International Trade Liberalization and Domestic Institutional Reform: E ects of. WTO Accession on Chinese Internal Migration Policy.

Transcription

1 International Trade Liberalization and Domestic Institutional Reform: E ects of WTO Accession on Chinese Internal Migration Policy Yuan Tian September 9, 2017 Abstract I study how the liberalization of international trade a ected domestic labor market regulations in China after the WTO accession. China has a rigid domestic labor market where mobility of labor across regions and across sectors (agricultural v.s. non-agricultural) are highly restricted by the Hukou registration system. China entered the WTO in 2001 and experienced substantial export growth afterward, and at the same time, local governments started to relax migration restrictions. I first build a simple political economy model to rationalize local government regulation decision. Then I test the empirical link between the trade shock and migration regulation change on 247 Chinese cities from the period of 2001 to Using local trade shocks generated from the interaction of industry level tari reduction and region level industrial composition, I study how the local labor demand shock drove the migration regulation changes. Using a unique newly-collected dataset on Chinese local migrant regulations, I construct a migration friendly index to measure local government s attitude towards migrant workers and use it as a proxy for the migrant amenity level in the theoretical model. 0.1 percentage point larger trade shock had 7 percent higher migration regulation index increase, and 1 percent higher migration regulation index increase increased the number of migrants by 0.07 percent. I also evaluate the e ects of regulation changes on migrant welfare and the overall e ect of trade liberalization on economic outcomes. This paper contributes to the literature on the linkage between trade and institutional and illustrates how goods mobility a ects factor mobility through the complementarity between trade policy and labor policy. Keywords: Trade liberalization, institution, migration policy. F16,O43,F63 Do mobility of goods and openness to trade a ect the mobility of workers? How do cities benefit from international trade liberalization when migration policies adjust at the same time? People s ability to move across geographic units and sectors is an important determinant of the Department of Economics, UCLA, 405 Hilgard Ave., Los Angeles, CA ( ytian14@ucla.edu). I thank Adriana Lleras-Muney, Pablo Fajgelbaum, Walker Hanlon, Moshe Buchinsky, and Youssef Benzarti for support and guidance. I thank participants in UCLA applied micro proseminar for comments. All errors are mine. 1

2 overall gains from trade and distribution of the gains. However, it has been documented that labor mobility is low both in developed countries like the U.S. and developing countries like Brazil and India (Kovak [2013], Topalova [2010], Autor et al. [2013]). China has a particular institution that restricts labor mobility: the Hukou registration system ties individuals to specific regions of origin, and urban public goods provision discriminates citizens based on worker s Hukou identity. However, the situation started to change after After China entered the WTO in 2001, with the exponential growth in export values, migration across regions and sectors has also been growing substantially. Local government started to relax migration restrictions, by including migrant workers in the social safety net, increasing access to local schools for migrant kids, and allowing some migrant workers to obtain local Hukou. Was trade liberalization a major driver of the migration regulation change? How does this regulation change a ect the relationship between trade liberalization and economic growth? In this paper, I study the e ect of international trade liberalization on city level migration regulations and migration flows and also evaluate the direct and indirect impact of trade on wage, employment, and economic growth. Tari and quota restrictions on Chinese goods fell after the WTO accession. Local government saw the opportunity to boost the local economy, and the incentive to attract and retain the migrant labor force became stronger. Di erent cities experienced di erential trade shock due to initial industrial composition, and I will investigate empirically whether cities with bigger trade shocks chose to liberalize the local labor market more, and the consequences of the labor market liberalization. There are several challenges to the study. The first challenge is the measurement of migration restriction changes. City level governments in China have substantial discretion in policies related to local economic growth, and migration policy is one important policy instrument. I built a novel data set on city level government regulations related to migrant workers and constructed a migration friendly index to measure local government s attitude towards migrant workers. Overall there are 175 cities that made di erent degrees of positive changes in migration restrictions from , and topics varied from work place training, wage arrears prevention, medical and social insurance and school access. The second challenge is sorting out the trade e ect. If some cities have certain advantages in both production and exporting in particular, and migration restrictions fall when there is economic growth, then it is hard to tell whether the regulation change is due to export growth or not. To solve this problem, I calculate a city s trade exposure or trade shock using the interaction of industry level tari reduction and city level industry employment shares. The tari reduction comes from importing countries of Chinese goods, and should not be correlated with city level economic condition. Overall, I find that regions that faced 0.1 percentage point larger trade shock had 7 percent higher migration regulation index increase, and 1 percent higher migration regulation index increase increased the number of migrants by 0.07 percent. Overall, at the mean value of the trade shock, the e ect of trade shock on migration regulation change was about 17% of the overall regulation change. Moreover, I find that cities with higher per capita income and with 2

3 greater migrant share of total employment responded more positively to the trade shock, while cities with more state-owned enterprises responded less. After discussing the e ect of trade liberalization on migration regulation and migrant flows, I also evaluate the overall impact of liberalization on employment, wage and GDP growth. Here the trade shock has two channels to a ect employment and wages: the direct channel where the tari reduction caused goods prices to go up, pushing up the labor demand, and the indirect channel where migration regulations are changed, and labor supply increases. I find that cities with 0.1 percentage point larger trade shock had 1.7 percent larger wage increase, 3.8 percent greater increase in total employment, and 5.4 percent larger GDP increase. For the indirect e ect, 1 percent higher migration regulation index led to 0.03 percent higher wage increase, 0.05 percent higher increase in total employment, and 0.08 percent larger GDP. Overall, at the mean value of the trade shock, the e ect of the trade shock explains 2% of the overall wage growth through the direct e ect, and 0.6% through the indirect e ect. For total employment growth, the numbers are 12% and 3%, respectively. For per capita GDP growth, the e ects are 6% and 1.5%, respectively. To sum up, the direct e ect is 3-4 times the indirect e ect. The paper contributes to several strands of the literature. The first set of literature is on the relationship between the institution and economic growth. Some studies how institutions a ect economic growth, such as Acemoglu et al. [2001]on the colonial institution and contemporary income per capita. Some studies are on how trade a ects institutions, like Khandelwal et al. [2013] on export quota elimination and firm productivity, Autor et al. [2016] on Chinese imports and U.S. elections. Acemoglu et al. [2005] studies how Atlantic trade intensity a ected the economic institution of property right protection, and how this institution benefited later economic growth. My paper focuses on labor mobility restriction which is an important institution for economic activities, and I investigate both how international trade shape the institution and how the institution a ects gains from trade. The second strand of literature is on trade and labor market. Autor et al. [2013], Kovak [2013] and Topalova [2010] study how import competition a ected wage, employment and poverty rate, in the case of U.S., Brazil and India, respectively. In most of the papers here, labor mobility is assumed to be fixed, while in my paper, labor mobility is also a ected by trade due to the change of migration restriction. The paper will be organized as follows. I will start with a description of background, where I talk about how Hukou system restricts people s mobility, and what is in local government s power to decide the stringency of the migration restrictions. I will also discuss how the trade liberalization in China before and after the WTO accession, and anecdotal evidence on the people seeing WTO accession as the opportunity to further economic reform. In Section 2, I will present a simple framework to show government decision in response to trade shock. Section 3 presents the detail of data collection, on measuring the regulation change and regional trade shock. Section 4 presents empirical results. The last section concludes. 3

4 1 Background 1.1 Hukou system and labor immobility The Hukou system was first established in 1953 as a method to count the total population. Later, when urban and rural division became bigger, the Hukou became a way to di erentiate citizens regarding economic and employment opportunities. Every Chinese citizen has a Hukou, which is a registration status fixed on a location, and also specified as agricultural vs. nonagricultural. The location assignment is based on parents status, not the place of birth. If people wish to live and work in cities other than the registration place, they need to get temporary registration and are still subject to di erent kinds of discrimination. Many of them are working as temporary employees without contracts and paid lower wages; there is no medical insurance and injury insurance; children of migrants are not able to attend a local school, or need to pay substantially higher fees for school. Employers and rental house owners are responsible for enforcing the temporary registration of migrant workers since if their employee or tenants do not have registration, they would be held responsible by the police department and face fines. The system creates artificial division across regions and rural and urban areas. Except for college graduates who find stable jobs in other cities after graduation, it is extremely di cult to change Hukou status. Before 2000, there are very few quotas every year for people to change from agricultural to non-agricultural Hukou, and most quotas are assigned to the people whose married spouse are with non-agricultural Hukou. The long-divided system created significant income gaps between regions. As a result, despite the restrictions, a lot of people migrant to other cities to work and live, mostly temporarily. According to the population census, in 2000 there are 144 million people whose residence city and Hukou city were not the same, accounting for 11% of the national population. Until 2010 the number almost doubled to 261 million. Together with the increase in migrant flows, the migration restriction seems to become laxer over time. The first observation is that from 1995 to 2007, migrant regulations on the city level became more pro-migration, and especially so after I collected all the city level migrant related regulations from 1995 to For each document, I assign a score in { 2, 1, 0, 1, 2} to measure how friendly the regulation is to migrants. 2 means that the regulation is very against migrants (for example, evicting all migrant workers who are living in rental houses and are without proper documents), and 2 means that the regulation favors migrants (for example, setting up guidelines to ensure that migrant workers should sign contracts when employed, and the firm should pay social insurance for the migrant workers). In Figure 1, I plot the city-level average number of new regulations on migrant related issues and the regulation score. Each dot on the dashed line is calculated by summing over the number of regulations in each year over 247 cities, and then divide the total number by 247. Each dot on the solid line is calculated using regulation scores instead of the number of regulations. 1 See details of the data collection and cleaning process in the data section. 4

5 Figure 1: Number of regulations and regulation score, city level total, *Note: each dot is a year. Sum over all city-level regulations related to migrants. From Figure 1, the trend shows that before 2001, there is about 0.1 regulation per city per year that is related to migrant issues, but concerning score or migrant friendliness, they are essentially 0. After 2001, both the number of new regulations and the score of the regulation per city per year increased substantially. In 2006 for example, there is about one regulation per city, and the average score is about 0.8, meaning that in later years there are more regulations and the regulations are more favorable to migrants. The break in trend observed in Figure 1 is related to the central government s overall attitude towards migration across regions and urbanization in general. The central government started Hukou reform around 2000, when in the 10th Five-year Plan, they specifically talked about reducing political barriers to migration. For example, it was written in the plan: We will adapt to the market-oriented employment mechanism...to have an orderly and reasonable allocation of rural and urban labor. Also, local discretion was allowed, meaning that local governments were allowed to design their own reforms following the central government guideline. In a document by State Council of China in 2001, it wrote: Local governments should take into consideration local economic and social development levels, and conduct reforms that balanced the population growth, infrastructure, employment and social security, and other welfare programs. Interestingly, the timing of the reform was correlated with China s WTO accession in I will briefly talk about the trade liberalization around the WTO accession, and show the connection between trade shocks and migration regulation change. 5

6 1.2 Trade liberalization China entered the WTO in November 2001, as the 143rd member country. In the accession agreement, both China and partner countries had commitments regarding import tari reduction, quota removal and reduction of other non-tari barriers. Specifically, China started to enjoy the Most-Favored-Nation status with other member countries, meaning that whatever import tari partner countries imposed on a certain product, the Chinese good would be eligible to the same tari level. From Figure 2, we can see that the tari on Chinese goods started as about 4.1 percentage points in 2001, and declined to about 3 percentage points in The tari decline happened before the WTO accession, but the industries that experienced bigger tari reductions before and after WTO were not the same, creating relative winners and losers. Another thing to notice from Figure 2 is that the increase in export value was faster in the period than in the period. This could be because the WTO accession brought both reduction in tari (which is plotted here) and the reduction in non-tari barriers (which is not plotted here but is highly correlated with the tari reduction). Figure 2: Time trend of average tari and overall export volume, *Note: each dot on the red curve is weighted average of industrial level tari s, where the weights are shares of exports in this industry. For the industry level tari, itisconstructedastheweightedaverageofdestinationcountrytari s on Chinese exports in the specific industry, where the weights are shares of exports in this destination country in the specific industry in In this paper, I will focus on the post-wto period of , and the comparison will be between the cities that had bigger versus smaller tari shocks. In Figure 3, I plot the change in 6

7 log export (in gray) and percentage change in tari (in yellow) in the period. Each cluster is an industry in 2-digit SIC code, and the clusters are sorted by the value of exports in the industry in We can see that changes in tari levels were very di erent across industries. 2 This implies that cities that specialized in di erent industries would experience di erent levels of tari shocks. Figure 3: Distribution of tari changes and export growth across industries, *Note: each bar is a industry. Horizontally sorted by value of exports in the industry in These di erential shocks could induce di erent cities to respond di erently. In the following section, I will connect the trade shock with the regulation change. 1.3 Trade shocks and migration regulation change In a commentary by Ministry of Agriculture in 2003, it pointed out: After WTO entry, protection of domestic agricultural sector declined, and lots of excess labor in rural areas moved to cities. And in a commentary by Minister of Public Security, Division of Hukou Management in 2001, WTO accession was directly described as a chance to reform the Hukou system:... employment system, education system, and social security system are all evolving, and it is about the time to partially liberalize internal migration. Entering the WTO is an opportunity to change the Hukou system from management to service. With the central consent to experiment and reform, local governments responded to the trade liberalization with di erential degrees of relaxation of the Hukou system. In Figure 4, I plot the trends of regulation scores, dividing cities into three groups: dot line for cities with big trade shocks, dashed line for cities with medium shocks, and the dotted line for cities with 2 Industry code and industry name could be found in Table 12. 7

8 small shocks. Here trade shock is calculated as tari reductions from 2001 to 2007 (post-wto period), with cities with bigger shocks experiencing the larger reduction in tari s. From Figure 4, it is clear that although the trends for the three groups of cities were very similar before 2001, places with bigger trade shocks after 2001 chose to relax the migration restrictions more. Figure 4: Regulation score, city level average, , by the size of trade shock in *Note: each dot is a year-shock group. Sum over all city-level regulations related to migrants. In Figure 5, I further divide cities into four groups, depending on the trade shock size in the post-wto period, and one other city characteristic. In the top left graph, the four groups are: big trade shock and rich cities (solid line with solid squares), big trade shock and poor cities (solid line with hollow squares), small trade shock and rich cities (dashed line with solid dots), and small trade shock and poor cities (dashed line with hollow dots). Here rich and poor are defined using the 2001 per capita GDP medium as the cuto. The first graph shows that high-income cities with big trade shocks experienced the greatest relaxation of regulations, and the low-income cities with small trade shocks experienced the smallest change, indicating a positive interaction e ect of trade shock size and city income level on migration regulation change. The top right graph uses 2001 city wage level as the cuto, and implications are similar. The lower two graphs are related to industry migrant intensity. I calculate the migrant share of total employment by industry using 2005 mini-census data and calculate the city level migrant intensity using the interaction of city level industry employment shares and industry level migrant employment share. Cities with higher migrant intensity are the ones with industries 8

9 that employ relatively more migrants, and the cities with lower migrant intensity are the ones with industries that employ relatively more locals. The bottom left graph shows that cities with big trade shocks and higher migrant share of employment (solid line with solid squares) were the ones that changed regulations the most, and cities with small trade shocks and higher local share of employment (dashed line with hollow dots) changed the regulation the least. The bottom right graph conveys the similar message. I use the 2000 firm data to calculate the share of employment in state-owned enterprises (SOEs), and cities with higher SOE shares are the ones with relatively more employment in SOEs relative to private firms. SOEs employ relatively fewer migrants than private businesses, and the bottom right graph shows similar patterns as the bottom left one. Figure 5: Regulation score, city level average, , by the size of trade shock in , and by subgroups *Note: each dot is a year-shock group. Sum over all city-level regulations related to migrants. Given the observation in the data, I will use a simple model to illustrate the decision-making process of the Chinese government and show how the trade shocks a ected migration regulation changes. 9

10 2 Theoretical framework 2.1 Model Overview I model an open economy with N+1 regions with n 2 {1, 2,...,N,r}. There are N cities, indexed by n i 2 {1,...,N}, and one rural area with n = r. Each city is endowed with immobile local labor L i and an immobile fixed factor R i. A mass of M rural residents can either live in the rural area, or move to one of the two cities and work as migrant workers. The production technology in each city use the migrant labor, local labor, and the fixed factor as inputs, and produces a unique product that is sold on the international market. The technology of production is di erent across cities. Each city has an endowed natural amenity C i. Rural residents who live in City i will enjoy the wage, natural amenity C i, and additional A M,i provided by the city government. If they choose to stay in the rural area, they will enjoy a fixed amenity level of C r and a fixed wage of w r. Given the technology and factor endowment, the city government chooses the amount of amenity it provides for the migrant labor, in each state of the world set by the central government. 3 The central government chooses one of the two states where S 2 {0, 1}. S =0isthe immobile state where the local government has to choose zero amenity for the migrant labor. S = 1 is the mobile state where local governments are free to choose the amenity level. 2.2 Preferences and Worker s Decision Problem The indirect utility of rural worker l living in region n is vn l = v n l,where l n represents worker l s idiosyncratic taste for living in region n, and v i is common for all rural workers who live in region n. v n is determined by the amenity level and wage level in the following way: v n =(C n + A n ) 1 w 2 n For rural worker l who remains in the rural area, A r = 0. Worker l will choose to live in the region n where n = argmax n 0v n 0 l n 0. There is an idiosyncratic taste draw l n for each worker and area, and the draw is i.i.d. across workers and areas from a Frechet distribution where Pr( l n <x)=e x,with > 1. Thus, the number of migrant workers who live in area n will be vn M n = v M where v ( P n v n) 1/. With Frechet distribution, the average utility of migrant workers in all states will be the same and proportional to v, thus,v will be used as the measure of worker utility. 3 The amenity is equilvalent to a subsidy. 10

11 Plugging in the expression for v n, the labor supply equation for migrant labor is M n = 2.3 Production Technology and Firm s Problem (Cn + A n ) 1 wn 2 M (1) v Firms in each city i use the local labor, the migrant labor and a fixed factor to produce a unique product. Both local labor and local fixed factor are supplied inelastically: L i = L i, R i = R i. The price of the product p i is determined on the international market. The production function is Cobb-Douglas with constant return to scale, and output Y i can be written as: Y i = p i µ 0 il 1,i i M 2,i 1 1,i 2,i i Ri This is equivalent to writing the production function as Y i p i µ i M i i The output is subject to a sales tax t. The firm maximizes the profit by choosing max M i (1 t)y i w i M i where w i is the wage of migrant workers in City i. First order condition of the firm gives: 2.4 City Government Problem w i = i (1 t)p i µ i M i 1 i (2) Local government objective is to maximize net fiscal profit, which is the tax revenue minus the expenditure on public services. In the literature on political economy in China, Li and Zhou (2005), Jia (2015) and Jia (2012) argue that local governments are driven by promotion incentive, and as a result, they seek to maximize GDP growth. I adopt the fiscal incentive approach as in Gordon and Li (2012). The rationale for the objective function is that only few local government o cials are promoted, and the rest of them stay in the system because they can get the rent or net fiscal profit. The advantage of this approach here is that I can model the direct benefit of migrant inflow and the direct cost explicitly in the objective function: benefits in terms of increasing the tax income, and costs in terms of amenity provision. These tradeo s are present not only in China, but also in cases of low-skilled immigration in the U.S. and Europe. 4 4 Alternative, suppose that there is no tax, but the city government prefers bigger economy size and smaller expenditure on migrants. 11

12 In addition, I assume that each city government does not take into account the action taken by the other city government. This means that each city government takes v as given. Mobile State In the mobile state where S = 1, the city government maximizes the net fiscal profit by choosing the amenity level A M,i for migrant workers and the number of migrants M i : max ty i A i M i M i,a i subject to labor supply and labor demand constraints. The city government has full information about the production and labor supply. From the labor supply equation (Equation 1), A i can be solved as a function of M i. Thus the first order condition of the local government is Immobile State i = A i i M i (3) In the immobile state where S = 0, the city government maximizes the same maximization problem subject to the constraint that A i = 0. Thus, M i is determined by the labor supply equation and labor demand equation and the city government will take it as given. 2.5 Central Government Problem The central government is interested in increasing the total output (or GDP), which is Y = P i=i,ii Y i + M r w r. at the same time, a large inflow of population into the city might cause regime instability, incur administrative and bureaucratic costs, and all these will act as a cost to the central government. The central government has full information about the economy and the decision process of the city governments and chooses to S 2 {0, 1}. Assume that the central government starts with the immobile state and the decision rule is as follows: S = 1(Y S=1 Y S=0 Y ) (4) where 1( ) is an indicator function. This means that if the cost of immobility (Y S=1 Y S=0 ) is big enough, the central government will switch to the mobile state. However, once in the mobile state, the cost of switching back to the immobile state is very high (suppose impossible for now). 12

13 2.6 General Equilibrium A general equilibrium of this economy consists of the distribution of workers {M n } n2{1,2,...,n,r}, city output values {Y i } i2{1,...,n}, wages {w i } i2{1,...,n}, amenities {A i } i2{1,...,n}, the migrant welfare measure v, the state of mobility S, and economy wide GDP Y such that i) firms make optimal decision about production; ii) rural workers make optimal location decision; iii) city governments make optimal decision of amenity provision; iv) the central government makes the optimal decision about the state of mobility; v) city level labor markets clear; vi) the national labor market clears, i.e. P n M n = M. 2.7 Comparative statics Mobile State S =1: Proposition: When there is a positive price shock in City 1, p 1,then 1. M 1, A 1, Y 1 ; 2. M 1, A 1 ; 3. M r. Immobile State S =0: Proposition: When there is a positive price shock in City 1, p 1,then 1. M 1, Y 1, but less than when S = 1; 2. M 1, but less than when S = 1; 3. M r, but less than when S = Implication for the overall economy When there is an economy-wide positive price shock and natural amenities {C i } N i=1 both Y S=1 and Y s=0 will increase, and Y S=1 will increase more. are small, 2.9 Equations for Estimation Iuseˆx d ln x to present percentage changes. The percentage increase in migrant inflow ( ˆM i )is S 1,i ˆM i = (1 i )( S 1,i )+ 1 ˆp i (5) where S 1,i = it(1 t)piµ i M i 1 i it(1 t)p iµ im i 1 = Bi i +C i B i+c i contribution in taxes, and B n is the baseline per capita migrant 13

14 The percentage increase in amenity level (Âi) is  i = 1 S 1,i 1 S 2,i (1 i )( S 1,i )+ 1 ˆp i (6) where S 2,i = Ai A i+c i.when i is bigger, the amenity is more responsive. The percentage increase in total employment (Êi) is where S 3,i = Ê i = Mi L i+m i. d L i + M i = S 3,i ˆMi = The percentage increase in migrant wages (ŵ i )is ( S 1,i )S 3,i (1 i )( S 1,i )+ 1 ˆp i (7) ŵ i =ˆp i +( i 1) ˆM i = (1 i )( S 1,i )+ 1 ˆp i 1 And the percentage increase in local wage (ŵ L i )is ŵi L S 1,i + 1 =ˆp i + i ˆMi = (1 i )( S i )+ 1 ˆp i When pooling the migrants with the local, the percentage change in mean wage for the total employment (ŵi T )is where S 3,i = ŵi T =(1 S 3,i )ŵi L + S 3,i ŵ i = ( S 1,i )S 3,i + 1 (1 i )( S 1,i )+ 1 ˆp i (8) Mi L i+m i. Define per capita GDP as G i = 3 Data and Measurement 3.1 Regulation Change Yi L i+m i. Then the percentage change in per capita GDP is Ĝ i =ŵ L i S 3,i ˆMi = ( S 1,i )(1 S 3,i )+ 1 (1 i )( S 1,i )+ 1 ˆp i (9) Government regulations are collected from the website The website is a feefor-service site where it contains databases including law and regulations (22,148 items), legal news (16,696 items), legal cases (1,955 items) and other law and regulation related information in China. Due to the rich and up-to-date information, subscribers of the website services include university libraries, government institutions, law firms, investment banks and other financial companies, and other big businesses. I will use the database regarding central and local government regulations. The documents are collected from o cial government websites, government gazettes, repositories of laws and 14

15 regulations, and documents provided by relevant government units, and all sources are recognized by the legislation law. The local government database includes governmental regulations, regulatory documents, judicial documents and government rules by all provinces, autonomous regions, municipalities, capital cities of provinces, 19 big prefectures designated by the State Council, and other prefectures. Out of 341 prefectures, there are 332 prefectures that have at least one regulation document included in the database. Among them, municipalities like Shanghai, Beijing and Chongqing have more than 25,000 items, while the median number of items is 861. The total number of items from 1995 to 2001 is **, and from 2001 to 2007 is **. Out of all local government regulations, I use keyword search in document titles for migration related items. The keywords that I use are: floating population, migrant worker, temporary residence, and Hukou. Overall, I find 138 items in and 673 items in Topics of these items spans various aspects of migrant life. 44% of them related to labor issues (wage payment, labor union, training, etc.), 18% about welfare programs (unemployment insurance, injury insurance, medical insurance, pension, etc.), 30% administrative (Hukou registration), 9% on birth control. Some regulations are beneficial for migrant workers, and some of them are against migrant workers. The overall time trend for the regulations is becoming more and more pro-migrant, in the period of Originally, the focus of the regulations was how to manage the floating population, for example, repatriation of migrant workers in rental houses. Starting from 2001, there were more regulations on regulating charges on temporary residence and working permits, providing migrant kids with compulsory schooling, urging firms to pay wages and sign contracts, and incorporating migrant workers into the social welfare system. To evaluate the migrantfriendliness of the documents, I create an index for each item and sum the item index by city to generate city level index or score. 5 Among 247 cities finally used in the analysis 6,themedium score in period is 0, and the maximum is 7; for the period, the medium is 2 and the maximum is 38. Besides municipalities, cities with very high scores include Ningbo and Guangzhou, which had very export-oriented growth. In Figure 6, I plot the regulation score from 1995 to Each dot is a year-city average score of new regulations on migrant issues. The solid blue line includes all new regulations, the red one for the work-related, the green for the welfare-related, and the other two for Hukouand-birth-control-related regulations. The figure shows that the increase in overall score of regulation is mainly driven by work-and-welfare-related regulations. In 2007, for example, the average score for all regulations is about 1, where 0.62 is from work-related regulations, and 0.25 is from welfare-related ones. 5 Details of the index construction can be found in the data appendix. 6 Because these cities have other useful measures in the City Yearbook. 15

16 Figure 6: Regulation score, city level average and by topic *Note: each dot is a year. Sum over all city-level regulations related to migrants. From Table 1, we can see that in the period, there are in total 673 new regulations on migrant issues, with a mean score of 1.08, and there were 175 cities with at least one new regulation. In the period, the numbers are much smaller: 138 new regulations in total, the mean score being 0.05, and 50 cities involved. Table 1: Descriptive statistics on number of regulations and number cities with certain regulations # of regulations Mean score # of cities # of cities with (+) (0) (-) Total Hukou related Birth control Work Welfare Total Hukou related Birth control Work Welfare Figure 7 shows the geographically distribution of the new regulations. Overall, coastal cities 16

17 had more changes, but many inland cities also made subtantial change. Figure 7: Geographic distribution of regulation changes, *Note: each dot is a year. Sum over all city-level regulations related to migrants. There is a question about the completeness of the set. To eliminate the availability bias in the sample, I also collect total number of regulation in each city (to address the concern that in some cities, more regulations are collected in the website), and number of regulation related to fiscal issues before and after the WTO accession (to address the concern that over time, more regulations are collected in the website, and the rate of growth might be di erent across cities). I will use this information in the robustness check. 3.2 Regional Trade Shock I will construct regional trade shocks to evaluate the e ect of trade shock on regulations. There are several ways to measure the trade shock, and I will discuss in detail the main measure that Iuse The WTO shock First, it is important to check the export supply elasticity. In Figure 8, I have percentage point changes in tari on the horizontal axis, and change in log exports on the vertical axis. Each dot is an industry. The red ones are for , and the blue ones are for For both periods, the fitted lines have a slope of 13-14, meaning that one percentage point reduction in tari on exports will induce percent increase in export values. 17

18 Figure 8: Trade elasticity, and *Note: each dot is an industry The red ones are for , and the blue ones are for Figure 8 indicates that tari reductions had a significant impact on exports, but the relationship existed for both the pre-wto period of and the post-wto period of However, my focus is on the post-wto period. If the industries that benefited more from the pre-wto tari reductions were the same as the ones that benefited more from the WTO-induced tari reductions, then I will not be able to separate the WTO e ect. In Figure 9, I plot the post-changes on the pre-changes. When I run a linear regression of percentage point changes in tari in on the change in , the coe cient is and is statistically insignificant. This means that the WTO created winner vs. loser industries concerning tari reduction, thus created winner and loser cities regarding trade shocks since cities were with di erent industrial compositions. 18

19 Figure 9: Correlation of changes of tari in the two periods, against *Note: each dot is a industry City level tari shock The first way is to use tari information directly following Kovak [2013]. The regional shock on wages is tariff it = X j ij ˆP j ij = Lij where ij = ij ij ij P ij 0 j 0 ij0 ij 0 and Li is the fraction of regional labor allocated to industry j, ij as the elasticity of substitution between the fixed production factor and labor, and 1 ij as the cost share of labor in industry j. With Cobb-Douglas production function, ij = 1, and ij and ij can be calculated from the firm level data. ˆPj is the price shock to industry i, and it is measured using export tari, import tari, and import tari on intermediate goods: tariff EXP it = X j ij ln(1 + tariff EXP j,t0 /100) ln(1 + tariffexp j,t /100) 19

20 tariffit IMP = X j ij ln(1 + tariffj,t EXP 0 /100) ln(1 + tariffexp j,t /100) tariff INTER it = X j ij ln(1 + tariff INTER j,t0 /100) ln(1 + tariffinter j,t /100) I construct these measures for two periods: , Industrial compositions are taken from the 2000 Industrial Enterprises Survey. The Survey is conducted on Chinese manufacturing firms with annual sales more than 500 million RMB, and the information includes basic firm information like name and address, financial information on sales, export values, fixed capital, wage payment and total sales cost, and also total employment. The first year with available export information is 2000 and I will use 2000 data to construct city industrial composition. There are in total 145, 546 firms in 2000 with positive sales revenue and wage information, more than 10 employees and valid industry code. The industry code is 4-digit GB code, and I aggregate it into the 2-digit level. The 2-digit GB code is slightly finer than the 2-digit SIC code, and the crosswalk between the codes are included in Table 12. To construct tari s on intermediate inputs, I use the 2002 province input-output table City level market access shock The second way is the market-access approach. The idea is that suppose the overall export (import) volume increases in a certain industry over time on the national level, and I can calculate per capita export growth by dividing the increase in export (import) volume by the total number of people employed in the industry. Then I can distribute the per capita export growth across regions according to the share of employment in the industry in a certain region, and then generate the overall regional trade shock by summing over industries. Specifically, following Autor et al. [2013], the formula to calculate regional export exposure is as follows: IPW M it = X j L ijt L jt M jt L it = X j L ijt L it M jt L jt IPW X it = X j L ijt L jt X jt L it = X j L ijt L it X jt L jt where L it is the start of period employment (year t) in region i and L jt is the start period employment in industry j, L ijt is the start of period employment in region i and industry j. M jt is the observed change in China s import from the rest of the world in industry j between the start and the end of the period. Labor market exposure to import competition is the change in import exposure per worker in a region (in Autor et al. [2013] is the change in Chinese import exposure), where imports are apportioned to the region according to its share of national industry employment. And the export exposure is calculated by replacing observed 20

21 change in China s import from the world M jt with China s export to the world X jt. The primary measure of interest in my paper is IPW X it. The Bartik instrument uses the overall national growth to generate regional growth, by interacting with initial conditions. The benefit here is that it will be free of other local shocks that are correlated with local export growth. However, using observed trade volume increase might still be problematic, since the overall trade increase might still be correlated with overall economic growth, and then the result will capture the economic growth e ect instead of trade growth e ect. Thus, I will use two ways to instrument the trade volume change further. The first is to use the importing country s income growth to instrument for China s export growth. The second is to use gravity dummies instead of GDP. The GDP based instrument is constructed in the following way. Suppose a country s fraction of income allocated to di erent industries consumption (import) do not change over time, and the fraction of import in an industry that comes from China also do not change over time, then the growth of demand for Chinese goods will be from the growth of importing country s income level. Specifically, the import value of Chinese goods in industry j (which I use as the super script instead of sub script) and year t is constructed as the following: X j ct = X n = X n = X n X j P cnt n P 0 Xj n 0 nt P GDP nt n 0 Xj n 0 nt n 0,j Xj n 0 nt X j cnt X j nt X j nt X nt GDP nt X j cnt X nt GDP nt where Xj cnt is the fraction of import in industry j and country n that comes from China X j nt in baseline year t, and Xj nt X nt is the fraction of import in industry j and country n out of the total import value. Then the export market access shock with the GDP measure in industry j between year t and t 0 is defined as X j,gdp t,t 0 = X n X j cnt X nt (log(gdp nt 0) log(gdp nt )) In Figure 10, I sort the industries by the value of export in 2000, and show the industry level shock constructed from the importing country GDP growth ( X j,gdp t,t ). 0 21

22 Figure 10: Industry level trade shock, using importing country GDP growth, *Note: each bar is a industry. Horizontally sorted by value of exports in the industry in Alternatively, I can use gravity dummies instead of GDP growth. First I run a regression of log pairwise country imports on origin and destination country dummies, controlling for geographic distances. Then I construct the export market access shock with gravity measure as X j,gravity t,t 0 = X n X j cnt X nt (D n,t 0 D n,t ) In Figure 11, I sort the industries by the value of export in 2000, and show the industry level shock constructed from the importing country gravity dummy changes ( X j,gravity t,t ). We can 0 see that the results are quite similar to the ones in Figure

23 Figure 11: Industry level trade shock, using gravity dummies, *Note: each bar is a industry. Horizontally sorted by value of exports in the industry in After getting X j,gdp t,t0 and Xj,gravity t,t, I use them instead of X 0 jt to calculate IPWit X. I also add a measure which is the intermediate goods market access shock. I will describe the procedure in detail in the appendix. IPW INTER it = X j L ijt L jt I jt L it = X j L ijt L it I jt L jt 3.3 Information of migrant The information of the number of migrants is from censuses (2000 and 2010) and 1% population surveys ( 2005). The 2000 and 2005 are from 0.095% data and 1% of the overall sample. For 2000 and 2010, there are also city level aggregate data. Here I define a person as a migrant as living in a place other than the Hukou registration place. On average, migrant workers are less educated than the local, but more educated than their hometown counterparts. From , migrants with families increase substantially compared to the previous 5 years. 23

24 Table 2: Summary of statistics for migration flow data Year Sample Population Sampling rate Sample Population Sampling rate Population #ofpersons 1,180,111 1,295,330, % 2,585,480 1,306,280, % 1,339,724,852 #ofhouseholds 345, ,370, % 996, ,190, % 401,517,330 #ofcities ? #ofmigrants 124, ,390, % 263, ,350, % 261,386, html html html Other city level measures Other city level measures are taken from the City Statistics Yearbook. Including total population, wage, employment, employment in private sector, employment share in agriculture, manufacturing and sevices, GDP, and share of GDP in three categories. To check the validity of wage data, I also construct wage using the IES. Correlation is Since there is concern on the quality of the Chinese city level GDP data, I use night light satellite data to double check. Correlation is 0.8. I also use city level government revenue and expenditure data. 4 Empirical Results 4.1 Scatter plots for trade-regulation-migrant relationship Were places with bigger trade shocks relaxing the migration restrictions more?in Figure 12, I focus on the post-wto period ( ). On the horizontal axis, I have the export tari shock in , and a bigger export tari shock means lower export tari s and e ectively higher export prices. The vertical axis is the change of log regulation score from 2001 to 2007, and each dot is a city. The dashed line is the linear fitted line with 2001 population size as weights, and the dotted line is the unweighted linear fitted line. Figure 12 shows similar information as in the previous trend graphs for the post period: cities that experience more positive trade shock had a bigger change in regulation score, meaning that they became more friendly to migrants. 24

25 Figure 12: Bigger trade shocks, more regulation change, , 247 Chinese cities Note: Each dot is a city. Figure 4.1 shows the relationship between the regulation change and migrant flow. The horizontal axis is the change in log regulation score from 2001 to 2007, the vertical axis is the change in migrant share of the population from 2000 to 2010, and each dot represents a city. We can see from the graph that the more relaxed the regulation on migrants, the bigger increase in migrant share. Mega cities like Beijing, Shanghai and Guangzhou are not outliers. The graph indicates that the regulations are indeed useful or the migration restrictions are indeed binding, in terms of a ecting migration flows. 25

26 Figure 13: More regulation change, great change in migrant share of population, , 247 Chinese cities Note: Each dot is a city. I will use regression to show the these trade-regulation-migration flow in details in the following sections. 4.2 Results for trade-regulation relationship and trade on various economic outcomes Main Results I am interested in estimating the equation ln(regulation Score) c = Trade Shock c + X c + c where c represents a city, and t represents a time period. X c,t is the vector of potential confounding factors that could be correlated with the trade shocks. Table 3 shows the regression results of changes in log migration regulation score on trade shocks. In Column (1)-(7), the outcome variable is the change in log regulation score from 2001 to 2007, and Column (8)-(9) are the changes from 1995 to All columns have standard errors clustered at province level. I start with Column (1) where I only add the export tari shock from 2001 to The coe cient 0.71 means that 1 percentage point change in export tari shock increases the change in log regulation score by 0.71, and it is equivalent to saying 1 percentage point higher increase 26

27 in export tari shock will result in 71 percent higher regulation score increase. The di erence between city level export tari shock in places at the 5th and 95th percentile was 0.48 percentage point. Then when evaluated at Column (1) estimates, regions at 95th percentile experienced 34 percent higher increase in regulation change. Table 3: Bigger trade shocks, more regulation relaxation Then I add potential confounding factors to check the robustness of the result. First, regarding trade shock, I focus on export tari shocks, since export tari s are imposed by importing countries of Chinese goods, and should not be correlated with city level economic factors. However, if industries that experienced big export tari reductions happened to be the ones that experienced big (or small) import reduction or big (or small) reduction in import tari on intermediate inputs, then the e ect of export tari shock will be biased. Thus, controlling for other types of trade shocks could be important. Second, since the export tari shock is calculated as the interaction of industrial level tari reduction and city level industry shares, if certain industries were correlated with other variables that were important determinants of regulation change, then not including those variables would cause omitted variable bias. Such variables could be the baseline value of regulation scores, the lagged change in regulation score, the lagged trade shock, or lagged economic growth rates. I will control for these variables in Column (2)-(7). Column (2)-(7) adds import tari shock and intermediate tari shock. In addition to three trade shocks, Column (3) and Column (4) control for initial level and past changes in regulation scores. Column (3) adds log regulation score in Column (4) adds lagged dependent variable, which is the change in log regulation score from 1995 to Column (5)-(7) add past trade shocks and past economic growth. Column (5) controls for the lagged trade shocks, 27

28 which are export tari shock, import tari shock and intermediate tari shock from 1995 to Column (6) controls for change in log wage from 1995 to 2001, Column (7) controls for changes in log per capita GDP from 1995 to Column (2)-(7) estimates for the e ect of export tari shock range from 0.84 to 1.18, but all of them are within 95% confidence interval of the estimate in Column (1), meaning that the e ect of export shock on regulation change is quite robust. Also, it is useful to see if the trade shocks from 2001 to 2007 predict changes in log regulation score in the pre-period ( ). If yes, then it means that industrial composition itself can predict the regulation change in any period, and tari shocks happened in the period are relevant for regulation changes before. In this case, tari changes in are not good measures for the time period specific trade shocks, and we cannot identify the WTO e ect. The result in Column (8) denies this possibility. Finally, I check if the trade-regulation relationship existed before the WTO accession. In Column (9), I regress the regulation change in on trade shocks in , and the coe cients are insignificant. This is consistent with the observation in previous sections about the timing of the regulation change. Although trade shocks happened in the previous period as well, they didn t result in big regulation changes across cities, because central government did not allow for the reform. Overall, the findings are consistent with the hypothesis that in the post-wto period, places with bigger trade shocks opened up migration restrictions more Heterogeneous e ect After investigating the trade-regulation relationship, I will check if there is interaction e ect of some baseline city characteristics with the size of trade shock. Table 4 shows the heterogeneous trade e ects. There are four variables I consider here: (1) log per capita GDP in 2001, (2) log wage in 2001, (3) migrant intensity in 2001, (4) state-owned enterprise (SOE) employment share in The first two are related to city income because richer cities might respond di erently from poor cities. The latter two are related to the demand for migrant labor. Migrant intensity is calculated using national industry level migrant shares of total employment interacted with city level industry employment share, and it is a proxy for 2,n in the model (output elasticity of migrant labor). The intuition is that places with higher migrant intensity have higher marginal returns to migrant inflow, thus choose to relax the regulation more. SOE usually hire more locals than migrants; thus it is similar to 1 minus the migrant intensity. Table 4 have two columns for each variable. Column (1)(3)(5) and (7) have export tari shock from 2001 to 2007, the variable I, and the interaction of export shock with I. Column (2)(4)(6) and (8) add additional baseline controls including log population, share of GDP in manufacturing and share of GDP in service in All columns control for import and intermediate trade shock, and log regulation score in

29 Table 4: Heterogeneous trade e ects based on city characteristics, In Column (1), there is positive interaction e ect for initial GDP and export tari shock (0.97) and the coe cient for export tari shock is negative (-8.28). Approximately at the mean value of log per capita GDP in 2001 (which is 8.9), the overall e ect of export tari shock become positive. This means that richer cities responded more positively to the export tari shock. Column (2) has similar interpretations. Column (3) and (4) use wage instead of per capita GDP, and the result is similar: richer cities respond more positively, and the overall e ect of export tari shock become positive at the mean value of log wages. In Column (5), there is positive interaction e ect for migrant intensity and export tari shock (11.76) and the coe cient for export tari shock is negative (-3.08). At the mean value of migrant intensity (0.3), the overall e ect of export tari shock become positive. This means that cities with bigger demand for migrants responded more positively to the export tari shock. Column (6) shows similar results. Column (7) and (8) use SOE share of employment, which is negatively correlated with migrant intensity, and there is a negative interaction e ect. It means that cities where SOEs dominated responded less positively to the trade shock. I also use log per capita fiscal revenue and expenditure for interaction e ects, and they show similar patterns as log per capita GDP and wage. Log total population in 2001, the share of GDP in manufacturing and the share of GDP in services do not have significant interaction e ects. Overall, the migrant intensity related measures show e ects confirming the theoretical predictions. 4.3 Results for migrant flow. After investigating the e ect of trade flows on regulation change, I will show how trade shocks and regulation changes a ected the flow of migrants. Here the trade shock can have an impact on the migrant flow through two channels. The first is the direct channel because bigger export 29

30 tari shock will directly a ect wages, and higher wages will attract more migrants. The second is the indirect channel because trade shocks a ect the regulation score on migrant issues, and places with higher regulation score provide migrants will more amenity, and this will attract more migrants as well. Thus, in Table 5, I use di erent measures of migrant flows, and control for regulation change, export tari shock and both. Column (1)(2) and (3) use change in migrant share of the population from 2000 to 2010 as the outcome variable. Column (1) shows that 1 percent larger increase in regulation score from 2001 to 2007 will result in percentage point larger increase in the migrant share of the population. Column (2) shows that 1 percentage point larger export tari shock results in 6.51 percentage point larger increase in the migrant share of the population. Column (3) includes both the regulation change and the tari shock, and both coe cient become smaller, and the coe cient for tari shock becomes insignificant. The di erence between city level export tari shock in places at the 5th and 95th percentile was 0.48 percentage point. Then evaluated at the coe cient estimate in Table 5 Column (3), cities at 95th percentile had 2.52 percentage point higher increase in migrant share of population than the ones at 5th percentile through the direct e ect. Using estimates from Table 5 Column (3) and Table 3 Column (1), 95th percentile ones had 0.55 percentage point higher increase in the migrant share of the population through the regulation e ect. The direct e ect is about 5 times of the indirect e ect. Given that the mean of change in the migrant share of the population was 6.99 percentage point from 2000 to 2010 and the standard deviation is 5.81, the overall trade e ect accounts for 44% of the mean change, and 53% of one standard deviation change. Table 5: Regulation change ( ) and the flow of migrants ( ), 247 cities Column (4)(5) and (6) use change in the log number of migrants across cities (but within a province) as the outcome variable. Column (4) shows that 1 percent higher increase in regulation score leads to 0.07 percent higher increase in the number of migrants across cities. Column (5) shows that 1 percentage point higher export tari shock results in 48 percent larger increase in number of migrants across cities. When including both the regulation change and the export shocks together in Column (6), both coe cients become smaller. Again, using the gap of export shocks in the 95th and 5th percentile cities and estimates in Table 3 Column (1) and 5 Column 30

31 (6), 95th percentile ones had 23 percent higher increase in number of cross-city migrants through the direct e ect, and 2 percent higher increase through the regulation change. The direct e ect is about 11 times the indirect e ect. Given that the mean value of the increase in the number of cross-city migrants being 160 percent and the standard deviation being 6 percent, the overall trade e ect accounts for 16% of the mean change, and 37% of one standard deviation change. Column (7)(8) and (9) use change in log number of migrants across provinces as the outcome variable. Here only the indirect e ect through regulation change is significant. Column (7) shows that 1 percent higher increase in regulation score leads to 0.15 percent higher increase in the number of migrants across cities. Column (9) gives similar estimates. Combining information in Table 3 Column (1) and Table 5 Column (9), 95th percentile cities had 4.78 percent higher increase in the number of migrant across cities. Then the overall trade e ect accounts for 5.6% of the mean change and 7.2% of one standard deviation change in the number of cross-province migrants. Combining the estimates in Table 5, it is clear that the overall e ect of trade shocks on migrant flows are statistically significant and economically sizable. 4.4 Results for wage, employment and per capita GDP After showing the results of migrant flows, I will discuss in this section how trade shocks a ected other economic outcomes like wages, employment and GDP growth. As discussed in the previous section, trade shocks can a ect these outcomes through both the direct and the indirect channels. Trade shocks will a ect wages directly given employment, then wages a ect the marginal return to the inflow of migrants, and will a ect regulation change. Then the regulation change and wage change feed back into the migrant inflow, and migrant inflow a ects wages. I will call the coe cients for the trade shocks as the direct e ect, and the coe cient on regulation changes the indirect e ect. In Table 6, Column (1)(2) and (3) use change in log wages from 2001 to 2007 as the outcome variable. Column (1) controls for regulation change, Column (2) controls for trade shocks, and Column (3) controls for both. We can see that the estimates in Column (3) are very similar to Column (1) and (2). Using the estimates from Column (3), 1 percent higher increase in regulation score results in 0.03 percent higher increase in wages, and 1 percentage point higher trade shock leads to 14 percent higher increase in wages. Combining with the estimates from Table 3 Column (1), cities at the 95th percentile had 6.7 percent higher increase in wages through the direct e ect than the cities at the 5th percentile, and 1 percent higher increase through the indirect e ect. The direct e ect is about 7 times the indirect e ect. Given the mean change in wages being 82 percent and the standard deviation being 14 percent, the overall trade e ect is 9.4% of the mean and 55% of one standard deviation for changes in wages. Column (4)(5) and (6) use change in log total urban employment from 2001 to 2007 as the outcome variable. Column (4) controls for regulation change, Column (5) controls for trade shocks, and Column (6) controls for both. Again, the estimates in Column (6) are very similar 31

32 to Column (4) and (5). Using the estimates from Column (6), 1 percent higher increase in regulation score results in 0.05 percent higher increase in wages, and 1 percentage point higher trade shock leads to 36 percent higher increase in wages. Combining with the estimates from Table 3 Column (1), cities at the 95th percentile had 17 percent higher increase in wages through the direct e ect than the cities at the 5th percentile, and 1.7 percent higher increase through the indirect e ect. The direct e ect is about 10 times the indirect e ect. Given the mean change in total urban employment being 32 percent and the standard deviation being 39 percent, the overall trade e ect is 58% of the mean and 48% of one standard deviation for changes in total urban employment. Table 6: Regulation change ( ) and change in log wage, employment and GDP p.c. ( ), 247 cities Column (7)(8) and (9) repeat the exercise using change in log per capita GDP from 2001 to 2007 as the outcome variable. Column (7) controls for regulation change, Column (8) controls for trade shocks, and Column (9) controls for both. Similar as before, estimates in Column (9) are very similar to Column (7) and (8). Using the estimates from Column (6), 1 percent higher increase in regulation score results in 0.08 percent higher increase in wages, and 1 percentage point higher trade shock leads to 54 percent higher increase in wages. Combining with the estimates from Table 3 Column (1), cities at the 95th percentile had 26 percent higher increase in wages through the direct e ect than the cities at the 5th percentile, and 2.7 percent higher increase through the indirect e ect. The direct e ect is about 10 times the indirect e ect. Given the mean change in wages being 87 percent and the standard deviation being 27 percent, the overall trade e ect is 33% of the mean and 100% of one standard deviation for changes in GDP per capita. Overall, the trade e ect on wage, employment and income is statistically significant and economically large. The e ect on per capita GDP is bigger than the e ects on wages and employment, potentially capturing other channels through which trade shocks a ected the economy (through payment to other factors for example). 32

33 4.5 Discussion on regulation e ects Did migrant flow drive regulation change, or the other way around? On the one hand, migration regulation change can a ect migrant amenities, making the city more or less attractive to migrants, and lead to bigger or smaller migrant inflow. On the other hand, big migrant inflow could put pressure on the city infrastructure or local employment, and lead to regulation changes. Since trade shocks a ect both the migrant flow and the migrant regulation, it could be useful to distinguish which happens first. In previous sections, I use the migration flow from 2000 to 2010, since I cannot observe the number of migrants in In Table 7 Column (1)-(4), I construct change in log regulation score for (as lagged two periods), (as lagged one period), (as current period) and (as one lead period), and check the e ect on migrant flows from For Column (5)-(7), I construct regulation changes for (as lagged one period), (as the current period) and (as one lead period), and check the e ect on migrant flows from Panel A uses change in log number of migrants as the outcome variable, and Panel B uses changes in migrant share of the population as the outcome variable. The results in Table 7 show that regulation change happened before the migration flow increase, meaning that the first channel dominates. For the period, the coe cients for regulation change decline from 0.40 in Column (1) to 0.9 in Column (3) in Panel A. This could be the mechanical e ect from the fact that mean change in regulation increases from 0.04 in Column (1) to 0.95 in Column (3). However, when we go to the lead period in Column (4), although there is still a sizable change in log regulation score of 0.53, there is no longer a positive e ect of regulation changes on migrant flow from Similar patterns can be found in Column (5)-(7). And Panel B conveys the same information. 7 The estimate in Table 7 Panel A, Column (6) can be roughly compared with the estimate in Table 5 Column (4) and (7).The estimate in Table 7 Panel B, Column (6) can be roughly compared with the estimate in Table 5 Column (1). 33

34 Table 7: Regulation change and migrant flows, lagged, current and lead, 254 cities Overall, I find a positive e ect of lagged or current regulation change on migrant flow, but no e ect of lead regulation changes. This reinforces the argument that the regulations are indeed binding, and change in regulation are important Did regulations improve the migrant welfare? From the data description part, we see that the regulations had very specific targets: increasing wage received by migrants, forcing firms to sign contracts, providing social insurances to migrants, and giving migrant kids access to local primary and secondary schools. Thus, it would be helpful to see if the regulations indeed help migrants and improve migrant welfare. Unfortunately, the only data available that include these measures is the 2005 mini-census data. This means that I cannot see how regulation changes a ect change in migrant welfare measure, but only a cross-section. In order to alleviate the concern that places with more generous regulations towards migrants are essentially di erent from other cities, I will control for local worker welfare measures, log per capita GDP and log number of migrant adults in In Table 8, I find that the places with higher regulation scores are also the ones with higher 34

35 migrant welfare, concerning social insurances, income levels and school enrollment rates for children. Column (1) shows that 1 unit increase in regulation score is related to 0.3 percent increase in unemployment insurance rate for migrants. Given the mean of insurance rate for migrant being 21 percent and for locals being 37 percent, 1 unit increase in regulation score will close 5.3% of the migrant-local gap. Column (2) and (3) show similar patterns. There is no significant e ect of regulation score on terms of contract (see Column (4)). Column (5) indicates that 1 unit increase in regulation score is related to income increase by 13 Yuan per month, which is 22% of the wage gap between locals and migrants. Column (6) has an additional control which is the log number of kids in school age. Thus the estimate in Column (6) means that 1 unit increase in regulation scores will increase the number of migrant kids enrolled by 0.2%, given the number of migrant kids at the school age. Column (7) is about whether regulation score increase is correlated with more migrant children brought to cities where their parents are working, and the result is insignificant. Table 8: Regulation score in 2005 and migrant welfare, 245 cities Overall, the results in Table 8 shows that higher regulation score places are also the ones with higher migrant being, although the estimates are relatively small. The shortcoming here is that there could be other factors that a ect the regulation score and the migrant welfare directly, and the coe cients might be biased. 4.6 Robustness check Alternative measure of regulation change. It is important to get regulation change measure. The main measure in the text is from manual coding. However, the manual coding could be subject to biases. Thus I will use text analysis in order to check if the regulation change matters. The first is to use negative, neutral and positive scale instead of -2, -1, 0, 1,2 scale. The second is for negative, neutral and positive documents, I count the total number of characters. The rationale here is on how much text is devoted to the discussion of the migrant issues. 35

36 Table 9: Alternative measure of regulation change Alternative measure of trade shock. The main regression uses the tari change. Third, use Autor et al. [2013] OLS measure. Fourth, use Autor et al. [2013] gravity and GDP measures to instrument. Table 10: Alternative measure of trade shock Adding industrial composition controls The regional tari shocks are generated using the interaction of city level industrial composition and industry level tari reductions. If certain industries are the driver of variation, and these industries are also correlated with other local factors that a ect regulation changes directly, then the estimates for regional tari shock e ects would be biased. To check if there exists such an industry, I add industry employment share one by one and run the regression in Table 3 Column (3). In Figure 14, I plot the coe cient estimates with 90% confidence intervals, and each bar is 36

37 from a regression including a specific industry employment share. We can see that the coe cient estimates are relatively stable around 1, which is the estimate in Table 3 Column (3). Thus, the results are not sensitive to specific industry e ects. Figure 14: Coe one cients from main regression by adding industrial composition controls one by Decomposition of migrant flow Table 11: Alternative measure of trade shock 37