SYR

TGO

THA

TJK

TUNTURUKR

URY

USA

VEN

YEM

ZAF

24

68

10Lo

g(F

AO

Pric

e / I

CP

Pric

e)

1000 3500 10000 35000 100000Real GDP Chain per worker

Plots the (FAO Relative Prices / ICP Relative Prices)

Ratio of FAO Producer Prices to ICP Expediture Prices (Food)

Chapter 2

Regions, Frictions, and Migrations in a

Model of Structural Transformation

Abstract

Why do some regions grow faster than others? More precisely, why do rates of convergence differ? Re-cent research points to labour market frictions as a possible answer. This chapter investigates how labourmarket frictions interact with regional migration. Motivating this are two important observations: (1)farm-to-nonfarm labour reallocation costs have fallen, disproportionately benefiting poorer agriculturalregions; and (2) migration flows vary dramatically by region, lowering (raising) marginal productivitiesin destination (source) regions. Using a general equilibrium model of structural transformation cali-brated to US regional data over time, I find that whether labour market improvements lead to incomeconvergence depends on migration barriers between regions. Specifically, a large labour migration costbetween regions magnifies the convergence impact of improved labour markets within the poor region.The model captures the unique convergence experiences of various US regions, such as between theMidwest and the Northeast and between the South and the North.

47

CHAPTER 2. REGIONS, FRICTIONS, AND MIGRATIONS 48

2.1 Introduction

International differences in productivity, particularly within the agricultural sector, are key drivers of

income and living standard differences across countries [Caselli, 2005, Restuccia et al., 2008]. Various

factors may drive these differences, from inefficient farm sizes [Adamopoulos and Restuccia, 2010],

poor domestic transportation infrastructure [Adamopoulos, 2010, Gollin and Rogerson, 2010], barri-

ers to labour or intermediate inputs [Restuccia et al., 2008], or barriers to trade in agricultural goods

[Tombe, 2010]. In this chapter, I focus on labour market frictions and labour migration costs. Given

the lack of international wage and price data over time, which prevents measurement of these frictions

for a large set of countries, I study different regions within the United States over time. Using a gen-

eral equilibrium model of structural transformation calibrated to US regional time-series data, I find

that whether labour market improvements lead to income convergence depends on migration barriers

between regions. Specifically, a large labour migration cost between regions magnifies the convergence

impact of improved labour markets within the poor region. The model captures the unique convergence

experiences of various US regions, such as between the Midwest and the Northeast and between the

South and the North.

Between region migration in the United States has been an important phenomenon with implications

for convergence. There was a massive post-war increase in the size of Southern and, to a lesser extent,

Midwestern labour forces relative to the Northeast. Such flows are normally thought of as a conver-

gent force, as labour responds positively to wage (marginal product) differences. If labour responds to

other factors, however, it may opt to migrate towards lower wage areas, which would increase average

income dispersion. I will show that improvements in the ability of workers to switch from agricultural

to nonagricultural occupations in one region will attract migrants from the other. If the region expe-

riencing the labour market improvement has lower wages, such as in the US South, then in-migration

results in increased regional income inequality. The impact of labour market improvements on regional

convergence is thus enhanced by migration restrictions, which is the key result of this chapter.

The key intuition behind this result is as follows. Labour reallocation shrinks farm labour sup-

ply and increases relative farm wages, disproportionately improving overall average income in the


agriculturally-specialized region, such as the Midwest1 or South2. Improvements in a region’s labour

market, however, make it a more attractive region in which to live. The resulting labour inflow offsets

some of the income gains, due to diminishing labour productivity. Thus, it is in the presence of regional

migration barriers that one would expect the largest convergence impact from labour market improve-

ments. This mechanism may provide additional insight into how the Southern and Midwestern states

experienced dramatically different rates of income convergence with the industrial Northeast3, despite

both being agriculturally specialized with equally distorted labour markets.

This chapter contributes to the structural transformation and growth literature and, in particular, joins

research dealing with frictions within a two-sector, multi-region framework. As a whole, this literature

examines the strong negative relationship between the share of output and employment commanded by

the agricultural sector and the overall level of economic activity - a phenomenon known as the “Kuznets

fact” of growth. Recently, various researchers have developed simple models to explain this, from

increasing consumer goods variety [Greenwood and Uysal, 2005, Foellmi and Zweilmueller, 2006] or

preference non-homotheticities [Kongsamut et al., 2001] to differential sectoral productivity growth

[Ngai and Pissarides, 2007] or capital deepening [Acemoglu and Guerrieri, 2006].4 While capturing

the output and employment facts quite well, these models cannot match a number of other observations

relating to regional incomes, sectoral wages, or internal migration patterns.

Recent attempts to capture regional convergence, a rising agricultural wage, and internal labour

flows show that it is important to move beyond frictionless market structures. In particular, Caselli and

Coleman (2001) incorporate labour market frictions between agriculture and non-agriculture to show

that improved ability of workers to acquire manufacturing skills can capture the rise in relative agricul-

tural wages observed in the data - a feature previous models could not. This channel, which will be

expanded on later, also leads to convergence in income levels between regions. Another recent paper

by Herrendorf et al. [2009] investigates the consequences of goods market frictions between regions.

The authors find that large reductions in transportation costs between regions are an important driving

force behind westward settlement patterns in the mid-1800s. Finally, in a recent and related piece, I

investigate to what extent labour and goods market frictions might interact to capture a broader set of

1Midwestern States (MW): IA, IL, IN, MI, MN, MO, ND, NE, OH, SD, WI2Southern States (S): AL, AR, FL, GA, KY, LA, MS, NC, OK, SC, TN, TX, VA, WV3Northeastern States (NE): CT, MA, MD, ME, NH, NJ, NY, PA, RI, VT4A concise review of the issues involved may be found in Matsuyama [2005].


regional convergence experiences [Tombe, 2008]. The results point to an important mitigating influence

of transportation cost reductions on regional convergence. Each of these papers establish important roles

for market frictions in matching historical data. This chapter is distinct from existing work by address-

ing how internal migration flows influence interactions between labour and goods market frictions and

regional convergence patterns.

2.2 Empirical Patterns, by Region

Data for three major regional groups in the United States display unique growth experiences that point

to the importance of migration, labour market frictions, and income convergence. Specifically, I inves-

tigate two regional pairs: (1) the Northeastern versus Southern states; and (2) the Northeastern versus

Midwestern states. Figure 2.1 illustrates the geographic location of each region.

There are three important observations that guide the analysis. First, barriers to labour reallocations

out of agriculture have dramatically fallen in all regions. Figure 2.5 shows how both regions begin

1880 with a majority of their workforce employed on the farm, which declines to insignificance by

2000.5 In addition, Figure 2.3 displays the agricultural wages relative to nonagricultural wages for both

regions - their experiences largely coincide. In 1880, agricultural workers earned approximately five

times less than their nonagricultural counterparts, while they earned only slightly more than 20% less

by 2000. From this wage data, I infer that labour markets were equally distorted in the sense of worker

occupational switching costs. One need not take a position on what these frictions might be, from poor

access to nonagricultural skills training to explicit restrictions on nonfarm labour recruitment policies,

all such distortions are captured by the model in a reduced-form manner.

The second feature in the data presents the main puzzle this chapter seeks to address: differential

rates of regional convergence. Figure 2.4 starkly illustrates a far higher degree of income convergence

between the South and Northeast than for the Midwest. The South’s relative overall earnings nearly

doubled between 1880 and 2000 while the Midwest’s rose by barely 10% from its already high level. It is

worth emphasizing that despite equally distorted local labour markets and very agriculturally specialized

workforces, the Southern states’ overall average income is almost half the Midwest’s. Two forces might

5It is certainly true that the Southern states experienced a greater degree of structural change but later analysis will findthis difference alone is incapable of explaining the unique regional experiences.


account for this: first, higher transportation costs lead to compensating nominal wage levels in the

Midwest (the topic of Tombe [2008]); and second, higher cost of regional migration facing Southern

residents suppresses labour productivity and wages in that region (a feature of this chapter).

I illustrate a third and final fact in Figure 2.5: the geographic distribution of employment has changed

dramatically through time. Between 1880 and 2000 the Southern states’ total employment relative to

the Northeast’s increased by nearly 80% and the Midwestern states’ increase by 20%.6 It is these large

internal labour flows that offset much of the convergence impact of improved labour markets. Simply

put, the relative incomes of the Southern and Midwestern states would have been higher but for the

in-migration that took place since 1880. It also appears that only the last half-century saw notable flows

towards the South, which might suggests a large initial cost of migration.7

2.3 The Model

At its core, this is a two-region, two-sector model. Both goods are available for consumption but one -

the agricultural good - faces a subsistence requirement, and therefore an income-elasticity below unity.

The two regions may engage in trade of either good by incurring an iceberg transportation cost. Workers

may also work in either sector. I outline the details below, with time subscripts omitted for brevity.

2.3.1 Firms

2.3.1.1 Goods Producing

An agricultural sector and a manufacturing sector exist in each of two regions, populated by perfectly

competitive firms. I assume that the competitive advantage of the one region - the “core” - is in manufac-

turing and will completely specialise in its production. Both agricultural and manufacturing activities

may be conducted in the other region - the “periphery”. Each produces output using input factors of

land, labour, and capital within constant returns to scale production technologies. Thus, for each region

6These patterns are not due to differential rates of child birth or mortality.7For instance, explicit restrictions to recruiting farm labour in many Southern states by agents outside the state by the

various enticement laws, emigrant-agent laws, and contract-enforcement laws passed in the 1890s and early 1900s. SeeRoback [1984] for more details.


i ∈ p,c and sector s ∈ f ,m

Y is = Ai

sNisαsLi

s1−αs (2.1)

where Y , N, and L denote output, land, and labour. By assumption, Apf > Ac

f = 0. To simplify notation,

peripheral agriculture is the numeraire (Ppf = 1). Regional land endowments are exogenous, with the

periphery’s share of land denoted by ω . The inclusion of land within the production functions ensures

a nondegenerate distribution of manufacturing production between the regions by creating diminishing

returns to scale in the regionally mobile factor (labour).

Each firm sell output and hire inputs from competitive markets; therefore, firms take output prices,

Pis , land rents, r, and wages, w, as given. They each use the production technology from Equation 2.1 to

maximize profits,

Πis = Pi

sYis −wi

sLis− ri

sNis ∀ i = p,c and s ∈ f ,m.

This implies firm input demands must satisfy standard first-order necessary conditions,

∂Y pf

∂N pf= Pp

m∂Y p

m

∂N pm

= rp (2.2)

Pcm

∂Y cm

∂Ncm

= rc (2.3)

∂Y pf

∂Lpf

= wpf (2.4)

∂Y im

∂Lim

=wi

m

Pim∀ i = p,c (2.5)

2.3.1.2 Transportation

Goods produced in one region may be transported to consumers in another region by incurring an

iceberg-cost, leaving fraction ∆ successfully delivered. This feature of the economy is modelled by

assuming there exists a perfectly competitive transportation sector, where firms maximize profits earned

through goods sold in one region that were purchased in another. This technology is similar to that

utilised by Herrendorf et al. [2009], who further allow distinct food and non-food transportation costs.

Formally, for Dis and Bi

s representing the quantity of good s delivered to (bought from) region i, we have


the objective for all i, j = p,c, i 6= j, and s ∈ f ,m

maxDi

s,Bis

πt = Pif D

if +Pi

mDim− p j

f Bjf − p j

mB jm.

The comparative advantage of the core region in manufacturing goods and the periphery in agriculture

ensures Dpf = Dc

m = Bpm = Bc

f = 0. Furthermore, given the nature of the transportation costs, it must be

the case that

Dis = ∆B j

s ∀i, j = p,c, i 6= j

which, together with zero profit condition, implies

∆Ppm = Pc

m, (2.6)

Pcf = 1/∆. (2.7)

2.3.2 Households

Each agent is endowed with preferences that treat consumer goods asymmetrically, with agricultural

goods contributing to utility only above a subsistence level. This results in an income inelastic demand

for agricultural goods that leads labour to shift to the manufacturing sector over time and for agriculture’s

share of consumption to decline. Each agent selects a region of residence and defers its consumption

decisions to a regional household. Household consumption is evenly divided amongst its members. I

normalize the total population across both regions to unity. Finally, non-labour income from land rent

is region-specific.

Formally, the household of region i ∈ p,c, with agricultural subsistence level a, faces the follow-

ing problem

maxci

f ,cim,L

if ,L

im

τ log(ci

f − a)+(1− τ) log(cim)

(2.8)

subject to

Pif c

if +Pi

mcim ≤ Li

f wif +Li

mwim +Niri. (2.9)


This leads to two simple equilibrium requirements. First, optimal allocation between consumption

goods is such that the marginal rate of substitution equal the output price ratio,

Um(cif ,c

im)

Ua(cif ,ci

m)=

1− τ

τ

cif − a

cim

=Pi

m

Pif∀ i ∈ p,c. (2.10)

Second, individual residency decisions maximize utility, subject to a migration cost proportional to

utility - denoted µ < 1. In equilibrium, migratory incentives will not exist, which implies,

(cpf − a)τcp

m1−τ = µ(cc

f − a)τccm

1−τ (2.11)

2.3.2.1 Occupational Choice

Given that the extent of labour market frictions is a key piece of this model, a detailed look at its assumed

features is important. Essentially, the frictions are assumed to exist between sectors in the peripheral

region. That is, labour is not freely mobile between agricultural and nonagricultural pursuits. This is

captured in a reduced form fashion, in that no particular source for the friction is explicitly modelled.

Instead, a cost proportional to wages is imposed on peripheral-region nonagricultural workers. One

might consider this cost as a payroll tax that is later rebated in a lump-sum fashion to the Southern

household.

Given the existence of this cost on nonagricultural employment in the periphery, the household there

will make the occupational allocation choice based only on its effect on total income. An agent will be

selected for manufacturing work only if the earnings in that sector are sufficient to compensate for the

cost. With ξ denoting the proportion of earnings lost due to the friction, the peripheral household selects

an agent to engage in manufacturing production if and only if

(1−ξ )wpm ≥ wp

f (2.12)

The value assigned to ξ will be set to match the observed sectoral wage ratio, 1− wpf

wpm

, from data.


2.3.3 Market Clearing Conditions

To close the model, the following standard market clearing conditions must hold.

Lpm +Lp

f +Lcm = 1 (2.13)

N pm +N p

f = Ω (2.14)

Ncm = 1−Ω (2.15)

These equations require: (1) labour in all regions and sectors sum to one, the normalized total; and (2)

peripheral-land sum to the amount exogenously allocated to that region, Ω.

In addition, goods markets must clear. Each region produces, consumes, exports, and imports goods.

The periphery imports manufactured goods and exports agricultural goods, while the core does the

opposite. This follows from the comparative advantages assumed. Finally, with the total population

normalized to unity, the population in the core and the periphery, respectively, is Lcm and (1− Lc

m).

Hence,

LcmCc

f = Dcf

(1−Lcm)C

pf +Bp

f = Y pf

LcmCc

m +Bcm = Y c

m

(1−Lcm)C

pm = Y p

m +Dpm

Combining these with the results implied by the transportation firm problem solved earlier, and observ-

ing Walrus law, we find the agricultural goods market clearing condition is sufficient, and expressed

as

(1−Lpm)∆Cp

f +LcmCc

f = ∆Apf N

pf

α f Lpf1−α (2.16)

2.4 Calibration

There are various parameters in the model that require calibration. Table 2.1 outlines the strategy and

the values to which each parameter is set.


Tabl

e2.

1:C

alib

ratio

nof

Mod

elPa

ram

eter

s

(a)C

omm

onTi

me-

Inva

rian

tPar

amet

ers

Para

met

erD

escr

iptio

nTa

rget

Val

ue

αN

onla

bour

Inco

me

Shar

eL

itera

ture

0.4

τA

gric

ultu

ralg

oods

’pre

fere

nce

wei

ght

Lite

ratu

re0.

01

(b)R

egio

n-Sp

ecifi

can

dTi

me-

Var

ying

Mod

elPa

ram

eter

s

Para

met

erD

escr

iptio

nTa

rget

MW

-NE

S-N

E

Dir

ectly

Cal

ibra

ted

Usi

ngO

bser

vabl

eD

ata

1880

1990

1880

1990

∆i

Bet

wee

n-re

gion

tran

spor

tatio

nco

stPr

ice

diff

eren

tials

0.70

0.90

0.95

0.99

ξi

Sect

oral

switc

hing

cost

Wag

edi

ffer

entia

ls0.

780.

270.

810.

28

Join

tlyC

alib

rate

dU

sing

Mod

elO

utpu

t18

8019

9018

8019

90

aiSu

bsis

tenc

ele

velf

orfo

odC

onsu

mpt

ion

shar

es0.

130.

21Ω

iM

W/S

imm

obile

fact

orsh

are

Reg

iona

linc

omes

0.38

0.33

µi

Eas

eof

betw

een-

regi

onm

igra

tion

Reg

iona

lem

ploy

men

t0.

580.

780.

370.

82

Initi

alG

row

thIn

itial

Gro

wth

AP f

MW

/Sag

ricu

ltura

lpro

duct

ivity

Nor

mal

izat

ion

1.00

2.91

%1.

003.

97%

AP m

Non

agri

cultu

ralp

rodu

ctiv

itySe

ctor

alE

mpl

oym

ent

0.98

1.97

%0.

972.

22%

AC m&

Reg

iona

lInc

omes

1.01

1.95

%1.

031.

93%

The

mod

elis

trea

ted

astw

ose

para

test

atic

exer

cise

s.Pa

ram

eter

sar

eca

libra

ted

toa

uniq

ueva

lue

for

each

regi

onal

grou

pan

dea

chye

ar.

Para

met

ers

can

becl

assi

fied

into

thre

egr

oups

:(1

)T

hege

nera

llyac

cept

edpa

ram

eter

valu

esth

atar

eco

nsta

ntth

roug

htim

ean

dco

mm

onac

ross

regi

ons;

(2)

thos

ew

ithob

serv

able

coun

ter

part

sin

the

data

,who

seva

lues

are

setd

irec

tlyto

mat

chth

eda

ta;

(3)

unob

serv

able

para

met

ers

calib

rate

djo

intly

such

that

mod

elou

tput

mat

ches

obse

rvab

leda

ta.

The

1990

valu

esof

the

vari

ous

para

met

ers

isre

port

eddi

rect

lyex

cept

for

the

prod

uctiv

itypa

ram

eter

s,as

the

grow

thra

tes

betw

een

1880

and

1990

ism

ore

info

rmat

ive.

Fina

lly,n

ote

that

indi

rect

lyca

libra

ted

para

met

ers

are

estim

ated

sim

ulta

neou

sly

and

the

liste

dta

rget

sar

eth

ose

that

are

mos

tse

nsiti

veto

the

corr

espo

ndin

gpa

ram

eter

s.


Two parameters are set identical across regions, remain constant through time, and take on values

generally accepted in the literature. Land’s (or the immobile factor’s) share of output, α , is set to 0.4

and the preference weight for agricultural goods, τ , is set to 0.01.8 Two additional parameters can also

be set to values directly observable in the data. The transportation costs parameter, ∆, is set to match

observed price differences for identical goods across each region and the wage wedge, ξ , is set to match

observed wage data. Interested readers can find details regarding the data used to calibrate these two

parameters in Section 2.6.

Productivity in the peripheral agricultural region is normalized to 1 in 1880 and productivity in the

nonagricultural sectors - both core and peripheral - are calibrated to match data on the distribution of

employment.9 The growth rates of sectoral productivities are determined, jointly with the other indirect

parameters, to help match the extent of labour reallocation and regional incomes in 1990. Finally,

for between-region migration cost, I relegate a detailed discussion to Section 2.6 and note here that

migration cost values are required to fully match the regional distribution of employment and relative

incomes in 1880 and 1990. The calibrated values in Table 2.1 suggest that migration costs from the

Southern region were far higher than from the Midwest. This is entirely consistent with many qualitative

historical analysis, such as Wright [1986]. The productivity growth of the agricultural sector is also

clearly higher than nonagricultural, consistent with findings of other researchers [Caselli and Coleman,

2001, Jorgenson and Gollop, 1992].

In summary, the fully calibrated model incorporates declining migration costs, declining transporta-

tion costs, rising productivity, among others, is reported in Table 2.1. The data targets selected to

calibrate the parameters are: (1) the peripheral region’s share of both region’s employment; (2) the aver-

age peripheral wages relative to the core; and (3) the agricultural labour share in the peripheral region.

To determine 1990 parameter values I calibrate the rate of productivity growth in each sector, as well

8Interpreting N as land might lead to a different value. For instance, the ratio of land rent to total nonagricultural capitalrent is 0.14, which implies land’s share when physical capital is included is 0.06. This represents 0.1 the value of the standardlabour share. Hense, αm = 0.1 when physical capital is abstracted from. The same argument is used for the agricultural sector,but with land’s share of capital rent at 0.5, which implies α f = 0.33. However, this model need not restrict N to land alone,as it represents any immobile factor used in the production function. While I use a common α = 0.4 throughout, the resultsremain largely unchanged under the alternative values. That being said, migration flows become increasingly sensitive to otherparameters, and the model more difficult to solve, as αm→ 0.

9This strategy of determining the regional productivity parameters by matching model output to targets, rather than directlycalculating TFP from the data, is a necessary consequence of the lack of a long series of region-specific sectoral output andinput data. The results are robust to an alternative approach discussed in Section 2.6.


Table 2.2: Calibration Performance vs. Data

(a) Midwest-Northeast

Observed Outcome in 1880 1990Peripheral Region Data Model Data Model

Relative Employment Size 1.05 1.05* 1.16 1.16*Agricultural Employment Share 0.55 0.55* 0.03 0.04

Relative Income 0.81 0.81* 0.86 0.86*

(b) South-Northeast

1880 1990Data Model Data Model

1.06 1.06* 1.59 1.59*0.73 0.73* 0.03 0.030.43 0.43* 0.83 0.83*

* denotes targets.

as the migration cost parameter, to match targets (1), (2), and the aggregate agricultural labour share.10

The calibrated model output is found in Table 2.2.

2.5 Counterfactual Experiments

2.5.1 Labour Market Frictions

The key contribution of this chapter is demonstrating how the presence of migration restrictions enhance

the contribution of labour market improvements to regional income convergence. To that end, I will

isolate the impact of an improvement in the labour market by adjusting only that parameters from its

initial 1880 value, leaving all other unchanged. This experiment will be repeated when labour flows

between regions is restricted; the results are presented in Table 2.3.

In the standard model as originally presented, the Midwest would experience a massive employment

inflow (becoming over twice as large as the Northeast region) as a result of the lower labour market

friction. The overall impact on relative regional incomes is nil. This results from, on the one hand,

improved labour markets increase regional average incomes while, on the other hand, an employment

inflow lowers labour’s marginal product and, therefore, earnings. The second experiment is identical

to the first but restricts employment to its 1880 allocation; that is, migration is restricted. In this case,

there is less structural transformation (in the sense of the agricultural labour share falls by less - to

39% instead of 31% in the first experiment) and a dramatic increase in the Midwestern relative income.

10The peripheral share is not targeted since moving from 4% to 3% requires an unrealistically high agricultural productivitygrowth rate. The model does not have agricultural production in the core region while there is in the data, so having a slightlyhigher peripheral agricultural share but matching aggregate shares seems reasonable. This is only important when agriculturalshares are low.


Table 2.3: Isolating the Effect of Labour Market Improvements

Reduce Labour MarketFriction by Two-Thirds

1880 Benchmark Model with Model withObserved Outcome Model Values Migration No Migration

Midwestern Region

Relative Employment Size 1.05 2.20 1.05Agricultural Labour Share 0.55 0.31 0.39

Relative Income 0.81 0.80 1.06Relative Utility 0.58 0.58 0.81

Southern Region



The wedge between agriculture and nonagricultural wages is reduced by two-thirds while all modelparameters are kept at their 1880 values. This roughly corresponds to the improvement observed between1880 and 1990. The impact of this change on the extent of regional income convergence is observedwhen migration is permitted and when it is not. The impact on convergence is completely offset by thein-migration triggered by the improved regional labour market.

Thus, the convergence impact of improving labour markets is largest for those regions that also have

high degrees of migratory restrictions, like the US South.

The results just highlighted continue to hold for alternative measures of convergence. For instance,

the data reveals relative GDP per capita measures by region behave nearly identically to relative wages.11

In the model, the two measures are equivalent given the identical labour shares across regions. A

measure of the real convergence impacts are displayed with relative utility levels implied by the model.

There is clear convergence along this dimension as well. One should not view these results as suggesting

that migration restrictions are beneficial. On the contrary, reductions in Northeastern utility and wage

levels contribute to the convergence.


Table 2.4: Isolating the Effect of Transportation Cost Reductions

Reduce TransportationCosts by Two-Thirds

1880 Benchmark Model with Model withObserved Outcome Model Values Migration No Migration

Midwestern Region



Southern Region



The transporation cost across regions is reduced by two-thirds while all model parameters are keptat their 1880 values. This roughly corresponds to the improvement observed between 1880 and 1990.The impact of this change on the extent of regional income convergence is observed when migration ispermitted and when it is not.

2.5.2 Goods Market Frictions

In order to demonstrate that the existence of the migration option is not important for the other major

friction in the model, I conduct a similar experiment for the fraction of goods that successfully arrive at

their destination, ∆. Results are displayed in identical format in Table 2.4. Of particular note, there is no

substantial difference between the impact of transportation cost reductions when migration is permitted

or not. In addition, there are two interesting observations made here that are entirely consistent with

Tombe [2008]. First, lower transportation costs lead to peripheral emigration. This is due to cheaper

means of satisfying the subsistence consumption for core-residents, which means migrants can take

advantage of cheaper nonagricultural products while still eating a sufficient amount. Second, there is

a sizable divergence impact. Lower transportation costs lower between region price differences and,

therefore, relative wages. Specifically, core-producers of the nonagricultural good earn a higher price

while the peripheral producers earn a lower one. This latter point follows directly from the nonagricul-

11There is a difference in levels, with both the Midwest and South displaying a lower relative GDP/Capita value thanrelative wages for all years. This may be due to a higher nonlabour share of income in the Northeastern states. Data utilizedfor this exercise is from Caselli and Coleman [2001]’s Data Appendix. My calculations available upon request.


tural goods prices contain a “delivery charge” to compensate for lost goods (recall Equation 2.6). In any

case, the migration channel appears to interact mainly with the labour market and not the goods market

frictions.

2.6 Discussion

2.6.1 Effects of Transportation and Migration Costs

This section will present a few derivations to highlight the underlying channels through which trans-

portation and migration costs influence the model’s equilibrium. First, Equation 2.11 - may be com-

bined with optimal consumption allocation conditions - Equations 2.10 and 2.9 - and regional pricing

conditions - Equations 2.6 and 2.7 - to arrive at the following,

Mc−Pcf a

Mp−Ppf a

= µ−1

∆1−2τ (2.17)

where Mi ≡ Lif w

if +Li

mwim +Niri is the total nominal income of region i. Note that for ∆ = 1 and µ = 1

we have income equalization, Mp = Mc. For ∆ < 1 we have Mp > Mc. Thus, as transportation costs

fall ((1-∆) ↓) peripheral earnings also fall relative to the core. A similar argument establishes that higher

migration costs, µ , lower peripheral incomes.

2.6.2 Calibration of Transportation Cost Parameter

This parameter specifies the fraction of shipped goods that successfully arrive at the destination. In

the model, price ratios between different locations depend exclusively on this parameter. Data from

the 1887 Report of the Senate Committee on Transportation Routes shows that to transport a bushel of

wheat between Atlantic ports to Great Lake ports by rail averaged 21 cents. This is a significant charge,

given the average price of a bushel of wheat was 104 cents over in 1870s.12 Harley [1980] compiles

additional evidence on wheat and freight prices. Depending on the route, the 1880 per bushel rate to

ship wheat from Chicago to New York at that time ranged between 8 to 15 cents. Further west, the rate

was nearly double, with an additional cost to ship from Kansas City to Chicago at 11 cents. The farm

price of a bushel of wheat was 118 cents in New York, 101 in Indiana, 93 in Wisconsin, 82 in Iowa,

12Average wheat prices available within the Statistical Abstracts of the United States


and 73 in Kansas. Thus, the further west one is relative to New York, the higher the transportation costs

and the lower the wheat price. While land-route rates between Southern and Northeastern locations are

not provided, the rate to ship from Odessa, TX or New York to Liverpool, UK were nearly identical

(10.4 versus 8.6 cents, respectively). This suggests that the ocean shipping rate from Southern ports to

Northeastern ones were substantially lower than land-based routes between MW and NE. Indeed, the

wheat price was very similar in Odessa to New York, with the wholesale bushel price at 112.13 Given

these price data, I settle on an 1880 value for ∆ of 0.7 between the Midwest and Northeast and 0.95

between the South and Northeast. In addition, the annual reduction in transportation costs will be set at

1% per year, for both regional groups, roughly consistent with findings of Glaeser and Kohlhase [2004].

2.6.3 Calibration of Peripheral Labour Market Frictions

As previously established, the cost of peripheral nonagricultural labour, denoted ξ , creates a wedge

in nominal wages. This implied wedge will be used from data to determine the size of ξ . Using

unadjusted data from Caselli and Coleman [2001], who derive results for the post-1940 period, and

spliced with data from Lee et al. [1957] (also provided by Caselli and Coleman (2001)), one can find

the relative agricultural wages since 1880. Specifically, I uniformly scale down the relative earning for

agricultural workers in the Lee et al. [1957] data in order to match the census results for the year 1940.

This procedure is identical to that employed by Caselli and Coleman [2001]. The underlying cause of

the difference between the two series is that Lee et al. [1957] includes the operator’s self-employment

income, not just the pure labour earnings.

I interpret agriculture’s low relative wage as a labour market friction rather than reflecting compo-

sitional differences in worker skill, age, or ability. Caselli and Coleman [2001] use census micro-data

from 1940 through 1990 to show age, education, or other differences in worker characteristics cannot

account for agriculture’s low relative wage.14 With narrow industrial sector controls, they further show

changes in industrial composition within the nonagricultural occupations (such as declining share of

high-skill manufacturing towards low-skilled services) do not account for the observed reduction in the

wage differences with agriculture.

13The farm price was not available for Odessa at this time, so the wholesale price was used. The New York wholesaleprice, at 120 in Winter and 117 in Spring, is nearly identical to the annualised average farm price of 118, which suggests thisis an acceptable approximation.

14See Table 6 of Section VII in Caselli and Coleman [2001] for more detail.


2.6.4 Calibration of Between-Region Migration Costs

I focus in this section on the Midwest-Northeast data, though all qualitative results hold for the South-

Northeast case as well. First, I investigate the model’s ability to match 1880 data without any migration

costs whatever. Second, after establishing the inability of the model to do so, I investigate migration

costs that do not fall over time. Finally, I determine to what extent migration costs must decline to match

the 1880 and 1990 data, with plausible values of the remaining parameters.

Given the peripheral region’s low paying agricultural sector, costless migration requires that this

region must differ in its nonagricultural productivity and land endowment sufficiently to ensure that

individuals (in the model) wish to reside there. That is, in the absence of lower peripheral utility (due to

migration costs) there must productivity premium to have equal levels of utility. However, this higher

productivity will increase the relative earnings of this region compared to the core. Table 2.5, Column

(1), displays the set of parameters and the model outputs that are closest to the data. The relative income

of the peripheral region is clearly far above that found in the data, with Midwestern average earnings

118% of the Northeast compared to the true data of 81%. If Northeastern productivity parameters were

to be increased (from their currently low value of 0.78 to something closer to the Midwestern value)

then model agents would migrate away from the Midwest, leading the model to miss along the relative

employment size dimension.

Given the importance of including some sort of Utility-wedge in the model to properly match the

data, I perform another experiment that sets all initial parameter values to match the 1880 values. I

then maintain the migration costs at their initial level but allow other parameters to evolve according to

observed data and calibrate the productivity parameters to give the model the best chance of matching

1990 data. Column (2) of Table 2.5 shows that without a reduction in migration costs, a large core-

premium productivity premium is still insufficient to match the data. Intuitively, as structural change

takes place the peripheral region becomes increasingly able to achieve higher utility as labour moves

to the nonagricultural sector. However, to maintain the initial 1880 Utility wedge the core region’s

productivity must grow substantially more than the periphery’s. Specifically, the core’s annual growth

rate is 1.75% while the periphery’s is 1.5%. This differential growth, however, leads to the model failing

to match the observed degree of regional convergence, with Midwestern incomes falling behind to 64%

of the Northeast. Clearly, a model with constant migration cost is unable to match the data.


Table 2.5: Model Performance under Various Migration Cost Assumptions

(a) Data vs. Model Output

Midwest-Northeast

1880 1990

No Constant Only DecliningVariable Data Cost Data Cost Cost

(1) (2) (3)

Lp/Lc* 1.05 1.08 1.16 1.16 0.19Lp

f /Lp* 0.55 0.80 0.03 0.04 0.92wp/wc* 0.81 1.18 0.86 0.64 1.05

(b) Calibrated for Model to Match Targets

Specification Year Ω µ ASf AS

m ANm a

(1) 1880 0.58 1 1.00 1.05 0.78 0.28(2) 1990 0.38* 0.58 26.12 5.11 6.84 0.13*(3) 1990 0.38* 0.78 1.00* 0.98* 1.01* 0.13*

* denotes targets.

Finally, a model with only declining migration costs will similarly be unable to match data. Column

(3) in Table 2.5 contains the result of holding all parameters at their 1880 values but for the migration

cost parameter, which is reduced to the baseline 1990 value of 0.78 (compared to 0.58). It clearly

illustrates that, without other parameter changes, a large number of workers must migrate out of the

peripheral region to sufficiently raise its relative utility levels. This leads to a far lower size and far

higher income than is actually observed. So, reducing the regional utility wedge (migration cost) to

slightly over 50% of its original level in addition to the other parameter changes is necessary to match

the data.15

This large reduction in migration costs appear entirely in agreement with existing literature. Quan-

titative comparisons are difficult but has long been recognised that the difficulty of migration between

regions or countries is decreasing in the stock of previous migrants in the destination. The initially large

utility differential suggested by the model may also be consistent with previous estimates. Greenwood

[1975], for example, conducts an interesting literature review and points out that black migrants out of

the Southern region could experience a fifteen to twenty percent earning increase. Moreover, there are

15The utility cost of migrating out of the peripheral region is 1-µ . So, (1-0.78)/(1-0.58)=0.524


Table 2.6: Average Annual Growth Rates of Key Variables, 1880-1990

Statistic Agriculture Nonagriculture HSUS Series

Employment Growth -1.01% 2.38% Ba652,Ba653,Ba814,Ba817Producer Price Growth 1.24% 2.41% Cc66,Cc68,Cc126,Cc127Nominal GDP Growth 2.14% 5.79% Ca216,Ca136,Da1117

Real GDP Growth 0.91% 3.39% Ca211,Ca136,Cc66

A 1.51% 1.96%

Source: Historical Statistics of the United States, Millenium Online Edition. Pre-1929 Nonfarm GDP Growth is implied from em-ployment weighted average from overall GNP and farm output growth. Pre-1929 Farm GDP Growth is assumed equal to agricultural outputgrowth.

substantial psychic costs of migration, which suggests the utility wedge suggested by the model for the

Southern region may not be too ridiculous. This is especially true given that psychic costs seem inversely

related to one’s level of education [Schwartz, 1973] and the Southern African-American population was

particularly disadvantaged along this dimension.

2.6.5 Alternative Productivity Calibration

The results of the original calibration strategy will be compared to a simple Solow-residual calculation

from the Historical Statistics of the United States that are based on national-level output. In addition,

recent BEA data will also be used to make reasonable regional-specific adjustments. The decomposition

of the data will proceed for each sector s. Denoting growth rates as γ , we have γAs = γYs − (1−αs)γLs .

The decomposition assumes the land input is fixed through time, both in total and in terms of its produc-

tivity, which is reasonable given that one finds a 0.03% annual growth in the index of cropland between

1910 and 1990.16 The values used for real GDP and employment growth in each sector are taken from

over a century of data (1880-1990) from the Historical Statistics, and assumed to be representative for

the period under which the model will be simulated: 1880-1990. The values and precise sources can be

found in Table 2.6, with 1.51% annual growth for A f and 1.96% for Am.

While I lack regional data sufficient to determine growth in Am by region since 1880, I can use recent

data from the Bureau of Economic Analysis to examine whether a 10-15% faster growth in Southern

nonagricultural productivity is reasonable. Specifically, I use data for the period 1969-1997 to estimate

that Am in the South grew at 7.75% per annum while only 6.75% in the North. Moreover, the Midwest

16HSUS Series Da665


had a growth rate remarkably similar to the North, with 6.5% per annum.17 Thus, the original calibration

results of 2.22% for Southern nonagricultural productivity growth compared to 1.93% for the Northeast

appears very reasonable.

The lower rate of agricultural productivity growth found here is different, however, from the original

calibration. Using this lower value for agricultural productivity growth, the model fails to fully capture

the labour reallocation out of agricultural - though it still results in a single-digit share in 1990. The

overall conclusion regarding the impact of migration restrictions on the relationship between labour

market and income convergence is unaffected by the alternative agricultural productivity growth rate.

2.7 Conclusion

Using a general equilibrium model of structural transformation calibrated to match historical data for

various US regions, this chapter finds that barriers to regional migration magnify the impact that im-

provements in the ability of workers to switch from agricultural to nonagricultural employment have on

regional convergence. Put another way, it finds that sectoral labour market frictions have effectively no

impact on a region’s relative earnings position unless they are coupled with explicit migration restric-

tions, as one would find historically in the US South.

17Note these figures are not directly comparable to earlier results, given there is no adjustment for price increases here.


Tables and Figures

Figure 2.1: US Census Regions

Source: Figure 2 of U.S. Business Travel, October 2003, Bureau of Transportation Statistics. Provided as a publicservice by the Research and Innovation Technology Administration of the BTS and considered public information and may bedistributed or copied. See http://www.rita.dot.gov/disclaimer.html.

Figure 2.2: Agricultural Share of Employment, by Region


Figure 2.3: Relative Agricultural Wages, by Region

.2.4

.6.8

1880 1900 1920 1940 1960 1980 2000year

South MidWest

Source: Lee et al. (1957), 1880−1920; Caselli and Coleman (2001), 1940−90; IPUMS Census, 2000

Relative Agricultural Wages

®

Figure 2.4: Overall Average Wage Relative to Northeast, by Region


Figure 2.5: Employment Relative to Northeast, by Region

Chapter 3

Structural Transformation and Regional

Convergence in Canada, 1901-1981

Abstract

In the early phases of economic growth, farmers switch to higher-paying nonagricultural employmentand average wages in agricultural regions may grow relative to industrial regions. Typical studies ofregional convergence in Canada do not consider a role for structural change, despite a clear geographicseparation between Canada’s industrial and agricultural regions. As structural change proceeds, agri-cultural employment shrinks and relative farm earnings rise, which may disproportionately benefit agri-cultural regions. Using early census volumes, I construct a unique dataset of census-division levelwage and employment levels in both agriculture and nonagriculture between 1901 and 1981. Withthis data, I quantify the contribution of structural change to Canadian regional income convergence.Specifically, I follow Caselli and Coleman [2001] and decompose average wage convergence across ge-ographic regions into two structural-change components (labour reallocation and inter-industry factors)and a residual (inter-regional factors). I find evidence for convergence of average overall earnings but,contrary to the American experience, this convergence is primarily due to inter-regional factors withstructural change playing little role.

70

CHAPTER 3. STRUCTURAL CHANGE AND CANADIAN CONVERGENCE 71

3.1 Introduction

Regional income inequality in Canada is high by international standards and is a central theme of

Canadian economic policy. Existing research finds demographic factors, work patterns, resource en-

dowments, or industrial structure have far less influence on inequality than labour productivity and

employment rates do [McInnis, 1968, Melvin, 1987, Coulombe and Lee, 1995, Lee, 1996, Coulombe

and Day, 1999]. During early phases of economic development, agricultural employment and output

shrink dramatically relative to other industries. This transformation may lead to changes in earnings

within agricultural regions relative to other parts of the country. In this chapter, I examine convergence

mechanisms related to structural change previously unexamined in the Canadian literature: (1) increased

agricultural wages, which disproportionately benefit poor agricultural regions, and (2) increased non-

agricultural employment, which has lower cross-region wage differences. I find evidence of overall

earnings convergence between 1901 and 1981 but structural change plays surprisingly little role.

To accurately contrast agricultural versus non-agricultural income patterns, I depart in two addi-

tion respects from the existing literature. First, I disaggregate sectoral earnings data not by province

but by counties within the industrial Windsor/Quebec-City corridor relative to those counties outside it.

Second, I gathered historical census data on sectoral earnings and employment, along with a number

of other relevant census-division level data, beginning in 1901. Existing studies of Canadian conver-

gence using census data begin in 1926, with the start of the five-year census cycle. Constructing a

usable dataset from earlier census records is necessary to capture a period of rapid structural change in

the Canadian economy. I perform accounting exercises on these data, following Caselli and Coleman

[2001], and decompose changes in regional earnings differences into changes in labour allocation, sec-

toral earnings, and region-specific factors. My results consistently indicate sectoral labour reallocation

and sectoral earnings convergence are not important drivers of convergence across Canada’s regions. To

be precise, of the thirteen percent reduction in the earnings gap between the Widsor-Quebec City corri-

dor and the rest-of-Canada nearly twelve resulted from region-specific factors. In contrast, the United

States experienced a more even split between the three factors - that is, inter-regional factors account

for only roughly one-third of the convergence [Caselli and Coleman, 2001].


3.2 Convergence Decomposition

Caselli and Coleman [2001] provide a means of decomposing regional convergence into component

parts. A brief derivation is provided here and further details may be found in Appendix B of their paper.

Given that a region’s average wage is a labour-force weighted average of the wages of its specific

sectors, it is clear that

wrt = wr

ag,tLrag,t +wr

na,tLrna,t

= wrag,tL

rag,t +wr

na,t(1−Lrag,t) (3.1)

where wrag,t , wr

na,t , Lrag,t , and Lr

na,t are, respectively, the average agricultural wage, average non-agricultural

wage, agricultural labour-force share, and non-agricultural labour-force share, for region r at date t.

In order to investigate a regional deviation from average, Equation 3.1 may be modified by adding a

quantity equal to zero on the right-hand side. That is,

wrt = wr

ag,tLrag,t +wr

na,tLrna,t +wag,tLr

ag,t −wag,tLrag,t +wna,tLr

na,t −wna,tLrna,t

= (wrag,t −wag,t)Lr

ag,t +(wrna,t −wna,t)Lr

na,t +wag,tLrag,t +wna,tLr

na,t (3.2)

As discussed earlier, the geographic groupings in this chapter are “Peripheral”, P, and “Core”, C. Sub-

stituting these labels into Equation 3.2 and taking their difference relative to the national average, one

finds that

wPt −wC

t

wt=

wPag,t −wag,t

wtLP

ag,t +wP

na,t −wna,t

wt(1−LP

ag,t)

−wC

ag,t −wag,t

wtLC

ag,t −wC

na,t −wna,t

wt(1−LC

ag,t)

+wag,t −wna,t

wt(LP

ag,t −LCag,t) (3.3)

Finally, one can take the difference between adjacent time periods and rearrange to arrive at the follow-

ing decomposition (which is found as Equation B3 in Caselli and Coleman [2001])

wPt −wC

t

wt−

wPt−1−wC

t−1

wt−1= ∆ω

Pag,t · LP

ag,t +∆ωPna,t · (1− LP

ag,t)


−∆ωCag,t · LC

ag,t −∆ωCna,t · (1− LC

ag,t)

+ωPt ·∆LP

ag,t − ωCt ·∆LC

ag,t

+∆ωt · (LPag,t − LC

ag,t) (3.4)

Where ωt =wag,t−wna,t

wt, ωr

t =wr

ag,t−wrna,t

wt, and ωr

j,t =wr

j,t−w j,t

wt, for r ∈ P,C and j ∈ ag,na. The first

two lines capture the extent to which average sectoral wages within a region change, the third line

captures the labour reallocation between sectors within a region, and the fourth line captures the change

in national average sectoral wages. Panel data on average sectoral wages and labour-force shares for

each region and each sector is required to employ Equation 3.4.

3.2.1 Core vs. Peripheral Classification

To clearly distinguish agricultural from non-agricultural regions of Canada, I divide census divisions

into “core” and “peripheral” groupings. Following closely the economic geography literature, I classify

census divisions within historically industrial areas as “core.”1 Figures 3.8b to 3.8c, extracted from the

Seventh Census of Canada, illustrate the economic geography of Canada. The central region bordering

the St. Lawrence River, Lake Ontario, and Southern Ontario through to Windsor, historically constitutes

the industrial heartland, while all other regional economies are typically resource extractive in nature.

The internal connections within the heartland regions of Canada are of very high quality relative to the

West or the Maritimes. The Grand Trunk Railway, for example, connected all major economic centres

of the core region but did not expand until long after regional industrial structures were established.

Finally, given adjustments to census division boundaries between each major census, I examine the

sensitivity of the main results to various alternative mappings of divisions to core or peripheral groups

in Section 3.5.3.

3.3 Data

I manually collect data from published historical census volumes between 1901 and 1981 [Domin-

ion Bureau of Statistics, 1901–1981]. Canadian census aggregates are reported for a wide-variety of

variables at levels of geographic aggregation ranging from individual cities or census subdivisions to

1This classification scheme closely follows that from McCann [1998].


Table 3.1: Key Statistics of the Data

1901 1911 1921 1931 1941 1951 1961 1971 1981

Without Farm Operators

Relative Agricultural Wage 0.5000 0.5150 0.7414 0.5556 0.5175 0.5700 0.5101 0.3064 0.3106

Agricultural Labour Share 0.0910 0.0745 0.1131 0.0712 0.0642 0.0343 0.0221 0.0194 0.0184

Periphery/Core Relative Wage 0.8532 0.9776 0.9640 0.9274 0.8864 0.9012 0.9157 0.9046 0.9776

Periphery Relative Ag. Wage 0.4609 0.4765 0.8861 0.6064 0.5276 0.5510 0.4568 0.2695 0.2893

Periphery Ag. Labour Share 0.1197 0.1000 0.1431 0.0958 0.0930 0.0482 0.0289 0.0265 0.0229

Core Relative Ag. Wage 0.5401 0.5752 0.5469 0.4929 0.5281 0.6174 0.5846 0.3620 0.3382

Core Ag. Labour Share 0.0764 0.0530 0.0855 0.0502 0.0427 0.0240 0.0171 0.0142 0.0148

With Farm Operators

Relative Agricultural Wage n/a n/a n/a 0.2117 0.6665 1.4951 0.5434 0.5808 0.6621


Periphery/Core Relative Wage n/a n/a n/a 0.7173 0.7944 1.0103 0.8899 0.8865 1.0367

Periphery Relative Ag. Wage n/a n/a n/a 0.1379 0.6775 1.7509 0.5151 0.6533 0.7332


Core Relative Ag. Wage n/a n/a n/a 0.3354 0.7139 1.1951 0.6072 0.5016 0.5175


Outlines key data provided or inferred from the decadal Census of Canada [Dominion Bureau of Statistics, 1901–1981].

provinces and the nation as a whole, depending on the variable and year. Certain variables, however,

are not reported in certain years and a number of inferences based on reported data must be performed.

Table 3.14 and Section 3.3.1 outline inferences I make. Beyond collecting and digitizing relevant vari-

ables, extensive manipulation of the data must be preformed. Most importantly, linking census divisions

across time requires tracking name changes, geographic boundary changes, and addition of new divi-

sions as the nation expanded.2

3.3.1 Variable Construction

In Table 3.14, I list relevant variables the Census explicitly reports and variables I must infer from other

sources. Selected variable construction procedures and data quality concerns are detailed below.

2This issue is particularly relevant for linking census-years between 1901 and 1931 censuses, since Saskatchewan andAlberta became Provinces in 1905, and a number of division name changes occurred.


3.3.1.1 Labour-Force Share

Given the lack of division level occupational data for certain years, the number of labourers employed

was constructed as follows. Labour employed in agriculture for cities with more than 30,000 residents

were explicitly reported and are used in conjunction with provincial aggregates to infer, based on census

division crop acreage, the labour employed in agriculture for all divisions. Intuitively, a census division

with a large city contained within is dominated by the city itself. Subtracting the urban from the province

totals gives a residual to be distributed amongst remaining divisions according to their crop acreage.3 All

occupational totals are distributed according to a division’s population. Specifically, for each non-city

division, i, in province, p, the following was calculated

Liag,t =

ω ip · (AGp−∑

K p

j=1 AG j)

ρ ip · (ALLp−∑

K p

j=1 ALL j)(3.5)

Where K p is the number of separately reported cities, j, in province, p; AG and ALL denote the num-

ber of farm labourers and total employment, respectively; ρ ip, is the share of the province’s residual

population contained in division i; and ω ip, the share of the province’s residual crop land contained

in division i. This procedure is valid if agriculture production functions are identical across divisions

within a province and have identical land-labour ratios, which follows under Cobb-Douglas production

with identical factor input shares with farmers facing identical input prices.

Subsequent to the above calculation, LPag,t and LC

ag,t are calculated as weighted averages,

LPag,t =

1ALLP

∑l∈ℜ

[ALLl ·Llag,t ]

LCag,t =

1ALLC

∑l /∈ℜ

[ALLl ·Llag,t ]

where ℜ is the set of divisions classified as Peripheral in year t and ALL is defined as before, with

subscripts P and C indicating totals for peripheral and core.

3Justifiable if the land-labour ratio is identical between farm operations within a province


3.3.1.2 Average Agricultural Earnings

Unfortunately, farm labour compensation is composed of cash and board. Certain census volumes

distinguish between the two forms of compensation while others do not. It is unclear whether “value

of farm labour” represents total monetary payments alone. Ideally, one would include both forms of

compensation in each year to prevent finding a rising average agricultural wage from mere monetization

of compensation schemes.

3.3.1.3 Average Non-Agricultural Earnings

Determining division-level average non-agricultural earnings will follow an approach similar to agri-

cultural wages. Establishment level data is given in certain years for number of employees hired and

total wage compensation paid. This gives an average annual earnings level by division. In other years

the average annual earnings of wage earners is reported directly for each division. From this, I infer an

average for non-agricultural wage earners algebraically from the labour force shares and overall average

earnings as follows:

Ena =Eall−Lag ·Eag

1−Lag

3.4 Results

Figures 3.1 and 3.4, which respectively include and exclude the Western provinces, illustrate the regional

convergence decomposition. More precise detail is provided in Tables 3.2, 3.7, and 3.13. The log-

earnings plots in the top of each figure illustrate earnings in agriculture, non-agriculture, and in total

over time; core, peripheral, and national earnings are reported with separate lines. Convergence is

observed when the gap between core and peripheral earnings shrinks. To visualize convergence, I plot

changes in earnings gaps for each of the three channels detailed in Equation 3.4 in the panel below the

earnings plots.

The convergence plots may be understood as follows. The solid line represents the percentage point

reduction in the overall Core-Periphery wage-gap between 1901 and the year ending indicated by the

horizontal axis. It follows that a positive slope indicates adjacent period convergence and a negative

slope indicates the opposite. Similarly, the red-dotted line is the portion of the wage-gap change due


Figure 3.1: Earnings and Convergence Plot

(a) Convergence Impact of Each Channel, Over Time

1900 1920 1940 1960 19804

5

6

7

8

9

10

Log−

Ear

ning

sAgricultural Earnings

1900 1920 1940 1960 19804

5

6

7

8

9

10Non−Agricultural Earnings

PeripheryCoreOverall

1900 1920 1940 1960 19804

5

6

7

8

9

10Overall Earnings

1900 1920 1940 1960 1980−0.05

0

0.05

0.1

0.15Inter−Regional

Gap

Poi

nt R

educ

tion

1900 1920 1940 1960 1980−0.05

0

0.05

0.1

0.15Labour Reallocation

1900 1920 1940 1960 1980−0.05

0

0.05

0.1

0.15Inter−Sectoral

(b) Share of Overall Convergence, by Channel

1900 1910 1920 1930 1940 1950 1960 1970 1980 1990-0.2

0

0.2

0.4

0.6

0.8

1

1.2Relative Importance of Convergence Factors

Inter-Regional

Labour Reallocation

Inter-Sectoral

Earnings plots illustrate agriculture, non-agriculture, and total over time; core, peripheral, and national earnings are reported withseparate lines. Convergence is observed when the gap between core and peripheral earnings shrinks. Below those, the convergence plotsrepresents the percentage point reductions in overall Core-Periphery wage gap between 1901 and the year ending indicated by the horizontalaxis. Positive slopes indicate adjacent period convergence and negative slopes indicate divergence. The red-dotted line is the portion of thewage-gap change due to each respective convergence factor. The final plot presents the proportion of the total wage-gap change resulting fromeach individual factor over time.


Table 3.2: Convergence Decompositions

Overall Measure 1911 1921 1931 1941 1951 1961 1971 1981

1901 0.1318 0.1179 0.0793 0.0351 0.0514 0.0670 0.0552 0.13191911 - -0.0140 -0.0525 -0.0967 -0.0804 -0.0649 -0.0767 0.0001

1921 - -0.0385 -0.0827 -0.0664 -0.0509 -0.0627 0.0140

1931 - -0.0442 -0.0279 -0.0123 -0.0242 0.0526

1941 - 0.0163 0.0319 0.0200 0.0968

1951 - 0.0155 0.0037 0.0804

1961 - -0.0118 0.0649

1971 - 0.0767

Inter-Regional 1911 1921 1931 1941 1951 1961 1971 1981

1901 0.1315 0.1078 0.0792 0.0379 0.0383 0.0467 0.0398 0.11301911 - -0.0271 -0.0552 -0.0956 -0.0959 -0.0875 -0.0951 -0.0219

1921 - -0.0226 -0.0635 -0.0582 -0.0481 -0.0540 0.0193

1931 - -0.0396 -0.0360 -0.0259 -0.0332 0.0401

1941 - 0.0025 0.0122 0.0046 0.0779

1951 - 0.0103 0.0016 0.0751

1961 - -0.0092 0.0645

1971 - 0.0737

Labour Reallocation 1911 1921 1931 1941 1951 1961 1971 1981

1901 -0.0006 -0.0029 -0.0028 -0.0040 0.0102 0.0195 0.0203 0.02331911 - 0.0007 0.0006 -0.0013 0.0131 0.0224 0.0243 0.0273

1921 - -0.0060 -0.0067 -0.0013 0.0052 0.0065 0.0089

1931 - -0.0027 0.0073 0.0146 0.0161 0.0188

1941 - 0.0115 0.0195 0.0219 0.0247

1951 - 0.0063 0.0070 0.0095

1961 - -0.0001 0.0025

1971 - 0.0030

Inter-Sectoral 1911 1921 1931 1941 1951 1961 1971 1981

1901 0.0009 0.0130 0.0029 0.0012 0.0029 0.0008 -0.0050 -0.00451911 - 0.0124 0.0021 0.0003 0.0024 0.0002 -0.0059 -0.0054

1921 - -0.0099 -0.0125 -0.0070 -0.0079 -0.0153 -0.0142

1931 - -0.0019 0.0008 -0.0010 -0.0071 -0.0064

1941 - 0.0023 0.0001 -0.0064 -0.0059

1951 - -0.0011 -0.0049 -0.0042

1961 - -0.0025 -0.0020

1971 - 0.0000

Each panel details the percent (in decimal form) change in the gap between core (industrial) and peripheral (agricultural) regions ofCanada. Each row represents a different starting date and each column represents an ending date. For example, 0.1319 in the top-right cellmeans the wage gap shrank by 13.19% between 1901 and 1981 - of which, 11.3% from inter-regional factors and the remainder from structuralchange.


to each respective convergence factor. Overall, they illustrate that regional convergence has taken place

over the past century for primarily inter-regional factors - that is, for reasons distinct from structural

change. To emphasize this point, the final plot in the figure presents the proportion of the total wage-gap

change resulting from each individual factor.

This is in sharp contrast to the results obtained by Caselli and Coleman [2001]. Their results for

the United States indicate a minor role for inter-regional factors. Thus, it appears structural change has

played a smaller role for the Canadian regional convergence experience than for the United States.

3.5 Discussion and Sensitivity of Results

3.5.1 Direct Comparison to US Experience

Using data from Caselli and Coleman [2001] for the years 1940 to 1990, I generate the following results

Table 3.3: Results from Table 2 of Caselli and Coleman [2001]

Period Total Inter-Regional Labor Reallocation Inter-Sectoral

1940–90 .312 .132 .110 .070

As illustrated in Figure 3.2 and Table 3.4, this analysis duplicates the Caselli and Coleman [2001]

numbers exactly. Note that the large, bolded numbers in Table 3.4 correspond to my calculations of their

reported numbers, which match perfectly with theirs, while the remaining numbers are my calculations

for the other annual pairs. For example, there was divergence in the 1970s. In contrast to this decade,

the 1980s and 1940s were periods of particularly high convergence.

Finally, I conducted the analysis on the Canadian data excluding Quebec to see if the non-Western

convergence pattern was affected; it was not. Including the western states in the US data generates far

greater volatility in the results, as it does for Canada. Thus, excluding western provinces makes the

Canadian analysis more comparable to the North-South convergence in the US. A selected comparison

is presented here in Table 3.5. Clearly, inter-regional factors are far more relevant to the Canadian

experience.


Table 3.4: Convergence Decompositions (US Data from Caselli and Coleman [2001])

Overall Measure 1950 1960 1970 1980 1990

1940 0.2163 0.2498 0.3005 0.4111 0.31211950 - 0.0334 0.0842 0.1948 0.0958

1960 - 0.0508 0.1613 0.0623

1970 - 0.1106 0.0116

1980 - -0.0990

Inter-Regional 1950 1960 1970 1980 1990

1940 0.1160 0.0963 0.1225 0.2292 0.13221950 - -0.0230 0.0017 0.1070 0.0084

1960 - 0.0248 0.1293 0.0298

1970 - 0.1042 0.0043

1980 - -0.1001

Labour Reallocation 1950 1960 1970 1980 1990

1940 0.0731 0.1169 0.1213 0.1122 0.11011950 - 0.0470 0.0602 0.0582 0.0578

1960 - 0.0182 0.0208 0.0215

1970 - 0.0040 0.0051

1980 - 0.0010

Inter-Sectoral 1950 1960 1970 1980 1990

1940 0.0272 0.0366 0.0567 0.0697 0.06981950 - 0.0095 0.0223 0.0296 0.0295

1960 - 0.0078 0.0112 0.0110

1970 - 0.0023 0.0022

1980 - 0.0001

Each panel details the percent (in decimal form) change in the gap between Northern and Southern USstates. Each row represents a different starting date and each column represents an ending date. For example,0.3121 in the top-right cell means the wage gap shrank by 31.21% between 1940 and 1990 - of which, 13.22%from inter-regional factors.


Figure 3.2: Earnings and Convergence Plot (Using the US Data)

1940 1950 1960 1970 1980 19906

7

8

9

10

11Lo

g−E

arni

ngs

Agricultural Earnings

1940 1950 1960 1970 1980 19906

7

8

9

10


SouthNorthOverall

1940 1950 1960 1970 1980 19906

7

8

9

10

11Overall Earnings

1940 1950 1960 1970 1980 19900

0.1

0.2

0.3

0.4

0.5Inter−Regional

Gap

Poi

nt R

educ

tion

1940 1950 1960 1970 1980 19900

0.1

0.2

0.3

0.4


1940 1950 1960 1970 1980 19900

0.1

0.2

0.3

0.4

0.5Inter−Sectoral

Earnings plots illustrate agriculture, non-agriculture, and total over time; core, peripheral, and national earnings are reported withseparate lines. Convergence is observed when the gap between core and peripheral earnings shrinks. Below those, the convergence plotsrepresents the percentage point reductions in overall Core-Periphery wage gap between 1901 and the year ending indicated by the horizontalaxis. Positive slopes indicate adjacent period convergence and negative slopes indicate divergence. The red-dotted line is the portion of thewage-gap change due to each respective convergence factor.

Table 3.5: Selected Comparison of US and Canadian Experience

Structural Change

Period Total Inter-Regional Labor Reallocation Inter-Sectoral

US Data 1940–80 .3121 .1322 .1122 .0697

Percent of Total (100) (42.34) (35.95) (22.33)

CDN Data 1941–81 .0968 .0779 .0247 -.0059

Percent of Total (100) (80.48) (25.52) (-6.00)

Compares the reduction in earnings gaps between Northern and Souther US states to the core (indus-trial) and peripheral (agricultural) regions of Canada for approximately the same time period. Interpretationof each number is as follows: 0.3121 means the North-South wage gap shrank by 31.21% between 1940and 1980. For Canada, the comparable number is 0.0968, which implies the gap in core-peripheral wagesshrank by almost 10% over the same time period. In parentheses, the fraction of convergence accountedfor by each channel. Strucutural change contributes less to convergence in Canadian than it does for theUS.


Figure 3.3: Foreign Born Population Share

(a) West Excluded

1900 1910 1920 1930 1940 1950 1960 1970 1980 19900.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

Periphery

Core

(b) West Included

1900 1910 1920 1930 1940 1950 1960 1970 1980 19900.12

0.14

0.16

0.18

0.2

0.22

0.24

0.26

0.28

0.3

Periphery

Core

Plots the fraction of core and peripheral populations that are foreign born. The evolution of this foreign born share over time is largelysimilar for core and peripheral regions of Canada if the Western provinces are excluded. When included, the early years of the analysis see amassive inflow of labour to Western Canada.

3.5.2 Exclusion of Western Provinces

Large external inflows of labour may drive wage changes independently of domestic factors. The West-

ern regions of Canada, illustrated in Figure 3.8c, experienced massive inflows of labour during the first

few decades of the 20th-Century - see Figure 3.3. The inclusion of Western provinces is not driving any

of the results of peripheral-core convergence.


Figure 3.4: Earnings and Convergence Plot - West Excluded


1900 1920 1940 1960 19804

5

6

7

8

9

10

Log−

Ear

ning

sAgricultural Earnings

1900 1920 1940 1960 19804

5

6

7

8

9


1900 1920 1940 1960 19804

5

6

7

8

9

10Overall Earnings

1900 1920 1940 1960 1980−0.05

0

0.05

0.1

0.15

0.2

Inter−Regional

Gap

Poi

nt R

educ

tion

1900 1920 1940 1960 1980−0.05

0

0.05

0.1

0.15

0.2

Labour Reallocation

1900 1920 1940 1960 1980−0.05

0

0.05

0.1

0.15

0.2

Inter−Sectoral


(b) Share of Overall Convergence, by Channel

1900 1910 1920 1930 1940 1950 1960 1970 1980 1990-0.2

0

0.2

0.4

0.6

0.8

1

1.2Relative Importance of Convergence Factors

Inter-Regional

Labour Reallocation

Inter-Sectoral

Earnings plots illustrate agriculture, non-agriculture, and total over time; core, peripheral, and national earnings are reported withseparate lines. Convergence is observed when the gap between core and peripheral earnings shrinks. Below those, the convergence plotsrepresents the percentage point reductions in overall Core-Periphery wage gap between 1901 and the year ending indicated by the horizontalaxis. Positive slopes indicate adjacent period convergence and negative slopes indicate divergence. The red-dotted line is the portion of thewage-gap change due to each respective convergence factor. The final plot presents the proportion of the total wage-gap change resulting fromeach individual factor over time.


Table 3.6: Key Features of the Data - West Excluded

1901 1911 1921 1931 1941 1951 1961 1971 1981

Without Farm Operators




Periphery Relative Ag. Wage 0.4579 0.4629 0.4156 0.4640 0.3758 0.3268 0.3839 0.2903 0.3197


Core Relative Ag. Wage 0.5401 0.5752 0.5469 0.4929 0.5281 0.6174 0.5846 0.3620 0.3382


With Farm Operators




Selected subset of variables from Table 3.1 with Western provinces excluded.


Table 3.7: Convergence Decompositions - Excluding the West

Overall Measure 1911 1921 1931 1941 1951 1961 1971 1981

1901 0.0568 0.1923 0.1377 0.1731 0.1633 0.1683 0.1912 0.2154

1911 - 0.1355 0.0809 0.1163 0.1066 0.1115 0.1344 0.1587

1921 - -0.0546 -0.0192 -0.0289 -0.0240 -0.0011 0.0232

1931 - 0.0354 0.0257 0.0306 0.0535 0.0777

1941 - -0.0097 -0.0048 0.0181 0.0424

1951 - 0.0049 0.0279 0.0521

1961 - 0.0229 0.0471

1971 - 0.0242

Inter-Regional 1911 1921 1931 1941 1951 1961 1971 1981

1901 0.0587 0.1869 0.1368 0.1766 0.1584 0.1623 0.1904 0.2152

1911 - 0.1299 0.0772 0.1186 0.1010 0.1035 0.1300 0.1544

1921 - -0.0526 -0.0127 -0.0329 -0.0299 -0.0019 0.0229

1931 - 0.0421 0.0257 0.0282 0.0541 0.0783

1941 - -0.0182 -0.0166 0.0092 0.0334

1951 - -0.0002 0.0237 0.0475

1961 - 0.0232 0.0467

1971 - 0.0235

Labour Reallocation 1911 1921 1931 1941 1951 1961 1971 1981

1901 -0.0024 0.0054 0.0012 -0.0030 0.0046 0.0057 0.0020 0.0015

1911 - 0.0059 0.0044 -0.0012 0.0056 0.0080 0.0061 0.0059

1921 - -0.0019 -0.0062 0.0038 0.0058 0.0012 0.0006

1931 - -0.0067 -0.0005 0.0022 - -0.0001

1941 - 0.0076 0.0112 0.0102 0.0103

1951 - 0.0051 0.0053 0.0056

1961 - 0.0002 0.0007

1971 - 0.0007

Inter-Sectoral 1911 1921 1931 1941 1951 1961 1971 1981

1901 0.0004 -0.0001 -0.0003 -0.0005 0.0003 0.0003 -0.0012 -0.0012

1911 - -0.0003 -0.0006 -0.0010 0.0001 -0.0016 -0.0016

1921 - -0.0001 -0.0002 0.0002 0.0001 -0.0003 -0.0003

1931 - - 0.0004 0.0002 -0.0005 -0.0005

1941 - 0.0008 0.0007 -0.0013 -0.0013

1951 - -0.0011 -0.0010

1961 - -0.0004 -0.0003

1971 - -0.0000

Each panel details the percent (in decimal form) change in the gap between core (industrial) and peripheral (agricultural) regions ofCanada. Each row represents a different starting date and each column represents an ending date. For example, 0.2154 in the top-rightcell means the wage gap shrank by 21.54% between 1901 and 1981 - of which, 21.52% from inter-regional factors and the remainder fromstructural change.


3.5.3 Alternative Regional Classifications of Census-Divisions

To determine if the results are sensitive to which region a census division was classified in, I generated

and plotted in an overlapping fashion in Figure 3.5 different combinations of which states were classified

as core or peripheral. This procedure switched bordering census divisions between one group and the

other, in various combinations. The overall pattern of convergence is not sensitive to how census-

divisions are classified into core and peripheral categories.

Figure 3.5: All North-South Combinations

1940 1960 19800

0.1

0.2

0.3

0.4

0.5Inter-Regional

Gap P

oin

t R

eduction

1940 1960 19800

0.1

0.2

0.3

0.4


1940 1960 19800

0.1

0.2

0.3

0.4

0.5Inter-Sectoral

These reproduce convergence plots from Figure 3.1 for various combinations of census division classifications, where counties borderingthe core-peripheral boundries are included in one or the other group (in various combinations). These plots represents the percentage pointreductions in overall Core-Periphery wage gap between 1941 and the year ending indicated by the horizontal axis. Positive slopes indicateadjacent period convergence and negative slopes indicate divergence. The red-dotted line is the portion of the wage-gap change due to eachrespective convergence factor. The results indicate my main results are not sensitive to group classification.

3.5.4 Agricultural Employment Definition

This subsection explores if the main results depend on whether the definition of farm labour includes

farm owner-operators, instead of hired wage labour only. Table 3.8 illustrates the large difference in

labour-force shares compared to the restricted definition.

Table 3.8: Labour Force Shares - With Farm Operators

1901 1911 1921 1931 1941 1951 1961 1971 1981

Peripheral 0.3880 0.3876 0.4209 0.3558 0.3406 0.2245 0.1432 0.0878 0.0633

Core 0.2850 0.3012 0.2993 0.2171 0.1712 0.1055 0.0691 0.0383 0.0287

Overall 0.3223 0.3435 0.3627 0.2882 0.2536 0.1602 0.1021 0.0599 0.0440

Displays the fraction of the Canadian labour force employed by the agricultural sector, including owner-operators.

The analysis, both with and without western provinces, is illustrated as earlier in Figures 3.6 and


Figure 3.6: Earnings and Convergence Plot, Including Farm Owner-Operators


1940 1950 1960 1970 19804

5

6

7

8

9

10

Log−

Ear

ning

s


1940 1950 1960 1970 19804

5

6

7

8

9



1940 1950 1960 1970 19804

5

6

7

8

9

10Overall Earnings

1940 1950 1960 1970 1980−0.1

0

0.1

0.2

0.3

0.4Inter−Regional

Gap

Poi

nt R

educ

tion

1940 1950 1960 1970 1980−0.1

0

0.1

0.2

0.3


1940 1950 1960 1970 1980−0.1

0

0.1

0.2

0.3

0.4Inter−Sectoral

These plots replicate those in Figure 3.1 when agricultural owner-operators are included. Earnings plots illustrate agriculture, non-agriculture, and total over time; core, peripheral, and national earnings are reported with separate lines. Convergence is observed when the gapbetween core and peripheral earnings shrinks. Below those, the convergence plots represents the percentage point reductions in overall Core-Periphery wage gap between 1901 and the year ending indicated by the horizontal axis. Positive slopes indicate adjacent period convergenceand negative slopes indicate divergence. The red-dotted line is the portion of the wage-gap change due to each respective convergence factor.

3.7. Data availability requires I restrict the time period to between 1931 and 1981. Inter-Regional

factors are still the most important of the three convergence factors by far. The number of farm owner

operators is known from 1901, allowing one to determine agricultural employment shares, but earnings

data by census-divisions does not begin until 1931. This result can be extended to earlier time periods

by assuming farm owners and family members earn the same return as recorded by farm labourers.

The results of that exercise are omitted but the main conclusion is maintained: inter-regional factors

dominate.


Figure 3.7: Earnings and Convergence Plot, Including Farm Owner-Operators - West Excluded


1940 1950 1960 1970 19804

5

6

7

8

9

10

Log−

Ear

ning

s


1940 1950 1960 1970 19804

5

6

7

8

9



1940 1950 1960 1970 19804

5

6

7

8

9

10Overall Earnings

1940 1950 1960 1970 1980−0.05

0

0.05

0.1

0.15

0.2

0.25

Inter−Regional

Gap

Poi

nt R

educ

tion

1940 1950 1960 1970 1980−0.05

0

0.05

0.1

0.15

0.2

0.25

Labour Reallocation

1940 1950 1960 1970 1980−0.05

0

0.05

0.1

0.15

0.2

0.25

Inter−Sectoral

These plots replicate those in Figure 3.1 when agricultural owner-operators are included and Western provinces are excluded. Earningsplots illustrate agriculture, non-agriculture, and total over time; core, peripheral, and national earnings are reported with separate lines.Convergence is observed when the gap between core and peripheral earnings shrinks. Below those, the convergence plots represents thepercentage point reductions in overall Core-Periphery wage gap between 1901 and the year ending indicated by the horizontal axis. Positiveslopes indicate adjacent period convergence and negative slopes indicate divergence. The red-dotted line is the portion of the wage-gap changedue to each respective convergence factor.


3.6 Concluding Remarks

Researchers have studied the contribution of structural change (rising agricultural wages and labour

switching towards non-agricultural activities) to regional convergence in other countries but not yet

for Canada. For instance, Caselli and Coleman [2001] find structural change accounts for most of the

convergence between the Northern and Southern US states between 1880 and 1980. In this chapter,

I depart from the existing Canadian convergence literature and examine the convergence impact of:

(1) increased agricultural wages, which disproportionately benefit poor agricultural regions, and (2)

increased non-agricultural employment, which has lower cross-region wage differences. To perform the

analysis, I collect earlier county-level Census data than is typically employed for Canadian convergence

studies. I find evidence of overall earnings convergence between 1901 and 1981 but structural change

plays surprisingly little role.


Tables and Figures

Table 3.9: Classifications of All 1901-1981 Census Divisions (P=Peripheral, C=Core)

Division Name Province Classification Division Name Province ClassificationAlberta AB P Calgary AB P

Division 1 AB AB P Division 2 AB AB PDivision 3 AB AB P Division 4 AB AB PDivision 5 AB AB P Division 6 AB AB PDivision 7 AB AB P Division 8 AB AB PDivision 9 AB AB P Division 10 AB AB PDivision 11 AB AB P Division 12 AB AB PDivision 13 AB AB P Division 14 AB AB PDivision 15 AB AB P Division 16 AB AB PDivision 17 AB AB P Edmonton AB P

MacLoed AB P Medicine Hat AB PRed Deer AB P Strathcona AB P

Victoria AB AB P Alberni-Clayquot BC PBulkley-Nechako BC P Burrard BC P

Capital BC P Cariboo No. 1 BC PCariboo No. 2 BC P Central Coast BC P

Central Fraser Valley BC P Central Kootenay BC PCentral Okanagan BC P Columbia-Shuswap BC PComox-Alberni BC P Comox-Atlin BC P

Comox-Strathcona BC P Cowichan Valley BC PDewdney-Alouette BC P Division 1 BC BC P

Division 2 BC BC P Division 3 BC BC PDivision 4 BC BC P Division 5 BC BC PDivision 6 BC BC P Division 7 BC BC PDivision 8 BC BC P Division 9 BC BC PDivision 10 BC BC P East Kootenay BC PFraser Valley BC P Fraser-Cheam BC P

Fraser-Fort George BC P Greater Vancouver BC PKitimat-Stikine BC P Kootenay BC P

Kootenay Boundary BC P Kootenay East BC PKootenay West BC P Mount Waddington BC PNanaimo No. 1 BC P Nanaimo No. 2 BC P

New Westminster BC P North Okanagan BC POcean Falls BC P Okanagan-Similkameen BC P

Peace River - Liard BC P Powell River BC PSkeena BC P Skeena A BC P

Skeena-Queen Charlotte BC P Squamish-Lillooet BC PStikine BC P Sunshine Coast BC P

Thompson-Nicola BC P Vancouver BC PVancouver Centre BC P Vancouver South BC P

Victoria BC BC P Yale BC PYale and Cariboo BC P Brandon MB P

Dauphin MB P Division 1 MB MB PDivision 2 MB MB P Division 3 MB MB PDivision 4 MB MB P Division 5 MB MB PDivision 6 MB MB P Division 7 MB MB PDivision 8 MB MB P Division 9 MB MB P

Division 10 MB MB P Division 11 MB MB PDivision 12 MB MB P Division 13 MB MB PDivision 14 MB MB P Division 15 MB MB PDivision 16 MB MB P Division 17 MB MB PDivision 18 MB MB P Division 19 MB MB PDivision 20 MB MB P Division 21 MB MB PDivision 22 MB MB P Division 23 MB MB P

Lisgar MB P Macdonald MB PMarquette MB P Portage La Prairie MB PProvencher MB P Selkirk MB P

Souris MB P Winnipeg MB PContinued on next page...



Division Name Province Classification Division Name Province ClassificationAlbert NB P Carleton NB NB P

Charlotte NB P Gloucester NB PKent NB NB P Kings and Albert NB PKings NB NB P Madawaska NB P

Northumberland NB NB P Queens NB NB PRestigouche NB P St. John NB PSt. John City NB P St. John City and County NB P

St. John County NB P Sunbury NB PSunbury and Queens NB P Victoria NB NB P

Westmorland NB P York NB NB PAnnapolis NS P Antigonish NS P

Cape Breton NS P Cape Breton North and Victoria NS PCape Breton South NS P Colchester NS P

Cumberland NS P Digby NS PGuysborough NS P Halifax NS PHalifax City NS P Halifax County NS P

Hants NS P Inverness NS PKings NS NS P Lunenburg NS P

Pictou NS P Queens NS NS PRichmond NS NS P Shelburne NS P

Shelburne and Queens NS P Victoria NS NS PYarmouth NS P Addington ON CAlgoma ON P Algoma East ON P

Algoma West ON P Bothwell ON CBrant ON C Brant North ON C

Brant South ON C Brantford ON CBrockville ON C Bruce ON CBruce East ON C Bruce North ON C

Bruce South ON C Bruce West ON CCardwell ON C Carleton ON ON CCochrane ON P Cornwall ON C

Cornwall and Stormont ON C Dufferin ON CDundas ON C Durham ON C

Durham East ON C Durham West ON CElgin ON C Elgin East ON C

Elgin West ON C Essex ON CEssex North ON C Essex South ON C

Frontenac ON ON C Glengarry ON CGrenville ON C Grenville South ON C

Grey ON C Grey East ON CGrey North ON C Grey South ON CHaldimand ON C Haldimand and Monck ON C

Haldimand-Norfolk ON C Haliburton ON PHalton ON C Hamilton City ON C

Hamilton City East ON C Hamilton City West ON CHamilton-Wentworth ON C Hastings ON C

Hastings East ON C Hastings North ON CHastings West ON C Huron ON CHuron Centre ON C Huron East ON CHuron North ON C Huron South ON CHuron West ON C Kenora ON P

Kent ON ON C Kent ON East ON CKent ON West ON C Kingston City ON C

Lambton ON C Lambton East ON CLambton West ON C Lanark ON CLanark North ON C Lanark South ON C

Leeds ON C Leeds and Grenville North ON CLeeds South ON C Lennox ON C

Lennox and Addington ON C Lincoln ON CLincoln and Niagara ON C London ON C

Manitoulin ON P Middlesex ON CMiddlesex East ON C Middlesex North ON C

Middlesex South ON C Middlesex West ON CMonck ON C Muskoka ON P

Muskoka and Parry Sound ON P Niagara ON CContinued on next page...



Division Name Province Classification Division Name Province ClassificationNipissing ON P Norfolk ON C

Norfolk North ON C Norfolk South ON CNorthumberland East ON C Northumberland ON ON CNorthumberland West ON C Ontario ON C

Ontario North ON C Ontario South ON COntario West ON C Ottawa City ON C

Ottawa-Carleton ON C Oxford ON COxford North ON C Oxford South ON CParry Sound ON P Peel ON C

Perth ON C Perth North ON CPerth South ON C Peterborough ON C

Peterborough East ON C Peterborough West ON CPrescott ON C Prince Edward ON C

Rainy River ON P Renfrew ON PRenfrew North ON P Renfrew South ON P

Russell ON C Simcoe ON CSimcoe East ON C Simcoe North ON C

Simcoe South ON C Stormont ON CSudbury ON P Sudbury Regional Municipality ON P

Thunder Bay ON P Thunder Bay and Rainy River ON PTimiskaming ON ON P Toronto ON CToronto Centre ON C Toronto City ON CToronto East ON C Toronto North ON C

Toronto South ON C Toronto West ON CVictoria and Haliburton ON P Victoria North ON C

Victoria ON ON C Victoria South ON CWaterloo ON C Waterloo North ON C

Waterloo South ON C Welland ON CWellington ON C Wellington Centre ON C

Wellington North ON C Wellington South ON CWentworth ON C Wentworth N. and Brant N. ON C

Wentworth North ON C Wentworth South ON CYork Centre ON C York East ON CYork North ON C York ON ON CYork South ON C York West ON CKings PE PE P Prince PE P

Prince East PE P Prince West PE PQueens East PE P Queens PE PE PQueens West PE P Abitibi QC PArgenteuil QC C Arthabaska QC C

Bagot QC C Beauce QC CBeauharnois QC C Bellechasse QC C

Berthier QC P Bonaventure QC PBrome QC C Chambly QC C

Chambly and Vercheres QC C Champlain QC CCharlevoix QC P Charlevoix-Est QC P

Charlevoix-Ouest QC P Chateauguay QC CChicoutimi QC P Chicoutimi and Saguenay QC PCompton QC C Deux-Montagnes QC C

Dorchester QC C Drummond QC CDrummond and Arthabaska QC C Frontenac QC QC P

Gaspe QC P Gaspe-Est QC PGaspe-Ouest QC P Gatineau QC PHochelaga QC C Hull QC C

Huntingdon QC C Iberville QC CIle-de-Montreal QC C Ile-de-Montreal et Ile-Jesus QC C

Ile-Jesus QC C Iles-le-la-Madeleine QC PJacques-Cartier QC C Joliette QC P

Kamouraska QC P Labelle QC PLac St. John QC P Lac St. John-Est QC P

Lac St. John-Ouest QC P Laprairie QC CLaprairie and Napierville QC C L’Assomption QC C

Laval QC C Levis QC CL’Islet QC P Lotbiniere QC C

Maisonneuve QC C Maskinonge QC PContinued on next page...



Division Name Province Classification Division Name Province ClassificationMatane QC P Megantic QC C

Metapedia QC P Missisquoi QC CMontcalm QC P Montmagny QC P

Montmorency QC P Montmorency No. 1 QC PMontmorency No. 2 QC P Montreal QC C

Montreal and Jesus Island QC C Montreal Centre QC CMontreal City QC C Montreal East QC C

Montreal St. Antoine QC C Montreal St. Jacques QC CMontreal St. Laurent QC C Montreal Ste. Anne QC CMontreal Ste. Marie QC C Montreal West QC C

Napierville QC C Nicolet QC COttawa County QC C Papineau QC P

Pontiac QC P Portneuf QC CQuebec QC C Quebec Centre QC C

Quebec City QC C Quebec County QC CQuebec East QC C Quebec West QC C

Richelieu QC C Richmond and Wolfe QC CRichmond QC QC C Rimouski QC P

Riviere-du-Loup QC P Rouville QC CSaguenay QC P Shefford QC C

Sherbrooke QC C Soulanges QC CSt. Hyacinthe QC C St. Jean QC C

St. Jean and Iberville QC C St. Maurice QC CStanstead QC P Temiscamingue QC P

Temiscouata QC P Terrebonne QC CTerritoire-du-Nouveau-Quebec QC P Timiskaming QC QC P

Trois-Rivieres QC C Trois-Rivieres and St. Maurice QC CVaudreuil QC C Vercheres QC C

Wolfe QC C Wright QC CYamaska QC C Assiniboia SK P

Assiniboia East SK P Assiniboia West SK PBattleford SK P Division 1 SK SK P

Division 2 SK SK P Division 3 SK SK PDivision 4 SK SK P Division 5 SK SK PDivision 6 SK SK P Division 7 SK SK PDivision 8 SK SK P Division 9 SK SK P

Division 10 SK SK P Division 11 SK SK PDivision 12 SK SK P Division 13 SK SK PDivision 14 SK SK P Division 15 SK SK PDivision 16 SK SK P Division 17 SK SK PDivision 18 SK SK P Humboldt SK P

Mackenzie SK P Moose Jaw SK PPrince Albert SK P Qu’Appelle SK P

Regina SK P Saltcoats SK PSaskatchewan SK P Saskatoon SK P


Figure 3.8: Illustration of Canada’s Census Divisions

(a) Eastern Region of Canada

(b) Industrial “Core” Region of Canada (c) Western Region of Canada

The above are for illutsrative purposes only. Map from Seventh Census of Canada, 1931 insert. Displays (a) eastern census divisions(the “core” regions are between Windsor and Quebec-City); (b) a closer perspective on core census-divisions; and (c) the Western region.


Table 3.10: Agricultural Employment Share - With Operators

Region 1901 1911 1921 1931 1941 1951 1961 1971 1981

Periphery 0.388 0.388 0.421 0.356 0.341 0.225 0.143 0.088 0.063

Core 0.285 0.301 0.299 0.217 0.171 0.105 0.070 0.038 0.029

Overall 0.322 0.344 0.363 0.288 0.254 0.161 0.102 0.060 0.044

Table 3.11: Annual Earnings, By Occupational Group, in Dollars

(a) Agricultural Workers

Region 1931 1941 1951 1961 1971 1981

Periphery 495 446 971 1312 1474 3898

Core 429 492 1191 1788 2165 4631

Overall 470 464 1060 1524 1767 4222

(b) Agricultural Operators

Region 1931 1941 1951 1961 1971 1981

Periphery 21 434 3009 1480 3906 13371

Core 268 551 2295 1854 3130 8894

Overall 111 474 2762 1615 3645 11980

(c) Non-Agricultural Workers

Region 1931 1941 1951 1961 1971 1981

Periphery 791 643 1540 2817 5013 13688

Core 893 757 1746 3028 5590 13163

Overall 846 708 1658 2938 5346 13391


Table 3.12: Convergence Decompositions - Including Farm Operators

Overall Measure 1941 1951 1961 1971 1981

1931 0.102 0.341 0.215 0.211 0.367

1941 - 0.239 0.113 0.110 0.265

1951 - -0.126 -0.130 0.026

1961 - -0.004 0.152

1971 - 0.156

Inter-Regional 1941 1951 1961 1971 1981

1931 0.030 0.238 0.101 0.139 0.273

1941 - 0.209 0.075 0.117 0.258

1951 - -0.120 -0.069 0.089

1961 - 0.032 0.168

1971 - 0.147

Labour Reallocation 1941 1951 1961 1971 1981

1931 0.010 -0.001 -0.021 -0.049 -0.017

1941 - -0.005 -0.019 -0.041 -0.031

1951 - -0.015 -0.035 -0.065

1961 - -0.016 -0.000

1971 - 0.002

Inter-Sectoral 1941 1951 1961 1971 1981

1931 0.062 0.166 0.065 0.080 0.075

1941 - 0.086 -0.013 -0.008 0.002

1951 - -0.122 -0.129 -0.096

1961 - 0.008 0.175

1971 - 0.013

Each panel details the percent (in decimal form) change in the gap between core (industrial) and peripheral(agricultural) regions of Canada. These results include earnings and employment data for farm owner-operatorswhen available. Each row represents a different starting date and each column represents an ending date. Forexample, 0.367 in the top-right cell means the wage gap shrank by 36.7% between 1941 and 1981 - of which,27.3% from inter-regional factors and the remainder from structural change.


Tabl

e3.

13:S

elec

ted

Ear

ning

san

dE

mpl

oym

entS

hare

Dat

a

Peri

pher

alE

arni

ngs

Cor

eE

arni

ngs

Nat

iona

lEar

ning

sA

gric

ultu

ralS

hare

s

Agr

icul

ture

Non

-Agr

icul

ture

Ove

rall

Agr

icul

ture

Non

-Agr

icul

ture

Ove

rall

Agr

icul

ture

Non

-Agr

icul

ture

Ove

rall

Peri

pher

yC

ore

Nat

iona

l

1901

$140

$305

$285

$187

$346

$334

$166

$333

$318

11.9

7%7.

64%

9.10

%

1911

$224

$472

$447

$269

$468

$457

$241

$469

$452

10.0

0%5.

30%

7.45

%

1921

$840

$948

$933

$550

$1,0

06$9

67$7

26$9

79$9

5114

.31%

8.55

%11

.31%

1931

$495

$817

$786

$428

$870

$847

$469

$846

$819

9.58

%5.

02%

7.12

%

1941

$446

$845

$808

$491

$931

$912

$463

$895

$868

9.30

%4.

27%

6.42

%

1951

$970

$1,7

61$1

,723

$1,1

91$1

,930

$1,9

12$1

,060

$1,8

60$1

,832

4.82

%2.

40%

3.43

%

1961

$1,4

05$3

,076

$3,0

28$1

,947

$3,3

30$3

,307

$1,6

43$3

,222

$3,1

872.

89%

1.71

%2.

21%

1971

$1,4

73$5

,467

$5,3

61$2

,164

$5,9

80$5

,926

$1,7

66$5

,766

$5,6

892.

65%

1.42

%1.

94%

1981

$3,8

98$1

3,47

3$1

3,25

4$4

,630

$13,

691

$13,

421

$4,2

21$1

3,59

3$1

3,42

02.

29%

1.48

%1.

84%


Tabl

e3.

14:L

isto

fKey

Var

iabl

esfo

rDec

ompo

sitio

n

Var

iabl

esR

awD

ata

Ava

ilabl

efo

rYea

rs...

Infe

rred

forY

ears

...N

otes

Ag.

Wag

e-E

arne

rs19

01R

epor

ted

wee

kshi

red;

1911

per-

wor

kerw

eeks

wor

ked

Ag.

Wag

e-E

arne

rs19

11-1

971

Prov

inci

alno

n-ci

tyre

sidu

als

(cro

p-w

eigh

t)

Ag.

Wag

e-E

arne

rs19

81R

epor

ted

Ag.

occu

patio

n;to

talf

arm

ers;

prov

inci

al-l

evel

unpa

id/p

aid

ratio

Non

-Ag.

Wag

e-E

arne

rs19

01-1

911

repo

rted

man

ufac

turi

ngw

orke

rs;i

ndus

tria

l/man

.rat

io†

Non

-Ag.

Wag

e-E

arne

rs19

21-1

981

Ari

thm

atic

Wag

e-E

arne

rs19

01-1

911

Ari

thm

atic

Wag

e-E

arne

rs19

21-1

931

Prov

inci

alno

n-ci

tyre

sidu

als

(pop

.-wei

ght)

Wag

e-E

arne

rs19

41-1

981

Ag.

Ear

ning

s19

01-1

941

Cas

h&

Boa

rd

Ag.

Ear

ning

s19

51-1

981

Cas

h

Non

-Ag.

Ear

ning

s19

01-1

911

assu

med

sam

eas

man

ufac

turi

ngav

erag

e

Non

-Ag.

Ear

ning

s19

21-1

981

Ari

thm

atic

All

Ear

ning

s19

01-1

911

Ari

thm

atic

All

Ear

ning

s19

21-1

931

Prov

inci

alno

n-ci

tyre

sidu

als

All

Ear

ning

s19

41

All

Ear

ning

s19

51re

port

edca

tego

rica

ldat

a

All

Ear

ning

s19

61-1

981

Des

crib

esda

taav

aila

bilit

yfo

rvar

ious

Cen

sus

ofC

anad

avo

lum

es.

†Se

eSt

atsC

an-S

elec

tInd

ustr

yW

age

Inde

x-C

atal

ogue

E19

8-20

8

Bibliography

Daron Acemoglu and Veronica Guerrieri. Capital deepening and non-balanced economic growth. NBER

Working Papers 12475, National Bureau of Economic Research, Inc, August 2006.

Tasso Adamopoulos. Do richer countries have higher distribution margins. 2008.

Tasso Adamopoulos. Transportation costs, agricultural productivity and cross-country income differ-

ences. International Economic Review, Forthcoming 2010.

Tasso Adamopoulos and Diego Restuccia. The size distribution of farms and international productivity

differences. April 2010.

Fernando Alvarez and Robert Jr. Lucas. General equilibrium analysis of the eaton-kortum model of

international trade. Journal of Monetary Economics, 54(6):1726–1768, September 2007.

James E. Anderson and Eric van Wincoop. Gravity with gravitas: A solution to the border puzzle.

American Economic Review, 93(1):170–192, March 2003.

James E. Anderson and Eric van Wincoop. Trade costs. Journal of Economic Literature, 42(3):691–751,

September 2004.

Costas Arkolakis, Arnaud Costinot, and Andrï¿œs Rodrï¿œguez-Clare. New trade models, same old

gains? (15628), December 2009.

Erhan Artuc, Shubham Chaudhuri, and John McLaren. Trade shocks and labor adjustment: A structural

empirical approach. American Economic Review, 100(3):1008–45, June 2010.

Martin Neil Baily and Robert M. Solow. International Productivity Comparisons Built from the Firm

Level. Journal of Economic Perspectives, 15(3):151–172, Summer 2001.

Andrew B Bernard and J Bradford Jensen. Exporting and productivity: The importance of reallocation.

Working Papers 01-02, Center for Economic Studies, U.S. Census Bureau, June 2001.

Andrew B. Bernard, Jonathan Eaton, J. Bradford Jensen, and Samuel Kortum. Plants and productivity

in international trade. American Economic Review, 93(4):1268–1290, September 2003.

99

BIBLIOGRAPHY 100

Andrew B. Bernard, J. Bradford Jensen, and Peter K. Schott. Survival of the best fit: Exposure to

low-wage countries and the (uneven) growth of u.s. manufacturing plants. Journal of International

Economics, 68(1):219–237, January 2006.

Loren Brandt and Xiaodong Zhu. Accounting for china’s growth and structural transformation. Working

Paper, 2010.

Loren Brandt, Chang-Tai Hsieh, and Xiaodong Zhu. Growth and structural transformation in china. In

China’s Great Economic Transformation, pages 569–632. Cambridge University Press, 2008.

Ariel Burstein and Jonathan Vogel. Globalization, technology, and the skill premium: A quantitative

analysis. Working Paper 16459, National Bureau of Economic Research, October 2010.

L. Caliendo and F. Parro. Estimates of the Trade and Welfare Effects of NAFTA. 2009.

Francesco Caselli. Accounting for cross-country income differences. In Philippe Aghion and Steven

Durlauf, editors, Handbook of Economic Growth, volume 1 of Handbook of Economic Growth, chap-

ter 9, pages 679–741. Elsevier, 2005.

Francesco Caselli and Wilbur Coleman. “The u.s. structural transformation and regional convergence:

A reinterpretation”. The Journal of Political Economy, 109(3):584–616, 2001.

Davin Chor. Unpacking sources of comparative advantage: A quantitative approach. Journal of Inter-

national Economics, Forthcoming 2010.

Juan Carlos Cordoba and Marla Ripoll. Agriculture, aggregation, and cross-country income differences.

Working Paper, Rice University and University of Pittsburg, 2006.

Arnaud Costinot, Dave Donaldson, and Ivana Komunjer. What goods do countries trade? a quantitative

exploration of ricardo’s ideas. Working Paper 16262, National Bureau of Economic Research, August

2010.

Serge Coulombe and Kathleen Day. Economic growth and regional disparities in canada and the north-

ern united states. Canadian Public Policy, 25(2):155–178, 1999.

Serge Coulombe and Frank Lee. Convergence across canadian provinces, 1961 to 1991. The Canadian

Journal of Economics, 28(4):886–898, 1995.

Robert Dekle, Jonathan Eaton, and Samuel Kortum. Unbalanced trade. American Economic Review, 97

(2):351–355, May 2007.

Dominion Bureau of Statistics. Census of Canada. 1901–1981.

Dave Donaldson. Railroads of the raj: Estimating the impact of transportation infrastructure. 2010.

Margarida Duarte and Diego Restuccia. The role of the structural transformation in aggregate produc-

tivity. The Quarterly Journal of Economics, 125(1):129–173, February 2010.

BIBLIOGRAPHY 101

Jonathan Eaton and Samuel Kortum. Trade in capital goods. European Economic Review, 45(7):1195–

1235, 2001.

Jonathan Eaton and Samuel Kortum. Technology, geography, and trade. Econometrica, 70(5):1741–

1779, September 2002.

Cristina Echevarria. Non-homothetic preferences and growth. Journal of International Trade and Eco-

nomic Development, 9(2):151–171, June 2000.

Robert Feenstra, editor. Per-capita income as a determinant of trade. MIT Press, Cambridge, 1988.

Robert C. Feenstra, Robert E. Lipsey, Haiyan Deng, Alyson C. Ma, and Hengyong Mo. World trade

flows: 1962-2000. Working Paper 11040, National Bureau of Economic Research, January 2005.

Anna Cecilia Fieler. Non-homotheticity and bilateral trade: Evidence and a quantitative explanation.

Working Paper, 2010.

Reto Foellmi and Josef Zweilmueller. Structural change and the kaldor facts of economic growth. 2006

Meeting Papers 342, Society for Economic Dynamics, December 2006.

Edward Glaeser and Janet Kohlhase. Cities, regions and the decline of transport costs. Papers in

Regional Science, 83(1):197–228, 2004.

Douglas Gollin and Richard Rogerson. Agriculture, roads, and economic development in uganda. 2010.

Douglas Gollin, Stephen L. Parente, and Richard Rogerson. Farm work, home work, and international

productivity differences. Review of Economic Dynamics, 7(4):827–850, October 2004.

Douglas Gollin, Stephen L. Parente, and Richard Rogerson. The food problem and the evolution of

international income levels. Journal of Monetary Economics, 54(4):1230–1255, May 2007.

Jeremy Greenwood and Gokce Uysal. New goods and the transition to a new economy. Journal of

Economic Growth, 10(2):99–134, 06 2005.

Michael J. Greenwood. Research on internal migration in the united states: A survey. Journal of

Economic Literature, 13(2):397–433, 1975.

C. Knick Harley. Transportation, the world wheat trade, and the kuznets cycle, 1850ï¿œ1913. Explo-

rations in Economic History, 17(3):218–250, 1980.

John R Harris and Michael P Todaro. Migration, unemployment & development: A two-sector analysis.


Yujiro Hayami and V. W. Ruttan. Agricultural Productivity Differences among Countries. The American

Economic Review, 60(5):pp. 895–911, 1970.

BIBLIOGRAPHY 102

Yujiro Hayami and Vernon Ruttan. Agricultural Development: An International Perspective. John

Hopkins University Press, Baltimore, MD, 1985.

Berthold Herrendorf and Akos Valentinyi. Which sectors make the poor countries so unproductive?

Journal of the European Economic Association, Forthcoming 2010.

Berthold Herrendorf, James Schmitz, and Arilton Teixeria. Transportation and development: Insights

from the us 1840-1860. Federal Reserve Bank of Minneapolis Research Department Staff Report 425,

May 2009.

Alan Heston, Robert Summers, and Bettina Aten. Penn World Table Version 6.3. Center for Inter-

national Comparisons of Production, Income and Prices at the University of Pennsylvania, August

2009.

Linda Hunter. The contribution of nonhomothetic preferences to trade. Journal of International Eco-

nomics, 30(3-4):345–358, May 1991.

Robert Inklaar and Marcel P. Timmer. GGDC productivity level database: International comparisons of

output, inputs and productivity at the industry level. Technical report, 2008.

Douglas Irwin. Trade restrictiveness and deadweight losses from us tariffs. American Economic Jour-

nal: Economic Policy, 2, 2010.

Dale Jorgenson and Frank Gollop. Productivity growth in u.s. agriculture: A postwar perspective.

American Journal of Agricultural Economics, 74:745–750, 1992.

Gueorgui Kambourov. Labour Market Regulations and the Sectoral Reallocation of Workers: The Case

of Trade Reforms. Review of Economic Studies, 76(4):1321–1358, October 2009.

Toshihiko Kawagoe, Yujiro Hayami, and Vernon W. Ruttan. The intercountry agricultural production

function and productivity differences among countries. Journal of Development Economics, 19(1-2):

113 – 132, 1985.

Hiau Looi Kee, Alessandro Nicita, and Marcelo Olarreaga. Import demand elasticities and trade distor-

tions. The Review of Economics and Statistics, 90(4):666–682, 07 2008.

Hiau Looi Kee, Alessandro Nicita, and Marcelo Olarreaga. Estimating trade restrictiveness indices. The

Economic Journal, 119:172–199, January 2009.

William Kerr. Heterogeneous technology diffusion and ricardian trade patterns. 2009.

Piyabha Kongsamut, Sergio Rebelo, and Danyang Xie. Beyond balanced growth. Review of Economic

Studies, 68(4):869–82, October 2001.

Simon Kuznets. Economic Growth of Nations. Harvard University Press, Cambridge, MA, 1971.

BIBLIOGRAPHY 103

David Lagakos and Michael E. Waugh. Specialization, agriculture, and cross-country productivity dif-

ferences. April 2010.

Everett Lee, Ann Miller, Carol Brainerd, and Richard Easterlin. Population Redistribution and Eco-

nomic Growth, United States, 1870ï¿œ1950., volume Methodological Considerations and Reference

Tables. Vol. 1. American Philosophical Soc., Philadelphia, 1957.

Frank Lee. Convergence in canada? The Canadian Journal of Economics, 29(1):S331–S336, 1996.

Andrei Levchenko and Jing Zhang. The evolution of comparative advantage: Measurement and welfare

implications. RSIE Discussion Paper, (610), 2010.

Staffan Linder. An essay on trade and transformation. Almqvist & Wiksell Boktr, Uppsala, 1961.

Robert E. Lucas. Trade and the diffusion of the industrial revolution. American Economic Journal:

Macroeconomics, 1(1):1–25, January 2009.

James R Markusen. Explaining the volume of trade: An eclectic approach. American Economic Review,

76(5):1002–11, December 1986.

Kiminori Matsuyama. Agricultural productivity, comparative advantage, and economic growth. Journal

of Economic Theory, 58(2):317–334, December 1992.

Kiminori Matsuyama. Structural change. Forthcoming entry in The New Palgrave Dictionary of Eco-

nomics, 2nd Ed., May 2005.

Lawrence D. McCann. Heartland and Hinterland: A Regional Geography of Canada. Prentice Hall

Canada, Scarborough, Ontario, 1998.

Marvin McInnis. The trend of regional income differentials in canada. The Canadian Journal of Eco-

nomics, 1(2):440–470, 1968.

James Melvin. Regional inequalities in canada: Underlying causes and policy implications. Canadian

Public Policy, 13(3):304–317, 1987.

L. Rachel Ngai and Christopher A. Pissarides. Structural change in a multisector model of growth.


Nina Pavcnik. Trade liberalization, exit, and productivity improvement: Evidence from chilean plants.

Review of Economic Studies, 69(1):245–76, January 2002.

Presada Rao. Intercountry comparisons of agricultural output and productivity. FAO Economic and

Social Development Papers, 112, 1993.

Diego Restuccia, Dennis Tao Yang, and Xiaodong Zhu. Agriculture and aggregate productivity: A

quantitative cross-country analysis. Journal of Monetary Economics, 55(2):234–250, March 2008.

BIBLIOGRAPHY 104

Jennifer Roback. Southern labor law in the jim crow era: Exploitative or competitive. The University of

Chicago Law Review, 51(4):1161–1192, 1984.

Paul Samuelson. The transfer problem and transport costs, ii: Analysis of effects of trade impediments.

The Economic Journal, 64(254):264–289, 1954.

T.W. Schultz. The Economic Organization of Agriculture. McGraw-Hill, New York, 1953.

Aba Schwartz. Interpreting the effect of distance on migration. The Journal of Political Economy, 81

(5):1153–1169, 1973.

Nancy L. Stokey. A quantitative model of the british industrial revolution, 1780-1850. Carnegie-

Rochester Conference Series on Public Policy, 55(1):55–109, December 2001.

Marc Teignier. The role of trade in structural transformation. Working Paper, University of Chicago,

2010.

C. Peter Timmer. Agriculture and economic development. In B. L. Gardner and G. C. Rausser, editors,

Handbook of Agricultural Economics, volume 2 of Handbook of Agricultural Economics, chapter 29,

pages 1487–1546. Elsevier, 2002.

Peter Timmer. The agricultural transformation. In Hollis Chenery and T.N. Srinivasan, editors, Hand-

book of Development Economics, volume 1 of Handbook of Development Economics, chapter 8,

pages 275–331. Elsevier, December 1988.

Trevor Tombe. Structural change and regional growth dynamics. Working paper, University of Toronto,

2008.

Trevor Tombe. The Missing Food Problem: How Low Agricultural Imports Contributes to International

Income Differences. University of Toronto, Department of Economics, Working Paper 416, 2010.

Dietrich Vollrath. The dual economy in long-run development. Journal of Economic Growth, 14(4):

287–312, December 2009.

Michael E. Waugh. International trade and income differences. American Economic Review, forthcom-

ing 2010.

Gavin Wright. Old South, New South: Revolutions in the Southern Economy Since the Civil War.

Louisiana State University Press, 1986.

Kei-Mu Yi and Jing Zhang. Structural change in an open economy. Working Papers 595, Research

Seminar in International Economics, University of Michigan, April 2010.

by trevor tombe - university of toronto t-space...penn state, simon fraser, toronto, wilfrid...

Documents