Trade, Migration and Productivity: A Quantitative Analysis of China

Transcription

1 Trade, Migratio ad Productivity: A Quatitative Aalysis of Chia Trevor Tombe Uversity of Calgary Xiaodog Zhu Uversity of Toroto First Versio: Jauary 2014; This Versio: September 2017 Abstract We study how goods- ad labor-market frictios affect aggregate labor productivity i Chia. Combig uque data with a geeral equilibrium model of iteral ad iteratioal trade, ad migratio across regios ad sectors, we quatify the magtude ad cosequeces of trade ad migratio costs. The costs were high i 2000, but declied afterward. The declie accouts for a third of the aggregate labor productivity growth i Chia betwee 2000 ad Reductios i iteral trade ad migratio costs are particularly importat. Despite the declie, migratio costs are still high ad potetial gais from further reform are large. JEL Classificatio: F1, F4, R1, O4 Keywords: Migratio; iteral trade; spatial misallocatio; gais from trade; Chia Tombe: Departmet of Ecoomics, Uversity of Calgary, 2500 Uversity Drive NW, Calgary, Alberta, T2N1N4. ttombe@ucalgary.ca. Zhu: Departmet of Ecoomics, Uversity of Toroto, 150 St. George Street, Toroto, Otario, M5S3G7. xzhu@chass.utoroto.ca. We thak Adrew Atkeso, Lorezo Caliedo, Kual Dasgupta, Joatha Digel, Jim Harriga, Chad Joes, Nicholas Li, Peter Morrow, Albert Park, Mike Peters, Chris Pissarides, Diego Restuccia, Aleh Tsyviski, Xiaobo Zhag, Fabrizio Zilibotti, Yaohui Zhao, ad especially Dael Trefler for very helpful commets ad suggestios. We have also beefited from commets of various semiar participats at the Bak of Caada, Brow, Calgary, Chicago Fed, Chiese Uversity of Hog Kog, Fuda, Coecticut, George Washigto, Hog Kog Uversity of Sciece ad Techology, IMF, Michiga State, Pekig, Pe State, Philadelphia Fed, Rochester, SAIS Johs Hopkis, Sa Fracisco Fed, SUFE, Texas A&M, Toroto, UBC, Virgia, Wester, Wuha, ad Yale; ad participats i may cofereces. Tombe ackowledges fiacial support from the Social Sciece ad Humaties Research Coucil IG ). A prelimiary versio of this paper was distributed uder the title of Trade, Migratio ad Regioal Icome Differeces: Evidece from Chia. Stable lik to paper:

2 1 Itroductio Chia s recet growth has bee othig short of remarkable. From 2000 to 2007 after oig the WTO ad before the fiacial crisis Chia s real GDP per capita early doubled. Chia s rapid asced as a key player i the world ecoomy is well kow, but equally dramatic has bee the growth of its iteral ecoomic itegratio. Trade betwee its provices has icreased more tha trade betwee Chia ad the rest of the world, ad the flow of workers across regios withi Chia represets the largest migratio i huma history. Policy chages may be a importat cause of these developmets. I the early 2000s, Chia had substatial policy-iduced migratio costs Pocet, 2006; Cai et al., 2008) ad iteral trade costs Youg, 2000; Pocet, 2005). Sice the, the Chiese govermet has udertake policy reforms ad ifrastructure ivestmets that reduced both migratio ad trade costs ad, at the same time, the Chiese ecoomy has experieced sigficat growth i aggregate productivity Zhu, 2012). What role did the policy chages play i Chia s rapid growth? I this paper we use a rich quatitative framework ad uquely detailed data to aswer this questio. We fid that the iteral trade ad migratio cost reductios, ad the associated icreases i trade ad migratio withi Chia, cotributed 27% of Chia s aggregate labor productivity growth betwee 2000 ad The reductio i iteratioal trade costs, o the other had, cotributed oly 7% of the growth. These results highlight the importace of iteral reforms for Chia s growth ad are i stark cotrast to the widely held perceptio that Chia s growth durig the period was a export-led experiece. Our quatitative framework builds o recet developmets i iteratioal trade research. We develop a two-sector multi-regio geeral equilibrium model featurig iteral trade, iteratioal trade, ad worker migratio. Followig Ahlfeldt et al. 2015) ad Reddig 2016), we itroduce withi-coutry trade ad worker mobility ito the Eato ad Kortum 2002) model ad explicitly model worker locatio choices i the presece of migratio costs ad heterogeeous worker prefereces regardig locatios ad sectors. We also icorporate ito the model collective owership of lad, a importat istitutioal feature of Chia that makes migratio difficult. Eve with these rich ad realistic features, the model is still aalytically 1

3 tractable ad ca be easily used for quatitative aalysis. To facilitate our quatitative aalysis, we compile a rich set of data o Chia s iteral ad exteral trade, iteral migratio, ad spatial distributio of icome. Usig the data ad our quatitative model we estimate both the levels of ad chages i trade ad migratio costs i Chia. We fid that trade costs were large i 2002, but they declied sigficatly betwee 2002 ad O average, iteral costs fell by 10-15% ad iteratioal costs fell by almost 10% i o-agriculture ad early 25% i agriculture. For migratio costs, we cosider them ogoig flow costs rather tha suk costs due to a uque istitutioal feature of Chia, the hukou household registratio) system that imposes large costs o workig ad livig outside oe s hukou locatio, primarily through restricted access to social services ad limited employmet rights. These costs are recurrig ad exist as log as migrats do ot have a local hukou. Accordig to our estimates for 2000, the average cost of movig from rural to urba areas withi a provice is equivalet to shrikig oe s real icome by a factor of early 3; betwee-provice moves are a order of magtude more costly. I additio, sice all rural lad ad some urba lad are collectively owed ad there is a lack of retal market for lad, migrats who leave their hukou locatio lose access to ad beefits from lad. High migratio costs ad restrictive lad markets mea oly a small proportio of workers move; those who do move ted to be youg workers facig lower migratio costs. Betwee 2000 ad 2005, however, migratio costs did declie substatially by 15% o average, ad by as much as 40% for betwee-provice moves. I a series of quatitative exercises usig our calibrated model, we evaluate how the measured cost chages affect trade flows, migratio, aggregate labor productivity ad welfare. We fid that the reductios i trade costs ca accout for most of the icreases i Chia s iteral ad exteral trade betwee 2002 ad They have relatively small effects o migratio, but large effects o aggregate labor productivity ad welfare, both of which icrease by more tha 13%. Because most provices i Chia trade more withi Chia tha abroad, the iteral trade cost chages cotribute more to the gais i aggregate labor productivity ad welfare tha the exteral trade cost chages. Similarly, the measured chages i migratio costs have small effects o trade shares, but large effects o migratio ad aggre- 2

4 gate labor productivity ad welfare. I respose to the migratio cost reductios, the stock of withi-provice ad betwee-provice migrats icrease by roughly 15% ad 82%, respectively. Most of the icreases are i the form of rural-to-urba migratio. Largely due to the reallocatio of labor from agriculture to o-agriculture, aggregate labor productivity ad welfare icrease by 4.8% ad 8.5%, respectively. Despite the recet declie i trade ad migratio costs, the scope for further cost reductios remais large. We fid that movig Chia s iteral trade costs to levels measured i Caada yields welfare gais of roughly 12%. Gais are eve larger if we lower the migratio costs to levels such that oe-third of Chiese workers move beyod their provice of registratio a level cosistet with U.S. migratio rates), with real GDP per worker icreasig by early 23% ad welfare by 15%. Fially, we quatify the effects of allowig for private lad owership ad a fully fuctiog lad retal market so migrats o loger give up the returs to lad whe they move. We fid that the umber of migrat workers would sigficatly icrease ad the resultig welfare gai would be early 10%. Our paper cotributes to two broad literatures. First, there is a growig literature likig iteratioal trade flows with the spatial distributio of labor ad ecoomic activity withi coutries, such as Cosar ad Fagelbaum 2012); Dix-Careiro ad Kovak 2014); Alle ad Arkolakis 2014); Brya ad Morte 2015); Reddig 2016) ad Caliedo et al. 2015). There are also papers ivestigatig migratio costs or iteral trade separately, such as Morte ad Oliveira 2014), Brya ad Morte 2015) or Gha et al. 2012), ad empirical ivestigatios of trade s effect o iteral migratio, such as McCaig ad Pavck 2012) for Vietam or Aguayo-Tellez ad Muedler 2009) ad Herig ad Paillacar 2012) for Brazil. There is also a large urba-ecoomics literature ivestigatig the role of iteratioal trade i alterig the spatial distributio of firms ad factors withi a coutry, such as Haso 1998). Little work has bee doe, however, ivestigatig the case of Chia perhaps the largest ad fastest expasio of trade ad iteral migratio ever recorded. Secod, the recet macro-developmet literature has emphasized differeces i aggregate total factor productivity TFP) as a key source of large cross-coutry icome differeces Kleow ad Rodriguez-Clare, 1997; Hall ad Joes, 1999; Caselli, 2005) ad misallocatio of iputs as a importat reaso for 3

5 low levels of aggregate TFP i poor coutries Baeree ad Duflo, 2005; Restuccia ad Rogerso, 2008; Hsieh ad Kleow, 2009; Bartelsma et al., 2013). We study specific sources of misallocatio i a importat developig ecoomy; a valuable research area highlighted by Restuccia ad Rogerso 2013). Bradt et al. 2013) use a geeral equilibrium model to quatify the aggregate productivity loss due to misallocatio of labor ad capital across space i Chia, but the sources of misallocatio are ot modeled. I cotrast, we model trade ad migratio costs as specific sources of misallocatio. Ngai et al. 2016) ivestigate the impact of the hukou system o labor mobility i Chia, but their aalysis is at a more aggregate level ad without detailed modelig of trade ad migratio across space. Caliedo et al. 2017b) ivestigate the impact of iteral ad exteral distortios i a world ecoomy with iput-output likages ad also fid that the impacts of iteral distortios are much larger tha iteratioal distortios. 2 Data, Facts ad Back-of-the-Evelope Calculatio I this sectio, we describe our data ad highlight some key facts about the Chiese ecoomy. We also discuss the migratio ad trade policies i place aroud 2000 ad how they subsequetly chaged. Fially, we coduct some back-of-the-evelope calculatios to illustrate the potetial gais from these policy chages to motivate our more comprehesive quatitative aalysis to come. There are 31 provices i Chia. We exclude Tibet i all that follows due to limited data. For the other 30 provices, we divide their ecoomies ito agricultural ad o-agricultural sectors. 2.1 Spatial Distributio of Icome Comparig real icomes across provices ad sectors is o trivial task. We eed data o GDP, employmet, ad price levels for each provice ad sector. Official statistics published i the aual Chia Statistical Yearbook CSY) reports omial GDP ad employmet data for agriculture, idustry ad services i each of Chia s provices, which we aggregate to agriculture ad o-agriculture. The CSY also reports both the rural ad urba cosumer price idices for each provice. I addi- 4

6 Figure 1: Spatial Distributio of Real Icomes ad Migratio i 2000 a) Real GDP/Worker, Relative to Mea b) Migrat Share of Total Employmet Heilogiag Heilogiag Xiiag Ier Mogol Jili Xiiag Ier Mogol Jili Beiig Hebei Tiai Liaog Beiig Hebei Tiai Liaog Qighai Gasu Nigxia Shaxi Shadog Qighai Gasu Nigxia Shaxi Shadog Shaaxi Shaaxi Xizag Hea Jiagsu Xizag Hea Jiagsu 1.5,3] 1.25,1.5] 1.1,1.25] 1,1.1].75,1].65,.75].55,.65].45,.55] [.25,.45] No data Yua Sichua Chogqig Guizhou Guagxi Haia Hubei Hua Guagdog Hog Kog Ahui Jiagxi Shaghai Fuia Zheiag Taiwa.1,.27].075,.1].04,.075].02,.04].01,.02].0075,.01].005,.0075] [.0025,.005] No data Yua Sichua Chogqig Guizhou Guagxi Haia Hubei Hua Guagdog Hog Kog Ahui Jiagxi Shaghai Fuia Zheiag Taiwa Note: Displays choropleths of relative real icome levels for each of Chia s provices ad the migrat share of total employmet. Dark reds idicate both high relative real icomes ad large migrat shares of employmet. tio, for a few years i the 1990s the CSY reported retail prices of maor cosumer products i provicial capital cities ad procuremet prices of agricultural products i rural areas by provice. Bradt ad Holz 2006) use these data ad the cosumptio basket weights published by Chia s Natioal Bureau of Statistics NBS) to costruct rural ad urba price levels i 1990 for each provice. We combie these 1990 price levels ad the published CPI idices to calculate the price levels i other years, ad the calculate real icomes by deflatig agricultural GDP ad o-agricultural GDP with rural ad urba price levels, respectively. With this data i had, we fid large regioal icome iequality i Chia. The ratio of the average real GDP per worker of the top five provices to that of the bottom five provices, for example, was 4:1 i I geeral, the coastal provices i the easter regio had substatially higher levels of real GDP per worker tha provices i the cetral ad wester regios. I Figure 1a we plot the real GDP per worker for the 30 provices i Chia i For compariso, we plot i Figure 2a the distributio of GDP per worker across US states alog with the distributio across Chia s provices. Sice we do ot have price level iformatio for the US, we use omial GDP per worker for both 5

7 Figure 2: Spatial Distributio of GDP per Worker 2000) a) Across Regios withi Chia ad the USA b) Across Chia s Regios withi Ag ad Noag Chia USA 0.9 Agriculture No Agriculture Desity Desity Log GDP/Worker Relative to Average Log GDP/Worker Relative to Average Note: Displays distributio of omial GDP per worker across regios. Pael a) compares aggregate values across Chia s provices relative to the distributio across US states. Pael b) displays values across regios of Chia withi agriculture ad o-agricultural sectors. Data for the US are from the BEA s state-level GDP ad employmet data. All data are for the year coutries i the plot. It is clear that the dispersio of icome is substatially larger i Chia tha i the US. We also plot i 2b the cross-provice distributio of GDP per worker withi Chia s agricultural ad o-agricultural sectors. The large dispersio of icome withi sectors ad higher icomes i the o-agricultural sector are evidet. Eve after cotrollig for price differeces betwee rural ad urba areas, the real GDP per worker i the o-agricultural sector was still much higher tha that i the agricultural sector i all the provices; the average ratio of the real GDP per worker i the two sectors withi a provice was 4:1. A importat reaso for the large real icome differeces across provices ad sectors i Chia is a hukou system that imposes severe restrictios o worker movemets withi Chia. 2.2 Migratio Policies ad Migratio Patters I 1958, the Chiese govermet formally istituted a household registratio system to cotrol populatio mobility. Cha 2010) provides a detailed discussio of the system; we summarize its key features here. Each Chiese citize is assiged a hukou registratio status), classified as agricultural rural) or o-agricultural urba) i a specific admistrative ut that is at or lower tha the couty or city 6

8 level. Idividuals eed approvals from local govermets to chage the category agricultural or o-agricultural) or locatio of hukou, ad it is extremely difficult to obtai such approvals. Before the ecoomic reform started i 1978, workig outside oe s hukou locatio or category was prohibited. This prohibitio was relaxed i the 1980s, but prior to 2003 workers without local hukou still had to apply for a temporary residece permit. This was difficult, so may did ot ad risked arrest ad deportatio by the local authorities. As the demad for migrat workers i maufacturig, costructio, ad labor itesive service idustries icreased, may provices, especially the coastal provices, elimiated the requiremet of temporary residece permit for migrat workers after There was also a atio-wide admistrative reform i 2003 that greatly streamlied the process for gettig a temporary residece permit i other provices. These policy chages made it much easier for a worker to leave his/her hukou locatio ad work somewhere else as a migrat worker. However, eve with a temporary residece permit, migrat workers without local hukou have very limited access to local public services ad face much higher costs for health care ad for their childre s educatio. So despite the reforms, the costs of beig a migrat worker remai high, especially for out-of-provice migrats ad older workers to whom havig access to public services is more importat. Not surprisigly, there are more withi-provice tha cross-provice migrats, most migrat workers are youg ad without childre, ad the average duratio of their stay outside the hukou locatio is oly seve years Meg, 2012). We costruct our ow data o withi- ad betwee-provice migratio. The mai sources of labor market data are the aual Rural Household Surveys ad Urba Household Surveys coducted by the NBS. However, these residece-based surveys are kow to uderestimate migratio. For studyig migratio, researchers have geerally used the idividual-level Populatio Cesus, as do we. 1 Specifically, we use the 1% sample of the 2000 Cesus ad the 20% sample of the % Mi 1 There is a ew Logitudial Survey o Rural Urba Migratio i Chia RUMiC) that is desiged to provide a more accurate picture of migratio i Chia. This survey covers oly e provices ad fiftee cities i Chia ad the survey results are largely cosistet with the data from populatio cesus. See Meg 2012), who provides a overview of the labor market data i Chia ad Cha 2013), who discusses migratio data i Chia. 7

9 Table 1: Stock of Migrat Workers i Chia Iter-Provicial Itra-Provicial Total Migrat Stock millios) Share of Total Employmet Total Migrats 4.2% 7.2% 14.3% 17.7% Ag-to-Noag Migrats 3.4% 5.6% 13.1% 16.4% Notes: Migrats are defied based o their their hukou locatio. Iter-provicial migrats are workers registered i aother provice from where they are employed. Itra-provicial migrats are workers registered i the same provice where they are employed, but are either o-agricultural workers holdig agricultural hukou or vice-versa. Cesus as our data source for migratio. Ay worker i a provice other tha the provice of his/her hukou is classified as a iter-provicial migrat. A worker withi his/her hukou provice but i a occupatio other tha his/her hukou category agricultural or o-agricultural) is classified as a itra-provicial migrat. Table 1 presets the total umber of iter-provicial ad itra-provicial migrat workers for 2000 ad 2005 ad their shares of total employmet. Most of the itra-provicial migrat workers are rural-to-urba migrats who have agricultural hukou but work outside agriculture. Partly due to the migratio policy chages, the umbers of iter- ad itra-provicial migrat workers have both icreased sigficatly betwee 2000 ad 2005, ad most of the icreases are rural-to-urba migrats. By 2005, there were 49 millio workers who moved across provicial boudaries ad 120 millio workers who switched sectors withi a provice. While migratio of this magtude is uprecedeted, as a share of total employmet it is less impressive. Despite large icome disparity across provices, iter-provicial migrat workers accouted for oly 4.2% of total employmet i 2000 ad 7.2% i There is heterogeeity across provices, of course. Figure 1b plots for each provice the migrat workers share of total employmet i Richer provices i coastal regios ted to have higher migrat worker shares tha poorer iterior provices, ad provices with more iter-provicial migrat workers also ted to have higher itra-provicial migrat workers. 8

10 2.3 Trade Policies ad Trade Patters Chia s iteratioal trade liberalizatio associated with its accessio to WTO i 2001 is well kow; it s sigficat iteral trade liberalizatio aroud the same time is less so. Several researchers have documeted high iteral trade costs i Chia i the 1990s Youg, 2000; Pocet, 2005). Others have also documeted that the degree of local market protectio i a provice was directly related to the size of the state sector i that provice Bai et al., 2004). Sice 2000, these trade barriers have bee reduced sigficatly. Some of the reductio was due to the govermet s deliberate policy reforms. For example, the state coucil uder the premier Zhu Rogi issued a directive at the ed of 2001 that prohibits local govermet from egagig i local market protectios. More importat, as a result of various SOE reforms, the size of the state sector has declied, ad therefore the local govermets have less icetives tha before to egage i local market protectios. Equally importat, improved trasport ifrastructure ad logistics also helped lower iteral trade costs. To costruct the trade data we use i our aalysis, we tur to the iter-regioal iput-output tables for 2002 ad 2007 costructed by Li 2010) ad Zhag ad Qi 2012). These tables are costructed based o the data from the NBS s Provicial Iput-Output Tables, Surveys of the Sources of Material Iputs for Idustrial Eterprises, ad the Surveys of Itial Destiatios of Idustrial Output, ad the iformatio o goods trasportatio by railways i Chia. Li 2010) reports bilateral trade flows for all provices ad for a variety of sectors i For chages i trade flows, Zhag ad Qi 2012) provide the bilateral trade flows betwee eight aggregate regios i both 2002 ad From these data, Table 2 reports aggregate bilateral trade betwee eight regios i Chia ad the rest of the world see the Appedix for a list of provices by regio). To ease comparisos, we ormalize all flows by the importig regio s total expeditures, resultig i a table of expediture shares π = x / N i=1 x, where x is the spedig by regio o goods from regio i. I additio, we report a regio s share of expeditures o goods from all other regios withi Chia i the last colum, ad each regio s home-share π which is the fractio of spedig allocated to local producers alog the diagoal. 9

11 Table 2: Iteral ad Exteral Trade Shares of Chia Exporter Total North- Beiig North Cetral South Cetral North- South- Other Importer east Tiai Coast Coast Coast Regio west west Abroad Prov. Year 2002 Northeast Beiig/Tiai North Coast Cetral Coast South Coast Cetral Regio Northwest Southwest Abroad Year 2007 Northeast Beiig/Tiai North Coast Cetral Coast South Coast Cetral Regio Northwest Southwest Abroad Note: Displays the share of each importig regio s total spedig allocated to each source regio. See the appedix for the mappig of provices to regios. The colum Total Other Prov. reports the total spedig share each importig regio allocated to producers i other provices of Chia. The diagoal elemets the home share of spedig), the share imported from abroad, ad the share imported from other provices will together sum to 100%. While the regios i Chia geerally import more from abroad tha from ay particular regio withi Chia, the total imports from the rest of Chia are still higher tha imports from abroad for most of the regios. The Cetral Coast ad South Coast regios are the exceptios. I 2002, their imports from abroad were sigficatly higher tha imports from the rest of Chia; they also had substatial iteratioal exports. Iterior regios of Chia have much higher home-shares tha coastal regios. I 2002, the cetral regio s home share is 0.88 compared to oly 0.72 for the south coast ad 0.63 for Beiig ad Tiai. Due to the iteral ad exteral trade liberalizatios, all regios i Chia became more ope betwee 2002 ad 2007, as evideced by the declie i their home shares. For most regios, this was due to icreases i import shares both from the rest of Chia ad abroad. But, agai, the Cetral Coast ad South Coast regios are the exceptios. Their shares of imports from abroad actually declied durig this period; all the declies i the two regios home shares were due to the icreases i their imports from the rest of Chia. Overall, iteral trade icreased more tha 10

12 exteral durig this period. O average, a regio s share of spedig allocated to imports from the rest of Chia icreased by early 7 percetage poits while the share imported from abroad icreased by oly 1 percetage poit Potetial Gais from Trade ad Migratio So far we have documeted that i 2000 there were large dispersio of icome across Chia s provices ad sectors, partly due to the high migratio ad trade costs. We also described the policy chages that helped reducig these costs betwee 2000 ad 2005 ad the icreases i trade ad migratio durig the period. What have these chages meat for Chia s aggregate ecoomy? Before turg to a full geeral equilibrium model for aswerig this questio, we first illustrate the potetial gais from the icreases i trade ad migratio with a back-of-theevelope calculatio based o a simple model. The calculatio will show that the cotributios of migratio flows ad icreases i iteral trade to aggregate GDP growth durig the period are much larger tha the cotributio from the icreases i iteratioal trade. We will latter show that this key quatitative result still holds whe the full model is used for the aalysis. Let y ad l be the real GDP per worker ad employmet share i regio ad sector, ad aggregate GDP per worker as y =, y l. Followig some shock that affects sectoral or regioal GDP ad employmet, the relative chage i aggregate GDP per worker is therefore ŷ = ω, ŷ ˆl = 1 + ω, g y + ω, g l +, ω g y g l where ˆx = x /x deotes relative chages i variable x, g x its growth rate, ad ω y l regio ad sector s share of itial aggregate GDP. Migratio affects aggregate GDP through the chage i employmet shares g l positively if workers migrate to relatively high GDP regios or sectors, ad egatively otherwise. Trade 2 Trade shares reported here are at the regioal level oly. For 2002, we compute trade shares for each provice ad sector ad fid, cosistet with the regioal data, iterior provices have higher home-shares tha coastal provices, ad most provices trade more withi Chia tha from abroad with the exceptio of the coastal provices. 11

13 affects aggregate GDP through its effect o g y. Though more difficult to quatify tha migratio, Arkolakis et al. 2012) show that withi a broad class of models aggregate gais from trade ca be captured by chages i a regio s home-share combied with a elasticity of trade parameter θ, which is typically estimated to be aroud 4. Specifically, ŷ = ˆπ ) 1/θ 1 π 1 θ π We ca further distiguish chages i home shares as due to chages i 1) spedig allocated to other Chiese provices, which we deote πc, ad 2) spedig allocated to iteratioal imports, which we deote π sum to oe, we have π = πc πw ad g y 1 θ π w π with the earlier expressio for ŷ, ad our data o trade ad migratio, we have. w. Sice all shares must ) πc + π. Together ŷ 1+ ω 1, θ π + ω 1, θ π + ωg l, }{{}}{{}}{{} Iteral Trade Exteral Trade Migratio π c π w 4.9% 0.5% 10.8%. Here we igored the quatitatively small iteractio term g y g l. This simple expressio decomposes aggregate growth ito cotributios from risig iteral trade, iteratioal trade, ad migratio. We fid migratio cotributes early 11% to Chia s aggregate labor productivity growth betwee 2000 ad 2005, holdig all other factors fixed. Icreases i iteral trade add 4.9% ad iteratioal trade 0.5%. Migratio matters because workers moved to higher productivity regios ad, more importat, sector. The large productivity gap betwee agriculture ad o-agriculture ad the reallocatio of labor from agriculture to o-agriculture accouts for about 90% of the gais from migratio. As for the gais from trade, iteral trade cotributes much more to growth tha exteral trade, because the icrease i the share of iteral trade for a provice is o average much larger tha the icrease i the share of exteral trade. While the decompositio is illustrative about the potetial gais from trade ad 12

14 migratio, it igores several importat issues that may have sigficat impact o the quatitative evaluatio of growth cotributios. First, it does ot takig ito accout the equilibrium relatioship betwee trade ad migratio: trade may iduce chages i migratio ad migratio may lead to chages i trade. Secod, without a structural model we caot quatify how much of the icreases i trade ad migratio were due to the reductios i migratio ad trade costs. Third, we treated agriculture ad o-agriculture symmetrically ad igored iput-output likages. Fourth, we may have overestimated gais from migratio by igorig differeces i fixed factor edowmets lad) ad eglectig the heterogeeous locatio prefereces of workers. Fially, we may have uderestimated the gais from iteratioal trade cost reductios because they may have icreased both iteral trade ad migratio as well as iteratioal trade. We tur ext to a geeral equilibrium model that explicitly deals with all these issues, ad use it to quatify the chages i the uderlyig migratio ad trade costs ad their cotributios to growth. 3 Quatitative Model Our geeral equilibrium model of trade ad migratio builds o the works by Eato ad Kortum 2002), Ahlfeldt et al. 2015) ad Reddig 2016). The model features two tradable sectors ad multiple regios of Chia betwee which goods ad labor may flow. Our mai departure from these papers is that we itroduce betweeregio migratio frictios ad withi-regio rural-to-urba migratios. There are N + 1 regios represetig Chia s provices plus the world. Each regio has two sectors: agriculture ad o-agriculture, deoted {ag, a}. Each regio is also edowed with a fixed factor lad, structures), deoted S, that is used for housig ad productio. Fially, there are L workers with hukou i regio ad sector ad L workers livig there. 13

15 3.1 Worker Prefereces Workers derive utility from fial goods ad residetial housig. The represetative worker i sector maximizes the Cobb-Douglas utility u = [ k {ag,a} C,k ) αε k ] S,h ) 1 α, 1) where C,k, = ag,a, are cosumptio of agricultural ad o-agricultural goods, respectively, ad S,h is housig structures. 3 The parameters α ad ε are preferece weights. Total cosumptio of k good i regio sector is C,k L. Households are subect to a budget costrait P,ag C,ag +P,a C,a +rs,h are cosumptio good prices, r the price of housig, ad v P,a v,ag, where P ad omial icome of a worker with hukou i regio ad sector. Let v =, v L /L be the average icome of workers i regio ad sector. The, it is straightforward to show that fial demad for good k i regio is Similarly, fial demad for housig is 1 α) v L. D k = αε k vl. 2), 3.2 Productio, Trade ad Goods Prices Agricultural ad o-agricultural goods are composites of a cotiuum of horizotally differetiated varieties y ν), = ag,a; v [0,1]. A perfectly competitive firm produces good usig the CES techology 1 σ/σ 1) Y = yν) dν) σ 1)/σ, 0 3 The homothetic prefereces greatly simplifies the aalysis. I the Appedix, we expad the model to allow subsistece food requiremets. All our key results go through with oly mior quatitative differeces. 14

16 where σ is the elasticity of substitutio across varieties. Each variety ν may be sourced from local producers or imported, whichever mimizes costs. The good Y is either cosumed directly by households or used as itermediate iputs by producers of differetiated varieties. These varieties are produced by perfectly competitive firms usig labor, itermediate iputs, ad lad with Cobb-Douglas techologies. The margial cost of productio for a firm with productivity ϕ is therefore c ϕ) 1 ϕ [ w β i r η i )] k {ag,a} P k σ k i, 3) where β ad η are labor ad lad shares, ad σ k share for itermediate iput from sector k such that β + η + k σ k = 1. Also, w is the wage, r the retal cost of lad, ad P k the price of itermediate iput from sector k, which is the same as the price of the fial good Y k. For otatio coveece, we follow Caliedo ad Parro 2012) ad defie c the term i the brackets i equatio 3. Producers i oe regio that sell to cosumers i aother icur a cost; those that sell withi a regio do t. Costs are iceberg ad oe must ship τ uts from regio i for oe ut to arrive i regio. Cosumer prices are therefore p ϕ) = τ c ϕ). Facig these prices, buyers opt for the cheapest source. As i Eato ad Kortum 2002), if ϕ is distributed Frechet with CDF F i ϕ) = e Ti ϕ θ the equilibrium trade shares are π = T i N+1 m=1 T m ) τ c θ i τ mc i ) θ, 4) where π is the fractio of regio spedig allocated to sector goods produced i regio i trade shares), ad fial goods prices are P = γ where γ is a costat, ad T i [ N+1 m=1 ] 1/θ Tm τ m cm) θ, 5) is a labor productivity parameter. I autarky, T is the 15

17 average labor productivity of firms i i, ). 4 Let X be total expeditures o good by regio. Total reveue is the R = N+1 π i X i. 6) i=1 Combied with demad for itermediates by producers, we have 3.3 Icomes of Workers X = D + k σ k R k. 7) Housig structures ad lad for productio are ot tradable ad are owed i commo by local residets. This assumptio is broadly cosistet with the istitutioal features of Chia, ad implies that migrat workers have o claim to fixed factor icome. As cosumer prefereces ad productio techologies are Cobb-Douglas, total spedig o the fixed factor is 1 α)v L + η R = 1 α)v L + η β 1 w L. Give a total fixed-factor edowmet of S, the market clearig coditio for the fixed-factor is r S = 1 α)v L +η β 1 w L. Add fixed-factor icome to labor icome we have v L = 1 α)v L +η β 1 w L +w L. Solvig for v L yields the followig equatio: v L = η + β αβ w L. Ad the total fixed-factor icome i regio sector is [ rs 1 α)β + η ] = αβ wl. 8) As oly workers with local hukou receive fixed-factor icome, the icome of a local worker i regio ad sector is v = w + rs /L ad the icome of a migrat 4 The labor productivity parameter reflects both TFP ad capital itesity. Bradt et al. 2013) show that average capital itesity does ot vary much across the Chiese provices, so spatial misallocatio of capital is ot a quatitatively importat issue ad we abstract it here. 16

18 worker is simply w. If we defie δ k i = α)β +η i f = i ad = k αβ L, 9) 1 i f i or k as the effective fixed-factor rebate rate to workers, the we ca write the icomes of workers i regio ad sector as v k i = δ k i w. Note this differs from rebates proportioal to wages Reddig, 2016) or lump-sum rebates Caliedo et al., 2017a) foud i the literature. Our assumptio is motivated by the actual lad owership istitutio i Chia, which has a importat egative effect o migratio. 3.4 Iteral Migratio Workers ca move across provices ad sectors withi Chia, but ot iteratioally. Each worker is registered to a provice ad assiged either a agricultural or a o-agricultural hukou. Let m k deote the share of workers registered i regio ad sector who migrated to regio i to work i sector k. There are costs to migratio. First, migrats must forego lad returs i their home regio, ad rely oly o labor icome aloe. Secod, workers icur a utility cost that lowers welfare by a factor µ k > 1. Fially, workers differ i their locatio prefereces εi k, which are i.i.d. across workers, regios, ad sectors. Give real wages Vi k P i workers from, ) maximize their welfare εi kδ k ) α r k 1 α i w k i / P ag ε a 1 ε i i all regios ad sectors, V i k k /µ. As εk i is a radom variable across a cotiuum of idividuals from, ), the law of large umbers esures that the proportio of the workers who migrate to regio i,k) is { m k = Pr z k i δ k V i k /µ k max z k i,k i δ k Vi k k /µ } ). This proportio ca be solved explicitly if prefereces over locatios follow a Frechet distributio with CDF F z x) = e x κ, where κ govers the degree of dispersio across idividuals. A large κ implies small dispersio. Propositio 1 Give real wages for each regio ad sector Vi k, migratio costs be- 17

19 twee all regio-sector pairs µ k, ad heterogeeous prefereces distributed F zx), the share of, )-workers that migrate to i,k) is m k = k N i =1 Vi kδ k ) k κ /µ ad total employmet at i,k) is L k i = N =1 m k L. Proof: See the appedix. V k i δ k /µ k ) κ, 10) While our assumptios about migratio costs are particular, they do ot drive the results. We could have alteratively modeled migratio costs as affectig worker productivity, or allowed for heterogeeous productivity. We explore these possibilities i the appedix, icludig cases where productivity is correlated or ucorrelated across regios ad sectors. Our mai quatitative results are robust to these differet modelig assumptios about migratio. 3.5 Solvig ad Calibratig the Model To ease our quatitative aalysis ad calibratio, we follow Dekle, Eato, ad Kortum 2007) ad solve for couterfactual chages. Let ˆx = x /x be the equilibrium relative chage i variable x i respose to some exogeous chage i model parameters. As this approach is icreasigly familiar i quatitative trade research, we provide the relevat expressios i the appedix. Here, we preset oly the relevat chages i aggregate welfare ad real GDP. Propositio 2 Give chages i migratio ad real icomes, the chage i aggregate welfare is Ŵ = N =1 ω ˆV ˆm 1/κ, 11) where ω L V m 1/κ is regio ad sector s itial cotributio to welfare. 18

20 Table 3: Calibrated Model Parameters ad Itial Values Parameter Set To Descriptio β ag,β a ) 0.40,0.24) Labor s share of output η ag,η a ) 0.13,0.02) Lad s share of output σ ag,a,σ a,ag ) 0.60,0.06) Itermediate iput shares α 0.87 Goods expediture share θ 4 Elasticity of trade κ 1.5 Elasticity of migratio π Data Bilateral trade shares m Data Bilateral migratio shares L Data Hukou registratios Notes: Displays model parameters, their targets, ad a descriptio. See text for details. Similarly, the chage i real GDP is Ŷ = N =1 φ ˆV ˆL, 12) where φ V L is the cotributio of regio ad sector to itial real GDP. Proof: See the appedix. Solvig the model i relative chages eases the model calibratio by elimiatig may fixed compoets of the model. Specifically, the model requires we calibrate parameters α,β,η,σ k,θ,κ) ad itial values π,mk, L i,v i ), othig more. I particular, we oly eed to kow the chages i trade ad migratio costs for all of our couterfactual aalysis; our results will ot be affected by ay potetial biases i our estimated trade ad migratio cost levels that are costat over time. Besides the two elasticities, θ ad κ, all the other parameters ca be easily calibrated. We briefly discuss our calibratio method here ad leave some more detailed discussio i the appedix. The itermediate iput shares σ k are directly calibrated usig Chia s Iput-Output Table. Because factor market are distorted i Chia, we choose the values of labor ad lad shares, β ad η, such that they are cosistet with the factor shares by sector reported by Caselli ad Colema 2001) for the US. See the appedix for a more detailed discussio about the calibratio of these productio share parameters.) For α, we use cosumer expediture data from 19

21 the most recet Chia Statistical Yearbook. The fractio of urba household spedig o housig is 11% ad for rural households is 15%. We set α = 0.87, implyig the housig share of expeditures is 13%. The total registrats by provice ad sector L ) ad itial migratio shares m k are observable i Chia s 2000 Populatio Cesus. Total atioal employmet for Chia is millio ad we ifer employmet for the rest of the world L N+1 ) at 2,103 millio usig the Pe World Table. Sice we do t have trade data i 2000, we use the trade shares geerated from the 2002 Chia Regioal Iput-Output Tables to approximate the values of the trade shares π i Fially, to get itial real icome per worker V, we use data o real GDP per worker by provice ad sector. I the model, trade balace esures the two are equivalet. 5 The calibratio results are summarized i Table Cost-Elasticity of Trade There is a large literature o the productivity dispersio parameter θ. This parameter govers productivity dispersio across firms ad, cosequetly, determies the sesitivity of trade flows to trade costs. Betwee-coutries, there are may estimates of this elasticity to draw upo. For example, Simoovska ad Waugh 2011) use cross-coutry price data to estimate θ 4. Parro 2013) estimates θ [4.5,5.2] for maufacturig usig trade ad tariff data. Based o this method, Tombe 2015) estimates θ = 4.1 for agriculture ad 4.6 for o-agriculture. Withi-coutries, however, there is little evidece to draw upo. Usig firm-level productivity dispersio i the US, Berard et al. 2003) estimates θ = 3.6. We set θ = 4 ad explore alterative values i the appedix Icome-Elasticity of Migratio We estimate the migratio elasticity empirically. The model implies the share of migrats from, ) to i,k), relative to the share of o-migrats who remai home, 5 I the appedix, we demostrate our results are robust to a alterative model where trade imbalaces match those observed i data. 20

22 is a fuctio of real wage differeces ad migratio costs, ) m k V k ) ) l m = κl i V κl δµ k. If icome differeces were exogeous, oe could estimate κ with a simple OLS regressio. But icome differeces are themselves iflueced by migratio flows ad factors that are potetially related to migratio costs such as remoteess or the istitutioal quality of the receivig regio. To deal with the edogeeity issue, we cosider two istrumets: 1) output exported iteratioally i 1997 ad 2) the average icome of eighborig provices. The ustificatios for the istrumets are as follows. Previous export-orieted provices will ted to have high icomes, but migratio i 2000 will ot itself affect past export behavior. Also, a regio whose eighbors have high-icome will ted to have high icome, but the eighbor s icome is plausibly exogeous to a give regio s migratio or icome shocks. Of course, either of these istrumets address cocers over uobserved istitutioal factors that affect a regio s attractiveess to migrats, icome, ad export share simultaeously especially if such uobserved factors are spatially correlated. Usig regio or regio-pair fixed effects i our regressio mitigates such cocers. We provide the resultig estimates for κ i Table 4 Though our estimates vary slightly, we opt to set κ = I the appedix we show our mai quatitative results are robust to a variety of values for κ betwee 1 ad 3. For cotext, there are recet estimates i similar models applied to the Uted States. Fagelbaum et al. 2015) use variatio i state-level taxes to estimate the elasticity of migratio. They specifically use a distace-weighted average of tax rates i all other states to istrumet for each state s ow taxes. While their estimates vary across a umber of model specificatios, the closest to our setup correspods to κ = I additio, i the appedix we explore a alterative model where workers differ i productivity rather tha prefereces. I that settig, κ maps directly ito observable momets of the idividual eargs distributio ad there is o eed to estimate the icome-elasticity of migratio. With the idividual 6 Usig migratio shares i 2005, we fid similar estimates with κ [1.09,2.22]. 21

23 Table 4: Estimates of the Icome-Elasticity of Migratio i Chia OLS 1997 Export Shares Neighborig Icome 1) 2) 3) 4) 5) 6) Icome Gap 1.47*** 1.39*** 1.98*** 1.59*** 1.45*** 1.45*** [0.05] [0.05] [0.08] [0.05] [0.06] [0.05] Distace -1.18*** -1.18*** -1.18*** [0.05] [0.05] [0.05] Costat -0.84** -9.86** -0.84** -9.80*** -0.84** -9.84*** [0.33] [0.64] [0.33] [0.55] [0.33] [0.55], i)-fes No Yes No Yes No Yes Obs First-Stage: Dep. Var. is Icome Gap 1997 Export Shares 0.22*** 0.38*** [0.004] [0.005] Neighborig Icome 0.44*** 0.44*** [0.005] [0.002] F-Stat Note: Displays the results of various regressios of the form lm /m ) = κlv i /V ) κlµ ). The first IV uses 1997 exports as a fractio of each provice s total output to istrumet for icome i The secod IV uses the distace-weighted average icome of all other provices. IV eargs data reported i Chia s 2005 populatio cesus, we fid κ rages from 1.5 to 2.5, depedig o the specificatio. 4 Iferrig Trade ad Migratio Costs I this sectio, we quatify the migratio ad trade costs withi Chia ad trade costs betwee Chia s provices ad the rest of the world. 4.1 Migratio Costs Equatio 10 provides a simple represetatio of migratio decisios through which we ifer migratio costs. Specifically, give migratio shares ad itial real i- 22

24 comes, migratio costs are µ k = 1 δ V k i V ) ) m 1/κ m k f or i, ad where δ ca be calculated directly from Equatio 9. We fid that µ k averages aroud 3.6 i We summarize these costs, their chages betwee 2000 ad 2005, ad the itial migrat stocks i 2000 i Table 5. Overall, migratio costs are the largest for those switchig both sectors ad provices with a average itial cost of early 38. I cotrast, those switchig sectors withi oe s home provice icur average costs of 2.9. For direct moetary costs of migratig the loss of implicit fixed factor icome from their home regio we fid that δ averages aroud 1.5. Thus, the overall cost of migratig δµ k averages aroud 5.3. How do these costs chage over time? We report the chage i average migratio costs i the last colum of Table 5. Overall, migratio costs declied to 84% of their itial level. Costs to switch provices fell the most, from 32.6 to Sectoral switches withi a worker s home regio also fell a lot, from 2.9 to 2.4. We use the full set of migratio cost chages betwee all provice-sector pairs i the quatitative aalysis to come. While the estimated migratio costs are ideed large, they are ot ureasoable. These magtudes are cosistet with idividual survey data from the 2002 Chia Household Icome Proect. This survey asks Rural-Urba migrats both what they curretly ear ad what they could ear if they were still i their home village. The typical respodet ears roughly 4 times what they believe they would ear i their home village, suggestig substatial migratio costs exist to maitai such a large gap. Ad these are costs for workers who did migrate. There is substatial heterogeeity across workers, ad the costs faced by those who did ot migrate could be much higher. Cesus 2005 provides sufficiet data so that we ca estimate µ k by age. Give a regio ad sector s real icome per worker V, we ca apportio this across workers i age-cohort ) c based o observed relative wages. Specifically, defie V,c = w,c/ w,c V where w is the employmet-weighted average wage across cohorts i the data. Give V,c ad cohort-specific migratio shares m k,c, 23

25 Table 5: Migratio Rates ad Average Migratio Costs Migratio Costs µ k Itial Share Level i Level i Relative of Employmet Chage Overall Agriculture to No-agriculture Migratio Cost Chages Overall Withi Prov Betwee Prov Betwee Provices Migratio Cost Chages Overall Withi Ag Withi Noag Notes: Displays migratio-weighted harmoc meas of migratio costs i 2000 ad The migrat share of employmet summarizes m k i We use itial period weights to average the 2005 costs to capture oly chages i costs ad ot migratio patters. we estimate cohort-specific migratio costs µ k,c. We fid cross-provice migratio costs average 2.15 amog those uder age 24, sigficatly lower tha the average costs across all age groups that we reported i Table 5. 7 We should ote that part of our measured migratio costs may also be attributed to workers locatio prefereces that vary systematically across provices. For example, if amety values are higher i some of the low icome provices tha other high icome provices, the holdig migratio costs costat, there would be less migratio from these low icome provices to high icome provices. Sice we have assumed i our model o amety differece across provices, the model implied migratio costs would be upward biased. However, sice all of our quatitative exercises will be about the impact of the chages i migratio costs, this bias i the level of migratio costs will ot affect our results as log as the systematic locatio differeces do ot chage over time. 7 We caot icorporate age-specific costs ito our aalysis, as idividual icome data are ot available i Cesus We therefore caot estimate chages i those costs. We report the levels here for perspective. 24

26 4.2 Modified Head-Ries Idex of Trade Costs To estimate trade costs, we follow Head ad Ries 2001) to back-out trade costs betwee regio ad i for sector goods usig oly observable trade shares ad the trade-cost elasticity θ. Specifically, ) τ τ τ i = ππ 1/2θ ii π π, 13) i which is a direct result of equatio 4 but geeralizes to a broad class of trade models. This method has a umber of advatages. I particular, τ is ot affected by trade volumes or by third-party effects ad applies equally well whether trade balaces or ot. Ufortuately, these trade cost estimates are symmetric i the sese that goods movig from i to is as costly as movig goods from to i. This matters, as Waugh 2010) demostrates that iteratioal trade costs systematically differ depedig o the directio of trade. To capture this, we presume trade cost asymmetries are exporter-specific such that τ = t t i, where t are symmetric costs t = t i ) ad t i are costs of exportig. This ad equatio 13 imply a Adusted-Head-Ries Idex τ = τ t i /t, as i Tombe 2015). To estimate asymmetric compoets of trade costs withi Chia, we closely follow the existig iteratioal trade literature ad therefore leave details to the appedix. Essetially, we use a stadard gravity regressio to ifer asymmetries from fixed effects. Overall, we fid that poor regios face the highest exporterspecific trade costs cosistet with existig cross-coutry evidece ad fid they are largely uchaged from 2002 to We report the relative chages i our trade cost estimates i Table 6; some otable patters emerge. Iteral trade costs are largely decreasig, with trade-weighted chage i trade costs withi Chia ag of ˆτ = 0.87 ad ˆτ a = For trade betwee Chia ad the world, the average ag chage i costs were ˆτ = 0.77 ad ˆτ a =