University of Southern California Dissertations and Theses

Essays on the dual urban-rural system and economic development in China

(USC Thesis Other)

Essays on the dual urban-rural system and economic development in China

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON THE DUAL URBAN-RURAL SYSTEM AND ECONOMIC DEVELOPMENT IN
CHINA
by
Jingyi Fang
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2024
Copyright 2024 Jingyi Fang

Acknowledgments
The author acknowledges the partial financial support of USC Graduate School Endowed Fellowship.
I would like to thank my advisor Monica Morlacco (co-chair) and Thomas Chaney (co-chair) for
their continuous support, persistent encouragement, and invaluable guidance throughout my entire
PhD journey. I am grateful for the expertise and knowledge of my committee members Paulina
Oliva and Neha Bairoliya, and faculty members Ayse Imrohoroglu, Andrii Parkhomenko, Augustin
Bergeron, Caroline Betts, Jack Hou, Jeff Nugent, Jefferey Weaver, John Strauss, Marianne Andries,
Pablo Kurlat, Robert Metcalfe, Romain Ranciere, Simone Quach, Vittorio Bassi and many audience
who gave helpful comments in reading groups, seminars, and conferences.
This endeavor would not have been accomplished without supports and help from my friends at
USC. My cohorts who graduated a year ahead of me and my peers Amanda Ang, Zhan Gao, Tao
Chen, Weizhao Huang, Wei Zhou and many others have been my dearest friends and allies against
numerous obstacles we faced during the PhD program. I’m especially grateful to my coauthor,
Kang Zhou, one of the most talented and humble applied micro-economists I’ve known, for his
continuous support and encouragement.
I could not have undertaken this journey without my husband Gang Qiu, who’s the most considerate
partner and an inspiring role model. I would also love to thank my daughter, Rylynn, for bring
all the tears and joy of life and teaching me about responsibility. Lastly, I’m deeply grateful
for the support of my parents, Dongying and Ronghe, and my in-laws, Hongbo and Wei. Their
unconditional belief and support in me gave me the strongest safe harbor. My furry family, Marley,
Puppy, and Grey, have brought me endless entertainment and comfort.
ii

Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Figures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Abstract. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
1 Cities with Invisible Walls: Labor Liberalization and Misallocation in China . . . . . . . . . . . . . . 1
Chapter One: Introduction ....................................................................... 1
Chapter One: Institutional Background ......................................................... 8
Hukou System and Migrants in China ..................................................... 8
Restrictions and Discrimination under Hukou ............................................ 9
The Hukou Reform in the 2000s............................................................ 13
Surge in Migrants following the Reform ................................................... 16
Chapter One: Methodology....................................................................... 19
Measurement of Baseline Distortion Level................................................. 19
Firms as Price-takers in the Local Labor Market ......................................... 20
Chapter One: Data and Variables................................................................ 21
Chapter One: Results ............................................................................. 23
Average Effects .............................................................................. 23
Differential Effects by the Baseline Distortion Level...................................... 26
Event Study.................................................................................. 28
Geography Competition .................................................................... 31
iii

Baseline Migration Networks as Substitutes............................................... 34
Alternative Measure of Policy Intensity: Industry-level Evidence ....................... 36
Robustness Analysis......................................................................... 39
Additional Findings on Firm Heterogeneity ............................................... 43
Capital Misallocation ....................................................................... 45
Chapter One: Aggregate Evidence and Policy Implication .................................... 46
Gain in Manufacturing Solow Residual ................................................... 46
Gain in Manufacturing Output ............................................................ 49
Chapter One: Conclusion ......................................................................... 52
2 Policy Assessment Related to Return Migrant Workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Chapter Two: Introduction ....................................................................... 54
Chapter Two: Institutional Background......................................................... 56
Migration in China .......................................................................... 56
The Return Migration Policy and Residents Increase .................................... 57
Chapter Two: Data and Variables ............................................................... 59
Chapter Two: Results............................................................................. 60
Work Type ................................................................................... 60
County-level Evidence ....................................................................... 63
Impact on Earnings ......................................................................... 65
Discussion: Medical Insurance .............................................................. 66
Chapter Two: Model .............................................................................. 67
Data and Variables .......................................................................... 70
Preliminary Results and Proposal.......................................................... 71
Chapter Two: Conclusion......................................................................... 74
3 Intervivos Transfers Change due to One Child Policy Intrigued Social Security Tax Increase 75
Chapter Three: Introduction ..................................................................... 75
Chapter Three: Model ............................................................................ 78
Households ................................................................................... 78
Firms ......................................................................................... 80
iv

Government .................................................................................. 81
Equilibrium .................................................................................. 81
Calibration ................................................................................... 82
Chapter Three: Results ........................................................................... 83
Steady States of the Initial and Final Economies ......................................... 84
Demographics ................................................................................ 85
Payroll Tax vs. Pension Payment to GDP ................................................ 86
Intervivos Transfers ......................................................................... 87
Chapter Three: Conclusions ...................................................................... 90
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Chapter One Appendix ........................................................................... 103
Institutional Background: the Hukou Reform in the 2000s .................................... 103
Methodology: Frims as Price-takers in the Local Labor Market............................... 107
Data and Variables: Summary Statistics ........................................................ 114
Results: Average Policy Effects .................................................................. 116
Results: Differential Effects by the Baseline Distortion Level ................................. 117
Results: Heterogeneity Analysis.................................................................. 119
Results: Robustness Analysis..................................................................... 120
Results: Additional Findings ..................................................................... 130
Aggregate Evidence and Policy Implication ..................................................... 134
v

List of Tables
1.1 Working Intensity by Rural and Urban Hukou............................................ 12
1.2 Ratio of Migrant Workers in Reformed and Non-reformed Cities ....................... 17
1.3 Average Effect of the Hukou Reform on Firm Outcomes from 1998-2007 .............. 25
1.4 Differential Impact of the Hukou Reform on Firm Outcomes from 1998-2007 ......... 28
1.5 Policy Impact by Competition Levels among Cities ..................................... 33
1.6 Policy Impact by Baseline Migration Networks .......................................... 36
1.7 Policy Impact on Industry-level Aggregate Outcomes .................................... 38
1.8 Solow Residual Gain from Different Methods ............................................. 48
1.9 GDP/Production Gain by Sources: Allocative Gain and Market Expansion .......... 52
2.1 Household Employment Type Change in Urban Regions ................................ 62
2.2 County-level Household Employment Type Change in Urban Regions.................. 64
2.3 Policy Impact on Work Status and Earnings.............................................. 65
2.4 Policy Impact on Earnings by Work Type................................................. 66
2.5 Summary Statistics.......................................................................... 71
2.6 Estimation of GMM Parameters .......................................................... 73
3.1 Calibrated Parameters ...................................................................... 83
3.2 Steady States ................................................................................ 84
3.3 Demographic changes: Data ................................................................ 85
3.4 Demographic changes: Model............................................................... 85
3.5 Pension to GDP fraction in the Data ...................................................... 86
3.6 Social Security Tax in the Model........................................................... 86
vi

7 Association Between Prefecture-level Characteristics and the Reform .................. 106
8 Summary Statistics by Groups ............................................................. 114
9 Summary Statistics of Pre-reform Observed Main Sample and Whole Sample ........ 115
10 Policy Impact on Output, Value Added and Value Added Tax .......................... 118
11 Policy Impact by Implement Years......................................................... 119
12 Summary Statistics of Exiters and Entrants .............................................. 120
13 Differential Policy Impact for Entrants and Exiters ...................................... 121
14 Average Policy Impact for Entrants and Exiters.......................................... 122
15 Policy Impact by Alternative Cutoffs: Quartiles of Baseline Distortion Level .......... 123
16 Differential Policy Impact using Alternative Specifications: with Weights and Alternative Distortion Classification ............................................................ 124
17 Policy Impact after Controlling for Concurrent Events .................................. 125
18 Policy Impact after Controlling Additional Fixed Effects ............................... 126
19 Policy Impact on Prefecture-level Aggregate Outcomes ................................. 127
20 Policy Impact by Alternative Baseline Distortion Cutoffs: Within Industry by Prefecture ........................................................................................ 128
21 Policy Impact after Trimming 5% data .................................................... 129
22 Policy Impact by State-Owned Status: Two Groups ..................................... 130
23 Policy Impact by Labor Intensity .......................................................... 131
24 Policy Impact by Firm Size: Two Groups ................................................. 132
25 Extension to capital misallocation ......................................................... 133
26 Pre-reform wedges of labor factor (2000) .................................................. 134
27 Predicted Change in Capital Wedges due to the Hukou Reform ........................ 135
29 First Stage of Instrumenting Employment with City-level Shift-share Migration Shock
................................................................................................ 136
28 Predicted Aggregate Solow Residual ....................................................... 137
30 Factor Price (Labor-Wage) Elasticity Estimation ........................................ 138
vii

List of Figures
1.1 Firm Employment Expansion for the Baseline More and Less Distorted Firms ........ 3
1.2 Hukou Reform Timeline..................................................................... 14
1.3 Map of Policy Implementation ............................................................. 15
1.4 Employment Expansion of an Average Firm .............................................. 18
1.5 Employment Expansion by Reform Status ................................................ 18
1.6 Differential Treatment Effect of Hukou Reform on Employment (left top), Labor
Productivity (right top), Revenues (left bottom) and Capital (rigth bottom). ......... 30
1.7 Manufacturing Output Increase due to Hukou Reform ................................... 49
2.1 Map of Treated Counties.................................................................... 58
3.1 Transfer by Types of Households: no TFP, fertility change, all transfer ................ 87
3.2 Transfer by Types of Households: no TFP, fertility change, per child transfer......... 88
3.3 Transfer by Types of Households: with TFP, fertility change, all transfer.............. 89
3.4 Transfer by Types of Households: with TFP, fertility change, per child transfer....... 90
5 Maps of Policy Implementation with 2000 baseline GDP ................................ 104
6 Maps of Policy Implementation with 2000 Population ................................... 105
7 Average Treatment Effect Event Studies (Treated Prefectures Only) ................... 116
viii

Abstract
This dissertation contributes to our understanding of the pull- and push- factors behind migration
patterns and their economic income under the existence of a special household registration system,
hukou, in China. In the first chapter, I studied the policy impact of a series of hukou reform
during the 2000s on the employment, marginal productivity of labor, capital, and revenues of local
manufacturing firms. I also quantify the productivity and manufacturing output gain due to this
relaxation. In chapter two, I investigate a series of policies that encourage return migration and
entrepreneurship. The lack of solid credit loaning advantages attracted people to return and then
trapped them in occasional jobs. Finally, chapter three relies on an overlapping generation model
to study the induced increase in social security tax due to One Child Policy and its impact on
intervivos transfers between parents and children. Each chapter addresses pivotal issues crucial for
the role of the rural-urban dual system in shaping economic development and consumer welfare.
ix

Chapter 1
Cities with Invisible Walls: Labor Liberalization and Misallocation
in China
with Kang Zhou
Chapter One: Introduction
Wages, incomes, and productivity vary enormously across regions and sectors even within a single country. One prominent source contributing to these differences is the presence of migration
barriers and associated costs Munshi and Rosenzweig (2016) Munshi and Rosenzweig (2016); Ngai
et al. (2019); Choudhury (2022); Nakamura et al. (2022); Foster and Zhou (2022). These barriers
can asymmetrically restrict labor market access for producers that differ in productivity, leading
to potential distortions in inputs away from their most efficient use and consequently a large loss
in aggregate productivity. Recent work by Tombe and Zhu (2019) and Bryan and Morten (2019)
has quantified considerable gains in aggregate productivity resulting from sectoral and geographic
mobility.1 These estimates align with theoretical predictions emphasizing migration barriers as
contributors to low productivity in less developed economies Easterly (1993); Jones (2013); Bartelsman et al. (2013), highlighting the importance of understanding specific barriers to mobility
and their impacts on misallocation.
1Tombe and Zhu (2019) estimate that reducing inter-provincial migration costs in China to levels comparable to
those of US individuals living outside their birth state would increase real GDP and welfare by nearly 13% and 46%,
respectively. Similarly, Bryan and Morten (2019) find that barriers to internal labor migration in Indonesia reduce
labor productivity by 22% relative to a counterfactual scenario without such barriers.
1

Yet, there remain many empirical obstacles to credibly identifying migration barriers and especially
estimating their impact on labor misallocation, which is relatively less explored compared to capital
misallocation.2 First, we can not observe the counterfactual performances of a firm that operates
in a market with fewer frictions to mobility, and this makes it empirically difficult to estimate the
change in allocative efficiency generated by labor liberalization. Second, policy-makers are particularly interested in the impacts of specific sources of distortions, which provide rich information
and implications for future policy design to mitigate the inefficiency incurred by them. However,
it is not easy to isolate the impact of one specific source of misallocation from others and measure
what fraction of the changes in misallocation over time can be attributed to the source due to
simultaneity bias.3
To tackle these challenges, we examine the allocative efficiency of firms’ labor inputs by exploiting the staggered introduction of a policy reform from China. Specifically, we focus on China’s
hukou system, a household registration arrangement that restricts migrants’ access to various public services like education, healthcare, and social welfare (Sieg et al., 2023).4 These discriminatory
practices result in substantial migration costs, hindering internal migration within China (Chang,
1996). However, the ”unified hukou reform” initiated in 2001 relaxed these restrictions by increasingly granting migrants full local urban residency rights. We examine whether this reform enhances
the allocation of labor across firms. To our knowledge, this is the first paper that provides causal
estimates of labor misallocation resulting from labor market liberalization.
Our analysis centers on input wedges tied to firms’ marginal products of labor (MRPL), as outlined
in the seminal work by Hsieh and Klenow (2009). Drawing from insights in the extensive misallocation literature, as summarized in Sraer and Thesmar (2023), we employ a Cobb-Douglas production
function, enabling the utilization of average labor products (Y/L), which are observable in data, as
2A growing body of literature has examined the extent and consequences of capital misallocation. Examples of
these studies include Hsieh and Klenow (2009), Song et al. (2011), Midrigan and Xu (2014), Gopinath et al. (2017),
and Bau and Matray (2023).
3To measure additional outputs generated by more efficient reallocation of inputs across producers, the literature
mainly relies on studying the extent to which the marginal products disperse across active producers. While this
approach allows for the quantification of misallocation, it falls short in identifying the specific sources responsible for
such misallocation. See Restuccia and Rogerson (2017) for a related discussion.
4The hukou system, also known as household registration, serves as a state institution regulating and constraining
population mobility in China. Refer to Section 1 for further discussion.
2

a measure of MRPL within a given prefecture.5 Firms are then categorized as more or less distorted
based on whether their average MRPL exceeds or falls below the pre-reform median MRPL within
a given prefecture. This categorization allows us to examine the evolution of labor misallocation by
comparing MRPL changes over time between initially more and less distorted firms. To illustrate
this, we plot the contrast in total employment growth between categories in Figure 1.1. Notably,
by 2007, the growth rate of the baseline more distorted manufacturing firms had surged to roughly
40% higher than their less distorted counterparts (58% versus 17%), suggesting a reduction in labor
misallocation during the reform decade.
0 .2 .4 .6
Growth Rate in Employment (Baseline in 2000)
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
High Productive Low Productive
Figure 1.1: Firm Employment Expansion for the Baseline More and Less Distorted Firms
Notes: The figure plots the employment growth of the baseline more distorted and less distorted firms from 1998 to
2007. Baseline more distorted firms are defined as firms with above median baseline labor productivity within the
local labor market, approximated with total output over year-end employment y/l in the data.
Our empirical analysis begins by employing the staggered introduction of the reform to estimate the
average policy effects across prefectures. On average, we find that the reform increased employment
size by 8.4% and revenues by 9.6% for an average firm, whereas the estimates on labor productivity
5More rigorously, this measure is based on the assumption that, given our controls, all firms in a prefecture
have the same labor share. Our results remain similar even when relaxing this assumption empirically, allowing for
variations in labor shares among firms within an industry in a prefecture. See Section 1 for a discussion.
3

and capital stock do not reach statistical significance. These results underscore two insights. First,
employment is the most direct input factor affected by the reform, which simultaneously increases
firms’ output (revenues) by allowing firms to grow on average. Second, the average expansion may
have masked the cross-firm reallocation of labor, leading to insignificant estimates if policy effects
are differential or even opposing between more and less distorted producers. Thus, it is important
to examine the differential policy effects based on firms’ pre-reform distortion levels.
To understand how more and less distorted firms responded differently to the reform, we estimate
the differential policy effects. Our empirical strategy is to compare within-city changes in employment between initially more and less distorted firms from reformed prefectures and those from observably identical but non-reformed prefectures. Our identification hinges on the assumption that,
given time-invariant firm characteristics and time-variant prefecture-level macroeconomic shocks,
the counterfactual trends in firm outcomes would have been similar between prefectures subjected
to the reform and those that were not or adopted the reform belatedly.6 We find a substantial
reduction in labor misallocation from reduced barriers to internal migration. In magnitude, the
reform increased employment by 26.0%, revenues by 4.1%, capital by 11.8%, and decreased MRPL
by 23.8% for previously more distorted firms compared to others. These findings imply that, following the reform, ex-ante more distorted firms experienced a larger expansion of employment, which
lowered their MRPL and led to a more concentrated distribution of MRPL among all firms.
We then assess the geographical variations in the policy’s impact, geographically paying attention
to how the policy impacts are amplified or attenuated by post-reform competition for potential
migrants and pre-reform migration networks. First, we focus on the geographical dynamics of competition because migrants are spatially allocated: cities closer to reformed prefectures experience
heightened competition for potential workers. We gauge this competition level by constructing an
index based on reform timing and geographic proximity. We find that, for a city facing a higher
geographic competition level by one standard deviation, the reform reduced the within-prefecture
6We investigate whether hukou reform reduced misallocation by testing whether the policy had differential effects
depending on firms’ initial labor productivity level. While the estimation does not require our treatment to be random,
it does require that initially more productive firms and less productive firms would have had similar employment
expansion in the absence of hukou reforms. We provide two pieces of indicative evidence to support that, even though
the counterfactual trends remain non-testable. First, we graphically demonstrate analogous pre-reform trends between
more and less distorted firms. Second, controlling for higher-order 2-digit industry-by-year fixed effects, which account
for many unobserved shocks at the industry level, does not change our results.
4

employment expansion by around 4.0% for the baseline more distorted firms compared to less
distorted firms. Next, we focus on migration networks because strong local networks introduce
greater de facto labor market flexibility, which can be particularly relevant in the presence of strict
labor regulations. We find that the decline in labor misallocation is more pronounced in areas with
weaker networks, suggesting the importance of networks in shaping the efficiency effects generated
by labor deregulation.
These findings provide new evidence that workers who would otherwise be more productive are
misallocated in firms that excessively hire and thus have low MRPL. To verify these findings, we
further estimate the impact of the reform on misallocation at the industry level. The industry-level
estimate takes advantage of the variation generated by the timing of the reform across prefectures
with varied industry compositions, which allows the construction of an alternative measure of the
reform. Specifically, we use the fraction of prefectures getting reformed in an industry as the
industry-level measure of reform exposure. We then re-estimate the effect of industry exposure to
the reform on misallocation measured by the industry variance of MRPL. Our results suggest that
labor misallocation declined by 45%-60% if all prefectures get treated, consistent with our main
results.
While our main results support the overall reduction in misallocation within prefectures, it is
important to identify the types of firms benefiting most from labor market deregulation, which has
important implications for future policy designs. In doing so, we conduct three firm heterogeneity
tests, focusing on firms’ ownership, labor intensity, and employment size. Overall, we find that the
impacts of the reform on allocative efficiency tend to be more pronounced for non-state-owned and
larger firms. We find also evidence of a stronger impact for labor-intensive firms compared with
capital-intensive ones.
Lastly, we move beyond establishing causal evidence and quantify the aggregate gain. The gain
can be boiled down into two components: the Total Factor Productivity (TFP) gain arising from
labor reallocation toward more productive firms and the output expansion arising from the shift
of agricultural workers to cities, subsequently absorbed into firms. To quantify the former, we
calculate the initial values of labor and capital distortions following conventional approaches (Hsieh
5

and Klenow, 2009), and estimate the reduction in firms’ overall labor distortion generated by
the reform using a generalized difference-in-difference estimator. Integrating these parameters
into our aggregation formulas from first-order approximations yields a 4.6% increase in aggregate
productivity, contributing roughly 15% to the total Solow residual gain in the manufacturing sector
over the period. To quantify the output expansion, we estimate the factor-price (labor-wage)
demand elasticity employing a shift-share migration supply instrument following Borusyak et al.
(2022).7 We then adopt a graph-based approach to quantify the gain in the spirit of Borjas and
Edo (2023). Quantitatively, the reform increased manufacturing output by 14.6%, or roughly 14.9
billion USD increase relative to China’s manufacturing output in 2000.
We subject our results to a variety of robustness tests. First, we address concerns regarding
firm entry and exit by flexibly restricting the sample. Second, we address the potential concern
of negative weights by adopting an approach proposed by De Chaisemartin and d’Haultfoeuille
(2020). Third, we test the sensitivity of our results to alternative classifications of ex-ante distorted
firms. Fourth, we additionally control for the reform year and the year preceding the reform fixed
effects to ensure that our results are not driven by unexpected unobservable shocks in or before
the reform. Additionally, we provide prefecture-level estimates to collaborate on our main results.
Lastly, we explore alternative functional forms and conduct sensitivity analysis by trimming our
sample. All results are robust to our main results.
This paper contributes to the literature in mainly three aspects. First, we extend the literature on
productivity and labor (mis)allocation in developing countries by studying a specific source of labor
market barriers–China’s hukou system, which discriminates against rural migrants (”in the city,
but not of the city”) (e.g., Munshi and Rosenzweig, 2016; Ngai et al., 2019; Zi, 2020; Choudhury,
2022; Nakamura et al., 2022; Foster and Zhou, 2022).8 Workers face various barriers when switching
across firms, sectors, or regions, leading to potential lower productivity (Lagakos et al., 2023). We
7Estimating factor-price elasticity with OLS can yield biased and inconsistent results due to the interdependence
of wages and employment. We instrument firm-level employment using a shift-share labor supply shock measure. To
construct this IV, we multiply the industry-level migration policy intensity (share of treated prefectures in a certain
year) by baseline industry composition in each prefecture, and we multiply the above by the indicator of pre-reform
more productive firms within prefectures. Our identification is based on the idea that positive migration inflow shock
affects firms’ marginal revenue product and thus labor demand. The validity of IV is discussed in Section 1.
8Examples of other barriers that impede labor from flowing to its highest return activity include job search
imperfections (Abebe et al., 2021), informality (Ulyssea, 2018), risks (Munshi and Rosenzweig, 2016), and institutional
constraints, e.g., minimum wages (Hau et al., 2020).
6

provide causal evidence of productivity gains resulting from hukou deregulation, shedding light
on the role of migration barriers in distorting firm-level labor allocation. Our focus on internal
migration is important given its importance in overall flows despite being underexplored relative
to international migration (Young, 2013).
Second, our results also contribute to work on migration’s impact on firms. Migration-induced
positive labor supply shocks mean a more flexible labor market, a larger pool of skills, and lower
costs of employment. Past work on migration has analyzed its effects within firms, including
various input adjustments and output changes such as factor use, product choice, and technological
adoption (e.g., Lewis, 2011; Peri, 2012; Dustmann and Glitz, 2015; Wang et al., 2021; Beerli et al.,
2021; Imbert et al., 2022). We instead examine the relative resource distribution shift in response
to migration. By focusing on the allocation of labor across different firms, our work highlights the
significant role of disproportionately increasing labor input for high-productivity firms as a crucial
driver of economic growth, serving as an additional channel for gains from migration.
Third, we augment the evidence on estimating misallocation, particularly leveraging experimental settings (e.g., Hsieh and Klenow, 2009; David and Venkateswaran, 2019; Alpysbayeva and
Vanormelingen, 2022; Adamopoulos et al., 2022; Sraer and Thesmar, 2023). Measuring misallocation remains a major challenge due to measurement errors and real frictions, making natural
experiments a compelling setting. While previous work has analyzed misallocation in capital and
product markets (e.g., Song et al., 2011; Midrigan and Xu, 2014; Gopinath et al., 2017; Varela,
2018; David and Venkateswaran, 2019; Peters, 2020; Bau and Matray, 2023; Blattner et al., 2023),
we employ a policy experiment to examine labor misallocation, which has received considerably
less but increasing attention.
The rest of our paper is organized as follows. Section 1 provides an overview of the institutional
background. Section 1 provides a discussion on methodologies for identifying misallocation and
some underlying assumptions. Section 1 introduces the data used in the paper and presents our
main empirical strategy. Section 1 reports our results, which contain the average policy effects, the
allocative policy effects, event studies, some additional findings, and robustness checks. Section
1 quantifies the aggregate allocative efficiency gain and the induced output growth. Section 1
7

concludes.
Chapter One: Institutional Background
Hukou System and Migrants in China
Hukou in China is a system of residency permits required by law to officially identify an individual’s
resident status within a specific administrative area (Zi, 2020). The hukou system, or household
registration system, was initially introduced in 1958 when local governments across the country
issued a hukou certificate to every resident. Over time, the certificate becomes a kind of ”passport”
for internal use and evolved into a system that regulates the movement of population (Chan, 2012).
Under the system, each citizen is required to register in one and only one place of (permanent)
residence. The hukou system is characterized by two features: hukou status and registration
location. First, a hukou status for an individual is defined as either urban or rural. Second, the
registration location for a person’s hukou is defined by his/her permanent residential location.
The hukou status and the registration location together define one’s rights and eligibility for public
services, including public education, healthcare, and social welfare, within a specific registered
administrative unit in China (Chen and Feng, 2013). Local governments lack incentives to provide
these services for migrants largely due to fiscal burdens (Sieg et al., 2023). Despite its initial
strict enforcement, the system was practically loosened as China started its transition to a marketoriented economy from a planning one. The deregulation and the rapidly growing demand for cheap
labor in the urban sector eventually catalyzed the greatest internal migration of people in China’s
history, albeit with the hukou system never being abolished (Cui and Jeffrey, 2015).
Migrants in China are typically defined as those who do not hold a local hukou in an area where
they work and live most (Zhou and Zhang, 2021; Jin and Zhang, 2023). The internal migrants,
often referred to as the ”floating population,” represent a huge number of rural people moving to
cities for employment opportunities while raising a family in their rural homes. The urban floating
population was around 130 million in 2000 and the number had reached 221 million by 2010,
roughly 17% of China’s total population, according to data from the 2000 and 2010 Population
8

Census (Liang et al., 2014). The world has never seen such a large-scale movement of labor force
within such a short time in history.
Restrictions and Discrimination under Hukou
Migrant workers in China face various restrictions and discriminatory practices governed by the
hukou system. First, migrant workers are typically ineligible for the full scope of local government services, including the most basic public services such as children’s education in cities where
they work and live most (Bao and Zhao, 2011; Song, 2014). For example, education is a crucial
consideration for China’s parents. Yet, migrant children are denied access to most urban public
schools that local children have access to (Sieg et al., 2023). This hukou-type-based provision of
public services adds costs to moving for rural migrants when workers seek to switch across firms,
sectors, and regions, creating a semi-urbanized and marginalized migrant population in China. The
phenomenon in China is called ”in the city, but not of the city” (Chan, 2011).
Beyond government services, migrant workers in cities have little access to social welfare benefits,
notably subsidized housing, unemployment insurance, and retirement pensions. Without access to
subsidized housing, which is accessible to local urban residents in the state-owned employment, few
migrants can afford homes in megacities where job opportunities are concentrated. Statistics suggest
that approximately 60% of migrants relied on the private housing market for accommodation in
2016.9 The disparity in other social welfare is also large. As of 2010, only 13.5% of migrants had
access to unemployment insurance compared to 66% of urban hukou holders (Meng, 2012). This
exclusion from the unemployment insurance system leaves rural migrants particularly vulnerable to
economic downturns, presenting high opportunity costs associated with unemployment. Notably,
their employment rate among the working-age population stands at 83% compared to 68% among
the local workers with urban hukou (Shen, 2015). Migrant workers, who are usually the bread
earners for their households, also tend to sustain longer job durations and shorter unemployment
periods (Zhang, 2010).
The unequal provision of public services and social welfare thus constructed something like invisible
9Data source: Discrimination Against Rural Residents in China.
9

walls in China cities that split urban residents into local hukou holders and migrant populations.
As a result of the separation, it is not surprising that the hukou system creates restrictions on
mobility. In the influential book on Chinese cities by Chang (1996), the author wrote, ”While the
1949 revolution led to the destruction of physical walls that delineated urban communities, it has
also constructed invisible walls, in an institutional sense, around every Chinese city, in order to
limit rural-to-urban migration...”.
Second, migrant workers face also various labor-market and social discrimination. For example,
migrants face restricted job choices based on their hukou status. In some major cities like Beijing,
migrants are not allowed to take jobs as taxi drivers, hotel front desk personnel, and in Kentucky
Fried Chicken stores until recent years (Kuhn and Shen, 2015). The jobs taken by rural migrant
workers are often 3D (Dirty, Dangerous, and Demeaning) types, which are typical jobs that local
hukou holders are unwilling to take (Shi et al., 2008; Bao and Zhao, 2011; Meng, 2012).
By 2009, only 7.3% of city migrant workers were employed in the state sector, while this figure stood
at 49.4% for urban hukou holders (Meng, 2012). This large gap partly reflects the institutionalized
discrimination against migrants in labor markets. Additionally, migrants also face barriers in the
lottery for a car license plate in some metropolises (Kuhn and Shen, 2015). In Zhao (2000), she
noted ”...as urbanites enjoyed more government subsidies, better protection, and higher incomes,
they came to perceive themselves as being superior to rural people. This became the historical and
psychological basis for the discrimination toward rural people....” (p.20).
While detecting discrimination is challenging, recent literature does find evidence of labor market
discrimination against migrants in China. For instance, in Shanghai, an important destination for
rural migrants, hourly earnings for migrants were only half of those for urban hukou holders in
1995, and importantly, around 47% of the wage gap cannot be explained by observed differences
in workers’ traits (Frijters and Lee, 2010). Similarly, Meng and Zhang (2001) find that most of
the wage gap between migrants and urban workers cannot be explained by productivity-related
traits, suggesting discrimination against migrants. Shi et al. (2008) provides further evidence that
migrants have a higher likelihood of wage arrears and lower average wages. In 2004, unpaid wages
for rural migrants in the construction industry alone were estimated at 3 billion Chinese Yuan
10

(around 0.412 billion USD) in Beijing (Shi et al., 2008). As supplementary evidence, Table 1.1
illustrates that rural workers typically put in substantially longer work hours per day and work
more days per week compared to urban hukou workers.
For firms, the administrative regulations embedded in the hukou system create additional complexities for firms hiring migrant workers, requiring permits and incurring additional costs. This is
because any individual who stays more than three days in a location outside their city of residence
needs to acquire a temporary residence permit, and it is not easy to obtain one. Undocumented
migrants, referring to internal migrants lacking a work permit and/or temporary residence permit, faced potential risk of detention and deportation under the Custody and Repatriation (C&R)
system, a derivative of the hukou system for its enforcement (Foster and Zhou, 2022). Notably,
Pomfret (2003) highlights in a report how the C&R system curtailed labor mobility, remarking that,
”......police distorted the now-abolished regulation, using it to fill arrest quotas during scheduled
roundups.” To gain documented status, firms shouldered various fees, encompassing a temporary
residence fee, migrant management fee, service fee, and city-entry fee. Furthermore, the ineligibility
of migrant workers to enroll in a wide range of work-related medical and unemployment insurance
also adds to extra costs for firms, leading to a higher cost of hiring migrants during that era (Song,
2014).
11

Table 1.1: Working Intensity by Rural and Urban Hukou
(1) (2)
Urban Rural
Days per week: 2002
Migration 5.88 6.61
Native 5.32 6.52
Hours per day: 2002
Migration 8.98 10.14
Native 8.13 10.09
Observations 10047 3454
Notes: The data is from the Chinese Household Income Project (CHIP).
Amidst these restrictions and discrimination, migrant workers face high friction when transitioning between jobs across firms, sectors, and regions. Consequently, labor costs exhibit significant
variations among heterogeneous firms. These differing labor costs encompass at least two distinct
components, as modeled in Section 1. The first component is modeled as wedges, which represent
firm-specific distortions on labor inputs. These wedges can be positive or negative, depending on
if they behave more like taxes or subsidies. The second one reflects the real labor cost that is
inherently tied to the physical productivity of labor. Consequently, heterogeneous firms exhibit
diverse profiles characterized by varied wedges, and reforming the hukou system can reduce the
friction for firms, which would lead to a relative reallocation of labor across firms.
12

The Hukou Reform in the 2000s
The increased demand for labor from urban firms and the emergence of a semi-capitalist market
system post-1992 accelerated the mobility of labor within China, forcing the liberalization of some
aspects of the hukou system (Cui and Jeffrey, 2015). In the early 2000s, China introduced a series
of reforms aimed at mitigating the negative societal impact of the hukou system (Pi and Zhang,
2016). Among these, the Tongyi Jumin Hukou, or ”unified residency permit,” emerged as one of
the most influential. The ”unified residency permit” reform was based on a pilot program that
tested the feasibility of easing hukou restrictions in select cities and yielded lots of positive social
feedback.10
The primary goal of the pilots and the subsequent hukou reform aimed to enable rural migrants
to acquire urban hukou, granting them access to urban social services and benefits (Liu, 2005).
Indeed, the reform was designed to provide equal access to social welfare benefits for all Chinese
citizens, irrespective of their origins. Consequently, migrants in treated prefectures gained the
choice to register in their current residing prefectures (Jin and Zhang,2023). The integration of
migrants into local urban society enabled them, primarily rural workers, to access similar benefits
as urban residents (Wang, 2004). Hence, while the hukou system itself is still in place and it still
works as a barrier to labor mobility, the reform does open the door for rural migrant workers to be
regularized and thus increase labor market liberalization in China.
To gather information on the reform’s rollout, we manually collected data on its timing and implementation in various prefectures using China’s most extensive law database guides.11 We then
matched these data to prefectures where firms operate each year. A prefecture is categorized as
”treated” if it has implemented the ”Tongyi Jumin Hukou” that allows rural workers to apply for
temporary residency permits.12 Figure 1.2 lists the treated prefectures for each year between 2001
10A Hukou pilot program initiated before the 2000s tested the relaxation of household registration restrictions
in cities like Shanghai, Guangzhou, and Shenzhen, which had experienced significant rural migrant influxes. Some
migrants could apply for temporary or permanent urban hukou if they meet specific criteria, like stable employment
and contributing to social security funds. These pilots yielded favorable results, leading to gradual hukou relaxation
in more cities during the 2000s and culminating in a nationwide hukou reform in 2015.
11We used pkulaw.cn to collect prefecture-level hukou reform documents. We supplemented the information with
https://www.wanfangdata.com.cn/..
12Some conditions are that the rural applicants must have stable jobs and places to stay in the urban city.
13

and 2007. As shown, the implementation of the Unified hukou system has been gradual and uneven
across different regions. It was first introduced in Shijiazhuang and Ningbo in 2001. By 2007, 74
out of 334 Chinese prefectures had undergone reform. The reformed prefectures are geographically
dispersed across the country, as shown in Figure 1.3.
A key question is whether the chosen reform prefectures differ from others and whether the differences impact the average performances of local firms. Since our empirical strategy allows us to
compare changes in outcome variables of more-distorted versus less-distorted firms within a prefecture, our main specification does not require that prefectures be randomly reformed. However, such
a requirement is related to the identifying assumption for the average change in city-level employment and our industry-level estimates. To explore the nature of the reform’s timing, we compare
city characteristics between reformed and non-reformed prefectures in terms of GDP and population in the initial year of our study period. Additionally, we employ various local characteristics to
predict the reform implementation timing.
2001 2002 2003 2004 2005 2006 2007
Hebei
Zhejiang
Shijiazhuang
Ningbo
Fujian
Zhejiang
Ningde
Quanzhou
Nanping
Sanming
Longyan
Fuzhou
Ezhou
Zhangzhou
Jiaxing
Putian
Shandong
Jiangsu
Sichuan
Chongqing
Hebei
Fujian
Guangxi
Linyi
Xuzhou
Suzhou
Meishan
Suqian
Yibin
Chongqing
Wuxi
Yancheng
Mianyang
Deyang
Suzhou
Hengshui
Luzhou
Zhenjiang
Dazhou
Xiamen
Nanning
Taizhou
Huanan
Nantong
Changzhou
Shanxi
Sichuan
Hunan
Gansu
Shandong
Guangdong
Neimenggu
Jincheng
Chengdu
Changde
Lanzhou
Dezhou
Rizhao
Zigong
Shenzhen
Xiangxi
Heze
Jiuquan
Nanjing
Jiayuguan
Yantai
Pinzhou
Weihai
Foshan
Jining
Dongying
Binzhou
Huhehaote
Shandong
Guangxi
Hubei
Hunan
Shaanxi
Gansu
Shanxi
Guangxi
Henan
Guangxi
Shanxi
Shandong
Zibo
Liaocheng
Taian
Beihai
Guigang
Huangshi
Xiangtan
Weifang
Jinan
Zaozhuang
Xi’an
Pingliang
Weinan
Yuncheng
Hechi
Hebi
Laibin
Taiyuan
Qingdao
2015
All prefectures
Figure 1.2: Hukou Reform Timeline
Notes: The figure shows the timeline of the staged reform. Province names are above the axis, and prefecture names
are below the axis.
14

3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2007
Figure 1.3: Map of Policy Implementation
Notes: The map shows how reforms spread out geographically. Yellow dots represent the treated prefectures. The
base color in red represents the baseline amount of population in 2000.
We begin by plotting, in Appendix 3, the year-by-year relationships between reform implementation and initial prefectural characteristics (GDP and population). As evident in Figure 5 and
Figure 6, the reform’s implementation in the years 2000-2007 displays geographical dispersion across
the country concerning either GDP or population in the initial year. This suggests no systemic
differences in reform timing across prefectures based on these two observable characteristics.
Moreover, we employ a rich set of prefectural characteristics to predict reform timing. The characteristics variables include the number of local firms, average capital of local firms, average employment of local firms, share of non-agricultural residents, local minimum wage, and the HerfindahlHirschman Index (HHI).13 We present the results in Table 7 in Appendix 3. Most variables are
insignificant. These results provide some reassurance against the idea that reform implementation
timing results from observed local economic and social conditions. Consequently, reformed and
non-reformed prefectures are likely to follow similar trends in firms’ outcomes of interest. Further
discussions on timing issues will be explored in Sections 1 and 1, where we test the pretend of the
reform by conducting an event study analysis.
13The Herfindahl-Hirschman Index is a widely used measure of local market concentration and competitiveness
from the industrial organization literature.
15

Surge in Migrants following the Reform
Hukou reform reduced migration barriers and discrimination against internal migrants substantially, as discussed above. Reduced barriers and discrimination accelerate the inflow of migrant
workers into the reformed cities relative to cities that did not receive the reform. To measure
this impact, we calculate and report the relative changes in migration flows between reformed and
non-reformed cities, serving as a proxy for decreased migration barriers and discrimination against
internal migrants. In doing so, we use the 1% sample of the 2000 population Census and the 20%
sample of the 2005 mini-census of the population at the individual level. Both datasets are widely
used for studying migration in China and are discussed in greater detail in Section 1.
We define cross-prefecture migration and cross-sector migration as our measure of internal migration, given that prefecture is widely used as a proxy of China’s local labor market (similar to the
US commuting zone) and rural-to-urban migration represents the most important component of
China’s internal migration over the period (Zhang and Zhou, 2023; Ge et al., 2023). Specifically,
any worker aged between 15-64 employed in a prefecture other than their hukou’s prefecture for
more than 6 months is classified as a cross-prefecture migrant, following Zhou and Zhang (2021).
Additionally, any agricultural worker aged between 15-64 engaged in a non-agricultural occupation,
distinct from their hukou category (agricultural), is classified as a cross-sector migrant, following
Tombe and Zhu (2019).14
Table 1.2 presents the ratio of migrant workers in reformed and non-reformed prefectures for 2000
and 2005, along with their changes over 2000-2005. Across both cities reformed by 2005 and
non-reformed cities, the proportion of cross-prefecture and cross-sector migrant workers increased
significantly between 2000 and 2005. However, due partly to the hukou reform, this growth is more
substantial for reformed cities, exhibiting an additional increase of cross-prefecture migrants by
6.3% and 4.6% for cross-sector migrants. This underscores the role of reduced migration barriers
and discrimination in expediting migrant arrivals in reformed cities. Although the percentage
changes in Table 1.2 might not appear striking in magnitude, the number changes are not trivial
14Tombe and Zhu (2019) consider both rural-to-urban and urban-to-rural migrants in identifying cross-sector
migration, whereas we consider only rural-to-urban migration given our interest of examining mainly rural outmigration.
16

given China’s vast population.
Table 1.2: Ratio of Migrant Workers in Reformed and Non-reformed Cities
Cross-prefecture migration Cross-sector migration
2000 2005 Change 2000 2005 Change
Share of total employment (%)
Non-reform cities 4.2 7.9 +3.7 14.1 20.2 +6.2
Reform cities 7.4 17.4 +10.0 22.9 33.8 +10.8
Notes: Migrants are defined based on their hukou location. Cross-prefecture migrants are workers aged between 15-64 and registered in another prefecture from
where they are employed. Cross-sector migrants are workers who are aged between
15-64 and hold agricultural hukou but work in non-agricultural occupations.
To further illustrate the year-by-year impact of reduced barriers and discrimination on migration
inflow, we leverage the Annual Survey of Industrial Firms (ASIF) data to calculate the employment
growth rate of an average manufacturing firm spanning 1998-2007. The ASIF is the most comprehensive data in China that covers all state-owned and non-state-owned enterprises with annual
sales of at least 5 million RMB and is detailed further in Section 1. Covering more than 91% of
total industrial firm sales in China, the ASIF is particularly suitable for analyzing manufacturing
employment trends. Figure ?? displays the evolution of an average manufacturing firm’s employment and its evolution categorized by groups of reformed and non-reformed prefectures relative
to 1998. The annual employment change in an average firm shows a sharp increase immediately
following the initiation of the reform in 2001 (Figure 1.4). However, the growth is more pronounced
for firms from reformed prefectures compared to those in prefectures without the reform (Figure
1.5).
This evidence strongly suggests that the reform led to an increased influx of rural migrants, consistent with the fact that the hukou reform reduced migration barriers and discrimination against
17

rural migrants. Given the reform’s induced influx of migrant workers, which could augment the
overall labor input in manufacturing, understanding how more and less distorted firms responded
differently to the reform becomes crucial for understanding the reform gains. Below we examine
both the changes in productivity and output in manufacturing outputs following the reform.
-.1
0 .1 .2 .3 .4
Growth Rate in Employment (Baseline in 2001)
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Figure 1.4: Employment Expansion of an Average Firm -.2
0 .2 .4 .6 Growth Rate in Employment (Baseline in 2001)
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Reformed Prefectures Non-reformed Prefectures
Figure 1.5: Employment Expansion by Reform Status
Notes: Each point represents the employment growth rate in a certain year relative to 1998. A firm is classified as
reformed if the prefecture of the firm location is reformed in a year. Data source: ASIF.
18

Chapter One: Methodology
Measurement of Baseline Distortion Level
Misallocation, though technically unobservable, is commonly modeled as wedges affecting input
factors in the literature (e.g., Hsieh and Klenow, 2009; Restuccia and Rogerson, 2017). These
wedges can be positive or negative, depending on if they behave more like taxes or subsidies.
Throughout the labor misallocation we discuss in the work, we consider cases with positive wedges,
which represent firm-specific distortions on labor inputs.15 Given that firms are price-takers of
inputs in narrowly defined local labor markets, these exogenously given wedges, combined with the
input prices, represent the allocative prices firms actually pay.
To be specific, a firm i produces using capital Ki and labor Li to maximize its profits given by:
π
i = pifi(Ki
, Li) −
P
x∈K,L(1 + τ
x
i
)p
xxi
. Here, the firm will keep consuming inputs Ki and Li until
the marginal returns to these inputs equal their marginal costs. The conditions can be expressed
as: pi
∂fi(Ki,Li)
∂xi
= (1 + σi)(1 + τ
x
i
)p
x
, where pi represents output price, σi
is firm specific markup,
τi denotes firm specific input distortions, and px is market price of inputs. We are not able to
distinguish markup distortions and input wedges in our empirical settings. Therefore, we treat
them as a summation of broadly defined firm-specific distortions on inputs.
Assuming that all firms within a narrowly defined market use the same production function to
produce, firms should have identical marginal returns to labor, without accounting for different
sources of labor distortions. Under this assumption, firms’ input wedges are proportional to the
marginal products of labor, which is observable in the data (Hsieh and Klenow, 2009). Assuming
a Cobb-Douglas production function, we use average labor products (Y/L) to approximate the
marginal products of labor, conditional on the same labor demand elasticity within the local labor market.16 Consequently, more distorted firms exhibit higher labor productivity due to their
15Our discussion about cases with positive wedges is without loss of generality.
16Throughout our analysis, we follow Hsieh and Klenow (2009) and use a Cobb-Douglas production function. If
we relax the Cobb-Douglas production function to a more generalized CES production function, all of our results
are still valid due to the monotonicity of marginal products of labor in average productivity measured as y/l from
the data. The alternative measure of policy exposure we construct in Section 1 nicely circumvents the Cobb-Douglas
functional form and provides consistent results that will stay valid across all other possible CES production functions.
19

limitations in employment expansion.
We classify a firm as more distorted in the baseline if its pre-reform marginal product of labor
is above the pre-reform median within a given prefecture. Our primary interest lies in how more
distorted firms, indicating higher labor wedges, are affected differently by the reform, relative to less
distorted firms. The potential disproportional larger impact on the expansion of more distorted
firms suggests a convergence of marginal product of labor toward their most efficient use. By
focusing on these relative changes, our strategy circumvents the commonly used approach that
attributes all cross-sectional dispersion in observed marginal returns to inputs as misallocation,
which may create an upward bias in measuring misallocation (Restuccia and Rogerson, 2017).
Specifically, to establish each firm’s baseline distortion level before the hukou reform, we average
each firm’s labor productivity over our period of interest till the last year before the first hukou
reform (1998-2000). We then classify a firm as more distorted if its marginal product of labor is
above the pre-reform median. This restricts our sample to firms that can be observed before the
reform. In 2000, the average labor productivity of the baseline more distorted firms was around 328
thousand CNY and that of low productive firms was around 52 thousand CNY for low productive
firms, making it only one-sixth of the labor productivity of more distorted firms, as shown in Table
8 in Appendix 3.
Firms as Price-takers in the Local Labor Market
We assume homogeneity among workers within local labor markets. Given that the majority of
workers in our manufacturing sample possess low levels of education, it’s reasonable to anticipate
low wage disparities across firms in local labor markets. For instance, in 2004, more than 85% of
manufacturing workers held only high school or lower degrees.17 Additionally, to maintain consistency with the hypothesis of firms operating as price-takers in local labor markets, we consistently
include wage rates as empirical controls throughout our analysis below.
17From the data of the 2004 ASIF, an average firm in 2004 had 245 male workers and 104 female workers. Among
245 male workers, 126 have middle school or below education, 86 have high school diplomas, 21 have associate degrees
and only 10 have college degrees. Among 104 female workers, 59 have middle school or below education, 34 have high
school diplomas, 8 have associate degrees, and only 3 have college degrees. Over 85% of workers have high school or
lower degrees, and less than 1% of all manufacturing workers have graduate degrees.
20

While we empirically control for wage rates throughout the analysis in the paper and do not
find contradictory evidence against the price-taking behaviors of workers from the literature, it is
valuable to theoretically consider the scenario where the price-taking hypothesis is violated. In
Appendix 3, we theoretically relax the assumption that firms are price-takers by extending our
framework of exogenous distortions to monopsony power as a source of distortions. The lack of
competitiveness in the labor market can lead to both efficiency losses and lower aggregate output.
Empirically, the recent work by Armangue-Jubert et al. (2023) studies whether variations in labor
market competitiveness explain disparities in GDP per capita across countries using a structurally
calibrated oligopoly model. Their finding that the labor market distortion, reflected as the increase
in the labor market power of producers, can decrease aggregate productivity is consistent with
our empirical findings in the next section. This again, suggests that our assumption of the firms’
price-taking behaviors is valid without loss of generality.
Chapter One: Data and Variables
Our analysis is based on data from the ASIF as well as a representative sample of the 2000 and the
2005 population censuses of China. Our firm-level ASIF data comes from the census of Chinese
manufacturing firms and spans the 1998-2007 period. The National Bureau of Statistics (NBS) of
China conducts an annual survey of manufacturing enterprises, which is one of the most important
sources of information on the performance of the manufacturing sector in China. The ASIF covers
all state-owned manufacturing enterprises and non-state manufacturing firms with sales exceeding
RMB 5 million (approx. $600,000). While smaller firms are excluded, this dataset accounts for
90% of total manufacturing output.
The survey covers a wide range of topics, including production, investment, technology, employment, and exports, and provides a comprehensive picture of the current state of the manufacturing
industry in China (Brandt and Khandelwal, 2015). Our baseline outcomes include employment,
revenue, average labor productivity, and capital. We also use the total wage bill and end-of-year
employment to construct the average wage rate per capita. Summary statistics are reported in Ta21

ble 9 in Appendix 3.
18 The output growth of the manufacturing sector was extremely fast during
our study period.
Our migration data stems from the individual-level census records from China’s 2000 and 2005
population censuses.19 These datasets provide a comprehensive overview of the demographic and
socioeconomic characteristics in China. The census data covers a wide range of topics, including
population size and composition, urbanization, education levels, employment patterns, hukou statuses, housing conditions, etc. The 2000 census shows that China’s population had reached 1.26
billion, making it the world’s most populous country. It also reveals that the urbanization rate
had increased significantly, with 36% of the population living in urban areas, up from 26% in 1990.
The 2005 mini-census builds on the insights from the 2000 census and provides updated data on
the country’s demographic and socioeconomic trends, which shows continued urbanization, with
the urban population growing to 43% of the total population. It also reveals improvements in education levels and living standards, with more people obtaining higher education and better access
to housing and healthcare. To focus on migrants in the labor force, we retain migrants aged 15 to
64, dropping those migrating for educational purposes.20
To obtain our final dataset for analysis, we first aggregate the individual-level census data to
calculate the prefecture-level migration flows and then merge it with the policy data. The merged
dataset is used for our analysis of migration changes in Section 1. Then, we merge the firm-level
data with the prefecture-level policy data. We keep our sample to manufacturing firms from 1998
to 2007. To classify the baseline labor distortion levels of manufacturing firms, we restrict the
sample to firms that were observable before the reform. Although it is due to data availability,
restricting the sample to the decade has two advantages. First, while firm entry and exit were
drastic in the 2000s, we can observe the policy effects on firm dynamics by comparing results using
the whole sample, the balanced sample, and the restricted sample. Second, focusing on 1998-2007
avoids potential bias from other labor market liberalization during the early 1990s and in the later
18To be able to classify firms’ ex-ante distorted level, our main sample is defined as firms that were observed in
the pre-reform years from 1998-2000.
19The representative 2005 1% Population Survey, collected also by NBS. The sampling frame of the 2005 MiniCensus covers the entire population at their current place of residence, regardless of whether they hold local hukou.
20As in Imbert et al. (2022), migrants whose goal is to study are less than 5% of the total. They estimate that 45
million rural workers migrated to cities from 2000-2005, that is 16% of the total urban population in 2000.
22

2010s. Prefecture-level summary statistics are reported in Panel C of Table 9 of Appendix 3.
Chapter One: Results
Average Effects
To evaluate the reform’s impact on employment, capital, average labor productivity, and revenues
of an average firm, we estimate the average policy impact using the following equation:
Yijct = β1Reformjt + ΓXjt + θi + δt + ϵijct, (1.1)
where i denotes a firm, j denotes a prefecture-level city, t denotes a year, c denotes an industry,
and Yijct denotes firm outcome variables. Reformjt is an indicator variable equal to one if the
reform has been implemented in prefecture j in or after year t. We control for city characteristics
including GDP per capita and total population in Xjt. Firm fixed effect θi controls for timeinvariant firm heterogeneity, such as establishment year. Year fixed effect δt controls for time-specific
macroeconomic conditions, such as year-specific demand shocks and price shocks. We include firmtype fixed effects and industry fixed effects, which are not controlled for by time-invariant firm
fixed effects since there are firms switching their ownership types and 2-digit industry codes during
our studied period. Firm types include state-owned enterprises, collective firms, domestic private
firms, and firms controlled by foreign firms. At the 2-digit level, there are 29 distinct industries in
our sample of manufacturing firms.21 Standard errors are clustered at the prefecture level.22 We
define our main sample as firms that were observed at least once in the pre-reform years and thus
were eligible for the classification of baseline more distorted or less distorted status. This leaves us
with roughly 640 thousand observations.
21The Chinese industries and associated codes are classified as follows: processing of foods (13), manufacture of
foods (14), beverages (15), textiles (17), apparel (18), leather (19), timber (20), furniture (21), paper (22), printing
(23), articles for culture and sports (24), petroleum (25), raw chemicals (26), medicines (27), chemical fibers (28),
rubber (29), plastics (30), non-metallic minerals (31), smelting of ferrous metals (32), smelting of non-ferrous metals
(33), metal (34), general machinery (35), special machinery (36), transport equipment (37), electrical machinery (39),
communication equipment (40), measuring instruments (41) and manufacture of artwork (42).
22Clustering at the prefecture by year level and controlling for higher order industry by year fixed effects give us
consistent results.
23

Table 1.3 reports the results by estimating Equation 1.1. The estimates in column (1) and column
(3) suggest that the policy has an overall positive impact on firm employment expansion. More
specifically, the average firm employs 8.4% more workers and generates 9.6% more revenues. We
do not see a significant impact of the reform on the average labor productivity and capital of the
average firm. One explanation is that there are differential policy effects on different groups of
firms, and these effects potentially offset each other. Thus, it is important to look at differential
policy effects by estimation Equation 1.3, and we provide more discussion in the next Section.
The central government might select prefectures that are culturally more inclusive and are economically of higher growing potential for reform. For example, Beijing, one of the cities with the
most stringent labor regulations, was not selected in these rounds of hukou reforms. Therefore,
the positive effects of the reform on employment and revenues could actually reflect the superior
ability of the government to identify prefectures with high growth potential. If this were the case,
firms in these selected cities would have expanded their employment and generated more profits
even in the absence of hukou reforms. To assess the proposed possibility, we estimate the following
equation:
Yijct =
X
k=6
k=−9
β2kI(t = k)Reformjt + ΓXit + θi + δt + ϵijct, (1.2)
where i, j, t, and c represent firm, prefecture-level city, year, and industry, respectively. I(t = k) is
an indicator that is equal to one if the period is k years before or after the reform. Other variables
are the same as those in Equation 1.1.
24

Table 1.3: Average Effect of the Hukou Reform on Firm Outcomes from 1998-2007
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy 0.084*** 0.004 0.096*** 0.042
(0.029) (0.030) (0.025) (0.036)
Observations 638,458 636,376 638,454 636,880
R-squared 0.102 0.242 0.175 0.031
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one
if a city has been treated. The sample has been kept to firms who have been
observed at least once before the reform. Control variables include prefecturelevel GDP, population, and firm-level wage rate. Standard errors are clustered
at the prefecture level.*, **, and *** denote 10%, 5%, and 1% statistical
significance respectively.
Event study plots are shown in Figure 7 of Appendix 3. By constructing dummies for being
observed in periods before and after the reform, we estimate dynamic policy impacts on firms in
treated prefectures relative to firms in untreated prefectures. The absence of a pre-trend suggests
that there are no differential trends between the reformed and unreformed cities. Moreover, the
graphs suggest two things. First, policy effects are strong on firm employment and average labor
productivity, which are the two most direct firm outcomes affected by the reform. Second, labor
productivity (graph b) starts to decline in the year of the reform (period 0) and is declining
progressively, suggesting a persistent gain of the reform in terms of misallocation decline.
25

Differential Effects by the Baseline Distortion Level
While the average policy effects suggest that the reform generates revenues for an average firm by
enabling its employment to grow, the positive impact of labor liberalization on treated prefectures
can be either amplified or attenuated depending on whether more distorted or less distorted firms
benefit more from the reform. To study the differential reform impact between more distorted firms
and less distorted firms within a treated prefecture relative to an untreated prefecture, we estimate
the following equation:
Yijct = β1Reformjt + β2Reformjt × Ii(HP) + ΓXit + θi + δt + ϵijct, (1.3)
where i, j, t, and c represent firm, prefecture-level city, year, and industry, respectively. Ii(HP) is
an indicator variable equal to one if a firm has been classified as a baseline more distorted firm in
pre-reform years based on its baseline above-median labor productivity within a prefecture. Other
variables are the same as those in Equation 1.1.
The regression specification tests whether the reform affects employment expansion and various
other performances of initially more distorted firms relative to less distorted firms in prefectures that
have opened up to migrants relative to non-reformed prefectures. Therefore, we are interested in β2,
which captures the differential effect of the reform. A positive β2 would suggest that employment
expands for the baseline more distorted firms relative to the less distorted firms, thus withinprefecture labor misallocation declines. Since β1 measures changes in outcomes of initially less
distorted firms, the summation of β1+β2 measures the policy effects on outcomes of more distorted
ones.23
Table 1.4 reports the estimated differential policy effects for the initially more distorted firms relative to the less distorted firms in the treated prefectures relative to untreated prefectures. Results
23We are interested in the long-term effects of the hukou reform. In practice, this is usually done by estimating
long-term first-difference specifications, for example, for five years or longer. However, due to the staggered roll-out
of the unified hukou reform, we are not able to differentiate the so-called long-term effects from spill-over effects and
this is a common empirical challenge for studies using staggered roll-out of a reform. More specifically, we are not
able to know if the policy effects are due to a hukou reform that was implemented five years ago or are due to some
neighboring prefectures that were treated two years ago. Answering this question requires (1) long-term data, e.g.
ten years or longer; (2) a one-time shock instead of staggered shocks.
26

show that after hukou relaxation the more distorted firms employ 26.0% more workers, earn 4.1%
more revenues, use 11.8% more capital, and their MRPL (or equivalently, labor productivity) decreases by 23.8% relative to the less distorted ones in the treated prefectures relative to untreated
prefectures. This indicates a decline in dispersion in the marginal product of labor that is proportional to labor productivity and thus reduced labor misallocation within a prefecture. The
reform generates higher revenue for both groups, and the initially more distorted firms generate
4.1% higher revenues relative to less distorted firms. Thus following the reforms, the more distorted firms expand and grow by employing more workers due to decreased labor distortions. As
a consequence, they also become more viable firms and generate higher income relative to the less
distorted ones.
To avoid the reform over-eliminate the gap of average labor productivity between initially more
and less distorted firms, leading to a case that by the end of the reform, the labor productivity
of the initially more distorted firms is lower than that of the less distorted firms, we provide prereform average labor productivity for both groups of firms respectively in Table 1.4. Misallocation
declines as the dispersion of average labor productivity (5.52-4.11=1.41) between two groups of
firms decreases by 30% after the reform ((5.52-0.238)-(4.11+0.135)¡1.41). Altogether, misallocation
within prefectures declines due to the reform, given the sharp expansion in employment for the
baseline more distorted firms relative to the less distorted firms.
Mean reversion is a potential concern for our estimates, which means that the differential policy
effects we estimate simply represent that the employment size of the more and less distorted firms
are reverting to their long-term mean or average. However, the fact that the coefficients for the
more distorted firms are almost four times those of less distorted firms, alleviates potential concerns
on mean reversion.24
24Given this evidence, the assumption of lognormality of physical productivity (TFPQ) and revenue productivity
(TFPR) as in Hsieh and Klenow (2009) is inevitably violated.
27

Table 1.4: Differential Impact of the Hukou Reform on Firm Outcomes from 1998-2007
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy -0.058** 0.135*** 0.073*** -0.023
(0.029) (0.028) (0.025) (0.035)
Policy x I(HP) 0.260*** -0.238*** 0.041** 0.118***
(0.014) (0.013) (0.017) (0.017)
Pre-reform Labor Productivity Level
Low Productive 5.27 4.11 9.30 8.34
High Productive 4.90 5.52 10.32 8.55
Observations 638,458 636,376 638,454 636,880
R-squared 0.108 0.245 0.175 0.031
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if a city has been treated.
I(HP) is an indicator equal to one if a firm was initially more distorted within a prefecture. Pre-reform
dependant variables are at 2000 CNY levels in logarithms. Control variables include prefecture-level GDP,
population, and firm-level wage rate. Standard errors are clustered at the prefecture level.*, **, and ***
denote 10%, 5%, and 1% statistical significance respectively.
Event Study
A threat to our identification strategy is that the central government could intentionally select
prefectures that show larger potential to improve efficiency, based on their misallocation changes
before the reform. Therefore, the positive effects of the reform could reflect the selective superiority
instead of a true positive policy impact. If this were the case, the reformed prefectures would have
28

seen a larger employment expansion of the initially more distorted firms relative to the less distorted
firms even in the absence of the hukou reform. To address the concern rigorously, we employ the
event study approach to investigate how the trends of misallocation change over time between two
groups of firms within reformed prefectures relative to untreated prefectures.
Specifically, we construct dummies for being observed k years before and after the reform and
interact these dummies with the indicator of being the baseline more distorted firms. Our key
outcomes to assess misallocation decline from event study plots are employment and average labor
productivity. The equation we estimate is below:
Yijct =
X
k=6
k=−9
β2kI(t = k) × Reformjt × Ii(HP) + ΓXit + θi + δt + ϵijct, (1.4)
where i, j, t, and c represent firm, prefecture-level city, year, and industry, respectively. I(t = k) is
an indicator that is equal to one if the period is k years before or after the reform. Other variables
are the same as those in Equation 1.3.
Figure 1.6 presents the event study graphs for employment, labor productivity, revenues, and
capital, respectively. In general, we find no pre-trends detected before the reform, suggesting that
being in the treated prefectures did not have a strong differential effect on initially more distorted
firms before the actual enactment. Therefore, our results are less likely to be driven by pre-trends
in the shrunk gap of marginal product of labor (or the shrunk labor productivity gap) between the
two groups.
29

Figure 1.6: Differential Treatment Effect of Hukou Reform on Employment (left top), Labor Productivity (right top), Revenues (left bottom) and Capital (rigth bottom).
Notes: The figure reports event study graphs for the relative effects of the hukou reform on initially more distorted
firms relative to less distorted ones. Hukou relaxation takes place in year zero. The figures plot coefficients on the
interaction between being observed t years from the reform and being a more distorted firm in treated prefectures.
Confidence intervals are at 95%.
The effect of the hukou reform is progressive over time. Figure 1.6 suggests that 5 years after
the reform, more distorted firms expand in employment size by roughly 19% relative to that of
less distorted ones in treated prefectures. Capital also increases faster for initially more distorted
firms by roughly 10% about 5 years after the reform, although the magnitude is not statistically
significant. The progressive changes in employment and labor productivity reflect the possibility
that it takes time for firms in the treated prefectures to respond and adjust after the reform.
30

Geography Competition
Thus far, we have provided evidence that labor misallocation declines following the reform by enabling the initially more distorted firms to grow. In this subsection, we further explore whether
these policy impacts are amplified or attenuated by post-reform geographic competition. Early
adopters of the reform face abundant labor supply, while later adopters must compete more intensely for labor. Similarly, proximity to reformed cities intensifies competition for migrant labor.
Understanding this geographical competition is crucial for interpreting the reform’s impact on labor
(mis)allocation.
To assess geographical competition, we take advantage of the variations in reform timing and
geographic proximity. In the spirit of Donaldson and Hornbeck (2016), we construct an index to
measure the competition level by using the following formula:
Indexjt =
X
k∈J/{j}
{
1
distancejk
}(P opulationnonagri,k) × Ikt(k is opened),
where distancejk is the geographical distance between prefecture j and k. Holding everything else
the same, the farther prefecture k is to prefecture j, the less impact it could have on prefecture
j. P opulationnonagri,k is the urban population amount within prefecture k in 2000, including local
workers and migrant workers. I(k is opened) is an indicator equal to one if prefecture k had been
already reformed before year t.
Interacting the competition index with all single and double terms, we use a triple difference
approach and estimate Equation 1.5 to assess the impact of geographic competition on differential
reform effects for initially more distorted versus less distorted firms in treated versus untreated
prefectures:
Yicpt = β0 + β1P olicypt + β2P olicy × I(HP) + β3Indexcomp,pt
+ β4P olicypt × Indexcomp,pt + β5I(HP) × Indexcomp,pt
+ β6P olicypt × I(HP) × Indexcomp,pt
+ δi + δt + δc + δownership,it + ϵicpt,
(1.5)
31

where i, j, t, and c represent firm, prefecture-level city, year, and industry, respectively. Yijct
denotes the outcome variables, including employment, average labor productivity, revenues, and
capital. Reformjt is an indicator variable equal to one if the reform has been implemented in
prefecture j in or after year t. I
HP
i
is an indicator variable equal to one if a firm has been classified
as a baseline more distorted firm in pre-reform years based on its baseline above-median labor
productivity within prefectures. Indexcomp,pt is an index that measures the competition level faced
by prefecture p at year t, depending on whether nearby cities have already been reformed or not,
as well as on the size of the local labor market of other competitive cities. Controls and fixed
effects are the same as in our main empirical specification. We are interested in β5 and β6, how
competition could affect and limit the policy benefit a prefecture could obtain.
Table 1.5 reports the results. This suggests that while the initially more distorted firms employ
23.4% more workers relative to less distorted firms in the treated prefectures relative to untreated
prefectures, one standard deviation (1.04) of increase in competition levels could limit the employment expansion by 4.0%. This is consistent with the underlying mechanism that competition
among cities limits the employment expansion of a single city by raising the market wage rate due
to labor demand increase. Consistent with this hypothesis, we should see a weaker effect of relative
employment expansion in late adopters of the reform. For example, a city that was reformed in 2001
should see a larger relative employment expansion than a city that was reformed in 2007. We estimate equation Yicpt = β0 +β1P olicypt +
Pt=2007
t=2001 β2tP olicyt,ptIndexcomp,pt +δct +δt+δi +Xpt +ϵicpt
and report results in Table 11 in Appendix 3. Results suggest that policy effects are diminishing
over time, and the relative employment expansion of more distorted firms relative to less distorted
firms is 6.4% smaller for a city reformed in 2007 than a city reformed in 2001.
32

Table 1.5: Policy Impact by Competition Levels among Cities
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy 0.123 0.070 0.147*** 0.085
(0.075) (0.078) (0.049) (0.082)
Policy x I(HP) 0.193*** -0.203*** 0.029 0.030
(0.037) (0.032) (0.054) (0.048)
Weight -0.058*** 0.090*** 0.010 0.007
(0.021) (0.019) (0.026) (0.029)
Policy x Weight -0.062* 0.009 -0.023 -0.045
(0.036) (0.041) (0.028) (0.043)
I(HP) x Weight 0.146*** -0.104*** 0.058*** 0.052***
(0.008) (0.007) (0.012) (0.009)
Policy x I(HP) x Weight -0.040** 0.035** -0.024 0.024
(0.018) (0.017) (0.028) (0.027)
Observations 636,099 634,017 636,095 634,524
R-squared 0.131 0.251 0.177 0.032
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if a city has
been treated. I(HP) is an indicator equal to one if a firm is classified as a pre-reform more
distorted firm. Weight is a logarithm inverse distance weighted average competition index to
measure the competition level a city faces when a nearby city has been treated or received the
treatment earlier on. Controls include city GDP, population and firm wage rate. Standard
errors are clustered at the prefecture level.*, **, and *** denote 10%, 5%, and 1% statistical
significance respectively.
33

Baseline Migration Networks as Substitutes
In order to further understand the role of hukou regulation in shaping misallocation, we assess if
the predetermined migration networks shape the migrants’ responses to hukou deregulation. More
specifically, we examine whether local migration networks work as a substitute for labor market
regulation (introduced by the hukou system). Previous literature suggests that strong local migration networks introduce greater de facto labor market flexibility (Foster and Zhou, 2022), which
can be particularly relevant in the presence of strict labor regulations. This is because networks
have the function of channeling migrants into destinations with lower moving costs. Reasons for
this include information spread, job-search assistance, and easier adaptation to an alien environment (Carrington et al., 1996). Additionally, migration networks provide persistent insurance to
hometown families and acquaintances (Rosenzweig and Stark, 1989; Giles and Yoo, 2007). As for
the case in China, Zhao (2008) shows that experienced migrants have a positive and significant
effect on subsequent migration through practical assistance in the process of migration.
If hukou reform works as a substitute for historical migration networks, we should expect the decline
in labor misallocation to be more salient in areas with weaker baseline networks. This conjecture
has an important implication for how one interprets the misallocation impact from reduced barriers
to moving and the gains from labor market regulations. We rigorously test the conjecture by using
China’s hukou reform. This reform provides migrant workers an option to register for local unified
hukou, making it possible and flexible for migrant workers to stay legally in the urban cities and
to switch to jobs that value their skills, such as sewing, heavy items handling, etc. Two advantages
arise following the implementation of unified hukou reform. First, the efficiency increase in firms
and workers matching decreases labor wedges in the local labor market as more people and firms
rely on publicly available information to search for jobs and employees, such as career fairs and
career plazas.25 Second, the eligibility for partial participation in the local social security system
for migrant workers with a unified hukou alleviates the pressure of fees and extra costs that firms
need to bear in order to hire these people instead of local workers.
25Don’t complain about things that you can’t change, the Economist. See https://www.economist.com/china/
2012/06/02/dont-complain-about-things-that-you-cant-change.
34

To test this hypothesis of the substitutability between hukou reform and migration networks rigorously, we construct a variable M igration Networksj , defined as the fraction of migration over
the number of residents with local hukou in 2000. Next, we interact this variable with all the
single and double terms in our main staggered DID specification. Our coefficient of interest is the
coefficient of the triple interaction term that represents the differential effects on the decline in
labor misallocation in areas with stronger and weaker local migration networks.
Table 1.6 reports the results. Our estimates suggest that one standard deviation (0.60) of increase
in migration networks corresponds to a 2.7% reduction in relative employment expansion. We
also find that the differential average labor productivity decline is roughly 1.4% lower and the
differential capital expansion is roughly 3.2% lower in high-network areas. This confirms that hukou
reform decreases migration costs for migrant workers and increases employer-employee matching
efficiency by decreasing labor distortions. Cities with weaker baseline migrant workers may be
more misallocated if certain firms have privileged access to the local labor market or are less likely
to be punished even if their workers do not have legal work permits. Hukou reform increases
available workers to employ for all firms and thus allows workers to be allocated to the baseline
more distorted firms.
35

Table 1.6: Policy Impact by Baseline Migration Networks
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy 0.046 0.013 0.045 -0.092**
(0.034) (0.034) (0.031) (0.043)
Policy x I(HP) 0.215*** -0.213*** 0.026 0.069***
(0.029) (0.014) (0.026) (0.022)
Policy x Migration Networks 0.108*** -0.127*** -0.029 -0.071***
(0.021) (0.022) (0.020) (0.023)
Policy x I(HP) x Migration Networks -0.045* 0.024* -0.017 -0.054***
(0.027) (0.013) (0.026) (0.018)
Observations 607,195 605,186 607,191 605,678
R-squared 0.905 0.848 0.893 0.866
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if a city has been
treated. I(HP) is an indicator equal to one if a firm is classified as a pre-reform more distorted firm.
M igrationNetworkj is the log baseline fraction of migration over the number of residents with local hukou.
Standard errors are clustered at the prefecture level.*, **, and *** denote 10%, 5%, and 1% statistical significance respectively.
Alternative Measure of Policy Intensity: Industry-level Evidence
Our approach so far has relied on the correct classification of firms into distinct ex-ante distorted
levels within a city. To offer an alternative perspective without the need for this classification, we
conduct an industry-level analysis by constructing a novel measure of policy intensity that focuses
on the share of treated prefectures within each operating industry.
36

Specifically, for an operating industry, we compute the policy intensity by using the ratio of treated
prefectures within that industry over the total number of prefectures with the industry in that year.
This ratio provides insight into reform intensity from an industry perspective. Subsequently, we
use the variance of marginal labor productivity within an industry as the measure of the observed
degree of misallocation, a common method adopted in the misallocation literature (e.g., Hsieh and
Klenow, 2009).
Then, we re-estimate the effect of labor liberalization on misallocation at the industry level and
present the results in Table 1.7. These results suggest that after all prefectures get treated, i.e.
the ratio increases from zero to one, employment and capital increase by almost 7 times. Labor
misallocation, as the variance of industry-level labor productivity, reduces by 45.2-66.7% if all
prefectures get treated. Moreover, we find no evidence of capital misallocation decline. These
findings echo our previous firm-level evidence. Overall, the industry-level evidence circumvents the
issues of potential unsolvable measure errors and unobserved within-prefecture changes in firms’
performances between groups and thus reinforces our main findings.
37

Table 1.7: Policy Impact on Industry-level Aggregate Outcomes
(1) (2) (3) (4)
Employment Employment Capital Capital
A: Input Factor
Share(Treated) 1.744*** 1.792*** 1.628*** 2.109***
(0.340) (0.411) (0.408) (0.488)
Observations 1,677 1,673 1,677 1,673
R-squared 0.831 0.852 0.817 0.842
Var(Lab.Prod) Var(Lab.Prod) Var(Cap.Prod) Var(Cap.Prod)
B: Variance
of Marginal Products
Share(Treated) -0.452* -0.667** -0.248 -0.618
(0.256) (0.294) (0.590) (0.633)
Observations 1,672 1,668 1,672 1,668
R-squared 0.640 0.478 0.432 0.281
Fixed Effects
Industry FE ✓ ✓
Year FE ✓ ✓
Industry x Year FE ✓ ✓
Notes: All dependent variables are in logs and aggregated to the industry level. Variable
”Share(treated)=Number of Treated Prefecture in Year t/Number of Prefectures that have industry c
in Year t measures the reform intensity at the industry level. Employment and capital are industry-level
aggregate labor and capital. Misallocation is measured by industry-level variance of labor and capital
productivity, which are calculated using average labor products y/l and average capital products y/k. Regressions control for 4-digit industry FE and year FE in columns (1) and (3), and regressions control for
2-digit industry by year fixed effects in columns (2) and (4).
38

Robustness Analysis
Chapter One: Firm Entry and Exit
Our analysis so far relies on the ASIF data that covers non-state manufacturing firms with sales
exceeding RMB 5 million and all state-owned firms. Since small firms are excluded from the ASIF,
firm exit could be a concern if exiting firms, who were initially less distorted, happened to be
smaller in employment size and lower in productivity. Similarly, firm entry could be a concern if
entering firms tend to be larger and more productive relative to the incumbent firms. We first
compare the baseline firm characteristics of entrants and exiters in Table 12, and they exhibit very
similar ex ante employment and capital level. Taking one step further to address the concerns,
we use three different samples to test the robustness of the results from estimating Equation 1.3:
the balanced sample, the balanced sample and exiters, and the whole sample which includes the
balanced sample, exiters, and entrants. We report results on differential policy impact in Table
13 in Appendix 3. We employ the same strategy to test the average policy impact by estimating
Equation 1.1 and report the results in Table 14 in Appendix 3. As shown, our results remain robust
across all samples, suggesting that our main results are not driven by firm entry or exit.
Negative Weights
Our empirical approach for estimating the main results is a staggered difference-in-differences design
with two-way fixed effects including multiple groups and time periods. De Chaisemartin and
d’Haultfoeuille (2020) point out that difference-in-different estimates can yield biased weighted
average treatment effects (ATT) when units are treated at different times and thus are subject to
different (or even negative) weights. To test the possibility that negative weights bias our estimation,
we compute the weights implied in the decomposition of the DID estimator. The results show that,
under the common trend assumption, DID specification estimates a weighted sum of 303 ATTs, all
of which receive a positive weight. This relieves our concern regarding negative weight.
39

Alternative Cutoffs of the Baseline Distortion Level
Quartile-Based Classification. To test the sensitivity of our estimates, we employ alternative
strategies to categorize firms based on their baseline distortion levels. First, we categorize firms
into four sub-groups by their baseline distortion levels instead of only two groups. In doing so,
we use quartiles of marginal product of labor to classify firms in a similar way we classify firms
into two groups in the baseline estimates. Then we interact the quartile dummies with the policy
indicator. The results are reported in Table 15 in Appendix 3. As expected, employment increases
progressively if a firm belongs to an initially more labor-distorted quartile. More specifically, for
firms belonging to the quarter of the most distorted quartile (75%-100%), the reform increased their
employment by 41.3% compared to the least distorted firms (0%-25%), and their labor productivity
decreased by 40.3% relatively.
Prefecture-Industry-Based Classification. Second, we classify firms as more or less distorted
by using a more stringent strategy. Specifically, for each industry within a prefecture, we construct a new cutoff: an indicator equal to one if a firm has initially above median average labor
productivity within prefecture by 2-digit industry unit. Next, we reestimate our main specification
using the new classification. The estimates, detailed in Table 16 in Appendix 3, are similar to
our main results. In magnitude, for more distorted firms relative to the less distorted firms in
the treated prefectures relative to untreated prefectures, employment size increases by 20.7% and
labor productivity decreases by 23.1%. This suggests a fall in misallocation following the reform
within prefectures by industry. The new classification can also partially address the concern that
the differential effects are solely driven by the sorting of workers across industries.
Concurrent Events
Potential concerns include significant concurrent events that happened around similar periods,
among which China’s entry to WTO and the abolition of agricultural taxes drives the most attention. Prefecture-level exports increased sharply after China joined the WTO (Brandt et al., 2017;
Dai et al., 2020). To rule out the concern that our estimated hukou reform impacts are confound40

ingly due to the potential rise in migration between cities resulted from tariff liberalization, we
deal with the potential regional export shock by constructing a measure of export tariff reduction
and controlling it in the main specifications. More specifically, literature (Pierce and Schott, 2016;
Handley and Limao, 2017) suggests that the reduction of the tariff gap following the granted normal
trade relations (NTR) to China correlates with higher growth of exports. Inspired by Dai et al.
(2020), we construct a variable of export uncertainty reduction ”T ariff shock” by calculating the
prefecture-level NTR tariff gap weighted by prefecture-level industry employment share. Results
controlling for important concurrent events reported in Table 17 in Appendix 3 are not affected by
the inclusion of tariff uncertainty reduction and still support our argument of labor misallocation
decline within prefecture.
To relieve the load of rural farmers and mitigate rural social unrest (Chen, 2017) unrestrained since
the 1990s, the central government has begun to subsidize local government through intergovernmental transfer since 2001. Till 2006, all agricultural taxes were abolished nation-wide. The abolition
of taxes incentivized rural farmers to stay in rural areas due to the increase in agricultural revenues,
leading to an underestimated impact of the hukou reform during our study period. We control for
this concurrent event by constructing a variable Agricultural Reform indicating when and where
was the tax reform implemented as well as controlling for the occurrence of the tax reform in our
main specification. Results are reported in Table 17 in Appendix 3, suggesting that the observed
employment expansion of the baseline more distorted firms and the labor misallocation decline is
still robust even after controlling for other significant events during the same period.
Additional Fixed Effects
To rule out bias from various unobserved time-varying shocks, we include more stringent industryby-year (2-digit) fixed effects or province-by-year fixed effects to capture these shocks. Results are
reported in Table 18 in Appendix 3. The relative larger employment expansion and misallocation
decline of the baseline more distorted firms are consistent to our main results.
Finally, if the timing of the reform is not random but driven by some unobserved events or shocks
that occurred in or before the reform years, the estimated effects would be biased by at least partially
41

picking up the effects of the events. To address the concern, we construct an indicator equal to
one if a year is the reform year or the year before the reform and re-conduct our main specification
controlling for the reform year fixed effect. Results are reported in Table 18 in Appendix 3. Even
after controlling for the reform year fixed effects, initial more distorted firms still hire 27.5% more
workers and their average labor productivity falls by 26.0% higher relative to less distorted firms
in the treated prefectures relative to non-treated prefectures.
Prefecture-level Evidence
Using firm-level data to construct our measures of labor productivity has several advantages. First,
the empirical specification allows us to control for firm fixed effects and allows the macroeconomic
trends to be different across industries by controlling for higher-order industry-by-year fixed effects,
enabling the results to be more stable by comparing firms within prefectures. Second, it avoids the
incident where there are systematic measurement errors to proxy misallocation as prefecture-level
dispersion of labor productivity. Comparing differential responses of more distorted firms relative
to less distorted ones differences the potential systematical measurement errors out. Yet, as an
additional robustness check, we also estimate the prefecture-level impact of the reform on misallocation. In doing so, we aggregate firm-level outcomes to prefecture-level aggregate employment,
capital, revenue, and number of firms, and employ the variance of labor productivity within a
prefecture as a proxy of labor misallocation. The prefecture-level results are reported in Table 19
in Appendix 3, which remain consistent with our main results.
Functional Forms
Within 2-digit industries, firms could have different output elasticity to labor and different production functions. Thus the faster expansion in employment of more distorted firms may actually
capture the effects of firms’ heterogeneous production functions. To allay this concern, we assume
that firms have similar production functions within a more narrowly defined industry and output
market within the prefecture by a 4-digit industry. We interact the indicator of more distorted firms
(equal to one if firms have above median productivity within prefectures by 4-digit industries) with
42

the policy indicator. The results are reported in Table 20 in Appendix 3. These results show a
differential policy impact on faster growth in employment for more distorted firms relative to less
distorted firms in the treated prefectures, consistent with our main results.
Trimming Outliers
To ensure that our results are not driven by outliers, such as the largest manufacturing firms hiring
80% of the workers in the labor market, we trim the top and bottom 5% data and re-conduct
our main specification as an additional robustness check. The results are reported in Table 21 in
Appendix 3. As shown, initially more distorted firms hired 24.6% more workers after the reform
and their average labor productivity fell by 30.5% more relative to less distorted firms in the treated
prefectures relative to untreated prefectures. The results suggest that the differential policy effects
lead to a more concentrated marginal product of labor and thus reduced labor misallocation within
prefectures.
Additional Findings on Firm Heterogeneity
Chapter One: State-owned Enterprises
State-owned enterprises (SOEs) in China possess distinctive characteristics compared to private
firms, stemming from differences in the quality of firm owners, the regulations it is facing related
to rural workers’ employment, and its goal of operational profits (Chow et al., 2010). In literature,
SOEs are documented to be less constrained by their total net worth and less likely to go bankrupt
even when they are not profitable (Chow et al., 2010). Motivated by these differences, we examine
whether SOEs are less affected by the hukou reform. We conduct DID analysis for SOEs and
non-SOEs separately and report the results in Table 22 in Appendix 3.
26 The estimates show a
more pronounced gain in allocative efficiency for non-SOEs relative to SOEs, while both types
experienced a reduction in misallocation. Specifically, the relative employment expansion of more
26We follow several classification standards for the definition of State-owned enterprises. First, we define firms as
SOEs if their registration type is State-owned enterprises. Second, we define firms as SOEs if their registration type
is either State-owned enterprises or collective enterprises. Third, we define firms as SOEs if their state-owned capital
exceeds 30% (or 50%) of all capital investment. Our results are robust across different SOE definitions.
43

distorted firms is more pronounced among non-SOEs (22.8% among non-SOEs vs. 19.8% among
SOEs). As a consequence, the relative decline in average labor productivity is also larger among
non-SOEs (24.7% among non-SOEs vs. 11.9% among SOEs).
Several reasons can potentially explain why SOEs are less affected by the reform. First, positions
at SOEs often require local hukou status, which is difficult for rural migrants to obtain when they
first come to the urban cities. Second, as discussed in Hsieh and Song (2015), SOEs were often
under external pressure to hire redundant workers, especially those with political connections, thus
the increased labor supply of rural workers is less attractive to SOEs, who already have more
than enough employment. Third, private firms benefit more from the reform due to the downward
pressure of rural migrant supply on local market wages, especially those firms with tight budget
constraints. By contrast, SOEs are less likely to face hard budget constraints due to their richer
capital stock and stronger connections to China’s SOE bank system (Qian and Roland, 1998; Bai
and Wang, 1998).
Labor Intensity
A natural prediction is that hukou reform affected firms with more labor-intensive production.
To measure the labor intensity of firms, we measure it by the ratio of firm compensation over
total production due to the lack of a perfect competitive local labor market, instead of production
elasticity. To understand if the relative employment expansion is larger among high labor-intensity
firms, We conduct diff-in-diff analysis in the high labor intensity and the low labor intensity groups
separately. Results are reported in Table 23 in Appendix 3. The differential employment expansion
of more distorted firms relative to less distorted firms is roughly 31% among firms with high labor
intensity, while this number is only 25% among firms with low labor intensity.27
27See Imbert et al. (2022) for further discussions on the development of high-labor intensity industries. For
example, the disproportional expansion of labor inputs might lead firms to over-hire workers, consequently resulting
in reduced capital investment and a shift toward less capital-intensive production, which could potentially hurt firms
in the long run.
44

Firm Size
Literature suggests that firm size plays an important role in misallocation. For example, Gopinath
et al. (2017) find that a decrease in interest rate leads to an increase in capital misallocation in
Spain because capital is going towards firms that have higher net worth but are not necessarily
more productive. To figure out which group of firms benefits more from the reform, we divide our
sample into two groups: small firms and big firms, depending on their ex-ante employment size,
and we conduct our main specification. The results are reported in Table 24 in Appendix 3, which
show that big and more distorted firms hire around 23.3% more workers relative to less distorted
firms in the treated prefectures. This number was only 16.1% for small firms. The misallocation
(as represented by changes in MRPL) decline among big firms is also slightly larger. Overall, the
results suggest that big firms benefit more from the hukou reform due to the faster expansion of
big and more distorted firms, consistent with Hsieh and Olken (2014) who find that large firms
have higher marginal products and face more severe distortions in Both China and India.
Capital Misallocation
Our main results suggest that the more labor-distorted firms expanded both their labor and capital.
This could be due to capital-labor complementarity, and could also be due to the scenario that
some highly capital-distorted firms see hukou reform as a chance to generally grow and expand.
For example, some clothing factories may expect a potential overall growth of the industry and are
willing to invest in both labor and capital. In this case, we should see that hukou reform directly
increases more labor and capital for the ex-ante capital-distorted firms. We use a similar empirical
specification and classify firms as high capital-distorted firms if their marginal capital productivity
is higher than the median within a prefecture in the pre-reform years. We then interact the policy
indicator with the high capital-distortion indicator. Results are reported in Table 25. As shown,
for ex-ante capital-distorted firms, capital increases 42.8% higher, and labor increases 10.6% higher
than less capital-distorted firms. This suggests that, besides labor-capital complementarity, hukou
reform directly increases capital for capital-distorted firms and decreases capital misallocation. This
is consistent with our aggregate results in Section 1 that around one-third of the aggregate Solow
45

Residual (1.6% out of 4.6%) is due to the decrease in capital misallocation.
Chapter One: Aggregate Evidence and Policy Implication
To understand the aggregate implications of these estimates, we quantify the reform’s contribution
to the overall change in productivity and output in the manufacturing sector in this section. The
hukou reform is significant for manufacturing expansion for at least two reasons. First, it triggers a
substantial reduction in migration costs, facilitating the influx of rural migrant workers into cities.
Second, by reallocating labor to initially more distorted and therefore more productive firms, this
reform enhances allocative efficiency. We quantify the increase in allocative efficiency through a
first-order approximation using a methodology proposed by Bau and Matray (2023). Subsequently,
we estimate the expansion in manufacturing output by adopting a graph-based back-of-the-envelop
approach and incorporating shift-share style instruments.
Gain in Manufacturing Solow Residual
The analysis in Section 1 provides reduced-form evidence of a decline in labor distortions. However,
it does not indicate whether the effects of deregulation have a substantial impact on aggregate
production increases and aggregate productivity gains. To quantify the aggregate policy effects,
following Bau and Matray (2023), we estimate the change in the aggregate Solow residual using a
first-order approximation equation, as illustrated in the equation below:
∆SolowI,t ≈
X
i∈I
λi∆logAi +
X
i∈I
X
x∈K,L
λiα
x
i
τ
x
i
1 + τ
x
i
∆logxi
.
where ∆logAi refers to the change in TFPQ for firm i, and λi represents the ratio of the firm’s
sales to industry I’s net output. τ
x
i
represents firm i’s input wedges to input factor denoted by x
(capital or labor) before the hukou reform, α
x
i
refers to the output elasticity to input x, and ∆logxi
captures firm i’s change in the log input x.
We rely on conventional measures of distortion. In the literature, the combined wedge is usually
46

modeled as the deviation of a firm’s factor use from its efficient level. In this paper, we calculate
capital and labor distortion using equations τil = αjL
piYi
ωili
−1 and τik = αiK
piYi
riki
−1. The underlying
assumption is that misallocation solely contributes to the ex-ante deviation of factor expenditure
shares from production elasticity.28 We use the average distortion from the years 1998-2000 as a
firm’s initial distortion before the reform. The estimated wedges for labor and capital are reported
in Table 26 in Appendix 3. The average labor wedge in 2000 was about one. For most industries,
wedges in 2000 are positive, indicating an ex-ante distortion in labor factor misallocation and an
under-hired employment size. The under-hired employment could be a result of labor market
frictions created by the hukou system, as discussed in Section 1, alongside various other factors
such as the Custody and Repatriation (C&R) system and the minimum wage regulation.
We predict a change in distortion due to the reform employing our main empirical specification.
Taking the initial labor and capital productivity distribution across firms into account, we investigate policy effects differing due to varying levels of firms’ ex-ante marginal labor and capital
productivity. We are interested in how each group responds to the shock and alters their input allocation accordingly. Therefore, we regress the logarithm of average labor production (and average
capital production) on group dummies interacted with the policy dummy controlling for firm fixed
effects, year fixed effects, industry fixed effects, and ownership fixed effects.29 The predicted value
is thus approximate to the percentage change in firm distortion.
Using a difference-in-difference specification has several advantages. First, it is less sensitive to
the issues that occur when cross-sectional data are used to estimate distortions (Bau and Matray,
2023). Second, by controlling for year-fixed effects and industry-by-year fixed effects, our specification absorbs firms’ time-invariant measure error and macroeconomic trends. Third, we are able
to disentangle the reduced distortion due to hukou reform from other impacts by predicting the
changes in marginal products using both the policy indicator and the pre-reform marginal productivity level of firms. Table 27 in Appendix 3 shows that labor wedges decreased by around 2.0% for
28For pre-reform capital expenditures, we follow Hsieh and Klenow (2009) and set the rental rate to 0.1. As in
Bau and Matray (2023), the choice of a relatively low value of r=10% is reasonable because both the capital wedges
and the estimated aggregate effect are decreasing r.
29As some firms have switched ownership and changed the industry they belong to during this period, firm fixed
effects alone might not suffice in controlling for those changes, so we additionally control for industry and ownership
fixed effects.
47

treated prefectures.30 We predict the input factor change using the same approach and regress the
logarithm of employment/capital on the policy indicator, firms’ baseline distortion level indicator,
their interaction, and the same set of fixed effects. Table 27 reports the predicted average increase
in labor at around 12.2% and the predicted average increase in capital at around 1.1%.
Using the first-order approximation equation, we quantify the aggregate productivity increase resulting from a reduction in misallocation for the treated prefectures. The number suggests that by
enabling workers to legally migrate across cities, the treated prefectures obtain an aggregate productivity gain of 4.6% as reported in Table 28 in Appendix 3, weighted by industry output. If we
adopt more strict assumption on the joint log-normal distribution of TFPQ and TFPR, we can use
from Hsieh and Klenow (2009) the residual gain measure −
θ
2∆V ar(MRP L)
31 and borrows from
Table 1.7 the policy impact on the dispersion of marginal factor products. This gives us an aggregate Solow residual gain of 15.0%, suggesting that ignoring the fact that some strict assumptions
do not apply could potentially lead to an overestimation of the aggregate Solow residual gain.
Table 1.8: Solow Residual Gain from Different Methods
(1) (2) (3)
Declined Labor Misallocation ∆Solow Assumptions
Method 1 -2.0% 4.6% N/A
Method 2 -15.0% 15.0% log-normal distribution
Notes: Method 1, following Bau and Matray (2023), estimates changes in labor inputs and
aggregates to the whole economy level using a first-order approximation in Equation 1.
Method 2, following Hsieh and Klenow (2009) uses predicted changes in variance in labor
productivities and values for variety elasticity. Method 2 needs to assume a log-normal
distribution of TFPQ and TFPR.
30Since the change in distortion is small, the interpretation for the predicted change in log marginal revenue product
is in percentage. We allow the difference-in-difference regression to include a flexible function, g(policyjt, Cit), where
Cit is firm characteristics. This is to say, ∆τ
l
i can be estimated by ∆ˆτ
l
i = exp(g(policyjt = 1, Cit)) − 1. When the
∆τ
l
i
is small, log (1 + ∆τ
l
i ) ≈ ∆τ
l
i
from Taylor series.
31We set θ to be 2.
48

Gain in Manufacturing Output
While the hukou reform has boosted aggregate Solow residual gain by allocating labor into the most
productive firms within prefectures, it could have an even bigger impact on aggregate manufacturing
production due to the induced labor market expansion and the resulting increased demand for goods
by the induced migrants. To quantify the policy effects on manufacturing output gain, we break
the gain into two parts: labor market expansion and allocative gain.32 We illustrate the hukou
reform gain in Figure 1.7 below.
S0
S2
D0
D2
MPL2
ω0
ω2
MPL0
L0 L1 L2
ΔMPL
Δ ω
E
A1
PQ A2
Figure 1.7: Manufacturing Output Increase due to Hukou Reform
As Figure 1.7 shows, we break the output gain into pure market expansion and gain from misal32In Borjas and Edo (2023), they find evidence of GDP increase due to labor market expansion in France. The
legalization process of immigrant workers expanded the labor market by decreasing inverse labor supply elasticity,
i.e. firms’ monopsony power decreased because of a more elastic labor supply.
49

location decline motivated by the present literature (e.g., Graham and Spence, 2000; Jones et al.,
2013; Borjas and Edo, 2023). The baseline labor L0, baseline marginal products of labor MP L0
and market wage rate ω0 constitute the pre-reform equilibrium with labor distortion MP L0 − ω0
and manufacturing output P Q. The reform has: (1) increased manufacturing output by moving
more workers L1 −L0 to urban cities while holding the labor distortion fixed (area E in the graph);
and (2) allocated more workers to high-productive firms within prefectures (area A1 and A2). However, we do not directly observe the exact amount of labor increase due to pure market expansion
L2 − L1 from the data, thus we do not know the output gain that can be solely attributed to
market expansion while holding labor wedges fixed. Nevertheless, leveraging the predicted input
change and wedge change alongside baseline wedges and estimated factor price elasticity allows us
to approximate the impact.
To estimate the factor price elasticity of labor, we employ the following regression specification :
ln(wageit) = α0 + ϵln(labor)it + δi + δt + δct + ownershipit + ηit. (1.6)
This equation allows us to estimate to what extent wage responded to labor input change after
controlling for firm, year, industry by year, and ownership fixed effects. However, using Ordinary
Least Squares (OLS) to estimate these parameters can lead to biased and inconsistent results
due to the inherent interdependence of wages and employment levels. Consequently, we employ
an instrumental variable approach, using a shift-share style measure of firm-level labor supply
shock.33 To construct the shift-share variable, we multiply the industry-level migration policy
intensity (measured as the number of treated prefectures in a specific industry for a given year)
by baseline industry composition in each prefecture and then multiply it by the indicator of prereform more productive firms within prefectures. Our identification is based on the idea that a
positive migration inflow shock impacts firms’ marginal revenue returns to labor, consequently
influencing labor demand. Prefectures face differential exposure to this shock through different
industry compositions, and firms can be exposed differently depending on their baseline constraints
in accessing migrant workers.
33Previous literature, such as (e.g., Amodio et al., 2023; Diamond, 2016a), uses shift-share demand shifters to
estimate the labor supply elasticity faced by firms. Here, we need a supply shifter to help us estimate firms’ demand
curve.
50

The validity of our IV rests on three key assumptions. First, this shift-share variable is correlated
to labor supply. As shown in Table 29 in Appendix 3, one standard deviation (0.029) increase in
city-level migration supply shock increases firm employment by 5%. Second, the IV is orthogonal
to firm-level wage and employment as it arises from national-level migration shock. This is likely to
be true as policy-induced migration supply shocks are exogenous. Finally, the exclusion assumption
requires that the IV affects wages only through labor supply. Much of these concerns have been
taken care of by including industry-by-year fixed effects. From Table 30 in Appendix 3, we set
the factor price elasticity to be -0.44. The aggregate expansion in labor is estimated to be 12.2%,
and the decreased labor wedge is estimated to be 2.0%. To obtain the aggregate output increase
in percentage term, we estimate the percentage increase in manufacturing output represented as
area R (= area E + A2 + A1 + A1) over the baseline manufacturing output P Q (as nominal price
P times real output Q):
R =
MRP L(L2 − L0)
P Q
≈ 14.6%,
A1 = 0.5∆MRP L(L2 − L1) ≈ 0.04%.
Since A1 is a small number, we take it approximately as zero. Now that the aggregate output
gain is 14.6%, we break it into allocative gain and market expansion. To do so, we estimate the
trapezoid area of allocative gain (area A1 + A2) as:
A1 + A2 ≈
MRP L1(L2 − L1)
P Q
≈ 4.8%
Table 1.9 reports the results. Since we do not observe a significant positive policy impact on firmlevel productivity, we assume that all Solow residual gain is solely from allocative efficiency. Results
suggest that the allocative gain of output is only half of the market expansion gain of output. If
we use GDP to measure output level, based on China’s manufacturing GDP in 2000 (4 trillion
CNY/0.48 trillion USD), the manufacturing output increase due to hukou reform is approximately
14.7 billion USD).34
34We take the exchange rate of USD to CNY as 8.28 in 2000. The manufacturing output increase is based on the
percentage expansion of roughly 14.6%, the fraction of prefectures treated (71/334), and the baseline manufacturing
output of around 4 trillion CNY/0.48 trillion USD in 2000.
51

Table 1.9: GDP/Production Gain by Sources: Allocative Gain and Market Expansion
(1) (2) (3)
Total Allocation Gain Market Expansion
GDP/Production 14.6 4.8 9.8
Notes: All numbers are in percentage.
Chapter One: Conclusion
We evaluate the decline in labor misallocation associated with a specific labor market liberalization
in China: the unified hukou reform, which substantially reduced labor market barriers to internal
migrants. The reform’s staggered implementation across prefectures triggered a considerable influx
of internal migrants into reformed cities. This allowed us to estimate shifts in input wedges across
firms within prefectures, reflecting firms’ deviations from efficient allocations. Our findings suggest
that, for initially more distorted firms relative to less distorted firms, the reform led to a 26%
increase in employment, a 4.1% increase in revenues, a 11.8% increase in capital, and a 23.8%
decrease in marginal revenue products of labor (or equivalently, labor productivity). Geographically,
the policy’s impacts are more pronounced in regions facing severe competition from ”early reform
adopters”. We also find that the gains from moving are enhanced in regions with weaker pre-reform
networks of migration. Additionally, we examine firm heterogeneity and find that non-state-owned,
labor-intensive, and larger firms benefit more from the reform relative to others.
In the second part of our analysis, we quantified the reform’s contribution to overall productivity
growth and aggregate manufacturing output. A conservative estimate suggests that the reform,
by reducing migration barriers, increased the aggregate Solow residual gain by 4.6% and the manufacturing output by 14.6% between 2000 and 2007.35 Our analysis provides insights into why
labor market liberalization reforms, especially in developing countries, can yield substantial bene35This accounts for roughly 15% of all aggregate productivity increases. Brandt et al. (2012) finds that the
weighted average annual productivity growth for incumbents is 2.85% for a gross output production function and
7.96% for a value-added production function over the period 1998–2007. Adopting a 2.85% annual growth rate, the
total growth during 1998–2007 is around 29%.
52

fits associated with a more efficient allocation of input factors across producers. While reforming
various institutions can be politically challenging, the gains resulting from these reforms are huge
and should motivate further research into conducting and understanding such reforms.
However, important questions on underlying mechanisms through which the liberalization reforms
take effect still remain to be explored further. One potential answer to this research question is that
the reforms can lower labor market friction by improving matching efficiency between employees and
employers. In the appendix, we provide a theoretical framework on how migration reform reduces
monopsony power differently across producers. Understanding such modeled mechanisms, alongside
unmodeled ones, is pivotal in understanding why heterogeneous firms that are more bounded in
labor employment benefit more relative to others, thus reducing the dispersion of marginal returns
to labor across firms. This is an important research question for further study.
53

Chapter 2
Policy Assessment Related to Return Migrant Workers
With Wei Zhou
Chapter Two: Introduction
China has witnessed one of the greatest internal migration across the world. Among hundreds of
millions of migrant workers, around 70% of which remains to be rural-to-urban migrants. However,
according to Su et al. (2018), as the labor demand declines following the maturity of internal and
international trade for urban firms, the growth rate of rural migrant workers falls sharply.
There is a growing body of literature on internal migration in China. However, few focus on return
migration, although there is an enlarging trends in its existence. Zhang et al. (2020) among these
papers studies how hukou barrier leads to an expanded probability of returning among low-skilled,
rural and inter-provincial migrants who moved from more developed working destinations back to
their home prefecture using the 2017 China Household Finance Survey (CHFS).1
This paper is among the first papers studying policy impact related to return migration encouragement. Specifically, a national plan called Summary of China Urbanization for 2014-2020 unveiled
on the Eighteenth National Congress of the Communist Party of China seeks to improve the business environment, labor market condition, and service functions for medium and small size cities,
making them sufficiently good places for return migration to live in. One subsequent series of poli1Hukou, a household registration in China, is attached to lots of social security, medical, education and accommodation benefits.
54

cies is to encourage return migration and entrepreneurship in some county-level cities.2 Support
and help is said to be provided to return migration and local residents on business permit granting,
factory site choosing, credit loans and general guidance. Three rounds of selected counties were
released by the National Development and Reform Commission and other related departments from
February 2016 to October 2017.
We study the policy impact by conducting a staggered difference-in-difference analysis. We found
that the benefits stated to be provided by local government to encourage entrepreneurship incentivized around 7.9% more residents to either move back or move to treated counties following the
policy implementation. However, we do not observe an increase in the number of people who initiate their own business. Instead, there is an increase in the number of people conducting occasional
job, which clearly provide less social security and benefits than an employed job. There are a
couple of reasons behind that. First, although the policy documents encourage local government
to establish leadership team and ease the credit loan for local entrepreneurs, the fraction of local
residents who successfully applied and received the loan is as low as only 5%, suggesting both the
factual difficulty in credit loans and the lack of knowledge and confidence in using loans to boost
small business. Second, the business environment indeed has become more vibrant and healthy for
pre-existing small family business, as their business profits have increased by about 43.9%. However, for employed workers, their wage rate has declined by around 24.6%-34.6% due to increased
local labor supply. The over-sufficient return migration flows put a downward pressure on local
employed wage rate and tighten the job market, crowding workers out to find occasional work.
We present an empirical spatial equilibrium model to quantify the entrepreneurship encouragement
policy in the future—how much (in percentage of original income) do you need to pay a worker for
him to be willing to return to smaller counties. And if at the end of the day, these workers found
out that they can not earn this fraction of incentivized income but in the short-run they won’t be
able to return to more developed urban cities either, what’s gonna be their welfare loss? We are
hoping to answer this question in the future.
This paper contributes to the literature in two aspects. First, this paper relates to work on migration’s impact on firms and local labor market. Migration-induced positive labor supply shocks
2
In China, county-level cities are usually smaller than prefecture-level cities.
55

improved labor market matching efficiency and introduced lower labor market friction. Past work
has analyzed its impact on the wage rate for local workers An et al. (2024), on the reallocation
of labors across industries Dustmann and Glitz (2015), labor adjustments of local manufacturing
firms Wang et al. (2021); Imbert et al. (2022). Although there are some papers Yu et al. (2017);
Wu et al. (2018); Wei and Zhu (2020) documenting the facts of return migration, less is know about
the return migration policy impact on local labor market and entrepreneurship.
Second, this paper presented unintended consequences of the return migration policy and seek to
leverage an empirical spatial equilibrium to address its importance. Past literature has studied
negative impact of the return migration supply on local rural workers Hu et al. (2023). This paper
reaches the same conclusion and suggests the use of an empirical spatial equilibrium Diamond
(2016a) to quantify the welfare loss of both return migration and local workers after program
launching.
The rest of our paper is organized as follows. Section 2 gives an overview of the institutional
background. Section 2 describes the data and main variables that we used in this chapter. Section
2 states our empirical strategies and reports our results, including individual-level results on work
type, county-level results, and potential mechanisms for increasing occasional workers. Section 2
suggests an empirical spatial equilibrium model to be used in the future to quantify policy impact.
Section 2 concludes.
Chapter Two: Institutional Background
Migration in China
As discussed in Section 1, migrants in China are defined as those who do not hold a local hukou
in an area where they work and live most (Zhou and Zhang, 2021; Jin and Zhang, 2023). The
internal migrants, often referred to as the “floating population”, represent a huge number of rural
people flooding to cities for job opportunities while keeping a family in their rural birthplaces. The
floating population makes up around 130 million in 2000 and the number had reached 221 million
56

by 2010, which is roughly 17% of China’s total population.3 Such a large-scale movement of labor
is truly once-in-a-lifetime.
Internal migration in China was minimal before 1980s due to strict central governmental control
over resource allocation. In many papers, such as Zhao (2000) and Foster and Zhou (2020), it
has been demonstrated that the household registration system is responsible for the rigidity of
migration flows. From 1980s and onwards, the total number of “floating population” has been
increasing ever since. Beginning in the late 1990s and early 2000s, especially following China’s
entry to WTO, the expanded demand from export-oriented industries was eager to hire massive
cheap unskilled rural migrant workers, particularly in the Eastern coastal regions.
After the 2010s, however, according to Su et al. (2018), the labor demand declines following the
maturity of internal and international trade for urban firms, and the growth rate of rural migrant
workers falls steeply from around 6% in 2010 to less than 1% in 2015 despite the continuous growing
total numbers of migrant workers.
The Return Migration Policy and Residents Increase
It’s an important question for policy makers to develop related policies potentially contributing to
migrating flows in various directions and understanding its consequences. The Eighteenth National
Congress of the Communist Party of China along with the State Council unveiled a national plan
called Summary of China Urbanization plan for 2014-2020 on March 16, 2014.4 According to
the plan, one of the major objectives is to increase the number of small and medium cities and
improve the service functions of small towns. Making these smaller cities comfortable and livable
should accelerate the fraction of urban residents with local hukou.5 The plan share similar spirit
as situ urbanization process in which marginal rural-urban places are transformed and improved
to fully urban places over time. As a series of succeeding policies, migrant workers are encouraged
to return home and initiate entrepreneurship in smaller towns. Three waves of treatment cities are
3According to data from the 2000 and 2010 Population Census (Liang et al., 2014).
4China Workshop 2014, National Plan.https://sustainability.wisc.edu/past-initiatives/
china-workshop-2014/national-plan/
5One of the goals of this plan is to have permanent urban residents reach 60% of the populace. Permanent urban
residents rose from 17.9 percent of the total population in 1978 to 53.7 percent in 2013.
57

selected: 90 county-level cities in February 2016, 116 county-level cities in December 2016, and 135
county-level cities in October 2017.
We map the geographic location of these treated counties in Figure 2.1. As indicated from the
map, the assignment of treatment was scattered geographically and voided to be concentrated in a
specific region.
Figure 2.1: Map of Treated Counties
To study whether an increase in the number of residents exists for treated counties, we follow the
following empirical specification at the county level: Yjt = α0 + α1 × P olicyjt + Controlsjt + δt +
δj + ϵjt, where j denotes a county-level city, t denotes a year, Yjt denotes county-level number of
residents, including fraction of households engaging in certain types of employment. Regressions
control for county-level male ratio, average number of household members, county fixed effects and
year fixed effects. Our results show that, the number of residents has increase by 15.9, while the
average number of surveyed residents is around 200 per county/neighborhood community. This is an
increase of around 7.9% in residents for treated counties. However, without necessary information
on one’s work history in urban cities and detailed information on one’s origin and destination, We
58

weren’t able to directly measure the number of people who returned. We seek to supplement this
section of analysis in the future should data available.
Chapter Two: Data and Variables
This paper relies on the data from the China Household Finance Survey (CHFS), a nationwide
sample survey project conducted by the China Household Finance Survey and Research Center
at the Southwestern University of Finance and Economics in China. The aim of this survey is
to collect comprehensive micro-level household data including demographics, employment, assets
and liabilities, income, consumption, social security, medical insurance and other insurance. The
baseline survey began in 2011. Following the baseline survey, four other rounds of survey projects
have been completed in 2013, 2015, 2017, and 2019. The sixth round of survey conducted in
2021 is open for in-university faculty use only. Over these rounds of survey, 34,643 representative
households from 29 provinces (autonomous regions and municipalities), 343 counties (districts),
and 1360 villages (neighborhood communities) are covered.6
The variables we are interested in include household demographics, type of employment (employed,
self-employed, farming, and occasional), engagement in small family business operation, earnings,
and related medical insurance participation, usage, and reimbursement. For policy treatment, we
manually collected the name of the counties and year of treatment from the documents issued
by the National Development and Reform Commission (NDRC) in 2016 and 2017. We map the
geographic location of these treated counties in Figure 2.1. The map suggests that the assignment
of treatment is scattered and spreading-out geographically and not concentrated in a specific area.
To obtain our final dataset for analysis, we first merge the individual-level data with household-level
data and village-level (neighborhood community level) master data to get one file for each year. We
then append survey data from 2013, 2015, 2017, and 2019 together to get panel data. This file is
then merged with policy treatment data using unique county-level geographical codes. We restrict
the sample to households with a clear hukou type and with age 18-80. 7 We are left with around
6CHFS China Household Financial Survey Data https://www.cnopendata.com/en/data/m/chfs/chfs.html
7When answering the question—what is your hukou type, a very few individuals answered that their hukou type
59

338,800 individual-year observations and 69,600 unique households. Around 77,700 individual-year
observations and 7,379 unique households stay through-out four rounds of survey from 2013-2019
in the balanced sample.
Chapter Two: Results
Work Type
To evaluate the impact of policy that aims to encourage return migration on work type switching
of local residents, we estimate the policy impact using the following equation:
Yijt = α0 + α1 × P olicyjt + Controlsijt + δt + δi + δprov,t + ϵijt (2.1)
where i denotes an individual or a household, j denotes a county-level city, t denotes a year, Yijt
denotes individual or household outcomes of employment type, including employed, self-employed,
farming and occasional job. Each is a dummy indicator equal to one if the individual or household’s
current employment type is such. The treatment dummy P olicyjt is an indicator equal to one if a
county has been treated in or after year t. Controls include gender, education, number of family
member, marital status, health condition, and retirement status. We control for time-invariant
household fixed effects and year fixed effects. This takes account of household heterogeneity and
time-invariant shock that applies to the whole country. We also control for more stringent province
by year fixed effects to allow province-level unobserved shocks and development. Standard errors
are clustered at the county level. We further keep the analysis sample to balanced sample in urban
areas, since the main economic activities in rural areas are solely farming. 8
Table 2.1 reports the results by estimating Equation 2.1. To avoid potential bias from household
entry and exit, we keep the analysis sample in Table 2.1 to balanced sample of households. Depenis undefined and unclear. We eliminated these people.
8Urban and rural area is an indicator depends on the survey interviewer’s judgement when conducting the
interview. Within the same county, there could be both urban and rural households. Rural households are mainly
live remotely and make their living by farming.
60

dent variable in Panel A of Table 2.1 is an indicator equal to one if at least one household member
is engaged in a certain type of employment. For example, F arming = 1 means that at least one
person in the household does farming for living. Dependent variable in Panel B is an indicator
equal to one if household head does certain type of work. From Table 2.1, the return migration
policy has increased the number of households with at least one member doing occasional work by
roughly 5.1% and that for household head by roughly 2.8%. The ratio of household members who
are engaged in a certain type of employment within a household does not show a significant increase
for occasional work, although the magnitude is positive (1.5%) and marginally significant. This
suggests that the above employment type switch is mainly achieved by extensive margin rather than
intensive margin. More households undertake occasional jobs following the encouraged migration
flows.
61

Table 2.1: Household Employment Type Change in Urban Regions
(1) (2) (3) (4)
A: Household Employed Self-employed Farming Occasional
policy -0.000 -0.010 -0.008 0.051**
(0.030) (0.028) (0.019) (0.026)
Observations 14,250 14,250 14,250 14,250
R-squared 0.989 0.988 0.994 0.815
B: HH Head Employed Self-employed Farming Occasional
policy -0.012 0.047** -0.013 0.028*
(0.021) (0.018) (0.016) (0.017)
Observations 14,250 14,250 14,250 14,250
R-squared 0.991 0.987 0.994 0.705
B: Ratio Employed Self-employed Farming Occasional
policy -0.003 -0.007 0.006 0.015
(0.014) (0.013) (0.011) (0.012)
Observations 14,250 14,250 14,250 14,250
R-squared 0.991 0.988 0.994 0.759
Household FE ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓
Province by Year FE ✓ ✓ ✓ ✓
Notes: Controls include age fixed effects, gender, education levels, retirement status,
number of household members, hukou types, marital status, and health condition.
Regressions are clustered at county level. Robust standard errors in parentheses. ***
p<0.01, ** p<0.05, * p<0.1
62

County-level Evidence
To evaluate the impact of return migration policy on county-level outcomes, we follow this equation:
Yjt = α0 + α1 × P olicyjt + Controlsjt + δt + δj + ϵjt (2.2)
where j denotes a county-level city, t denotes a year, Yjt denotes county-level outcome, including
fraction of households engaging in certain types of employment. Regressions control for county-level
male ratio, average number of household members, county fixed effects and year fixed effects.
Results are reported in Table 2.2. County-level fraction of households with at least one member
conducting occasional jobs has increased by about 3.4% and the fraction of households with household head conducting occasional jobs has increased by about 2.6%. This is consistent to the results
from previous section that the policy impact is at extensive margin making more households into
occasional jobs.
A central questions is, while the goal of this policy is to encourage entrepreneurship among return migration, hoping that they could initiate their own business with the help from the local
government on permits granting, on-site factory and office choices, and easier credit loan services,
and also with the social capital, work experience, and accumulated assets brought back from more
developed urban cities, why have more households gone into occasional work, where the pay is
unstable, and social benefits are nearly non-exist? To figure out this question, we raised a couple
of potential hypotheses and test them in the following sections. An upfront potential explanation
is that although stated in the policy documents, credit loans are not easy as they seem for small
county residents. According to the CHFS data, only around 5% of households actually rely on loans
and the rest of the households started their family business by all cash or by private unprofessional
loans under the table. Various reasons have been recorded for the lack of successful loans, including
that they have applied but didn’t get approved, they have not had the confidence to apply, and
they have not needed loans, meaning that the scale of the business they have been engaging with
falls under a more efficient size.
63

A second hypothesis states that with the return migration flows to these under-developed counties,
the local labor market has been stuck with an over-sufficient supply of labor. Find a stable job
has become more difficult than before and a decline in the wage rate for employed workers should
be observed. Switching to occasional work is a seemingly easier choice with fast earnings. We will
test this hypothesis in Section 2. A third potential is that due to the dual rural-urban system
embedded in China’s medical system, return migration enjoy cheaper enrollment, easier hospital
visit and inpatient stay when they are at their smaller home counties. This could be attractive to
the return migration. We discuss this issue in Section 2.
Table 2.2: County-level Household Employment Type Change in Urban Regions
(1) (2) (3) (4)
A: Household Employed Self-employed Farming Occasional
policy -0.00329 -0.00328 -0.0222 0.0336**
(0.0149) (0.0104) (0.0138) (0.0149)
Observations 1,183 1,183 1,183 1,183
R-squared 0.841 0.736 0.883 0.829
B: HH Head Employed Self-employed Farming Occasional
policy 0.00192 0.00899 -0.0158 0.0259**
(0.0121) (0.00895) (0.0116) (0.0108)
Observations 1,183 1,183 1,183 1,183
R-squared 0.856 0.685 0.866 0.748
County FE ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓
Notes: Controls include gender and number of household members. Robust
standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
64

Impact on Earnings
As previously discussed in Section 2, it’s important to study the wage rate changes after the policy
implementation for each employment category. We started first from small family business and
other employment. The essence is to observe the entry and exit from existing entrepreneurship.
Equation similar to 2.1 is estimated with different outcomes, including indicators for small family
business or other type of employment, and logarithm of the wage rate or family business profits.
Results reported in Table 2.3 suggest that although there is no significant changes to the probability
of initiating small family business, there is an 18.1% decline in wage rate for non-family-business
employment due to increased return migration flows, on the other side, the small business profits
have increased by 43.9% thanks to better business and loan environment for existing businesses
with very few new entrants.
Table 2.3: Policy Impact on Work Status and Earnings
(1) (2) (3) (4)
I(work) ln(wage) I(small business) ln(profit)
policy 0.023 -0.181** 0.030 0.439**
(0.024) (0.079) (0.021) (0.216)
Observations 32,421 9,400 14,451 1,492
R-squared 0.990 0.984 0.989 0.990
Household FE ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓
Province by Year FE ✓ ✓ ✓ ✓
Notes: Sample is kept to balanced sample. Controls include age fixed effects, gender,
education levels, retirement status, number of household members, hukou types, marital status, and health condition. Regressions are clustered at county level. Robust
standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
To dig deeper into how does the wage rate for each type of work changes after the return migration
65

policy implementation. We restrict the sample into employed people at other firms and people who
do occasional work separately and re-conduct the empirical analysis. Results reported in Table 2.4
suggest that the decline in wage rate for employed workers ranges from 24.6%-34.6% depending on
if we restrict the sample to urban areas, no significant impact on wage rate for occasional workers
is detected and the coefficient magnitude is small.
Table 2.4: Policy Impact on Earnings by Work Type
Employed Occasional
(1) (2) (3) (4)
ln(wage) ln(wage) ln(wage) ln(wage)
policy -0.246** -0.346* -0.086 0.066
(0.109) (0.196) (0.206) (0.115)
Observations 5,739 6,938 2,005 4,007
R-squared 0.988 0.986 0.751 0.727
Region Urban Urban
Household FE ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓
Province by Year FE ✓ ✓ ✓ ✓
Notes: Sample is kept to balanced sample. Controls include age fixed effects,
gender, education levels, retirement status, number of household members,
hukou types, marital status, and health condition. Regressions are clustered
at county level. Robust standard errors in parentheses. *** p<0.01, **
p<0.05, * p<0.1
Discussion: Medical Insurance
Medical insurance could be an additional attraction to residents in smaller counties due to the
type of insurance they can enroll in and the cheap premium they have to pay compared to the
insurance they are eligible to enroll in big cities which is high-cost and low-coverage. We examine
66

if the policy has an impact on the enrollment and reimbursement for four major types of medical
insurance, including urban employment medical insurance, urban residential medical insurance,
new cooperative medical scheme and new unified residential medical insurance. We do not find
significant policy impact is observed following the return migration policy.
After exploring three potential reasons for the increasing number of residents working as occasional
workers, our results provide indicative evidence supporting the argument that the seemingly encouraged entrepreneur for return migration in fact set higher barriers for starting small family
businesses in county-level cities, meanwhile the return migration flows have struck the local labor
market by imposing a downward pressure on wage rate for employers, pushing workers to occasional
type of work which is the informal sector. Without further data we are not able to infer about the
consumer welfare change, which we are hoping to incorporate into an empirical spatial equilibrium
model and answer in the future.
Chapter Two: Model
This section presents an empirical spatial model of local labor markets where wages for high and
low educated workers, housing rents, and population are determined in equilibrium (Diamond,
2016b). We are interested in using the model to answer the question, how to quantify (e.g. equally
percentage of increase in wage rate) the attractiveness of this encouraged entrepreneurship induced
return migration to match the actual fraction of return migration for each county. And, a more
compelling and intriguing question, if after these migration returned to their birth counties and
found about the contradictory high barriers to start their own business and eventually they are
trapped in occasional work, how much aggregate welfare should be decreased?
We introduce the migration choice under the Chinese hukou restriction and the housing markets
following the Chinese housing assignment program as well as endogenous local amenities into our
model.
Each prefecture j in China has a set of homogeneous firms which employ college (“high”) and noncollege (“low”) educated workers and wages are recovered from separate market clearing conditions
67

for high and low education workers and a random component ϵj of exogenous productivity.
w
H
jt = γHHlnHjt + γHLlnLjt + ϵ
H
jt
w
L
jt = γLHlnHjt + γLLlnLjt + ϵ
L
jt
(2.3)
A worker i of education z ∈ (H, L) lives in a prefecture j, inelastically supplies his unit of labor
and earns a wage ω
(H,L)
j
. The worker consumes tradable national goods O (consumption) and nontradable local goods M (housing) with prices P and R respectively and also values local amenities
Aj according to his valuation function si(). The individual’s problem of utility maximization subject
to a budget constraint:
max
M,O
ln(Mζ
) + ln(O
1−ζ
) + si(Aj )
s.t. P O + RjM ≤ Wedu
jt
⇒ max(1 − ζ)lnO + ζlnM + si(Aj )
= ln
Wedu
j
P
− ζlnRj
P
+ si(Aj )
(2.4)
We can rewrite the individual’s utility as a linear equation:
Vij = δ
zi
j + x
pro
j
βprozi + x
regiregiiβ
regizi (2.5)
where:
δ
z
j = β
wln(ω
edu
j
) + β
r
ln(rentj ) + β
aAj (2.6)
Utility depends on the wage, rentals, local amenities in prefecture j, and depends on which birth
68

place one was born in.9 The individual’s location choice follows a conditional logit model.10 According to (McFadden et al., 1973; Diamond, 2016b) a worker chooses the prefecture with the
highest utility among all possible destinations:
Hj =
X
i∈H
exp(δ
zi
j + x
pro
j
βprozi + x
regiregiiβ
regizi)
P
k
exp(δ
zi
k + x
pro
k
βprozi + x
regi
k
regiiβ
regizi)
Lj =
X
i∈L
exp(δ
zi
j + x
pro
j
βprozi + x
regiregiiβ
regizi)
P
k
exp(δ
zi
k + x
pro
k
βprozi + x
regi
k
regiiβ
regizi)
(2.7)
For the housing market, construction cost and land cost pin down the housing prices and rental
prices. Housing demand only applies to both local hukou holders and non-hukou people.
HDij =
ζWH
j
Rj
+
ζWL
j
Rj
rj =γ
ccln(CCj ) + γ
HDln(HDj ) + γ
landln(Landj )
(2.8)
Amenities are treated as either endogenous if we assume that the higher ratio of high educated
workers in a prefecture-level city, the higher the tax revenues which the local governments or urban
planning departments can distribute to build local amenities. We also consider the case where
amenities are treated as exogenous.
aj = γ
a
ln(H/Lj ) + γ
migln(M igrationj ) + ϵ
a
j
(2.9)
We instrument for endogenous amenities using two Bartik-style instruments, the first is nation level
productivity shock for each industry multiplied by the industry composition for each prefecture,
9The regions are based on geographical and economic clusters, and are commonly used in analysis, they are—
Northeastern: Heilongjiang, Jilin, Liaoning, Inner Mongolia; Central Area: Shanxi, Henan, Hubei, Hunan, Jiangxi,
Anhui; Eastern: Beijing, Tianjin, Hebei, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, Hainan; Western: Chongqing, Sichuan, Guangxi, Guizhou, Yunnan, Shaanxi, Gansu, Ningxia, Xinjiang, Qinghai, and Tibet.
10We use Knitro solver in Matlab and utilize first order conditions of the utility function to do the convex optimization exercise.
69

the second is export tariff decrease as export shock (demand shock) following the event that China
joined the WTO in 2001. We argue that these instruments are directly impacting the wage rate
other than impacting the labor demand. The model incorporates both labor and supply sides of the
markets, takes into account the impact of the housing assignment program on the housing demand,
and the impact of migration on local amenities, and we extend the Diamond’s model to fit the
institutional background and historical facts in the Chinese context.
We retrieve indirect utility with conditional logit estimates and then use the recovered utilities
in the empirical spatial model with a General Methods of Moments (GMM) to estimate model
parameters for insights.
Data and Variables
Our main analysis relies on 2005 census. It includes around 2 million observations, representing
5% of the population of China in 2005.11 In this paper, we restrict our sample to the population
of people between the ages of 18 and 55 who report a non-zero wage.12 This leaves us with around
700,000 observations. Questionnaire includes questions about education level, hukou registration,
wage, home construction cost, home characteristics (e.g. whether or not your home has running
water, material used for fuel, construction materials), and etc. In this paper, we only account for
migration between prefectures. hukou locations are designated as “urban” or “rural” depending on
the administrative unit of the location. Some people have a “rural” hukou but live in a relatively
urbanized environment e.g. in the main town in their prefecture. Table 2.5 shows summary statistics
for people in urban or rural areas by hukou status. Hukou and non-hukou residents are similar in
terms of number of years of education but differ greatly in wage on average. This suggests that
non-hukou workers are “compensated” for moving to a location where they do not have access to
local amenities, etc. Non-hukou residents are much less likely to own a house. In urban areas,
hukou holders are more likely to have a State job whereas in rural areas, non-hukou holders are
11The actually number of observations makes around 0.95% of the populace, which is a re-shuffling out of 5%
observations.
12We restricted our sample to people who work at least 40 hrs but less or equal to 56 hrs, which is 8 hrs per day
for 7 days a week. We drop observations who are still at college, and who lives in the same house which provide retail
services to avoiding driving the average rent low in a prefecture.
70

more likely to have a State job. This suggests a mechanism by which hukou and non-hukou holders
receive different benefits in different locations.
Table 2.5: Summary Statistics
Panel: urban citizens
College Educational years Wage (2005 RMB) House owner State job
Non-hukou 0.47 13.00 1568.62 0.61 0.58
Local hukou 0.36 12.27 1153.61 0.83 0.73
Total 0.39 12.44 1250.03 0.78 0.69
Obs. 214858
Panel: rural citizens
College Educational years Wage (2005 RMB) House owner State job
Non-hukou 0.03 8.88 894.01 0.30 0.14
Local hukou 0.00 7.42 382.04 0.98 0.04
Total 0.01 7.63 453.94 0.89 0.06
Obs. 493008
Notes: Source Chinese Census in 2005. We keep the sample to people age 15-64 in the labor force.
Preliminary Results and Proposal
Estimation of parameters is reported in Table 2.6, Model 1 takes prefecture-level amenity score as
exogenous while Model 2 takes that as endogenous and assumes that amenity changes as the ratio of
high-skilled versus low-skilled workers changes. Due to our Cobb-Douglas assumption of the utility
function, the ratio of the parameters of rent versus wage reflects the share of income that workers
allocate to housing. According to results in Model 1, this share for college workers is around 48.9%
while that for non-college workers is around 90.2%. The parameters from “Non-hukou” panel is even
larger, compared to hukou holders, non-hukou college workers spend 8.0% more (56.9%-48.9%) and
no-hukou non-college workers spend 14.4% more (1.046-90.2%) on housing. These number reflect
71

the fact that low-skilled workers bear a significant higher cost to live in big cities considering their
levels of labor productivity and the high price of accommodation in urban cities.
Labor demand results suggest that the wage rate for high-skilled and low-skilled workers depend on
a balanced mixture of college and non-college workers. Generally, an increasing demand for highskilled workers increases wage rate for both college and non-college workers due to both demand
increase and complementarity of high-skilled and low-skilled workers.
The level of rent is positively associated with housing demand and construction cost and negatively
associated with land availability. Amenity score (in Model 2) is positively associated with the ratio
of high- versus low-skilled workers, indicating that the relative fraction of more educated workers is
also endogenously contributing to local amenities. However, it’s difficult to interpret the rent share
of income in Model 2 if we assume endogenous amenity. For that reason, we are slightly inclined
to Model 1 for now.
In order to use this model to quantify the welfare change for return migration, we first assume
that the entrepreneurship policy provided x% of income increase should migrant workers return to
their hometown. The subsequent return migration flows should match what we observe from the
data for each county. This should let us quantify the incentive magnitude of this entrepreneurship
policy. Next, while it is in fact difficult for return migrant workers to start their own business,
they remained employed as workers. The new labor market clears and equilibrium wage rates are
set. Following the old utility function parameters, we should get new utilities of workers in each
prefecture for each type of workers. The difference between this new utility and our old utility
should give us the aggregate welfare decrease.
72

Table 2.6: Estimation of GMM Parameters
Model 1 Model 2
College Non-college College Non-college
Wage 2.064 ( 3.27) -45.088 ( 188.79) 4.443 ( 8.11) 68.932 ( 689.54)
Rent -1.010 ( 2.48) 40.667 ( 175.53) -5.308 ( 3.22) -65.146 ( 696.29)
Share 0.489 0.902 1.195 0.945
Non-hukou
Wage 2.401 ( 4.99) 60.123 ( 257.97) 3.380 ( 6.78) 27.470 ( 365.65)
Rent -1.367 ( 4.14) -62.902 ( 239.92) -4.638 ( 6.83) -17.508 ( 379.21)
Share 0.569 1.046 1.372 0.637
Rural
Wage 0.718 ( 4.50) 18.555 ( 66.81) 2.114 ( 8.46) 3.710 ( 98.86)
Rent -0.162 ( 3.54) -17.734 ( 62.36) -0.265 ( 4.51) -3.181 ( 102.84)
Share 0.225 0.956 0.125 0.857
Model 1 Model 2
College Non-college College Non-college
Labor Demand
College 0.171 ( 0.24) 0.738 ( 0.23) 0.191 ( 0.12) 0.738 ( 0.23)
Non-college -0.302 ( 0.66) -0.035 ( 0.57) -0.343 ( 0.11) -0.336 ( 0.30)
Housing
House Demand 0.201 ( 0.27) 0.061 ( 0.18)
Land Availability -0.079 ( 0.32) -0.173 ( 0.27)
Construction Cost 0.764 ( 0.42) 0.909 ( 0.28)
Amenity
High/Low 2.374 ( 0.41)
Notes: This table reports estimated parameters from the empirical spatial equilibrium model in the paper. Model 2
assumes endogenous amenity and Model 1 takes amenity as given.
73

Chapter Two: Conclusion
In this paper, we evaluate the program impact of a series of return migration policies launched in
2016-2017 to encourage entrepreneurship. Under the background of the slowing growth of ruralto-urban migration and urban labor demand, migrant workers are encouraged to return home
and start their own small business with the help of local government officials on factory, permit
granting, and credit loans. We leverage the staggered implementation of this program and use a
staggered difference-in-difference to analyze its impact. Our results suggest that this program has
incentivized about 7.9% more residents to move back to the treated counties following the program
launching. However, we do not observe an increased number of small family business. The reasons
behind that could be low credit loans granting rate and the lack of confidence in entrepreneurs by
government officials. Subsequently, the increased labor supply due to the return migration results
in lower wage rate in the employed sector and more workers starting to work as occasional workers.
This crowds out return migrant workers to the informal sector which potentially reduced their
welfare. We present an empirical spatial model to quantify the overall welfare loss due to this
spatial misallocation of workers following the policy implementation and aim to finish this in the
future.
74

Chapter 3
Intervivos Transfers Change due to One Child Policy Intrigued Social Security Tax Increase
Chapter Three: Introduction
As a result of the world-wide rapid ageing, the pension systems of many countries are facing with
unprecedented challenges. With an increasing dependency ratio, less young agents are able to
support the living of old persons. The situation is more severe in China due to the One Child
Policy (OCP) which was enacted to address the population growth in 1980, allowing couples to
only have one child (with some exceptions). According to Green book of China’s social security
system (2019), by 2015, there were already 6 provincial governments paying more pension payment
than payroll tax revenues, and it predicted that 14 out of 31 provinces would be struggling with
pension deficit by 2022. This calls the government to raise the payroll tax rate.
Demographics play an important role in intra-household risk smoothing (e.g., Hayashi et al., 1996;
Dercon and Krishnan, 2000; De Weerdt and Dercon, 2006; Fafchamps and Lund, 2003). Intergenerational transfer is one of the most important sources of consumption smoothing, especially in Asian
countries. A natural question to ask is, what is the impact of demographic change to intervivos
transfers? And then, are we going to hurt the welfare of young workers if the social security tax
rate are forced to increase in order to maintain a fixed replacement rate for the elderly.
In this article, we examine the role of demographics on intergenerational transfers with an overlapping generation model with altruism. As many studies (Cox and Rank, 1992; Altonji et al.,
75

1997) have shown, altruism is an important mechanism of intra-household intervivos transfers.
Motivated by literature (Nishiyama, 2002;
˙Imrohoro˘glu and Zhao, 2018a,b), Our analysis relies on
an overlapping generation research design to take account of demographic transitions caused by
OCP. As Abhijit Banerjee and Qian (2014) suggests, general equilibrium framework outperforms
partial equilibrium framework which overestimate the impact of fertility on household decisions
by introducing a feedback channel of equilibrium prices. We find a substantial impact of OCP on
the shrinkage of labor force by more than 60%, inducing a significant social security tax increase
from 2% to almost 5% to maintain the same replacement rate and living standards for the senior
generation. Real wage rate increases by 50% because of a less tight labor market conditional on
3% TFP growth. This suggests that although the burden to take care of old parents significantly
increases, forcing the government to set more aggressive strategies and increases the social security
tax rate, low fertility rate in an economy with high productivity growth could still benefit the young
generation by generously increasing their real income and real output per capita. We next show
that in households with low human capital accumulation the unequal wealth status between the
young and the old generations accelerates intervivos transfers.
This article makes three main contributions to the literature. First, it adds to the literature of
demographic transitions and aging. Samuel H. Preston (1989) and Foster and Walker (2014) finds
that the mortality rate decline is responsible for population aging in the US, our paper departs
from these sources and mainly studies a policy driven reason for China’s population aging — the
One Child Policy that was implemented in 1980. Literature (e.g., Zhang and Goza, 2006; Smith
et al., 2014; Zhong, 2011; Zeng and Wang, 2014) finds that OCP caused a rapid population aging
and is responsible for income inequality and middle-age burden, particularly in rural China, where
many people migrate to urban cities for waged jobs. While aging is a prominent problem in
many countries, there are different approaches to tackle with the issue. Foster and Walker (2014)
examines the “active aging” idea that has emerged in Europe as the foremost policy response to
the challenges of population aging by prioritizing the extension of working life. As for China, Zeng
and Wang (2014) calls for a nation-wide policy expansion of “more children policy” and expansion
of rural pension coverage. This paper departs from the literature and addresses a couple of benefits
from the OCP, which is relieved raising burden for the old generation when they are still young,
76

and relieved burden taking care of the elderly for the young generation when their parents get old.
Second, it bears a direct connection to the literature examining social security system and pension
deficit. The Chinese government has been fighting with low pension coverage and insufficient
individual contributions for at least two decades. Fang (2018) finds that without government
subsidies, 14 out of 31 provinces would have been in deficit in 2010. Literature (e.g., Feldstein,
1999; Blundell et al., 2002; Zeng, 2011) suggests several policy solutions to this problem, including
increasing retirement age from current level (white-collar men 60 and women 55) to 65 years old,
expand the multiple child policy as fast as possible, some benefits cuts and and coverage expansion
to all firms. This paper provides a new angle and suggest that sustainable economic growth with
lower fertility rate could be possible.
Finally, it adds to the literature that looks at intra-household financial transfers and resources
allocation. Past literature (e.g., Hayashi et al., 1996; Dercon and Krishnan, 2000; De Weerdt and
Dercon, 2006; Fafchamps and Lund, 2003) use microdata to test full risk-sharing and complete insurance mechanism within households and within villages. This paper departs from these literature
by relying on a general equilibrium model to test the underlying mechanism. Similar to Ortigueira
and Siassi (2013), households with agents of unequal status bear heavier risk-sharing activities.
In their paper, average hours worked by wives of unemployed husbands are 8% higher than those
worked by wives of employed husbands for poor households. Consistent with our findings, children
give more intervivos transfers to parents in households with low human capital accumulation.
The rest of the paper is organized as follows. Section 3 presents the model economy, including
household agents, firms, and a government who collects income tax for the pension payment.
Section 3 also calibrates parameters and benchmarks the model to represent the demographic
transitions from 1980 to 2015 in China. Section 3 presents the steady states results before and
after the demographic transition. Results suggest that consistent with real data, the fraction of old
people above 65 years old doubles. Social security tax in the model is forced to increased to 5%
from around 2% because of the increasing dependency ratio. Model suggests that the direction of
intervivos transfers depend on the relative status of the young and the old agents within households,
i.e. within a household with poor parent and rich kid, the kid transfer money to his/her parent at
77

an earlier age and also at a higher amount. With a 3% annual productivity growth, the average
kid earn 50% more after the transition due to a shrinkage labor force, indicating that the kid is
still better off even if he needs to take care of more old parents due to OCP. Section 3 concludes.
Chapter Three: Model
Altruistic households with overlapping generations lived in the model economy following (˙Imrohoro˘glu
and Zhao, 2018a), firms are owned by households, the government provides a pay-as-you-go (PAYG)
social security system and pays a fixed replacement rate for the old generation as a fraction of their
income.
Households
The economy is composed of overlapping generations of agents, whose lifespan is 2T periods. When
an agent is born in the economy when his parents are T years old, he starts to work at once, and
co-live with his parent for T periods. During these T periods, his parent retires at age R. After the
retirement, his parent faces a risk of death, and dies at the age of 2T when the young agent is T
years old. The young agent then will then give birth to his own children and starts to co-live with
his children for T years until he dies. The generation overlapping continues in this economy. T is
set to be 35 in this economy. Each young agent enters the economy at the age of 20 and is ready
to work, by then, his/her parent is 55 and will face mandatory retirement at age 60. The parents
need long-term care after retirement.
After retirement, there are naturally two types of households in the economy, one has both the
old agent and the young, the other has only the young agent and the old one dies. The size
of the population evolves over time exogenously at the rate gt
, and the population growth rate
is g = n
1/T , where n is the fertility rate.1 To model the intervivios transfers (Laitner, 1992),
households maximize the summation of their utilities and are subject to a total budget constraint
1Some literature studies endogenous fertility, where households decide their optimal number of children (e.g.,
Morand, 1999; Schoonbroodt and Tertilt, 2014). However, we set the fertility rate to be exogenous due to OCP.
78

by pooling their resources together. Young agents are endowed with a unit of labor and they
supply their labor inelastically. Each agent faces a shock z at birth that determines his life-time
productivity to be high or low, and we denote that of the children z
′
. Children’s productivity
relates to that of their parent z by a Markov chain, the transition matrix is denoted as Π(z, z′
).
Beside the agent’s life-time ability, there are two other components for variation across agents: a
deterministic age efficiency ϵj , and a stochastic income shock µj . We model this to a human capital
accumulation or experience accumulation. Agents also face a stochastic income shock, it could be
a positive productivity shock or a negative “rainfall” shock. The stochastic income shock follow an
AR(1) process:
log(µj ) = Θlog(µj−1) + νj (3.1)
νj ∼ N (0, σ2
µ
), Θ < 1
The AR(1) process is discretized into a 3-state Markov chain using Tauchen (1986) method, and
the transition matrix is denoted as Ω(µ, µ′
). The original µ at birth is from a random draw from
distribution Ω( ¯ µ). All children are born at the same time.
Parents face a health risk h, determining whether he needs long-term care (LTC) or not. The LTC
requires children to spend time ξ and money m on their parent. The health transition matrix for
agents is given by Γ(h, h′
). After retirement, the parent is facing with a risk of death d = 1, the
transition matrix is given by Λj+T (d, d′
), and Λj+T (1, 1) is the survival probability of the old agent.
If an old agent is retired and is alive, he receives a security payment SS.
Therefore the earning of the household is given by:
ej =



[ωϵjµjzs(n − ξh) + ωϵj+T zf (1 − h)](1 − τss − τe) if j + T ≤ R
ωϵjµjzs(n − ξh)(1 − τss − τe) + dSS if j + t > R,
(3.2)
where τss is the payroll tax and τe is the income tax. The household’s budget constraint is:
79

aj+1 + (n + d)cj + mh = ej + aj [1 + rt(1 − τk)] + κ (3.3)
where τk is the capital gain tax, r is the interest rate, cj is the per capital consumption at t = j,
aj+1 is the asset saving for next period, mh is goods consumption if the parent needs LTC, and κ is
government transfers consisting of a proportional earnings on top of the consumption floor. 2 The
household needs to decide on its optimal consumption and savings given equilibrium prices and
taxes. States variables include age j, current savings aj , ability draws zf and zs, labor productivity
shock µ, health status h, and mortality indicator d, let x = (a, zf , zs, µ, h, d) be the state vector.
Given prices and discount factor β, the household is to maximize his life-time utility:
Vj (x) = max
c,a′
[(n + d)u(c) + βE[V˜
j+1(x
′
)]] (3.4)
subject to the budget constraint, and aj ≥ 0, c ≥ 0, and
V˜
j+1(x
′
) =



Vj+1(a
′
, z′
f
, z′
s
, µ′
, h′
, d′
)if j = 1, 2, ..., T − 1
nV1(
a
′
n
, z′
f
, z′
s
, µ′
, h′
, d′
)if j = T
(3.5)
after the old agent’s death, he leaves a bequest and assets to all his children equally. Each child
will then have his own household and live on.
Firms
The representative firm, owned by households, produces a single good and has Cobb-Douglas
production function Yt = AtKα
t N
1−α
t with capital share of output α, and total factor of productivity
2κ = κ1ej + κ2, the first element is a proportional earnings of the household used to balance the government
budget constraint, the second is a consumption floor for the poorest households, following Imrohoroglu and Zhao
(2018), De Nardi, French and Jones (2010) and Hubbard, Skinner, and Zeldes (1995),
κ2 = max n
0,(n + d)c
¯
+ mh − [ej + aj [1 + rt(1 − τk)] + κ1ej ]
o
the assumption is that when the household is on the consumption floor (κ2 > 0), aj+1 = 0 and c = c
¯
, i.e., the
consumption is on the minimum level to maintain their survival, and they fail to save anything for the next period.
80

(TFP) At
, the growth rate of TFP γt =
At+1
At
1/(1−α)
. Capital depreciates at rate δ, the firm
maximizes its production which determines the equilibrium prices:
rt = αAt
Kt
Nt
α−1
(3.6)
ωt = (1 − α)At
Kt
Nt
α
(3.7)
Government
The government taxes capital at τk and labor income at τe, and uses the tax revenues for government
expenditures, i.e. infrastructures, transportation, etc. Government taxes the working labor force
at τss and spend the tax revenues on social security payment only.
Equilibrium
The steady state stationary equilibrium is defined as: given fiscal policy G, τss, τe, τk, SS, and
exogenous fertility rate n, a stationary equilibrium is a set of value functions, household decision
rules on consumption and next period’s savings, time-invariant measures of households {Xj (x)
T
j=1},
and relative prices {ω, r}, such that:
1. Given the prices and fiscal policies, the households solve their optimization problem as in
equations (3.2) - (3.5).
2. Firm solves its profit maximization problem and determine the factor prices.
3. Consistency:
K =
P
j,x aj (x)Xj (x)
N =
P
j,x(ϵjzs(n − ξh) + ϵj+T zf (1 − h))Xj (x)
(3.8)
81

4. Measures of households:
Xj+1(a
′
, zf , zs, µ′
, h′
, d′
) = 1
n1/T P
a,µ,h,d:a
′ Ω(µ, µ′
)Γ(h, h′
)Λ(d, d′
)Xj (a, zf , zs, µ, h, d),
for j < T
(3.9)
X1(a
′
, zs, z′
s
, µ′
, 1, 1) = n
X
a,µ,h,d,zf :a
′
Ω( ¯ µ
′
)Π(zs, z′
s
)XT (a, zf , zs, µ, h, d) (3.10)
5. The government’s budget:
X
j,x
κ1ejXj (x) = τkrK + τeωN − G (3.11)
6. PAYG system is self-financing:
X
T
j=R−T +1
X
x
d(SSj + κ2)Xj (x) = τss[
R
X−T
j=1
X
x
ejXj (x) + X
T
R−T +1
X
x
ωϵjµjzs(n − ξh)Xj (x)]
(3.12)
Calibration
Calibration follows ˙Imrohoro˘glu and Zhao (2018a), the economy started from a steady state around
1980, the year that the OCP has been implemented. As fertility rate changes, the economy converges
to a steady state with the new fertility. The parameters calibrated are below:
82

Table 3.1: Calibrated Parameters
Parameter Definition Value
α capital income share 0.5
δ capital depreciation rate 0.1
σ risk aversion parameter 3
A Total Factor Productivity 0.32
β discount factor 0.99
m goods cost of LTC 33% of GDP per capita
ξ time cost of LTC 0.42
z ∈ {H, L} permanent labor ability {1.79, 1.0}
G government purchases 14% of GDP
SS social security payment 15%
γ
1−α
initial − 1 initial steady state TFP growth rate 3.1%
ninitial initial steady state fertility rate 2.0
nf inal final steady state fertility rate 0.8
Notes: We follow Imrohoroglu and Zhao (2018) for calibration.
Chapter Three: Results
This section presents results for the impact of social security tax increase induced by fertility drop
from 2 children per person to 0.8 child per person. Young agents are taxed at a higher rate to keep
the same replacement rate after the fertility rate decline. We compare the two steady states and
the model demographics change in China from 1980. We next investigate the change of the social
security tax. The final subsection presents the relative change in intervivos transfers between two
steady states.
83

Steady States of the Initial and Final Economies
The results for steady states in the initial and final economies are 3
:
Table 3.2: Steady States
Initial Final
r 1.15 1.06
ω 0.10 0.15
LTC 0.12 0.14
Aggregate capital 2.12 1.79
Aggregate labor 4.91 1.74
Output per capita 0.38 0.45
Payroll tax 2.11% 5.04%
Replacement rate 15% 15%
Notes: Initial steady state economy has
a fertility rate of 2 children per person.
Final steady state economy has a fertility
rate of 0.8 children per person.
The interest rate goes down from 1.15 to 1.06 and the wage rate goes up from 0.10 to 0.15. This
suggests that capital is valued less and labor is valued more in the final steady state, consistent
with the decline of the ratio of aggregate capital to aggregate labor ( 2.12
4.91 → 1.79
1.74 ). Due to one child
policy, there is smaller aggregate labor supply, so output per capital increases from 0.38 to 0.45.
Social security tax is 2.11% in the initial steady state. After the fertility decline, dependency ratio
goes up so it is forced to be raised to 5.04% to hold the replacement rate fixed.
3The only difference between the initial and final economies is that the fertility rate goes from 2 to 0.8, there is
no change in TFP factor in this experiment.
84

Demographics
Table 3.3: Demographic changes: Data
Data Population Growth Age 0-14 Age 15-64 Age 65+
1981 1.014 33.60% 61.50% 4.90%
2011 1.005 16.50% 74.4% 9.10%
Notes: Data Source: China Statistical Yearbook.
Table 3.4: Demographic changes: Model
Model Age 20-29 Age 30-64 Age 65+
Initial n=2 26.67% 60.66% 12.67%
Final n = 0.8 14.90% 59.98% 25.11%
Notes: Results are separately from the model in
the initial and final economies.
The demographic changes from China Statistical Yearbook4
suggests that the fraction of people
aged 65 and above goes up from 4.90% to 9.10%, almost doubles. This is consistent to the model
results that fraction of people 65+ goes up from 12.67% to 25.11%, although the magnitude is higher
due to the assumption that all agents are born at the age of 20 and start to work immediately in
the model. For the very young (≤ 20 years old), it can be seen that from the data the fraction of
people aged 0-14 years old goes down from 33.6% to 16.5%, almost shrinks to its half. Consistent
results are presented with the model that the fraction of people aged 20-29 declines from 26.67%
to 14.90%, suggesting a rapid ageing process.
4National Bureau of Statistics of China (1981,2011). China Statistical Yearbook.
http://www.stats.gov.cn/tjsj/ndsj/2011/indexch.htm
85

Payroll Tax vs. Pension Payment to GDP
Table 3.5: Pension to GDP fraction in the Data
Data Pension/GDP
1981 1.30%
2018 4.86%
Data Source: China
Statistical Yearbook.
Table 3.6: Social Security Tax in the Model
Model Tax
Initial n=2.0 2.11%
Final n=0.8 5.04%
Notes: Results are
separately from the
model in the initial
and final economies.
The fraction of pension over GDP from the data shows that for the Chinese government the
“burden” of pension payment is increasing from 1.30% in 1981 to 4.86% in 2018. Model results
suggest that in the final steady state the social security tax rate in a pay-as-you-go (PAYG) pension
system goes up from 2.11% in the initial economy to 5.04% due to the fertility rate decrease. The
ratio of pension over GDP is a proxy of payroll tax. Although current Chinese pension system
is not PAYG but rather a mixture of a defined benefit PAYG portion and an investment-based
defined contribution portion,5
the model still addresses the potentiality of pension deficit without
5China’s New Rural Pension Scheme was launched in 2009 by the Chinese government. Zhang, Luo, and Robinson
in 2019 shows that this pension income crowds out approximately 27.9% of the monetary support from adult sons
and decreases the likelihood that adult sons live with their parents by 6.5%.
86

raising the social security tax rate.
Intervivos Transfers
Intervivios transfers from the model are plotted in such way: the horizontal axis is the age of the
household, which is defined as the age of the youngest agent in the household. For example, the
point 30 on the vertical axis means that the youngest agent in the household is 30 years old and
his/her parent is 65 years old. The horizontal axis is the amount of transfer — positive means
transfer from the parent and negative means transfer from the children. The blue curve is the
transfer in the initial economy with fertility rate 2.0, and the red line is the transfer in the final
economy with fertility rate 0.8.
Figure 3.1: Transfer by Types of Households: no TFP, fertility change, all transfer
87

Figure 3.2: Transfer by Types of Households: no TFP, fertility change, per child transfer
Figure 3.1 and 3.2 shows that, without tfp change, as the social security tax rate increases due
to decreased fertility rate, the transfer to children goes down. Little change has been observed of
the total transfer from children due to a fixed replacement rate. Figure 3.2 departs from 3.1 as it
illustrates per-capita transfer from and to parents, while Figure 3.1 shows the total transfer from
or to parents.
Figure 3.2 shows that although transfer to per child goes down, transfer from per child goes up a
lot. Key reason for that is the high dependency ratio in the final steady state, to keep replacement
rate fixed, children need to transfer more to his/her parent than the case with more siblings. In
the final steady state, transfers from the parents to children go down, suggesting that in the final
steady state the relative financial position of the child is “better” due to the shrinkage of aggregate
labor force and the increse in real wage rate.
Figure 3.3 and 3.4 illustrate the results from the initial and final economies with TFP being 0.32
and the growth rate being 3.1%. In Figure 3.3, results show that the transfer to children goes down
in the final steady state, indicating that economy growth helps young agents to earn more income
88

and rely less on their parents. Figure 3.4 illustrates that young agents needs less transfer from their
parents while they are young and give more transfer to their parents when they are old.
Figure 3.3: Transfer by Types of Households: with TFP, fertility change, all transfer
89

Figure 3.4: Transfer by Types of Households: with TFP, fertility change, per child transfer
Overall the results say, as fertility rate goes down, government needs to tax the young agents at a
higher social security tax rate to maintain a fixed replacement rate. Real wage rate goes up due to
shrinkage of aggregate labor force. The increased income is sufficient for young agents to overcome
the higher tax rate. The relative financial status of the young agent is higher than the old agent
within household. As a result, although social security tax increase and dependency ratio increases,
young agents are not hurt due to a shrinkage labor force and increasing real income.
Chapter Three: Conclusions
This paper studies the impact of fertility decline and social security tax increase on intervivos
transfers within households. In section 3 we presents an OLG with altruism model to simulate
the increase in social security tax from 2.11% to 5.04% following a decline in fertility rate from
2 children per adult (in 1980) to 0.8 child per adult (in 2010). Data from the China Statistical
Yearbook has shown that ratio of pension payment to GDP has increased from 1.30% in 1981 to
90

4.86% in 2018, suggesting that for the government the pension payment burden keeps increasing.
Consistent results from the model show that transfers to children decrease and transfers to parents
increase, suggesting a relative higher financial status of the young agent within households. This
suggests that although the burden to take care of old parents increases, forcing the government to
set more aggressive strategies and increases social security tax rate, low fertility rate in an economy
with high productivity growth actually benefits the young generation by generously increasing their
real income and real output per capita.
We leverage an OLG with altruism model to separate the effect of fertility rate and the social
security tax increase from many other impact factors. If OCP was not the reason for “death
of despair” in China, then how it interacts with many other factor and leads to so many social
problems, such as insufficient long-term care, left-behind children in rural China, and rocketing
unemployment rate for the 20-year-old.6 These are important research questions for further study.
6Claire Fu and Daisuke Wakabayashi. 2023. Don’t Be So Picky About a Job, China’s College Graduates Are
Told. The New York Times.
91

References
Abebe, G., Caria, A. S., Fafchamps, M., Falco, P., Franklin, S., and Quinn, S. (2021). Anonymity
or distance? job search and labor market exclusion in a growing african city. The Review of
Economic Studies, 88(3):1279–1310.
Abhijit Banerjee, Xin Meng, T. P. and Qian, N. (2014). Aggregate fertility and household savings:
A general equilibrium analysis using micro data. NBER Working Paper 20050.
Adamopoulos, T., Brandt, L., Leight, J., and Restuccia, D. (2022). Misallocation, selection, and
productivity: A quantitative analysis with panel data from china. Econometrica, 90(3):1261–
1282.
Alpysbayeva, D. and Vanormelingen, S. (2022). Labor market rigidities and misallocation: Evidence
from a natural experiment. Labour Economics, 78:102229.
Altonji, J. G., Hayashi, F., and Kotlikoff, L. J. (1997). Parental altruism and inter vivos transfers:
Theory and evidence. Journal of Political Economy, 105(6):1121–1166.
Amodio, F., Medina, P., and Morlacco, M. (2023). Labor market power, self-employment, and
development. Working Paper.
An, L., Qin, Y., Wu, J., and You, W. (2024). The local labor market effect of relaxing internal
migration restrictions: Evidence from china. Journal of Labor Economics, 22.
Armangue-Jubert, T., Guner, N., and Ruggieri, A. (2023). Labor market power and development.
Unpublished Manuscript.
Bai, C. and Wang, Y. (1998). Bureaucratic control and the soft budget constraint. Journal of
Comparative Economics, 26(1):41–61.
Bao, Shuming, B. O. H. J. and Zhao, Y. (2011). The regulation of migration in a transition
economy: China’s hukou system. Contemporary Economic Policy, 29(4):564–579.
92

Baqaee, D. R. and Farhi, E. (2019). The macroeconomic impact of microeconomic shocks: Beyond
hulten’s theorem. Econometrica, 87(4):1155–1203.
Bartelsman, E., Haltiwanger, J., and Scarpetta, S. (2013). Cross-country differences in productivity:
The role of allocation and selection. American economic review, 103(1):305–334.
Bau, N. and Matray, A. (2023). Misallocation and capital market integration: Evidence from india.
Econometrica, 91(1):67–106.
Beerli, A., Ruffner, J., Siegenthaler, M., and Peri, G. (2021). The abolition of immigration restrictions and the performance of firms and workers: Evidence from switzerland. American Economic
Review, 111(3):976–1012.
Blattner, L., Farinha, L., and Rebelo, F. (2023). When losses turn into loans: The cost of weak
banks. American Economic Review, 113(6):1600–1641.
Blundell, R., Meghir, C., and Smith, S. (2002). Pension incentives and the pattern of early retirement. The Economic Journal, 112(478):C153–C170.
Borjas, G. J. (2017). The labor supply of undocumented immigrants. Labour Economics, 46:1–13.
Borjas, G. J. and Edo, A. (2023). Monopsony, efficiency, and the regularization of undocumented
immigrants. Working Paper 31457, National Bureau of Economic Research.
Borusyak, K., Hull, P., and Jaravel, X. (2022). Quasi-experimental shift-share research designs.
The Review of Economic Studies, 89(1):181–213.
Brandt, L. and Khandelwal, A. (2015). Comments and discussion: Grasp the large, let go of the
small: The transformation of the state sector in China. Brookings Papers on Economic Activity,
2015(SPRING):295–366.
Brandt, L., Van Biesebroeck, J., Wang, L., and Zhang, Y. (2017). Wto accession and performance
of chinese manufacturing firms. American Economic Review, 107(9):2784–2820.
Brandt, L., Van Biesebroeck, J., and Zhang, Y. (2012). Creative accounting or creative destruction?
firm-level productivity growth in chinese manufacturing. Journal of Development Economics,
97(2):339–351.
Bryan, G. and Morten, M. (2019). The aggregate productivity effects of internal migration: Evidence from indonesia. Journal of Political Economy, 127(5):2229–2268.
93

Brzezinska, I. (2021). The working dread? analyzing the impact of the hukou reform on firms’
monopsony power in china. Technical report.
Carrington, W. J., Detragiache, E., and Vishwanath, T. (1996). Migration with endogenous moving
costs. The American Economic Review, 86(4):909–930.
Chan, K. W. (2011). In the city, but not of the city: the myth of china’s urbanization. China-US
Focus, 16.
Chan, K. W. (2012). Migration and development in china: trends, geography and current issues.
Migration and Development, 1(2):187–205.
Chang, S. D. (1996). Cities with invisible walls. China Review International, 3(2):375–376.
Chen, S. X. (2017). The effect of a fiscal squeeze on tax enforcement: Evidence from a natural
experiment in china. Journal of Public Economics, 147:62–76.
Chen, Y. and Feng, S. (2013). Access to public schools and the education of migrant children in
china. China Economic Review, 26:75–88.
Choudhury, P. (2022). Geographic mobility, immobility, and geographic flexibility: A review and
agenda for research on the changing geography of work. Academy of Management Annals,
16(1):258–296.
Chow, C. K., Song, F. M., and Wong, K. P. (2010). Investment and the soft budget constraint
in china. International Review of Economics Finance, 19(2):219–227. Special Issue: Recent
Developments in China’s Financial Markets.
Cox, D. and Rank, M. R. (1992). Inter-vivos transfers and intergenerational exchange. The Review
of Economics and Statistics, 74(2):305–314.
Cui, R. and Jeffrey, C. (2015). Reform and the hukou system in china. Migration Letters.
Dai, M., Huang, W., and Zhang, Y. (2020). Persistent effects of initial labor market conditions:
The case of china’s tariff liberalization after wto accession. Journal of Economic Behavior Organization, 178:566–581.
David, J. M. and Venkateswaran, V. (2019). The sources of capital misallocation. American
Economic Review, 109(7):2531–2567.
De Chaisemartin, C. and d’Haultfoeuille, X. (2020). Two-way fixed effects estimators with heterogeneous treatment effects. American Economic Review, 110(9):2964–96.
94

De Weerdt, J. and Dercon, S. (2006). Risk-sharing networks and insurance against illness. Journal
of Development Economics, 81(2):337–356.
Dercon, S. and Krishnan, P. (2000). In sickness and in health: Risk sharing within households in
rural ethiopia. Journal of Political Economy.
Diamond, R. (2016a). The determinants and welfare implications of us workers’ diverging location
choices by skill: 1980-2000. American Economic Review, 106(3):479–524.
Diamond, R. (2016b). The determinants and welfare implications of us workers’ diverging location
choices by skill: 1980-2000. American Economic Review, 106(3):479–524.
Donaldson, D. and Hornbeck, R. (2016). Railroads and american economic growth: A ”market
access” approach. The Quarterly Journal of Economics, 131(2):799–858.
Dustmann, C. and Glitz, A. (2015). How do industries and firms respond to changes in local labor
supply? Journal of Labor Economics, 33(3):711–750.
Easterly, W. (1993). How much do distortions affect growth? Journal of Monetary Economics,
32(2):187–212.
Fafchamps, M. and Lund, S. (2003). Risk-sharing networks in rural philippines. Journal of Development Economics.
Fang, H. (2018). The chinese pension system. NBER Working Paper 25088.
Feldstein, M. (1999). Social security pension reform in china. China Economic Review, 10(2):99–
107.
Foster, A. and Zhou, K. (2020). No More Lethal Deportations: Migration Access, Household Size,
and Agricultural Development in China.
Foster, A. D. and Zhou, K. (2022). No more lethal deportations: institutional barriers, household
size, and economic development in china. Unpublished Manuscript, Zhejiang University.
Foster, L. and Walker, A. (2014). Active and Successful Aging: A European Policy Perspective.
The Gerontologist, 55(1):83–90.
Frijters, P. and Lee, L. (2010). Jobs, working hours, and remuneration packages for migrants and
urban residents, in xin meng, chris manning, s. li and tn effendi (eds.) the great migration:
rural-urban migration in china and indonesia. cheltenham: Edward elgar publishing.
95

Ge, S., Zhang, J., and Zhou, K. (2023). Aging in a dual economy: Urban aging, massive migration,
and agricultural development. Unpublished Manuscript, Zhejiang University.
Giles, J. and Yoo, K. (2007). Precautionary behavior, migrant networks, and household consumption decisions: An empirical analysis using household panel data from rural china. The Review
of Economics and Statistics, 89(3):534–551.
Gopinath, G., Kalemli-Ozcan, S¸., Karabarbounis, L., and Villegas-Sanchez, C. (2017). Capital ¨
allocation and productivity in south europe. The Quarterly Journal of Economics, 132(4):1915–
1967.
Graham, D. J. and Spence, N. (2000). Manufacturing employment change, output demand, and
labor productivity in the regions of britain. International Regional Science Review, 23(2):172–
200.
Handley, K. and Limao, N. (2017). Policy uncertainty, trade, and welfare: Theory and evidence
for china and the united states. American Economic Review, 107(9):2731–83.
Hau, H., Huang, Y., and Wang, G. (2020). Firm response to competitive shocks: Evidence from
china’s minimum wage policy. The Review of Economic Studies, 87(6):2639–2671.
Hayashi, F., Altonji, J., and Kotlikoff, L. (1996). Risk-sharing between and within families. Econometrica, 64(2):261–294.
Hsieh, C.-T. and Klenow, P. J. (2009). Misallocation and manufacturing TFP in China and India.
The Quarterly Journal of Economics, 124(4):1403–1448.
Hsieh, C.-T. and Olken, B. A. (2014). The missing ”missing middle”. Journal of Economic Perspectives, 28(3):89–108.
Hsieh, C.-T. and Song, Z. M. (2015). Grasp the large, let go of the small: The transformation of
the state sector in china. Brookings Papers on Economic Activity, pages 295–346.
Hu, Y., Wang, Y., and Zhang, P. (2023). Anti-urbanization and rural development: Evidence from
return migrants in china. Journal of Rural Studies, 103:103102.
Imbert, C., Seror, M., Zhang, Y., and Zylberberg, Y. (2022). Migrants and firms: Evidence from
china. American Economic Review, 112(6):1885–1914.
Jin, Z. and Zhang, J. (2023). Access to local citizenship and internal migration in a developing
country: Evidence from a hukou reform in china. Journal of Comparative Economics, 51(1):181–
215.
96

Jones, C., Daron Acemoglu, M. A., and Dekel, E. (2013). Misallocation, input-output economics,
and economic growth. Advances in Economics and Econometrics.
Jones, C. I. (2013). Input-output economics. In Advances in Economics and Econometrics: Tenth
World Congress, volume 2, page 419. Cambridge University Press.
Kuhn, P. and Shen, K. (2015). Do employers prefer migrant workers? evidence from a chinese job
board. IZA Journal of Labor Economics, 4:1–31.
Lagakos, D., Mobarak, A. M., and Waugh, M. E. (2023). The welfare effects of encouraging rural–
urban migration. Econometrica, 91(3):803–837.
Laitner, J. (1992). Random earnings differences, lifetime liquidity constraints, and altruistic intergenerational transfers. Journal of Economic Theory, 58(2):135–170.
Lewis, E. (2011). Immigration, skill mix, and capital skill complementarity. The Quarterly Journal
of Economics, 126(2):1029–1069.
Liang, Z., Li, Z., and Ma, Z. (2014). Changing patterns of the floating population in china, 2000–
2010. Population and development review, 40(4):695–716.
Liu, Z. (2005). Institution and inequality: the hukou system in china. Journal of Comparative
Economics, 33(1):133–157.
McFadden, D. et al. (1973). Conditional logit analysis of qualitative choice behavior.
Meng, X. (2012). Labor market outcomes and reforms in china. Journal of Economic Perspectives,
26(4):75–102.
Meng, X. and Zhang, J. (2001). The two-tier labor market in urban china: occupational segregation and wage differentials between urban residents and rural migrants in shanghai. Journal of
Comparative Economics, 29(3):485–504.
Midrigan, V. and Xu, D. Y. (2014). Finance and misallocation: Evidence from plant-level data.
American economic review, 104(2):422–458.
Morand, O. F. (1999). Income distribution, and growth. Journal of Economic Growth.
Munshi, K. and Rosenzweig, M. (2016). Networks and misallocation: Insurance, migration, and
the rural-urban wage gap. American Economic Review, 106(01):46–98.
97

Nakamura, E., Sigurdsson, J., and Steinsson, J. (2022). The gift of moving: Intergenerational
consequences of a mobility shock. The Review of Economic Studies, 89(3):1557–1592.
Ngai, L. R., Pissarides, C. A., and Wang, J. (2019). China’s mobility barriers and employment
allocations. Journal of the European Economic Association, 17(5):1617–1653.
Nishiyama, S. (2002). Bequests, inter vivos transfers, and wealth distribution. Review of Economic
Dynamics, 5(4):892–931.
Ortigueira, S. and Siassi, N. (2013). How important is intra-household risk sharing for savings and
labor supply? Journal of Monetary Economics.
Peri, G. (2012). The effect of immigration on productivity: Evidence from us states. Review of
Economics and Statistics, 94(1):348–358.
Peters, M. (2020). Heterogeneous markups, growth, and endogenous misallocation. Econometrica,
88(5):2037–2073.
Pi, J. and Zhang, P. (2016). Hukou system reforms and skilled-unskilled wage inequality in china.
China Economic Review, 41:90–103.
Pierce, J. R. and Schott, P. K. (2016). The surprisingly swift decline of us manufacturing employment. American Economic Review, 106(7):1632–62.
Pomfret, J. (2003). China voids rule on jailing vagrants: Policy blamed for hundreds of feaths.
Washington Post, 19.
Qian, Y. and Roland, G. (1998). Federalism and the soft budget constraint. American Economic
Review, pages 1143–1162.
Restuccia, D. and Rogerson, R. (2017). The causes and costs of misallocation. Journal of Economic
Perspectives, 31(3):151–174.
Rosenzweig, M. R. and Stark, O. (1989). Consumption smoothing, migration, and marriage: Evidence from rural india. Journal of political Economy, 97(4):905–926.
Samuel H. Preston, C. H. . M. E. (1989). Demographic conditions responsible for population aging.
Demography.
Schoonbroodt, A. and Tertilt, M. (2014). Property rights and efficiency in olg models with endogenous fertility. Journal of Economic Theory, 150:551–582.
98

Shen, P. K. . K. (2015). Do employers prefer migrant workers? evidence from a chinese job board.
IZA Journal of Labor Economics.
Shi, L. et al. (2008). Rural migrant workers in China: scenario, challenges and public policy.
Citeseer.
Sieg, H., Yoon, C., and Zhang, J. (2023). The impact of local fiscal and migration policies on human
capital accumulation and inequality in china. International Economic Review, 64(1):57–93.
Smith, J. P., Strauss, J., and Zhao, Y. (2014). Healthy aging in china. The Journal of the Economics
of Ageing, 4:37–43. The Economic Implications of Population Ageing in China and India.
Song, Y. (2014). What should economists know about the current chinese hukou system? China
Economic Review, 29:200–212.
Song, Z., Storesletten, K., and Zilibotti, F. (2011). Growing like china. American Economic Review,
101(1):196–233.
Sraer, D. and Thesmar, D. (2023). How to use natural experiments to estimate misallocation.
American Economic Review, 113(4):906–938.
Su, Y., Tesfazion, P., and Zhao, Z. (2018). Where are the migrants from? inter- vs. intra-provincial
rural-urban migration in china. China Economic Review, 47:142–155.
Tauchen, G. (1986). Finite state markov-chain approximations to univariate and vector autoregressions. Economics Letters, 20(2):177–181.
Tombe, T. and Zhu, X. (2019). Trade, migration, and productivity: A quantitative analysis of
china. American Economic Review, 109(5):1843–72.
Trottner, F. (2023). Unbundling market power. Technical report.
Ulyssea, G. (2018). Firms, informality, and development: Theory and evidence from brazil. American Economic Review, 108(8):2015–2047.
Varela, L. (2018). Reallocation, competition, and productivity: Evidence from a financial liberalization episode. The Review of Economic Studies, 85(2):1279–1313.
Wang, F., Milner, C., and Scheffel, J. (2021). Labour market reform and firm-level employment
adjustment: Evidence from the hukou reform in china. Journal of Development Economics,
149:102584.
99

Wang, F.-L. (2004). Reformed migration control and new targeted people: China’s hukou system
in the 2000s. The China Quarterly, 177:115–132.
Wang, Y. (2019). Green Book of China Social Security System: China Social Security System
Development Report.
Wei, X. and Zhu, H. (2020). Return migrants’ entrepreneurial decisions in rural china. Asian
Population Studies, 16(1):61–81.
Wu, C., Fu, Q., Gu, J., and Shi, Z. (2018). Does migration pay off? returnees, family background,
and self-employment in rural china. China Review, 18(1):59–78.
Young, A. (2013). Inequality, the urban-rural gap, and migration. The Quarterly Journal of
Economics, 128(4):1727–1785.
Yu, L., Yin, X., Zheng, X., and Li, W. (2017). Lose to win: entrepreneurship of returned migrants
in china. The Annals of Regional Science Article, pages 341–374.
Zeng, Y. (2011). Effects of demographic and retirement-age policies on future pension deficits, with
an application to china. Population and Development Review.
Zeng, Y. and Wang, Z. (2014). A policy analysis on challenges and opportunities of population/household aging in china. Population Ageing.
Zhang, H. (2010). The hukou system’s constraints on migrant workers’ job mobility in chinese
cities. China Economic Review, 21(1):51–64.
Zhang, J., Huang, J., Wang, J., and Guo, L. (2020). Return migration and hukou registration
constraints in chinese cities. China Economic Review, 63:101498.
Zhang, J. and Zhou, K. (2023). Quota removal, destination-specific export shocks, and skill acquisition in china. Journal of Development Economics, 165:103149.
Zhang, Y. and Goza, F. W. (2006). Who will care for the elderly in china?: A review of the
problems caused by china’s one-child policy and their potential solutions. Journal of Aging
Studies, 20(2):151–164.
Zhao, Y. (2000). Rural-to-urban labor migration in china: the past and the present. Rural labor
flows in China, 2:15–33.
Zhao, Y. (2008). The role of migrant networks in labor migration: The case of china. Contemporary
Economic Policy.
100

Zhong, H. (2011). The impact of population aging on income inequality in developing countries:
Evidence from rural china. China Economic Review, 22(1):98–107.
Zhou, K. and Zhang, J. (2021). Trade normalization, export quality, and in-migration of skilled
workers: Evidence from china. Journal of Economic Behavior & Organization, 184:375–387.
Zi, Y. (2020). Trade liberalization and the great labor reallocation.
˙Imrohoro˘glu, A. and Zhao, K. (2018a). The chinese saving rate: Long-term care risks, family
insurance, and demographics. Journal of Monetary Economics, 96:33–52.
˙Imrohoro˘glu, A. and Zhao, K. (2018b). Intergenerational transfers and china’s social security
reform. The Journal of the Economics of Ageing, 11:62–70. Macroeconomics of Aging.
101

Appendices
102

Chapter One Appendix
Institutional Background: the Hukou Reform in the 2000s
In this section, we plotted the spreading out of hukou reform during the 2000s in maps. More
specifically, we used GDP level and population level in 2000 as baseline prefecture feature, and
plotted the reformed prefectures as yellow dots till the end of each year. The scattered mapping
suggests that there is less concern about potential correlation between prefecture characteristics
and treatment assignment.
Next, we use a set of prefecture-level characteristics to predict which prefecture has ever been
treated or not. Except for the number of firms, all other characteristics, including average capital,
average employment, share of non-agricultural residents, minimum wage floor, and prefecture-level
Herfindahl–Hirschman index, can not successfully predict the upcoming reform.
103

4.6e+07
134971
No data
Baseline GDP in 2000
Prefectures Reformed by 2001
4.6e+07
134971
No data
Baseline GDP in 2000
Prefectures Reformed by 2002
4.6e+07
134971
No data
Baseline GDP in 2000
Prefectures Reformed by 2003
4.6e+07
134971
No data
Baseline GDP in 2000
Prefectures Reformed by 2004
4.6e+07
134971
No data
Baseline GDP in 2000
Prefectures Reformed by 2005
4.6e+07
134971
No data
Baseline GDP in 2000
Prefectures Reformed by 2006
4.6e+07
134971
No data
Baseline GDP in 2000
Prefectures Reformed by 2007
Figure 5: Maps of Policy Implementation with 2000 baseline GDP
Notes: The maps show the spreading out of policies across prefectures in China. Yellow dots represented the treated
prefecture by a certain year. The base color was filled using prefecture-level GDP in 2000. We year-by-year label
the reformed and non-reformed prefectures and visually allow the prefectures to vary in GDP and population in the
initial year.
104

3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2001
3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2002
3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2003
3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2004
3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2005
3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2006
3091.09
15.96
No data
Baseline Population in 2000
Prefectures Reformed by 2007
Figure 6: Maps of Policy Implementation with 2000 Population
Notes: The maps show the spreading out of policies across prefectures in China. Yellow dots represented the treated
prefecture by a certain year. The base color was filled using prefecture-level population in 2000. We year-by-year
label the reformed and non-reformed prefectures and visually allow the prefectures to vary in GDP and population
in the initial year.
105

Table 7: Association Between Prefecture-level Characteristics and the Reform
(1) (2) (3) (4) (5) (6)
VARIABLES policy policy policy policy policy policy
Number of Firms 0.0832***
(0.0164)
Average Capital 0.0574
(0.0438)
Average Employment 0.0434
(0.0519)
Share of Non-agri. Residents -0.0142
(0.0562)
Minimum Wage 0.151
(0.142)
HHI -0.000554
(0.0606)
Observations 292 292 292 250 249 292
R-squared 0.081 0.006 0.002 0.000 0.005 0.000
Notes: The dependent variable policy equals one if a prefecture has ever been treated in the study period, that
is from 1998-2007. All dependent variables are in logs and restricted to the pre-reform period from 1998-2000,
averaged over three years. HHI is industry concentrated which has been aggregated to the prefecture level,
weighted by pre-reform firm capital stock.
106

Methodology: Frims as Price-takers in the Local Labor Market
In this appendix, we extend our framework of exogenous distortions to monopsony power as a
source of distortions.
Labor wedges contribute to two sources of misallocation. First, it deviates firms’ marginal labor
productivity from the per unit factor price (wage rate), leading to aggregate dead weight loss;
Second, distorted input allocation leads to labor misallocation across firms if reallocating labor
to high marginal labor productivity firm is welfare-improving. With the hukou system in China,
firms have monopsony power because workers are faced with the cost of switching jobs. As a result,
they may not leave their jobs when employers cut their wages slightly. Firms’ monopsony power
over migrant workers is generally higher than that over local workers (Borjas, 2017; Shen, 2015).
Having restricted job opportunities and facing a risk of getting caught and sent to C&R facilities,
migrant workers without local hukou are less likely to quit their jobs even if employers decrease
their wages. In our framework, labor supply can instead be elastic and firms are allowed to face
different labor supply curves if we assume that workers’ utility is affected by exogenous differences
in non-wage amenities at different firms. Additionally, firms do not observe workers’ idiosyncratic
workplace preferences and thus can not be wage discriminating. This allows labor supply elasticity
to vary exogeneously at the firm level (Trottner, 2023).
Hukou reform changes several aspects. It allows migrant workers to migrate and work, responding
to destination wage rate change. It also puts an extra cost for employers who now need to provide
accommodations, food, employment insurance, and medical insurance for legalized migrant workers.
Thus, we model hukou reform as a (an inverse) labor elasticity decrease and an increased migrant
input tax (Brzezinska, 2021; Borjas and Edo, 2023). For simplicity, suppose that initially hukou
reform only decreases inverse labor elasticity. The production technology uses three inputs: native
workers Ln, migrant workers Lm, and capital K. Production function Q = f(Ln, Lm, K) is concave
and linear homogeneous. The inputs supply for firm i is Lij = Pijw
1/ϵij , with j = n, m, k. In the
factor supply function, ϵij is the (inverse) supply elasticity of input j and of firm i and is used to
measure firm’s monopsony power, and Pij is the baseline stock of input j for firm i even if the price
107

of input is zero or firm level factor supply is inelastic as ϵij goes to infinity (input supply does not
respond to input price change at all). Rewriting the upward-sloping input supply function gives
input price function wij = P
ϵij
ij L
ϵij
ij .
A cost-minimizing firm equates its marginal product of input i to its marginal cost:
fij = (1 + ϵij )wij = (1 + ϵij )P
−ϵij
ij L
ϵij
ij , (13)
where j = n, m, k for firm i. The first-order conditions indicate that the marginal product of
labor is equal to its wedged wage in a market with monopsonistic producers, and the wedge is
the inverse factor of supply elasticity. This relationship shows that the greater the monopsony
power, the greater the labor input wedge. Under the assumption of Cobb-Douglas production,
Qi = AiL
αn
ni L
αm
mi K
αk
i
, the marginal productivity of labor AiL
αn
ni αmL
αm−1
mi K
αk
i = αmQ/Lmi =
(1+ϵmi)ωmi. This indicates that the labor wedge is proportional to the inverse labor share (revenues
over total wage bill).
To decompose the change in output into labor market expansion and allocative effect, we follow
Baqaee and Farhi (2019) and decompose the change in output as:
dlogY =
∂logY
∂logL dlogL
| {z }
∆InputChange
+
∂logY
∂logAdlogA
| {z }
∆T echnology
+
∂logY
∂logχ dχ
| {z }
∆AllocativeEf f iciency
,
where Y is aggregate output, A is total factor productivity index and χ is allocation rule.7 Without hukou reform affecting the technical efficiency, the increase in output can be decomposed into
changes in external input increase and a reallocation of resources across producers. This decomposition leads to a natural definition of changes in aggregate productivity:
dSolow = dlogY −
∂logY
∂logL dlogL =
∂logY
∂logAdlogA
| {z }
∆T echnology
+
∂logY
∂logχ dχ
| {z }
∆AllocativeEf f iciency
.
This measures the change in aggregate productivity as a combination of the effects of changes in
7We define the allocation matrix χ, where χij = yij/yj is the share of the physical output of producer j used by
producer i.
108

technical and allocative efficiency. Assuming hukou relaxation does not affect technology, the Solow
residual gain comes purely from the allocative effect. We use both a reduced-form method and a
more structural first-order approximation method to get the aggregate productivity gain in Section
1.
To unbundle the allocative effect and market expansion effect in production gain, we first consider
the hukou reform as a reduction in inverse labor supply elasticity thus a reduction in monopsony
power employers have over migrant workers (ϵim). Before the reform, only a limited number of
workers could respond to new job opportunities and to overall wage increases in urban firms. After
the implementation of hukou relaxation, more workers migrate to work in urban firms and can
obtain local hukou if they satisfy some conditions (have a legal temporary address and have a job).
The decline in inverse labor supply elasticity allows migrant workers to respond more strongly to
wage change and force employers to increase wage if they want to attract these migrant workers to
work. The decline in inverse elasticity leads to a decline in the marginal cost of labor:
dMCm
dϵm
= MCm(logLm
Pm
+
1
1 + ϵm
) > 0 (14)
To be more clear about the policy effects, we write the element X change due to hukou reform as
dX/dR = −dX/dϵm. To estimate the change in all other input factors due to hukou reform R, we
rewrite the first order conditions as a system of equations,8 det(

H

) = ∆:






fnn − ϵnfnL
−1
n
fnk fnm
fkn fkk − ϵkfkK−1 fkm
fmn fmk fmm − ϵmfmL
−1
m












dLn/dϵm
dK/dϵm
dLm/dϵm






=






0
0
dMCm
dϵm






(15)
Solving the system of equations, the change in migrant demand due to the decline in labor supply
elasticity is:
dLm
dϵm
=
1
∆
dMCm
dϵm
(F1F2 − f
2
nk) < 0.
8We denote fnn − ϵnfnL
−1
n as F1, fkk − ϵkfkK−1
as F2, fmm − ϵmfmL
−1
m as F3 and dMCm
dϵm
as MC˜
109

We can easily get the other two equations:
dLn
dϵm
=
1
∆
dMCm
dϵm
(fnkfmk − fnmfkk + fnmϵkfkL
−1
k
)
dK
dϵm
=
1
∆
dMCm
dϵm
(fnkfnm − fkmfnn + fkmϵnfnL
−1
n
) < 0.
This gives that by decreasing inverse labor supply elasticity, the marginal cost of hiring migrate
workers decreases, and demand for all input grows, given that capital is complementary to migrant
workers:
dLm
dR > 0,
dLn
dR ?0,
K
dR > 0. (16)
We aim to understand, after the hukou reform, how labor misallocation changes from two aspects:
1) if the variance of marginal products decreases with the decreases of inverse labor supply elasticity;
and 2) if more labor is reallocated to more labor-productive firms. To explore this, we first consider
an extreme case, in which hukou reform fully eliminates the monopsony power of firms over migrant
workers (ϵm goes to zero). In this scenario, firms now become price takers and the variance of
marginal cost shrinks to zero. Consequently, misallocation goes down as the variance of marginal
labor products decreases:
var(fij ) − var(f
af ter
ij ) = var((1 + ϵij )ωij ) > 0,
where j = n, m, k. We now compare the change in employment between two firms, a firm with
ex-ante monopsony power ϵim over migrant workers and a firm with ex-ante zero monopsony power.
The difference in change in employment is given by:
∆L
High − ∆L
Low =
dLmHigh
ϵ
High
m
dϵHigh
m − 0 > 0.
This suggests that firms with higher initial monopsony power experience larger employment expansion post-reform due to a more pronounced decline in marginal cost, prompting increased hiring of
110

migrant workers:
dLm
dR
dMCm
=
−1
∆
(F1F2 − f
2
nk)(log(Lm/Pm) + 1
1 + ϵm
) > 0.
Considering the upward-sloping supply curve for both native workers and capital, the impact of
reform on native wages and capital rent can be written as follows:
dlogωn
dR = ϵn
dlogLn
dR ?0
dlogrk
dR = ϵk
dlogK
dR > 0.
Given capital’s complementarity with migrant workers, the demand for capital increases. However,
the demand for native workers might increase or decrease depending on whether they complement
or substitute them. Analyzing the impact of hukou reform on migrant workers’ wages involves two
key components:
dlogωm
dR = −
dϵm(logLm − logPm)
dϵm
= −logLm
Pm
+ ϵm
dlogLm
dR .
The first term is negative because migrant workers are greater than its baseline amount following
hukou relaxation. This negative effect is consistent with labor supply shock: when the supply of
migrant workers increases, there is a downward pressure on their wage rate. Conversely, the second
term represents a positive impact, where employers, facing reduced monopsony power, increase
wages to attract migrant workers. For the overall effect of reform on the migrants’ wage rate to be
positive, the second effect must outweigh the supply shock effect. The influence of hukou reform
on the average wage rate is a composite of its effect on native and migrant workers:
˜ logω
dR =
Lm
Lm + Ln
dlogωm
dR +
Ln
Lm + Ln
dlogωn
dR .
Notably, firms with higher initial monopsony power experience a more substantial employment
111

expansion post-reform. Consequently, the supply shock effect from increased migrant employment
dominates the impact on the average wage rate. This is consistent with our empirical findings in
Section 1, where firms with higher initial labor productivity employ more workers due to decreased
marginal labor costs, causing a decrease in the average wage rate. Meanwhile, ex-ante low labor
productivity firms do not see a significant change in overall employment size, but their average
wage rate increases slightly because of labor demand increase.
Let’s now consider the other parameterized change of hukou reform—costs firms incur to employ
migrant workers, such as accommodation, dining, insurance, etc. After incorporating these expenses, the marginal cost transforms to: MCm = (1 + τm)(1 + ϵm)wm = (1 + τm)(1 + ϵm)P
−ϵm
m L
ϵm
u
.
For simplicity, let’s assume hukou reform increases costs by τm and reduces migrant labor supply
elasticity by ϵm, and both changes are by k percentage. Thus, the new cost becomes:
dMCm
dRϵτ
=
dMCm
dlogτ
dlogτm
dR +
dMCm
dlogϵm
dlogϵm
dR
= k(
dMCm
dτm
τm −
dMCm
dϵm
ϵm)
= kMCm(
τm − ϵm
(1 + τm)(1 + ϵm)
− ϵmlogLm
Pm
).
The second term, representing the impact of labor supply elasticity, is negative. For marginal cost
to decrease post-reform, the first term must also be negative. This condition suggests τm < ϵm.
In cases where firms possess considerable initial monopsony power and a relatively low additional
cost due to hukou relaxation, their new marginal cost decreases after the reform. Under these
circumstances, if native workers and capital complement migrant workers, the demand for these
two inputs increases, and their prices surge. However, if native workers substitute for migrant
workers, the demand for native labor falls (dLn/dR < 0), even though overall employment increases
(
dLn
dR +
dLm
dR > 0). This is because the second effect (increase in migrant worker employment) is a
major effect.
We show that this elasticity decline leads to a decline in the producer’s marginal cost of migrant
workers, and thus increased demand for all inputs. We next show that ex-ante high marginal cost
firms, thus high labor-wedged firms, have a bigger drop in the marginal cost of migrant workers
and experience larger employment expansion after the reform. The consequences are production112

increasing because more workers are hired by high marginal labor productivity firms. To approximate the production gain specifically from labor market expansion, we follow Borjas and Edo (2023)
and estimate the increase in a firm’s production solely from the decline in inverse labor elasticity
(without reallocation across producers). We compare a reduced form aggregate production gain
with the production gain from market expansion based on micro-model estimates (Section 1), and
we get a difference in production gain. The difference illustrates the allocative effect.
113

Data and Variables: Summary Statistics
Table 8: Summary Statistics by Groups
(1)) (2) (3)
AVG SD N
A: Initially Less Productive
Employment 376.76 804.56 53972
Capital 15622.35 65031.61 53972
Intermediate Input 17979.32 46417.95 53972
Revenues 21327.70 60389.70 53972
Avg. Lab. Prod 52.45 40.03 53972
Avg. Cap. Prod 8.88 133.88 53623
Wate Rate (in thousands) 7.70 5.85 52335
Total Wage (in thousands) 2763.65 7924.65 53972
A: Initially More Productive
Employment 281.37 1222.68 60892
Capital 32547.29 273618.06 60892
Intermediate Input 65651.93 377436.57 60892
Revenues 81829.22 485732.76 60892
Avg. Lab. Prod 327.57 586.46 60892
Avg. Cap. Prod 22.03 327.14 60679
Wate Rate (in thousands) 13.75 40.94 59019
Total Wage (in thousands) 3559.20 18741.20 60892
B: Non SOE
Employment 254.24 595.87 80818
Capital 16915.72 116650.98 80818
Intermediate Input 40997.16 251088.62 80818
Revenues 49584.95 305079.11 80818
Avg. Lab. Prod 232.40 502.46 80818
Avg. Cap. Prod 18.53 290.55 80476
Wate Rate (in thousands) 11.63 32.89 78271
Total Wage (in thousands) 2575.11 9839.89 80818
B: SOE
Employment 616.01 1961.35 24365
Capital 56538.90 387099.40 24365
Intermediate Input 58313.53 389106.83 24365
Revenues 75986.02 536706.14 24365
Avg. Lab. Prod 90.70 251.90 24365
Avg. Cap. Prod 5.98 75.91 24188
Wate Rate (in thousands) 8.59 24.95 23581
Total Wage (in thousands) 5778.79 26092.70 24365
Notes: All variables are in levels. The table reports descriptive statistics in
pre-reform year 2000.
114

Table 9: Summary Statistics of Pre-reform Observed Main Sample and Whole Sample
(1)) (2) (3) (4) (5)
A: In Logarithm AVG SD N MIN MAX
Employment 5.08 1.17 711742 2.3 11.5
Capital 8.44 1.78 709749 0.0 17.9
Intermediate Input 9.71 1.52 710591 0.0 18.3
Revenue 9.91 1.50 711727 0.0 18.7
Avg. Lab. Prod. 4.67 1.31 711694 -6.9 11.7
Avg. Cap. Prod 1.31 1.63 709702 -13.9 11.9
Wage Rate 2.17 0.82 691655 -7.4 9.0
Total Wage Bill 7.26 1.43 691655 0.0 15.8
B: In Levels
Employment 261.76 875.03 1732991 10.0 99800.0
Capital 22214.72 228102.25 1732991 0.0 57895456.0
Intermediate Input 60017.73 518213.99 1732991 0.0 1.2e+08
Revenue 76212.88 656761.10 1732863 1.0 1.4e+08
Avg. Lab. Prod 313.79 793.69 1732863 0.0 358847.4
Avg. Cap. Prod 29.27 855.31 1726315 0.0 911128.3
Wate Rate (in thousands) 14.21 28.69 1688133 0.0 26278.6
Total Wage (in thousands) 3710.58 23278.66 1732991 0.0 7508160.0
Year 2001.32 2.59 711742 1998.0 2007.0
City Code 3498.22 1107.23 711742 1101.0 6543.0
Industry Code (2-digit) 27.49 8.58 711742 13.0 42.0
C: City-level Stats
Population 5.82 0.71 2596 2.7 8.1
GDP 15.00 0.94 2559 12.0 18.6
Nonagricultural Residents Share -1.25 0.51 2580 -2.6 0.0
Fiscal Expenditure 12.16 1.24 2579 8.4 16.9
Minimum Wage 5.83 0.34 2569 5.0 6.7
Per capita GDP 9.18 0.77 2559 7.0 13.4
Herfindel Concentration Index 4.61 0.29 2997 2.7 6.8
D: Whole Sample
Employment 4.77 1.11 1732991 2.3 11.5
Capital 8.17 1.73 1726394 0.0 17.9
Intermediate Input 9.68 1.36 1730817 0.0 18.6
Revenue 9.92 1.33 1732863 0.0 18.7
Avg. Lab. Prod. 5.02 1.23 1732805 -8.1 12.8
Avg. Cap. Prod 1.62 1.57 1726258 -13.9 13.7
Wage Rate 2.37 0.76 1688133 -7.4 10.2
Total Wage Bill 7.13 1.30 1688133 0.0 15.8
Notes: Summary statistics are reported in logarithms in Section A and in levels in Section B. The table
includes the pre-reform observed main sample from Section A-C and the whole sample from Section D.
115

Results: Average Policy Effects
Figure 7: Average Treatment Effect Event Studies (Treated Prefectures Only)
Notes: The figure reports event study graphs for the average effects of the hukou reform on firms. Hukou
relaxation takes place in year zero. The figures plot coefficients on the dummies of being observed t years from the
reform. Confidence intervals are at 95%.
116

Results: Differential Effects by the Baseline Distortion Level
To further confirm that if these initially more distorted firms indeed generate slightly higher revenues relative to the less distorted firms, we examine the differential effects on a wide range of
variables, including output, industrial value added, and value-added tax besides revenues we used
in our main results. Table 10 in Appendix 3 reports the results.9 Regression results with ex-ante
employment weights suggest that the baseline more distorted firms that are initially bigger in size
disproportionally increased their revenue relative to the less distorted firms. More specifically, after
weighting our main regressions with ex-ante employment, the baseline more distorted firms increase
their output and value-added by 8.4% and 7.8% respectively, and they also pay higher value-added
tax by roughly 24.2%. This suggests that the reform benefits bigger firms, and they become more
profitable by expanding and growing in the market. The estimates also suggest that the reform allows the baseline more distorted firms to relax their distortion and expand by hiring more workers,
enabling these producers that were ex-ante more constrained in labor input to grow and reach a
more efficient size in labor employment.
9We also compare results with and without baseline employment weights in order to see the role that firm size
plays.
117

Table 10: Policy Impact on Output, Value Added and Value Added Tax
(1) (2) (3)
VARIABLES Output Value Added Value Added Tax
A: Pre-reform Main Sample
Policy 0.074*** 0.058* -0.071**
(0.025) (0.030) (0.031)
Policy x I(HP) 0.021 0.045** 0.168***
(0.017) (0.020) (0.025)
Observations 636,376 575,768 557,185
R-squared 0.152 0.126 0.059
B: Employment Weighted
Policy 0.072** 0.040 -0.041
(0.029) (0.034) (0.042)
Policy x I(HP) 0.084** 0.078* 0.242***
(0.036) (0.040) (0.045)
Observations 586,891 570,742 514,016
R-squared 0.176 0.132 0.056
C: No year 2004
Policy 0.077*** 0.058* -0.062*
(0.026) (0.030) (0.032)
Policy x I(HP) 0.020 0.045** 0.178***
(0.017) (0.020) (0.026)
Observations 592,060 575,768 518,538
R-squared 0.169 0.126 0.064
Fixed Effects
Firm ✓ ✓ ✓
Year ✓ ✓ ✓
Industry ✓ ✓ ✓
Ownership ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if a city
has been treated. Value added is calculated by gross output net intermediate input
plus value added tax”. Panel B are weighted regressions with pre-reform employment
weights. Panel C regressions get rid of the year 2004 due to missing data issues. All
regressions control for city GDP, population, and firm wage rate. Standard errors are
clustered at the prefecture level.*, **, and *** denote 10%, 5%, and 1% statistical
significance respectively.
118

Results: Heterogeneity Analysis
Table 11: Policy Impact by Implement Years
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy = 1 0.043*** -0.013 0.043*** -0.003
(0.014) (0.012) (0.014) (0.026)
1.policy#1.i highyl pref -0.021 -0.011 -0.035 0.007
(0.015) (0.022) (0.041) (0.013)
1.policy#2002.year 0.010 -0.002 0.022 0.029
(0.023) (0.011) (0.025) (0.024)
1.policy#2003.year 0.038 0.012 0.053** 0.025
(0.025) (0.016) (0.021) (0.033)
1.policy#2004.year 0.051* 0.046* 0.048** 0.012
(0.029) (0.025) (0.024) (0.042)
1.policy#2005.year 0.066* 0.056* 0.112*** 0.021
(0.036) (0.031) (0.036) (0.046)
1.policy#2006.year 0.072 0.029 0.100*** 0.052
(0.044) (0.040) (0.035) (0.049)
1.policy#2007.year 0.113** 0.017 0.133** 0.125**
(0.049) (0.055) (0.052) (0.056)
1.policy#1.i highyl pref#2002.year 0.029 -0.004 0.021 0.007
(0.029) (0.030) (0.052) (0.018)
1.policy#1.i highyl pref#2003.year 0.013 -0.013 0.006 0.002
(0.019) (0.024) (0.045) (0.023)
1.policy#1.i highyl pref#2004.year 0.002 -0.030 0.004 0.017
(0.024) (0.031) (0.049) (0.026)
1.policy#1.i highyl pref#2005.year -0.020 0.000 -0.011 0.023
(0.032) (0.027) (0.058) (0.033)
1.policy#1.i highyl pref#2006.year -0.001 0.007 0.010 0.010
(0.032) (0.033) (0.059) (0.037)
1.policy#1.i highyl pref#2007.year -0.035 0.016 -0.014 0.027
(0.038) (0.034) (0.062) (0.061)
Observations 638,458 636,376 638,454 636,880
R-squared 0.139 0.253 0.180 0.033
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. I(HP) × year dummies are not reported to save space. Controls include city GDP, population, and firm wage rate. Standard errors are clustered at the prefecture
level.*, **, and *** denote 10%, 5%, and 1% statistical significance respectively.
119

Results: Robustness Analysis
Table 12: Summary Statistics of Exiters and Entrants
(1)) (2) (3) (4) (5)
A: Whole Sample AVG SD N MIN MAX
Employment 4.77 1.12 1760989 2.3 11.5
Capital 8.18 1.73 1754366 0.0 17.9
Intermediate Input 9.65 1.39 1757821 0.0 18.6
Revenue 9.88 1.38 1760863 0.0 18.7
Avg. Lab. Prod. 5.16 1.16 1757290 -8.1 12.8
Avg. Cap. Prod 1.76 1.51 1750797 -13.1 13.8
Wage Rate 2.35 0.78 1714678 -7.8 10.2
Total Wage Bill 7.13 1.31 1714678 0.0 15.8
Observations 1760989
B: Entrants
Employment 4.78 1.09 840721 2.3 11.5
Capital 8.34 1.66 838800 0.0 17.7
Intermediate Input 9.93 1.27 840343 0.0 18.6
Revenue 10.19 1.23 840658 0.0 18.7
Avg. Lab. Prod. 5.43 1.03 840325 -5.7 11.7
Avg. Cap. Prod 1.88 1.35 838438 -10.2 13.8
Wage Rate 2.53 0.68 822850 -7.4 8.1
Total Wage Bill 7.32 1.25 822850 0.0 15.8
Observations 840721
C: Exiters
Employment 4.89 1.13 521868 2.3 11.1
Capital 8.11 1.77 520135 0.0 17.9
Intermediate Input 9.46 1.45 520622 0.0 17.8
Revenue 9.66 1.44 521843 0.0 17.9
Avg. Lab. Prod. 4.85 1.21 520512 -6.4 12.8
Avg. Cap. Prod 1.64 1.64 518796 -12.1 12.0
Wage Rate 2.13 0.81 506769 -7.4 10.2
Total Wage Bill 7.02 1.32 506769 0.0 14.9
Observations 521868
Notes: Summary statistics are in logarithms. Entrants are defined as
firms that entered the sample and stayed in the sample, and exits are
defined as firms that stayed for at least three years in the sample, exited
the sample and never came back. 120

Table 13: Differential Policy Impact for Entrants and Exiters
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
A: Balanced Sample
Policy -0.077* 0.143*** 0.065** -0.017
(0.041) (0.041) (0.033) (0.042)
Policy x I(HP) 0.283*** -0.261*** 0.039 0.168***
(0.025) (0.016) (0.027) (0.027)
Observations 96,059 96,027 96,059 96,022
R-squared 0.125 0.376 0.433 0.020
B: Balanced Sample + Exiters
Policy -0.050 0.133*** 0.080*** -0.016
(0.031) (0.029) (0.026) (0.035)
Policy x I(HP) 0.248*** -0.242*** 0.027 0.105***
(0.013) (0.012) (0.017) (0.017)
Observations 631,513 630,095 631,509 629,957
R-squared 0.108 0.245 0.176 0.031
C: Balanced Sample + Exiters + Entrants
Policy -0.096*** 0.167*** 0.065*** -0.049
(0.031) (0.028) (0.025) (0.033)
Policy x I(HP) 0.233*** -0.214*** 0.042** 0.124***
(0.017) (0.017) (0.018) (0.018)
Observations 1,213,418 1,211,562 1,213,391 1,208,666
R-squared 0.103 0.223 0.205 0.021
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if a city has been treated. Panel A
only includes firms that stayed in our sample from 1998-2007. Panel B is our main analysis sample, which includes
firms that were observed at least once in pre-reform years. Panel C includes all the firms. To keep the ex-ante
high and low productive classification standards consistent in all panels, we use a constructed labor productivity for
pre-reform classification, which equals the predicted residual of labor productivity y/l after taking out all-time fixed
effects. Controls include city GDP, population, and firm wage rate. Standard errors are clustered at the prefecture
level.*, **, and *** denote 10%, 5%, and 1% statistical significance respectively.
121

Table 14: Average Policy Impact for Entrants and Exiters
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
A: Balanced Sample
Policy 0.082** -0.004 0.088*** 0.078**
(0.039) (0.042) (0.032) (0.037)
Observations 96,651 96,613 96,651 96,614
R-squared 0.114 0.371 0.432 0.019
B: Balanced Sample + Exiters
Policy 0.084*** 0.004 0.096*** 0.042
(0.029) (0.030) (0.025) (0.036)
Observations 638,458 636,376 638,454 636,880
R-squared 0.102 0.242 0.175 0.031
C: Balanced Sample + Exiters + Entrants
Policy 0.048** 0.043 0.098*** 0.024
(0.022) (0.029) (0.024) (0.026)
Observations 1,540,410 1,537,728 1,540,380 1,535,101
R-squared 0.084 0.266 0.244 0.016
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if a city has been treated. Panel A
only includes firms that stayed in our sample from 1998-2007. Panel B is our main analysis sample, which includes
firms that were observed at least once in pre-reform years. Panel C includes all the firms. Controls include city
GDP, population and firm wage rate. Standard errors are clustered at the prefecture level.*, **, and *** denote
10%, 5%, and 1% statistical significance respectively.
122

Table 15: Policy Impact by Alternative Cutoffs: Quartiles of Baseline Distortion Level
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy -0.142*** 0.205*** 0.048 -0.062*
(0.032) (0.029) (0.030) (0.033)
Policy × Distortion Level 2 0.149*** -0.135*** 0.033 0.032
(0.011) (0.014) (0.020) (0.026)
Policy × Distortion Level 3 0.276*** -0.226*** 0.075*** 0.136***
(0.015) (0.019) (0.022) (0.021)
Policy × Distortion Level 4 0.413*** -0.403*** 0.044* 0.136***
(0.018) (0.022) (0.026) (0.029)
Observations 605,238 603,205 605,234 603,722
R-squared 0.111 0.250 0.177 0.028
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. Quartile groups of initial labor productivity are
dummies for ex-ante average labor productivity from 0-25%, 25%-50%, 50%-75%, and 75%-
100% level in the pre-reform years. Standard errors are clustered at the prefecture level.*, **,
and *** denote 10%, 5%, and 1% statistical significance respectively.
123

Table 16: Differential Policy Impact using Alternative Specifications: with Weights and Alternative
Distortion Classification
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
A: Within Prefecture Weighted
Policy 0.012 0.059** 0.062** -0.011
(0.031) (0.029) (0.029) (0.034)
Policy x I(HP) 0.211*** -0.122*** 0.105*** 0.142***
(0.029) (0.026) (0.035) (0.034)
Observations 632,377 630,300 632,373 630,811
R-squared 0.189 0.367 0.173 0.067
B: Within Industry x Prefecture
Policy -0.023 0.124*** 0.099*** 0.000
(0.031) (0.031) (0.025) (0.038)
Policy x I(HP) 0.207*** -0.231*** -0.006 0.081***
(0.015) (0.014) (0.015) (0.019)
Observations 638,458 636,376 638,454 636,880
R-squared 0.106 0.245 0.175 0.031
C: Within Prefecture Weighted
Policy 0.012 0.059** 0.062** -0.011
(0.031) (0.029) (0.029) (0.034)
Policy x I(HP) 0.211*** -0.122*** 0.105*** 0.142***
(0.029) (0.026) (0.035) (0.034)
Observations 632,377 630,300 632,373 630,811
R-squared 0.189 0.367 0.173 0.067
D: All Sample with
Constructed I(HP)
Policy -0.096*** 0.167*** 0.065*** -0.049
(0.031) (0.028) (0.025) (0.033)
Policy x I(HP) 0.233*** -0.214*** (0.0158) (0.0218)
(0.0190) (0.0178)
Observations 1,213,418 1,211,562 1,213,391 1,208,666
R-squared 0.103 0.223 0.205 0.021
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry&Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if a city has been treated.
I(HP) is an indicator equal to one if a firm has above median baseline labor productivity within a
prefecture. I(HP) indicator for the whole sample in Panel D is based on predicted labor productivity
after taking out year-fixed effects. Regressions A and C are weighted by ex-ante employment. Standard
errors are clustered at the prefecture level.*, **, and *** denote 10%, 5%, and 1% statistical significance.
124

Table 17: Policy Impact after Controlling for Concurrent Events
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
A: China’s Entry to WTO
Policy -0.060** 0.131*** 0.065** -0.038
(0.028) (0.028) (0.026) (0.031)
Policy x I(HP) 0.261*** -0.239*** 0.041** 0.118***
(0.014) (0.013) (0.017) (0.017)
Tariff Shock -0.023*** 0.011** -0.011* 0.012**
(0.007) (0.005) (0.006) (0.005)
Observations 564,627 562,672 564,623 563,239
R-squared 0.905 0.846 0.892 0.866
B: Abolition of Agricultural Taxes
Policy -0.060** 0.137*** 0.074*** -0.020
(0.028) (0.027) (0.025) (0.036)
Policy x I(HP) 0.260*** -0.239*** 0.041** 0.118***
(0.014) (0.013) (0.017) (0.017)
Agricultural Reform -0.025 0.029** 0.019 0.046**
(0.017) (0.011) (0.018) (0.022)
Observations 607,195 605,186 607,191 605,678
R-squared 0.905 0.848 0.893 0.866
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. Controls include city GDP, population, firm wage rate and
concurrent events. Standard errors are clustered at the prefecture level.*, **, and *** denote 10%, 5%,
and 1% statistical significance respectively.
125

Table 18: Policy Impact after Controlling Additional Fixed Effects
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
A: Industry by Year FE
Policy -0.064** 0.148*** 0.079*** -0.016
(0.026) (0.025) (0.025) (0.035)
Policy x I(HP) 0.266*** -0.259*** 0.028* 0.111***
(0.013) (0.013) (0.016) (0.017)
Observations 607,195 605,186 607,191 605,678
R-squared 0.906 0.850 0.894 0.867
B: Province by Year FE
Policy -0.105*** 0.107*** -0.001 -0.045**
(0.019) (0.027) (0.028) (0.023)
Policy x I(HP) 0.255*** -0.233*** 0.041** 0.118***
(0.014) (0.012) (0.017) (0.017)
Observations 607,194 605,185 607,190 605,677
R-squared 0.908 0.851 0.895 0.868
C: Control for Reform Year
Policy -0.075** 0.176*** 0.089*** -0.002
(0.030) (0.033) (0.029) (0.040)
Policy x I(HP) 0.275*** -0.260*** 0.039** 0.098***
(0.015) (0.017) (0.018) (0.023)
Reform Year -0.048** -0.008 -0.046** -0.045*
(0.019) (0.020) (0.018) (0.024)
Observations 607,195 605,186 607,191 605,678
R-squared 0.905 0.848 0.893 0.866
D: Control for Year before Reform
Policy -0.081*** 0.181*** 0.094*** -0.018
(0.029) (0.029) (0.027) (0.038)
Policy x I(HP) 0.278*** -0.260*** 0.042** 0.100***
(0.015) (0.018) (0.018) (0.023)
Year Before 0.042** 0.022 0.072*** 0.011
(0.019) (0.020) (0.017) (0.021)
Observations 607,195 605,186 607,191 605,678
R-squared 0.905 0.848 0.893 0.866
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. Controls include city GDP, population, and firm wage rate.
Standard errors are clustered at the prefecture level.*, **, and *** denote 10%, 5%, and 1% statistical
significance respectively.
126

Table 19: Policy Impact on Prefecture-level Aggregate Outcomes
(1) (2) (3) (4) (5)
VARIABLES Employment Capital Revenue Var(Lab.Prod.) No. firms
Policy 0.203*** 0.163*** 0.148*** -0.150** 0.280***
(0.027) (0.033) (0.033) (0.068) (0.027)
Observations 2,997 2,997 2,997 2,985 2,997
R-squared 0.964 0.952 0.966 0.804 0.961
City FE ✓ ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs and aggregated to the prefecture level. Labor misallocation is approximated by prefecture-level variance in marginal products of labor measured
as y/l. We include all observations in prefecture-level analysis.
127

Table 20: Policy Impact by Alternative Baseline Distortion Cutoffs: Within Industry by Prefecture
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy -0.023 0.124*** 0.099*** 0.000
(0.031) (0.031) (0.025) (0.038)
Policy x I(HP) 0.207*** -0.231*** -0.006 0.081***
(0.015) (0.014) (0.015) (0.019)
Observations 638,458 636,376 638,454 636,880
R-squared 0.106 0.245 0.175 0.031
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. I(HP) is an indicator equal to one if a
firm has ex-ante above median labor productivity within the prefecture by 2-digit
industry. Controls include city GDP, population, and firm wage rate. Standard
errors are clustered at the prefecture level.*, **, and *** denote 10%, 5%, and
1% statistical significance respectively.
128

Table 21: Policy Impact after Trimming 5% data
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy -0.098*** 0.189*** 0.080*** -0.021
(0.021) (0.024) (0.021) (0.029)
Policy x I(HP) 0.246*** -0.305*** -0.017 0.095***
(0.013) (0.010) (0.017) (0.020)
Observations 544,059 540,574 544,375 540,915
R-squared 0.874 0.814 0.845 0.839
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. Data are trimmed at the top and
bottom 5% level. Controls include city GDP, population, and firm wage rate.
Standard errors are clustered at the prefecture level.*, **, and *** denote 10%,
5%, and 1% statistical significance respectively.
129

Results: Additional Findings
Table 22: Policy Impact by State-Owned Status: Two Groups
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
A: SOE
Policy 0.029 0.051 0.076 -0.028
(0.026) (0.041) (0.051) (0.034)
Policy x I(HP) 0.198*** -0.119** 0.078 0.143***
(0.030) (0.048) (0.060) (0.050)
Observations 111,805 110,293 111,802 111,309
R-squared 0.946 0.824 0.910 0.945
B: Non SOE
Policy -0.081*** 0.150*** 0.068*** -0.018
(0.027) (0.032) (0.026) (0.040)
Policy x I(HP) 0.228*** -0.247*** 0.005 0.099***
(0.014) (0.013) (0.017) (0.017)
Observations 491,065 490,586 491,064 490,059
R-squared 0.896 0.824 0.876 0.842
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry by Year ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if
a city has been treated. I(HP) is an indicator equal to one if a firm belongs to
high productive group. Controls include city GDP, population and firm wage rate.
Standard errors are clustered at the prefecture by year level.*, **, and *** denote
10%, 5%, and 1% statistical significance respectively.
130

Table 23: Policy Impact by Labor Intensity
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
High Labor Share
Policy 0.007 0.077** 0.090*** 0.028
(0.030) (0.030) (0.024) (0.035)
Policy×I(HP) 0.310*** -0.292*** 0.021 0.087***
(0.029) (0.030) (0.019) (0.020)
Low Labor Share
Policy -0.156*** 0.283*** 0.140*** -0.014
(0.036) (0.037) (0.035) (0.044)
Policy×I(HP) 0.250*** -0.312*** -0.064** 0.055**
(0.025) (0.021) (0.029) (0.028)
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. Labor intensity is measured by the share of
compensation over production. ”High Labor Share” means that the share of compensation over production is above median in the baseline within prefecture. Regressions
control for city characteristics such as GDP and population, as well as firm, year,
industry, and ownership fixed effects. Standard errors are clustered at the prefecture
level.*, **, and *** denote 10%, 5%, and 1% statistical significance respectively.
131

Table 24: Policy Impact by Firm Size: Two Groups
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
A: Small Firms
Policy -0.096*** 0.207*** 0.098*** -0.024
(0.023) (0.036) (0.026) (0.043)
Policy x I(HP) 0.161*** -0.246*** -0.047*** 0.090***
(0.020) (0.015) (0.018) (0.033)
Observations 206,910 206,801 206,908 206,507
R-squared 0.175 0.235 0.319 0.017
B: Big Firms
Policy -0.030 0.101*** 0.069*** 0.004
(0.029) (0.029) (0.024) (0.040)
Policy x I(HP) 0.233*** -0.236*** 0.019 0.091***
(0.020) (0.016) (0.019) (0.021)
Observations 285,354 285,199 285,353 285,004
R-squared 0.104 0.298 0.226 0.048
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator equal to one if
a city has been treated. I(HP) is an indicator equal to one if a firm belongs to
more productive firms in the pre-reform years. Firms are split into two ex-ante
size groups by pre-reform employment. Controls include city GDP, population and
firm wage rate. Standard errors are clustered at the prefecture level.*, **, and ***
denote 10%, 5%, and 1% statistical significance respectively.
132

Table 25: Extension to capital misallocation
(1) (2) (3) (4)
VARIABLES Employment Avg.Lab.Prod Revenue Capital
Policy = 1 0.033 0.070** 0.102*** -0.163***
(0.033) (0.028) (0.024) (0.036)
policy x I(High Y/K) 0.106*** -0.136*** -0.012 0.428***
(0.021) (0.016) (0.025) (0.021)
Observations 607,195 605,186 607,191 605,678
R-squared 0.904 0.847 0.893 0.867
Fixed Effects
Year ✓ ✓ ✓ ✓
Firm ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Controls ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. P olicy is an indicator variable equal to one
if the prefecture has been treated in or before year t and zero otherwise. I(High Y/K) is
an indicator equal to one if baseline Revenue/Capital is above median within prefecture.
Controls include city GDP, population and firm wage rate. Standard errors are clustered
at the city level, *, **, and *** denote 10, 5, and 1% statistical significance respectively.
133

Aggregate Evidence and Policy Implication
Table 26: Pre-reform wedges of labor factor (2000)
(1) (2) (3) (4)
Wedges in Treated Prefectures in Untreated Prefectures
Industry Name labor capital labor capital
Processing of foods 1.98 4.77 1.84 4.39
Manufacture of foods 1.79 3.61 1.65 3.35
Beverages 2.20 3.80 1.91 3.83
Tobacco 2.78 3.51 1.93 3.88
Textiles 0.89 3.79 0.83 3.31
Apparel 0.40 6.56 0.49 6.14
Leather 0.43 6.51 0.70 6.35
Timber 1.14 6.25 1.22 5.47
Furniture 1.09 5.66 1.25 5.41
Paper 1.37 4.24 1.42 4.29
Printing 1.09 1.66 0.95 1.78
Articles for Cultures and Sports 0.46 6.29 0.46 6.21
Petroleum 1.92 2.91 1.82 2.75
Raw chemicals 1.51 5.03 1.43 4.21
Medicines 1.74 4.65 1.63 4.12
Chemical fibers 1.31 3.04 1.29 2.80
Rubber 0.84 4.21 0.69 4.77
Plastics 1.03 4.23 1.01 4.26
Non-metallic minerals 0.92 4.77 0.82 4.44
Smelting of ferrous metals 1.51 5.20 1.62 5.50
Smelting of non-ferrous metals 1.58 6.35 1.52 5.69
Metal 0.87 4.81 0.95 4.94
General machinery 0.77 6.38 0.78 5.51
Special machinery 0.92 4.83 0.89 4.31
Transport equipment 1.02 4.99 0.85 4.42
Electrical machinery -0.29 -0.33 -0.25 -0.09
Communication equipment 1.00 6.68 1.08 6.17
Measuring instruments 1.04 5.06 1.02 4.89
Manufacture of artwork 0.73 6.80 0.54 5.93
Unweighted Average 1.08 5.01 1.06 4.63
No. of Industries 29
Notes: . The Chinese industries and associated codes are classified as follows: processing of foods (13),
manufacture of foods (14), beverages (15), textiles (17), apparel (18), leather (19), timber (20), furniture
(21), paper (22), printing (23), articles for culture and sports (24), petroleum (25), raw chemicals (26),
medicines (27), chemical fibers (28), rubber (29), plastics (30), non-metallic minerals (31), smelting of ferrous
metals (32), smelting of non-ferrous metals (33), metal (34), general machinery (35), special machinery (36),
transport equipment (37), electrical machinery (39), communication equipment (40), measuring instruments
(41) and manufacture of artwork (42).
134

Table 27: Predicted Change in Capital Wedges due to the Hukou Reform
(1) (2) (3) (4)
Industry Name Change in Labor in Labor Wedge Change in Capital in Capital Wedge
Processing of foods 0.170 -0.068 0.071 0.028
Manufacture of foods 0.142 -0.043 0.067 0.029
Beverages 0.155 -0.053 0.053 0.046
Tobacco 0.172 -0.063 -0.002 0.109
Textiles 0.085 0.013 -0.006 0.102
Apparel 0.077 0.020 0.006 0.087
Leather 0.088 0.007 0.040 0.051
Timber 0.136 -0.034 0.032 0.066
Furniture 0.108 -0.008 0.008 0.090
Paper 0.128 -0.025 -0.001 0.102
Printing 0.115 -0.017 0.045 0.049
Articles for Cultures and Sports 0.056 0.038 0.008 0.083
Petroleum 0.173 -0.062 -0.008 0.115
Raw chemicals 0.139 -0.032 -0.026 0.131
Medicines 0.153 -0.048 0.025 0.076
Chemical fibers 0.161 -0.049 -0.040 0.149
Rubber 0.092 0.009 -0.016 0.114
Plastics 0.135 -0.032 0.019 0.081
Non-metallic minerals 0.082 0.017 -0.010 0.106
Smelting of ferrous metals 0.164 -0.054 -0.015 0.122
Smelting of non-ferrous metals 0.172 -0.059 -0.037 0.148
Metal 0.131 -0.030 0.025 0.074
General machinery 0.090 0.009 -0.003 0.100
Special machinery 0.091 0.006 0.019 0.075
Transport equipment 0.127 -0.024 0.009 0.090
Electrical machinery 0.021 0.062 0.065 0.015
Communication equipment 0.142 -0.039 0.020 0.080
Measuring instruments 0.153 -0.052 0.056 0.043
Manufacture of artwork 0.107 -0.008 0.022 0.074
Aggregate 0.122 -0.020 0.011 0.088
No. of Industries 29
Notes: . The Chinese industries and associated codes are classified as follows: processing of foods (13), manufacture of foods (14),
beverages (15), textiles (17), apparel (18), leather (19), timber (20), furniture (21), paper (22), printing (23), articles for culture
and sports (24), petroleum (25), raw chemicals (26), medicines (27), chemical fibers (28), rubber (29), plastics (30), non-metallic
minerals (31), smelting of ferrous metals (32), smelting of non-ferrous metals (33), metal (34), general machinery (35), special
machinery (36), transport equipment (37), electrical machinery (39), communication equipment (40), measuring instruments (41)
and manufacture of artwork (42).
135

Table 29: First Stage of Instrumenting Employment with City-level Shift-share Migration Shock
(1) (2)
VARIABLES Employment Employment
city migration shock 1.687***
(0.578)
0b.i highyl pref#c.city migration shock 0.132
(0.532)
1.i highyl pref#c.city migration shock 3.271***
(0.545)
GDP per capita 0.010 0.010
(0.014) (0.014)
Concentration HHI 0.014*** 0.013***
(0.002) (0.002)
Shift-share Instrument Mean SD
0.015 0.029
Observations 538,371 538,371
R-squared 0.902 0.903
Fixed Effects
Firm ✓ ✓
Year ✓ ✓
Industry ✓ ✓
Ownership ✓ ✓
Notes: All dependent variables are in logs. Standard errors are clustered at the
prefecture level. City-level shock is constructed by multiplying baseline prefecturelevel industry composition (prefecture-level share of production of each industry)
by national-level industry-specific migration policy treatment shocks (industrylevel share of treated prefectures in each year) in a shift-share style. Regressions
control for firm and year-fixed effects. Controls include prefecture-level GDP per
capita and prefecture-level HHI index. Standard errors are two-way clustered at
the city by year level, *, **, and *** denote 10, 5, and 1% statistical significance
respectively.
136

Table 28: Predicted Aggregate Solow Residual
(1) (2) (3) (4)
Industry Name Industry Output Share ∆ Solow Elasticity βL (OP) βK (OP)
Processing of foods 0.046 0.070 0.630 0.315
Manufacture of foods 0.020 0.082 0.790 0.270
Beverages 0.021 0.094 0.799 0.372
Tobacco 0.003 0.085 0.842 0.294
Textiles 0.105 0.029 0.504 0.197
Apparel 0.045 0.019 0.516 0.195
Leather 0.025 0.026 0.450 0.236
Timber 0.011 0.044 0.523 0.270
Furniture 0.007 0.032 0.583 0.270
Paper 0.030 0.056 0.602 0.260
Printing 0.013 0.061 0.691 0.121
Articles for Cultures and Sports 0.013 0.028 0.532 0.243
Petroleum 0.009 0.061 0.550 0.165
Raw chemicals 0.088 0.053 0.629 0.271
Medicines 0.023 0.077 0.767 0.306
Chemical fibers 0.016 0.060 0.536 0.256
Rubber 0.014 0.029 0.483 0.216
Plastics 0.039 0.038 0.506 0.212
Non-metallic minerals 0.070 0.035 0.562 0.346
Smelting of ferrous metals 0.029 0.055 0.606 0.329
Smelting of non-ferrous metals 0.025 0.047 0.574 0.360
Metal 0.050 0.036 0.521 0.227
General machinery 0.055 0.035 0.604 0.336
Special machinery 0.037 0.045 0.691 0.283
Transport equipment 0.056 0.047 0.646 0.283
Electrical machinery 0.004 0.017 0.538 0.197
Communication equipment 0.070 0.046 0.590 0.334
Measuring instruments 0.062 0.056 0.653 0.200
Manufacture of artwork 0.013 0.040 0.567 0.295
Aggregate - 0.046 0.603 0.264
No. of Industries 29
Notes: The Chinese industries and associated codes are classified as follows: processing of foods (13), manufacture of
foods (14), beverages (15), textiles (17), apparel (18), leather (19), timber (20), furniture (21), paper (22), printing (23),
articles for culture and sports (24), petroleum (25), raw chemicals (26), medicines (27), chemical fibers (28), rubber (29),
plastics (30), non-metallic minerals (31), smelting of ferrous metals (32), smelting of non-ferrous metals (33), metal (34),
general machinery (35), special machinery (36), transport equipment (37), electrical machinery (39), communication
equipment (40), measuring instruments (41) and manufacture of artwork (42).
137

Table 30: Factor Price (Labor-Wage) Elasticity Estimation
(1) (2) (3) (4)
Pre-reform Whole Sample
Wage Rate Wage Rate
VARIABLES OLS IV OLS IV
Log(L) -0.352*** -0.441*** -0.313*** -0.329***
(0.009) (0.040) (0.007) (0.037)
GDP per capita 0.008 0.009 -0.005 -0.000
(0.012) (0.011) (0.013) (0.013)
Concentration HHI 0.005*** 0.006*** 0.005*** 0.006***
(0.002) (0.002) (0.002) (0.002)
Observations 522,730 522,730 1,324,851 1,324,851
R-squared 0.703 0.680 0.683 0.665
Fixed Effects
Firm ✓ ✓ ✓ ✓
Year ✓ ✓ ✓ ✓
Industry ✓ ✓ ✓ ✓
Ownership ✓ ✓ ✓ ✓
Notes: All dependent variables are in logs. Standard errors are clustered at the prefecture level. Standard errors are two-way clustered at the city by year level, *, **,
and *** denote 10, 5, and 1% statistical significance respectively.
138

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The impact of economic shocks on firm behavior: insights from three studies

PDF

The impact of digital transformation on urban communities, welfare, and distributional outcomes

PDF

Essays on urban and real estate economics

PDF

Essays in development economics

PDF

Location choice and the costs of climate change

PDF

The decisions of migration and remittances in rural China

PDF

Migration, location, and economic opportunity

PDF

From first principles to machine intelligence: explaining charge carrier behavior in inorganic materials

PDF

Essays in climate change adaptation: role of market power in incentivizing adaptation behavior

PDF

Essays on wellbeing disparities in the United States and their social determinants

PDF

Essays on treatment effect and policy learning

PDF

Essays on environmental economics: education, employment, and experiments

PDF

Essays in environmental economics

PDF

Essays on the platform design and information structure in the digital economy

PDF

Growth and development in Africa: macroeconomic and microeconomic perspectives

PDF

Essays on narrative economics, Climate macrofinance and migration

PDF

Essays on the economics of climate change adaptation in developing countries

PDF

Three essays on strategic commuters with late arrival penalties and toll lanes

PDF

Essays in the economics of education and conflict

PDF

Essays on development economics and adolescent behavior

Asset Metadata

Creator Fang, Jingyi (author)

Core Title Essays on the dual urban-rural system and economic development in China

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Degree Conferral Date 2024-05

Publication Date 04/17/2024

Defense Date 03/20/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag China,Economic development,migration,OAI-PMH Harvest,urban-rural system

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Morlacco, Monica (committee chair), Bairoliya, Neha (committee member), Chaney, Thomas (committee member), Oliva, Paulina (committee member)

Creator Email fangjy1218pku@gmail.com,jingyif@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113877824

Unique identifier UC113877824

Identifier etd-FangJingyi-12816.pdf (filename)

Legacy Identifier etd-FangJingyi-12816

Document Type Dissertation

Format theses (aat)

Rights Fang, Jingyi

Internet Media Type application/pdf

Type texts

Source 20240417-usctheses-batch-1141 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu