A Passage to India: Quantifying Internal and External Barriers to Trade

K.7 A Passage to India: Quantifying Internal and External Barriers to Trade Van Leemput, Eva Please cite paper as: Van Leemput, Eva (2016). A Passage to India: Quantifying Internal and External Barriers to Trade. International Finance Discussion Papers 1185. https://doi.org/10.17016/IFDP.2016.1185 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1185 November 2016

BoardofGovernorsoftheFederalReserveSystem InternationalFinanceDiscussionPapers Number1185 December2016 APassagetoIndia: QuantifyingInternalandExternalBarrierstoTrade EvaVanLeemput NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other thananacknowledgmentthatthewriterhashadaccesstounpublishedmaterial)shouldbecleared withtheauthororauthors. RecentIFDPsareavailableontheWebatwww.federalreserve. gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science ResearchNetworkelectroniclibraryatwww.ssrn.com.

APassagetoIndia: QuantifyingInternalandExternalBarrierstoTrade EvaVanLeemput* Abstract: This paper quantifies the size of internal versus external trade barriers and assesses the impact on trade and welfare. I develop a quantitative multi-sector international trade model featuring nonhomothetic preferences in which states trade both domestically and internationally. I discipline the model using rich micro data on price dispersion as well as foreign and domestic tradeflowsattheIndianstatelevel. Ifindthat(1)state-basedpricedatapredictinternaltradeflows well; (2) internal trade barriers make up 40% of the total trade cost on average, but vary substantiallybystatedependingonthedistancetotheclosestport;and(3)thewelfareimpactsofdomestic integration are substantial: reducing trade barriers across states to the U.S. level increases welfare bymore(13%)thanfullyeliminatinginternationalimportbarriers(7%). Keywords: InternalTradeBarriers,ExternalTradeBarriers,Welfare,India JELclassification: F13,F14,O18,O24. *The author is a staff economist in the Division of International Finance, Board of Governors of the Federal Reserve System, Washington, D.C. 20551 U.S.A. The views in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or of any other person associated with the Federal ReserveSystem. Contact: eva.vanleemput@frb.gov. IwouldliketothankJosephKaboski,WyattBrooks,KevinDonovan,TimothyKehoe,andNelson Mark for valuable comments, suggestions, and support. I also thank seminar participants at Notre Dame, the Federal Reserve Bank of Chicago, the Federal Reserve Bank of Minneapolis, the IMF, USITC, the Federal Reserve Board, Toulouse School of Economics, Goethe, Williams College, UC Louvain, VU Amsterdam, Florida International U, Delaware, Vienna, Georgetown SFS for helpfulcommentsandsuggestions.

1 Introduction International barriers to trade, and the benefits from reducing them, have been both a policy and research focus for decades. But recently, policy has focused more on domestic trade barriers and their role in improving the economy. For example, in 2014, the World Bank increased their financing commitments to roads, bridges, energy, and other infrastructure projects by 45%, to $24 billion, compared to 2013. However, the question remains whether domestic integration has a largereffectontradeandwelfarethanreducinginternationaltradebarriers. This paper quantitatively addresses two main questions: (1) how large are internal versus external trade barriers in India? and (2) how large are the welfare impacts of internal compared to external integration? I contribute to the literature in two ways. First, whereas the literature has mostly focused on estimating the impact of decreasing either internal or external trade barriers, this paper quantifies both barriers in a unified framework and compares the welfare impacts of relaxing them. Second, this framework allows for a richer deconstruction of different types of internal trade barriers. With respect to foreign market access, I model an additional barrier–an internalbarrierofreachinganinternationalport. Withrespecttodomestictrade,themodeldeconstructs two types of trade barriers in India: across–state and rural–urban within state. Therefore, as a secondary question I ask: does India have more to gain from better integrating regions across orwithinstates? Idevelopaquantitative,two-sector,internationaltrademodelthatexplicitlyaccountsforinternal trade barriers. I model India as a multistate country in which each state trades both with other states and with the rest of the world. Each state consists of a rural and an urban region, which produces agricultural and manufacturing goods, respectively, and consumers have nonhomothetic preferences for these goods. The model’s innovation is twofold. The first is to include domestic trade barriers in an international trade framework. More specifically, I model two types of domestic trade barriers in India: a trade barrier across Indian states and one between the urban and rural region within an Indian state. The second is to allow states within India to have different internationalportaccess. Standardinternationaltrademodelsimplicitlyassumethattheentirepopulation has free access to an international port. In this paper, however, I allow for Indian states with and 1

withoutinternationalports. Thisimpliesthatstateswithoutaninternationalportfaceanadditional domestic trade barrier when trading internationally: the cross–state barrier to move goods to and fromthenearestport. To map the model to the Indian economy, I combine two rich micro-level datasets: (1) the State Movement/Flows of Goods and (2) the Foreign Trade Statistics for India. Together, they detail good-specific trade flows bilaterally between Indian states and internationally between each Indian state and the world. On the export side, it includes the state of origin, the destination for cross–state trade, and the port of exit for international trade. For imports, it includes the state of origin in the case of cross–state trade and the port of entry for international trade. These data overcome two major common issues, namely, distinguishing domestic from international trade flowsandidentifyingthespecificportofinternationaltrade. Todisciplinethetradebarriersinthemodelbothwithinandacrossstates,Iusedatacontaining measuresofpricedispersionofwholesaleagriculturalpricespublishedbytheDataPortalofIndia. This data set consists of daily state-, market-, and variety-specific wholesale prices of 50 different agricultural goods across 1,831 markets in all Indian states. Using a no-arbitrage argument, I compute cross–state trade barriers for all agricultural prices observed in urban markets, and I find they are five times larger than in the United States, which is around the same magnitude as estimated using micro evidence for other developing countries (Atkin and Donaldson, 2016). To discipline the rural–urban trade barriers in each state, I use the same no-arbitrage argument using thepricevariationacrossurbanandruralmarketswithinastate. Ifindthatthemedianrural–urban trade barrier is almost double the size of the median cross–state trade barrier in the United States. Finally, I compute international trade barriers by matching state-wise international import shares foragricultureandmanufacturing. Using price dispersion to discipline trade barriers instead of direct measures such as distance has the advantage of encompassing more than just infrastructure barriers. In fact, I find that the pricedispersionishighlycorrelatedwithdistancebutalsowithnon-infrastructuralbarrierssuchas the level of corruption across Indian states, the arduousness of the tax administration, and the tax level. In addition, I show that price dispersion both across and within states predicts internal trade flowswell. 2

First,Ifindthat,onaverage,internaltradebarriersinIndia,whichincludeboththerural–urban and cross–state barriers, make up 40% of the total trade barrier, but there is a large heterogeneity across states: the 90–10 percentile is 70–13%. Most of the heterogeneity is driven by port versus non-port states, where internal barriers make up 17% and 51% of the total trade barrier, respectively. Second, I find larger welfare gains from domestic integration in India than international; reducingtradebarriersbetweenIndianstatestotheU.S.levelincreaseswelfareby13%comparedto a welfare gain of 7% when fully eliminating international import barriers. Finally, deconstructing internal trade barriers in India, I find larger welfare gains associated with removing cross–state barriers than rural–urban barriers. To summarize, I find that (1) India has more to gain from becomingmoreinternallyintegratedthanfromfullyeliminatinginternationalimportbarriersand(2) thelargestinternalgainscomefrombetterconnectingIndianstates. Related Literature ThispaperbuildsontheseminalmodelofgeographyandtradeofEatonandKortum(2002)andon Fieler’s (2011) two-sector extension, and it contributes to two main literatures. The first examines the size, nature, and impact of international trade barriers in explaining international trade patterns. Waugh(2010) studiesthesizes oftrade barriers acrossdeveloped anddevelopingcountries. He finds that trade frictions between rich and poor countries are systematically asymmetric, with poorcountriesfacinghighercoststoexportrelativetorichcountries. Tombe(2015)alsoestimates larger trade costs for developing countries, but argues that these countries face especially large trade costs in agriculture. Regarding the components of international trade barriers, Lima˜o and Venables (1999) find that infrastructure is an important determinant of transport costs, especially for landlocked countries. They estimate that halving transport costs would increase trade volume by a factor of five. Hummels and Schaur (2013) estimate that time in transit poses a large international trade barrier. They find that each day in transit is equivalent to an ad valorem tariff of 0.6 to 2.3%. The focus of this literature is on border costs. My paper complements these works by disentangling border costs from intracountry costs and assessing their respective impact on trade 3

andwelfare. Thesecondisanemergingliteratureexaminingtheroleofinternaltradecostsontheproduction andtradepatternsindevelopingcountries. Basedondetailedmicrodata,Donaldson(2016),Atkin andDonaldson(2016),Allen(2014),Asturiasetal.(2016),andAdamopoulos(2011)quantifythe sizeandnatureofintracountrytradefrictionsfordevelopingcountriesbutdonotcomparethemto international trade frictions. Other work has estimated the regional effects of differential market access within countries on the gains from trade. Behrens et al. (2006) find that for regions that are “geographically disadvantaged” with high intracountry trade costs, being remote acts as a barrier tocompetitionfromabroad. Storeygard(2016)showsthatregionsinsub-SaharanAfricawithbetter access to port cities benefit more from terms of trade shocks. Coc¸ar and Fajgelbaum (2016) developaframeworkforthespatialdistributionofeconomicactivityinanarbitrarytopographyof trade costs where products are differentiated by origin. They find that in the presence of high domestic frictions, international trade creates a partition between a commercially integrated coastal region with a high population density and an interior region where mobility is limited. Gollin and Rogerson (2014) examine the interaction of intracountry trade frictions and employment patterns between agricultural and manufacturing sectors, taking as given that developing countries import very little agriculture. My paper shows the effect of these intracountry trade frictions on international trade patterns. The distinctive aspects of this paper are the state-based domestic and international trade data and the state-based characteristics. Previous studies have differentiated intracountry trade flows, but only aggregated international trade flows. In this paper, I can distinguish international trade by state and also through which ports trade happens. As I will show, this is quantitatively important in explaining aggregate trade flows and in determining the effect of internationaltradepoliciesintermsofheterogeneityacrossstates. Moreover,thisframeworkallows metocomparedomesticandinternationalintegrationandtheirwelfareimplications. Thequestion arises, what does India have to gain from trading with the rest of the world versus becoming more integrateditself? The remainder of the paper is organized as follows. Section 2 outlines the international trade model. Sections 3 and 4 describe the data and the quantitative analysis, respectively. Section 5 presentstheresults. Section6presentsadditionalrobustnesschecks. Section7concludes. 4

2 Model The model extends the Eaton-Kortum (2002) framework to include many states within a country that consist of both a rural and urban region. More concretely, I model trade between a multistate country and a second uniform country representing the rest of the world. The quantitative part is basedonIndiaand,therefore,themodelispresentedaccordingly. 2.1 Environment India consists of multiple states denoted by s = 1...S. Each state consists of two regions, rural and urban, indexed by r = {R,U}, where R denotes the rural and U the urban region. This economy produces two types of goods, agriculture and manufacturing, indexed by g ∈ {a,m}, where a denotes agricultural and m manufacturing goods. All agricultural production takes place intheruralregionandallmanufacturingproductionintheurbanregion.1 Foreachgood,thereisa continuum of individual varieties, j , defined on the interval [0,1]. Both good types are produced g withRicardiantechnologyandrequireonlylaborforproduction,whichisimmobileacrossregions andstates.2 Iassumethattheshareofworkersinruralandurbanregionsisβ ∈ [0,1]and(1−β ), s s respectively. Internal trade for all agricultural and manufacturing varieties in India consists of two types: (1)within–statetradebetweenruralandurbanregionsand(2)cross–statetrade. 2.1.1 Consumption Arepresentativehouseholdwithconstantrelativeincomeelasticity(CRIE)preferences,asinFieler (2011),choosesthequantityconsumedofbothagriculturalandmanufacturinggoodsofvarietyj , g denotedby{q(j )} , g ∈ {a,m}tomaximizeutility: g jg∈[0,1] 1 1 (cid:18) (cid:19)(cid:90) (cid:18) (cid:19)(cid:90) σ (cid:104) (cid:105) σ (cid:104) (cid:105) U s r = σ − a 1 q(j a ) σa σ − a 1 dj a + σ m −1 q(j m ) σm σm −1 dj m (1) a m 0 0 1This is consistent with the data: in India, on average, 69% of the population is located in the rural region, of which91%areemployedinagriculture. 2Cross–statemigrationislowinIndia. The2001Censusreportsthattotalcross–statemigrationisaround1.6%, ofwhich43%isduetomarriage. Iallowforrural–urbanlabormobilityinSection6.1,andIfindthatthequalitative resultsarerobust. 5

subjectto 1 1 (cid:90) (cid:90) qr(j )pr(j )dj + qr(j )pr(j )dj = wr, s a s a a s m s m m s 0 0 where wr is the regional wage in state s. The good-specific parameter σ > 1 represents the s g elasticity of substitution across varieties. It also represents the income elasticity. Define xr as g,s the expenditures on a good of type g ∈ {a,m} in region r and state s. Then, from the first-order conditionsfromthehouseholdproblem,expenditurestowardagriculturerelativetomanufacturing satisfy xr (cid:0) Pr (cid:1)1−σa a,s = (λr)σm−σa a,s , xr s (cid:0) Pr (cid:1)1−σm m,s m,s where λr is the Lagrange multiplier on the budget constraint and the shadow value of income. s It is strictly decreasing in the consumer’s total income. If σ − σ > 0, then expenditures on m a agriculture relative to manufacturing, xr a,s, increase with λr and, therefore, decrease with income. xr s m,s In other words, as a household’s income increases, the percentage of spending on manufacturing goodsincreasesrelativetoagriculturalgoods. Hence,CRIEpreferencesallowforconsumerswith different income levels to concentrate their spending on different types of goods and, therefore, allowformatchingcross-countryagriculturalandmanufacturingconsumptionpatterns. 2.1.2 Production Eachregiondrawsagood-specificproductivityz fromastate-specificdistributionforallvarieties s of the good produced in that region, that is, rural regions only have productivity draws for all agricultural varieties and the urban region only for all manufacturing varieties. As in Eaton and Kortum(2002),IassumethisdistributionisFrchet: (cid:0) (cid:1) z (j ) ∼ F (z) = exp −T z−θg s ∈ {a,m}. (2) s g s g,s s TheparameterT isstate-andgood-specificandgovernstheabsoluteadvantage: highervaluesof g,s T indicateproductivitydrawsfromahighermeandistribution. Thegood-specificparameter,θ , g,s g governs the comparative advantage. Higher values imply that there is a smaller range of productivity draws. For example, for a high θ , agricultural productivity varies little across states, which a 6

implies small gains from trade. Production in both sectors has constant returns to scale (CRS). Because labor is immobile, wages are both state- and region-specific. Denoting the regional wage instatesbywr,theunitcostforgoodvarietyj is s g wr s . z (j ) s g Unit costs are higher as wages increase or productivity decreases. Next, I describe the three types oftrade: (1)rural–urbantradewithinastate,(2)cross–statetrade,and(3)internationaltrade. 2.1.3 Rural–UrbanTradewithinState WithineachstateinIndia,productionisregion-specific. Hence,tradewithinastateoccursbetween the urban and rural region. I assume that consumers in the rural region can buy agricultural goods attheunitcostofagriculturalproduction. However,iftheywanttobuymanufacturinggoods,those need to be transported from the urban to the rural region at an iceberg transportation cost, δ > 1. s Consumers in the urban region can buy manufacturing goods at the unit cost of manufacturing production,butbuyingagriculturalproductsrequiresthesameicebergtransportationcost,δ . Note s thattheseicebergtransportationcostsarestate-specificandIwillrefertothemasrural–urbantrade barriers. Given these barriers, prices for both goods are different across regions. Table 1 presents theregionalpricesforagriculturalandmanufacturinggoodsinstates. Table1: PriceswithinState Agriculture Manufacturing R U w w δ s s s Rural pR(j ) = pR(j ) = s a z (j ) s m z (j ) s a s m R U w δ w s s s Urban pU (j ) = pU (j ) = s a z (j ) s m z (j ) s a s m 7

2.1.4 Cross–StateTrade I also model trade across states in India. I assume that they can only trade via their urban regions at an additional cross–state iceberg transportation cost, d > 1, which represents the cost of sl importingfromtheurbanregioninstatel tothatofstates. ThisimpliesthatifthestateofGujarat ships rice to the state of Rajasthan, the rural region in Gujarat first needs to ship the rice to its urban region at a rural–urban trade barrier δ , after which it can be shipped to the urban region G of Rajasthan at a cross–state trade barrier d . If the rice is consumed in the urban region of RG Rajasthan, the total barrier is δ ∗ d . If the rice is consumed in the rural region of Rajasthan, G RG it needs to be transported at an additional rural–urban trade barrier of δ , and the total barrier is R δ ∗d ∗δ . This entire barrier is what I will refer to as the internal trade barrier that includes G RG R thecross–statetradebarrierandtherural–urbantradebarrier. Table2: Cross–StateTradeBarriers FromState1 Rural Urban Rural δ ∗d ∗δ δ ∗d 1 21 2 1 21 ToState2 Urban d ∗δ d 21 2 21 Within India, the four scenarios that can occur are detailed in Table 2, which summarizes the total trade barrier of shipping one unit of variety j from state 1 to state 2. Section 4 explains g how these barriers are separately identified. This concludes the multistate economy model. Next, Idescribethethirdtypeoftrade: internationaltrade. 2.1.5 InternationalTrade AllIndianstatesnotonlytradewitheachother,butalsowiththerestoftheworld(RoW),modeled as one country. The preferences and production processes for both types of goods in the RoW are also described by (1) and (2). As is standard in the literature, international trade is subject to an 8

international border iceberg transportation cost, which is good-specific and depends on whether India is the importer or exporter.3 Delivering one unit of a good-specific variety j from the RoW g to India requires the production of τ > 1 units, and delivering one unit of j from India to the g,imp g RoWrequirestheproductionofτ > 1units. g,exp The innovation in the model is that international trade can only occur through international ports. The trade literature typically assumes that the entire population consumes and produces at the port, whereas this paper distinguishes Indian states as having an international port or not. This isquantitativelyimportantalongtwodimensions. First, it is important for the price of imported goods. Consider an Indian state with an international port, Gujarat. If Gujarat imports rice from the RoW, it incurs a border import barrier, τ . If the rice is consumed in the urban region of Gujarat, the total tradebarrier is τ . If the a,imp a,imp rice is consumed in the rural region, it needs to be transported at an additional rural–urban trade barrier, δ , making the total trade barrier τ ∗δ . Now consider a non-port state, Rajasthan. If G a,imp G RajasthanimportsricefromtheRoW,itfirstneedstoimportthegoodthroughtheportofGujarat, whichistheclosestport,ataborderimportbarrierτ . Then,thericeisshippedfromGujaratto a,imp Rajasthanatanadditionalcross–statebarrierofd . Ifthericeisconsumedintheurbanregionof RG Rajasthan,thetotaltradebarrierisτ ∗d . IfitisconsumedintheruralregionofRajasthan, a,imp R,G itsrural–urbantradebarrier,δ ,isincurredaswell,andthetotaltradebarrierisτ ∗d ∗δ . R a,imp RG R Second,itisimportantforthepriceofexportedgoods. Considerthesameexample. IfGujarat exports rice, it first needs to be transported from the rural to the urban region with a rural–urban tradebarrier,δ ,afterwhichitisexportedwithaborderexportbarrier,τ ,makingthetotaltrade G a,exp barrierδ ∗τ . IfRajasthanexportsrice,similarly,itfirstneedstobetransportedfromtherural G a,exp to the urban region with a rural–urban trade barrier, δ . Rajasthan does not have an international R port. Thus, the rice needs to be transported to the urban region of the nearest port, Gujarat, with a cross–statebarrier,d ,afterwhichitisexportedtotheRoWwiththeborderexportbarrierτ . G,R a,exp The total trade barrier is δ ∗ d ∗ τ , which is weakly higher than if Rajasthan would have R GR a,exp hadaninternationalport. 3Thebordertradebarriersaregood-specifictobeconsistentwithTombe(2015)andIallowforimportandexport coststodiffer,thatis,τ (cid:54)=τ tobeconsistentwithWaugh(2010) g,imp g,exp 9

Thus, differential international market access affects not only import but also export prices. Moreover,amoreflexiblelocationmodelofconsumptionandproductionallowsforheterogeneous effectsontradeandwelfareasaconsequenceofinternalandexternaltradeliberalization. Tosummarize,defineDr asthetotaltradebarrierfacedinregionr instatesforgoodvariety g,sl j imported from state l. Dr depends on the region of consumption: rural or urban, the type g g,sl of good: agriculture or manufacturing, and whether state s has an international port. Define state t as a non-port state and state s as its nearest port state in India. The total barriers to trading internationally for state s and t are summarized in Table 3, which consist of the border trade barriersandthedomestictradebarriers. Table3: TotalInternationalTradeBarriers PortStates Non-PortStatet Agriculture Manufacturing Agriculture Manufacturing Imports Rural δ ∗τ δ ∗τ δ ∗d ∗τ δ ∗d ∗τ s a,imp s m,imp t ts a,imp t ts m,imp Urban τ τ d ∗τ d ∗τ a,imp m,imp ts a,imp ts m,imp Exports Rural δ ∗τ τ δ ∗d ∗τ d ∗τ s a,exp m,exp t st a,exp st m,exp Urban δ ∗τ τ δ ∗d ∗τ d ∗τ s a,exp m,exp t st a,exp st m,exp Tocomputetradeflows,Ineedtodefineregion-specificpriceswithinstatesgiventhepreviously outlined trade barriers. Following the literature, I assume that markets are perfectly competitive.4 Then,thepriceforgoodvarietyj inregionr instatesis g pr(j ) = min{pr (j );l = 1...S,RoW}, s g sl g wherepr (j )variesacrossregionsandstates. Pricescanbesummarizedasfollows: sl s (cid:26)wrDr (cid:27) pr(j ) = min s g,sl ;l = 1...S,RoW . s g z (j ) s g 4GiventhatIdisciplinetradebarriersusingpricedata, anypricedifferenceinthedatacouldalsoreflectmarket poweracrossstates. Therefore,thisassumptionisnotasstrictinthedata. 10

Notethatifbothrural–urbanandcross–statetradebarriersaresettoone,themodelreducestothe EatonandKortum(2002)model. 2.2 Equilibrium (cid:8) (cid:9) Foragivensetofvaluesfor{β } ,{T } , Dr ,and{L } s s=1...S,RoW g,s s=1...S,RoW g,sl s,l=1...S,RoW s s=1...S,RoW (cid:8) (cid:9) anequilibriumisasetofregion–good–state-specificpriceindexes, Pr ,region–stateg,s s=1...S,RoW specificwages,{wr} ,andgood-specificbilateraltradeflows,{X } ,where s s=1...S,RoW g,sl s,l=1...S,RoW g ∈ {a,m}andr ∈ {R,U},suchthat: 1. Sector–state-specificproductivitiesfollowaFrchetdistributiongivenby(2). 2. Consumersmaximizeutility(1). 3. Consumerspurchasefromthe(tradecostinclusive)minimumcostproducer. 4. Labormarketsclear,thatis,totalshipmentsfromstatesequaltotalproductioninstates. AllIndianstatesandtheRoWhaveacontinuumofindividualswhoinelasticallysupplyoneunitof labor. LetL bethepopulationofstates,ofwhichβ L isemployedinagricultureand(1−β )L s s s s s in manufacturing where β ∈ [0,1]. The price index associated with the objective function (1) is s Pr = (cid:20) Γ (cid:18) θ g +1−σ g (cid:19)(cid:21) 1− 1 σg (cid:34) (cid:88)(cid:104) T (cid:0) wrDr (cid:1)−θg (cid:105) (cid:35)− θ 1 g , (3) g,s θ g,l l g,sl g l where Γ is the gamma function. A necessary condition for a finite solution is (θ + 1) > σ , g g g ∈ {a,m}. Thespendingofaconsumerinregionr instatesongoodg is xr = (λr)−σg (cid:0) Pr (cid:1)1−σg , g,s s g,s where λr > 0 is the region- and state-specific shadow value of income. This is implicitly deters mined by the budget constraint in each region: (cid:80) xr = wr. Following the methods in Eaton g,s s g∈{a,m} and Kortum (2002), the probability that state l provides a good at the lowest price in region r in 11

states,πr ,is g,sl xr T (cid:0) wrDr (cid:1)−θg πr = g,sl = g,l l g,sl , (4) g,sl xr Φr g,s g,s with Φr = (cid:80) T (cid:0) wrDr (cid:1)−θg. To close the model I need to solve for productivity parameters g,s g,l l g,sl l {T } by assuming that labor markets need to clear at given wages. Total agricultural imports in g,s statesfromstatel arethesumofruralandurbanimports: X = xR β L +xU (1−β )L , (5) a,sl a,sl s s a,sl s s whereβ ∈ [0,1]isthelaborshareintheruralregion. Totalagriculturalproductioninstatel is s wRβ L . l l l Byequatingsupplytodemand,Igetthelabormarketclearingconditionintheruralregion: (cid:88) wRβ L = X . (6) l l l a,sl l Similarly,totalmanufacturingimportsinstatesfromstatelarethesumofruralandurbanimports: X = xR β L +xU (1−β )L . (7) m,sl m,sl s s m,sl s s Totalmanufacturingproductioninstatel is wU (1−β )L . l l l Byequatingsupplytodemand,Igetthelabormarketclearingconditionintheurbanregion: (cid:88) wU (1−β )L = X . (8) l l l m,sl l Productivityparameters{T } canbecomputedbysolvingequations(6)and(8).5 g,s s=1...S,RoW This completes the statement of the model. To summarize, an economy is defined by a set of S IndianstatesandtheRoW,eachwithtworegions: aruralandanurbanregion. Theruralregion, with a population β L , only produces agricultural goods, and the urban region, with population s s 5TheruralregionoftheRoWwaschosenasthenumeraire. 12

(1−β )L , only produces manufacturing goods. Each region within a state has a specific wage s s wr andalocationimpliedbytradebarriersDr . Goodg ∈ {a,m}ischaracterizedbytechnology s g,sl parameters T and θ . An equilibrium is a set of productivity parameters {T } that g,s g g,s s=1...S,RoW satisfythemarketclearingconditions(6)and(8). 3 Indian Trade Data and Patterns The analysis is based on annual state-level data collected by the Directorate General of Commercial Intelligence and Statistics (DGCIS) from April 2011 to March 2012.6 There are two main datasets that describe internal and external trade flows in India. The first is the “Inter-State Movements/FlowsofGoodsbyRail,RiverandAir.” Thisdatasetcontainsstate-leveltradeflowsamong 27 states and 3 union territories in India. The second is the “Foreign Trade Statistics of India,” which contains international import and export data by country and details the ports of entry and exitinIndia. 3.1 Indian Interstate Trade The DGCIS collects data on intra-Indian trade flows in the “Inter-State Movements/ Flows of Goods by Rail, River and Air,” which provides data on the movement of goods by rail, river and air across 37 trade blocks in India for 70 good categories, which represent the main traded goods. These trade blocks contain 27 states, three union territories, and seven major port blocks.7 The trade statistics are collected through invoices of commodity consignments dispatched from each railway station to other trade blocks. Each railway and steamer company consolidates the figures withrespecttothestationswithwhichitisconcernedandsubmitsquarterlyreturnstotheDGCIS.8 6ThedatawaspurchasedfromtheDGCIS(http://www.dgciskol.nic.in/). 7The 27 states are Andhra Pradesh, Assam, Arunachal Pradesh, Bihar, Chattishgarh, Goa, Gujarat, Haryana, Himachal Pradesh, Jammu and Kashmir, Jharkhand, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Manipur, Meghalaya, Mizoram, Nagaland, Orissa, Punjab, Rajasthan, Tamil Nadu, Tripura, Uttarakhand, Uttar Pradesh, and WestBengal. ThethreeunionterritoriesareChandigarh,Delhi,andPuducherry,andthesevenmajorportblocksare Andhra Pradesh, Gujarat, Maharashtra, Karnataka, Kerala, Tamil Nadu, and West Bengal. Figure A.1 in the online AppendixshowsamapofIndiawiththeallthestatesandunionterritoriesincludedinthisanalysis. 8Indian states have a strong incentive to monitor these interstate flows given that they are subject to the Central Sales Tax (CST). All of the revenue accruing under the CST is collected and kept by the state in which the sale originates. 13

These data are unique, as they provide greater insight into domestic trade flows, which are rare, especially in developing countries. A second great advantage of these data is that they allow differentiating trade between Indian states from trade between Indian states and the rest of the world. For example, I can distinguish trade flows that originate in Rajasthan and are destined for the stateof Gujarat fromtrade flows thatoriginate in Rajasthanand are destinedfor the rest ofthe world but pass through the ports of Gujarat. The same is true for internal versus external imports. Typically, intracountry trade data do not make this distinction.9 However, for the purpose of this study, it is important to be able to distinguish these trade flows and in fact, I find that doing so is quantitativelyimportantwhenassessingthegainsfromtradeliberalization. There are, however, two caveats when using the data. The first is that interstate trade is measuredinvolumesandnotinvalues,astheyarenotavailable. Thesecondcaveatisthattheinterstate trade data do not include trade via roads. It is known that trade via roads has become increasingly important in India over the last decade with the construction of the national highways and the Golden Quadrilateral. Hence, the trade flows via railroads are an underestimate of the total trade across Indian states.10 In order to alleviate these two concerns, I merge these data with the portlevel data. I describe the port-level data in the next section, and then I describe the methodology ofmergingthesetwodatasets. 3.2 Indian International Trade Theinternationaltradedatacomefromthe“ForeignTradeStatisticsofIndia-PrincipalCommodities & Countries,”. This dataset contains all Indian international exports and imports and details the ports of exit and entry, respectively. In total there are 20 seaports that account for 50% of the total trade value, eight airports that account for 23%, and other ports such as Inland Container Distributions and land ports in the northern and northeastern part of India that account for 5%. The data report the exact port for 78% of total international trade, whereas the other 22% of total international trade occurs through “other ports.” Given that this still reflects a substantial part of 9For example, the Commodity Flow Survey data report trade flows across U.S. states but do not disentangle domesticallyproducedgoodsfrominternationalimportsorexports. 10Nevertheless, no comprehensivedata onfreightmovement bymotorized transporton roadsisavailable, which wouldhaveindicatedtheirimportanceintotaltradevalue. 14

totaltrade,Isupplementtheportdatawithtwoothersourcesofdata. The first is the “Agri Exchange” data for 2011–12. These data are from the Agricultural and ProcessedFoodProductsExportDevelopmentAuthority(APEDA)inIndiaandreportmoredetail ontheportsofexit(mainlythesmallerones)foragriculturalgoods. AgriExchange,however,does notincludedataontheportofentryofinternationalagriculturalgoods. The second source of data comes from the “Open Government Data Platform in India,” published by the Indian government. These data include port-level international import- and exportlevel data for the most important ports in India, of which 17 are not included in the Foreign Trade Statistics of India.11 When I supplement the original port data with these two datasets, I can explain 87% of total international port-level trade. The remaining 13% of port trade is distributed accordingtoeachstate’simportandexportshareforeachindividualgood.12 3.3 Matching Interstate with International Data I now describe how I merge the interstate trade data with the international trade data. First, I match each commodity in the interstate trade data to the export and import commodity in the international trade data. As explained in Section 3.1, trade across Indian states is differentiated by exports destined for a specific state and exports going to the port of a state, for international export. Thisallowsmetocomputetotalinternationalexportsandimportsforaspecificcommodity inthosestateswithaninternationalport. Then, I aggregate all commodity-specific international exports and imports at the ports in the interstatetradedata,thatis,theportsofAndhraPradesh,Gujarat,Karnataka,Kerala,Maharashtra, TamilNadu,andWestBengal. Theinternationaldatadescribehowmuchistradedinvaluesatthe portlevel. Hence,I“price”theinterstatetradedatasuchthatthetotaltradevalueattheportsfrom theinterstatetradedataamountstotheactualport-leveltradevalues. As an example, consider trade in wheat. The interstate trade data report trade in volumes. To convert the interstate trade volumes for wheat to values, I use the international trade information. 11ThereasonwhyIdonotusetheseportdataasthemaindataistiming. Thedataarefor2012–13,whichdonot coincidewiththeyearsfromtheinterstatedata: 2011–12. WhenIsupplementtheoriginaldatausingthisdatasource, Iassumethefractionoftotaltradeeachportrepresentsisconstantacross2011–12and2012–13. 12Theconcentrationofsmallerportsiscorrelatedwiththeexistenceofalargeportandthus,internationaltrade. 15

Morespecifically,IaggregatethetotalamountofwheattransportedtoallIndianportsforinternationalexport. Thisneedstoadduptothetotalport-levelexportvalueforwheat,whichisobtained from the international trade data. I then price wheat trade in India such that total port trade (volume*price) at all port blocks adds up to the port value in the international trade data. I do this for eachindividualcommodityintheinterstatetradedataset. Thisalleviatesthetwoissuesdiscussedearlier. First,alltradevolumeareconvertedtovalueby matching the port values. Second, the concern that trade via roads is not included in the interstate data is alleviated, as I aggregate up the interstate data to match the international trade flows. One underlying assumption is that trade via roads is highly correlated with trade via railroads. To provide some evidence for this, I correlate the density of railway networks within each state with that of road networks. I find that that the rail and road densities are highly positively correlated, withacorrelationcoefficientof0.90.13 3.4 Trade Patterns 3.4.1 InternationalTrade 76% of India’s international trade goes through 21 main ports: 14 seaports and 7 airports. Figure 1 shows their location. Clearly, the landlocked states in the northern and northeastern regions of India are the most remote from these major ports. In addition, Figure 1 shows the population in eachstate. Itshowsthatalargefraction(50%)ofthepopulationlivesinstateswithoutaccesstoan international port. This is important, as these patterns are also reflected in the states’ international tradeopenness. ThetoppanelinTable4showstheinternationaltradeopennessforagricultureandmanufacturing. Importandexportopennessaredefinedastotalinternationalimportsandexports,respectively, dividedbytotalsectoralproduction. Overall,Indiatradeslittleagricultureandmostoftheinternationaltradeisinmanufacturing. Agriculturalimportsareespeciallylow,at3%oftotalagricultural production. 13Source: TheMinistryofStatisticsandProgrammeImplementationinIndia(MOSPI). 16

Figure1: InternationalPortsandPopulationinIndia The top panel in Table 4 then breaks down the results for port and non-port states, where the port states are those states with a major international port as shown in Figure 1.14 The first thing to note is that the observed trade patterns in the aggregate data are still present: both port and non-portstatesaretradinglittleagricultureandaremostlytradingmanufacturinggoods. Thesecondthingtonoteisthatmostoftheinternationaltradeisdrivenbytheportstates,both for agriculture and manufacturing. For example, total manufacturing export openness is 34% for 14The10portstatesareAndhraPradesh,Delhi,Goa,Gujarat,Karnataka,Kerala,Maharashtra,Orissa,TamilNadu, andWestBengal. The20non-portstatesareArunachalPradesh,Assam,Bihar,Chandigarh,Chattishgarh,Haryana, Himachal Pradesh, Jammu and Kashmir, Jharkhand, Madhya Pradesh, Manipur, Meghalaya, Mizoram, Nagaland, Puducherry,Punjab,Rajasthan,Tripura,UttarPradesh,andUttarakhand. 17

port states and only 7% for non-port states. These international trade openness results, thus, show thereissubstantialheterogeneityinhowconnectedIndianstatesareinternationally. Table4: InternationalandDomesticTradeOpenness Agriculture Manufacturing Imports Exports Imports Exports International AllStates 3% 6% 27% 24% PortStates 4% 9% 39% 34% Non-PortStates 1% 3% 6% 7% Domestic AllStates 7% 7% 10% 10% PortStates 7% 6% 8% 9% Non-PortStates 6% 7% 13% 12% 3.4.2 DomesticTrade The bottom panel in Table 4 reports Indian states’ domestic import and export openness. Similar totheinternationaltradeopenness,thedomestictradeopennessforagricultureandmanufacturing isdefinedastotaldomesticimportsandexportsdividedbytotalsectoralproduction,respectively. Total domestic trade openness in India amounts to 7% for agricultural and 10% for manufacturing goods.15 In addition, these trade patterns are similar for port and non-port states, where the non-port states trade marginally more domestically. Combined with the results from the top panel in Table 4, port states trade more internationally than domestically, especially for manufacturing goods. Fornon-portstates,theoppositeistrue;theytradebothgoodsmoredomestically. Furthermore,theoverallmagnitudeofIndiandomestictradeisrelativelysmallcomparedtothe 15TheInternaltradedataaresymmetric,whichimpliesthetotalexportandimportopennessresultsforeachgood arealsosymmetric. 18

UnitedStates. UsingtheCommodityFlowSurveydata(CFS),Caliendoetal.(2014)findthattotal domestic trade as a fraction of GDP is 30% in the United States.16 In India, it is only 9%, which indicatesthatIndianstatestraderelativelylittlewitheachothercomparedtotheUnitedStates. 4 Quantitative Analysis Inowdescribehowthemodelmapstothedata. Theparameterstakenfromthetradeliteratureare the elasticity of substitution and the trade elasticity for manufacturing {θ ,σ }. What I measure m m inthedataaretheelasticityofsubstitutionandthetradeelasticityforagriculture{θ ,σ },thestate a a population L with β ∈ [0,1] in the rural and (1−β ) in the urban regions, regional state wages s s s wr, and both rural–urban and cross–state trade barriers. In the context of the model, I compute s fourinternationaltradebarriersτ andτ withg ∈ {a,m}. g,imp g,exp 4.1 Equilibrium Conditions States are indexed by s = 1...S,RoW where regions within a state are indexed by r = R,U (rural and urban) and good types by g = a,m (agriculture and manufacturing). Given parameters {σ ,σ ,θ ,θ },populationdata,L ,ofwhichthefractionβ ∈ [0,1]islocatedintheruralregion, a m a m s s (cid:8) (cid:9) (cid:8) (cid:9) and state wages in the rural and urban regions, wR and wU , I solve for s s=1...S,RoW s s=1...S,RoW productivity parameters , {T } and {T } , such that labor markets clear, a,s s=1...S,RoW m,s s=1...S,RoW whichisshownbyequations(6)and(8). Parameters As was shown in Section 2, the elasticities of substitution for agriculture and manufacturinggoverntheincomeelasticityofdemandformanufacturinggoodsrelativetoagricultural goods,whichdecreaseswithincomeifσ −σ > 0. IintroducedtheseCRIEpreferencestomatch m a agriculturalandmanufacturingconsumptionpatterns. Hence,Ifixσ at5followingFieler(2011) m andIcalibrateσ tomatchtheshareoftotalagriculturalexpendituresinIndia,whichis39%. a 16TheCFScontainsthetotaltradevalueoftradeacrossallU.S.states,whichamountedto5.2trillionUSDin2007. Caliendoetal.(2014)correctforinternationalimportsintheCFSandfindthattheCFSdomesticshipmentofgoods islargerthanthedomesticconsumptionmeasureforallsectors,byafactorrangingfrom1to1.4. 19

The trade elasticity for manufacturing, θ , is taken from Simonovska and Waugh (2014),17 m and the trade elasticity for agriculture, θ , is estimated following their method. The parameter a values are given in Table 5. Note that the estimated trade elasticity for agriculture is comparable to international trade elasticities, which suggests that, even within India, gains from trade are substantial. Table5: ParameterValues Parameter Value Explanation {σ ,σ } {3.3,5} ElasticityofSubstitution A M {θ ,θ } {5.6,5} TradeElasticity a m State-based Indian Data The state-specific Indian data needed for the analysis are population, includingthefractioninruralandurbanregions,andregion-specificwages. The population data are obtained from the 2011 Indian census, which contains statistics on the population in each state and the fraction in rural and urban regions. Overall, 69% of the total populationlivesinruralregionsrangingfrom2%inDelhito90%inHimachalPradesh. Rural and urban wages are proxied by total production in the agricultural and manufacturing sectors divided by the population that lives in each region, respectively. Gross agricultural output in each state is taken from the state-based estimates of the value of output from agriculture and allied activities, which are published by MOSPI. The same state-based data for gross manufacturing output are not available. Therefore, I first obtain the total manufacturing output in India for 2011 from the United Nations Industrial Development Organization’s (UNIDO) Industrial Statistics Database. Next, I use state-based GDP, which is obtained from the State Domestic Product series in 2011–12, also published by MOSPI, as a proxy to distribute output across states.18 For theRoW,theoutputdataaretakenfromUNIDO.19 17Thetradeelasticityof5formanufacturingisslightlyhigherthaninSimonovskaandWaugh(2014)becauseof theconditiononσ forafinitesolution,i.e.,(θ +1)>σ . g g g 18AsacheckIcorrelatetotalmanufacturingoutputwithtotalmanufacturingemployment,whichcanbeobtained at the state level from MOSPI. I find a high positive correlation of 0.84, suggesting that state-based GDP is a good proxyformanufacturingoutput. 19TableA.1intheonlineAppendixprovidesanoverviewofallthestate-baseddata. Onethingtonoteisthat,on 20

4.2 Trade Barriers ThissectiondescribeshowinternalbarrierstotradeinIndia,Dr ,arecomputed. Idisciplineboth g,sl cross–state and rural–urban trade barriers by applying a no-arbitrage condition to disaggregated pricedata. Theideaisthatthepricedifferentialforanygivengoodg inanytwodifferentlocations (s and l), is bounded from above by the bilateral trade cost, that is, ps(jg) ≤ d . If this condition p l (jg) sl didnothold,therewouldbeanarbitrageopportunity. Thus,anestimateofatradebarrierbetween anytwolocationsisthemaximumofrelativepricesovergoodvarietyj : g ˆ log d = max2log{p (j )−p (j )}, (9) sl s g l g jg wheremax2representsthesecondhighestpricedifferencetoreducemeasurementerror. I use disaggregated price data to discipline trade barriers across and within Indian states. The DataPortalofIndiapublishesdailywholesalepricesforagriculturalgoodvarietiesacrossdifferent markets in India (Mandi).20 It provides daily estimates for the minimum, maximum, and modal prices in a given market. The analysis in this paper is based on price data for 50 agricultural commodities across 1,831 markets in India during the fiscal year in India from April 2011 until March2012.21 Using the price data, I compute a yearly region- and state-specific price for each commodity. I first average out the daily modal prices in each market to monthly prices, after which I drop the top 99th and bottom 1st percentile to reduce measurement error. Next, I average these prices to a yearly price for a specific commodity variety in a given market in India. Given that the model differentiates between a rural and urban region within a state, I map this feature into the data. I differentiatemarketsineachstateasanurbanorruralmarket,whereanurbanmarketisdefinedas being located in a city with a population of one million or more.22 Finally, I compute an average urban and rural price for each commodity, which will be used to measure both cross–state and rural–urbantradebarriers.23 average, wages (proxied by output per worker) in the urban regions are three times larger than in the rural regions, whichisconsistentwithRestucciaetal.(2008). 20ThisdatasetisgeneratedthroughtheAGMARKNETPortal,http://agmarknet.nic.in. 21The50agriculturalcommoditiesarelistedinTableA.2intheonlineAppendix. 22TheIndiancensusof2011definesanurbanagglomerationasacitywithapopulationofonemillionormore. 23For two states, Arunachal Pradesh and Bihar, however, there is no price information. I use the average price informationofthesurroundingstates. ForArunachalPradesh,IusetheaveragepricedatafromManipur,Meghalaya, 21

Cross–State Trade Barriers As outlined previously, I will use a no-arbitrage condition to measure the cross–state trade barriers {d } . Given that trade between states can occur only sl s,l=1...S through the urban regions, I use the calculated average urban prices to measure the cross–state tradebarriers. Icomputethecross–statetradebarriersforeachstatepairinIndiausingthesecond highest price differential across each good variety. Hence, I calculate (30∗30−30) cross–state tradebarriers,andPanel(a)inFigure2showsahistogramoftheresults. Themediantradebarrier is2.52,whichimpliesatradebarrierof150%. (a)India: Mediand =2.52 (b)U.S.: Mediand =1.32 sl sl Figure2: Cross–StateTradeBarriers To assess the magnitude of these cross–state trade barriers in India, I compare them to the UnitedStates. TheUnitedStatesDepartmentofAgriculture(USDA)collectspricesonagricultural goodsacross50statesandtheDistrictofColumbia.24 Iapplythesameno-arbitrageconditionand measurecross–statetradebarriersasthesecondhighestpricedifferentialacrosseachgoodvariety. Panel (b) in Figure 2 shows a histogram of the computed cross–state trade barriers in the United States. I find a median trade cost of 1.32, which is around one-fifth of the Indian trade costs and aroundthesamemagnitudeasothermicroevidencehasestimated. Thisissuggestiveofrelatively large trade frictions across Indian states, which corroborates the relatively small domestic trade flowspresentedinthebottompanelofTable4. Mizoram,Nagaland,andTripura. ForBihar,IusetheaveragepricedatafromUttarPradeshandAssam. 24TableA.3intheonlineAppendixshowsthe40agriculturalgoodsusedintheUnitedStates. 22

Rural–Urban Trade Barriers Next, I compute the rural–urban trade barriers, {δ } , using s s=1...S the price variation across rural and urban markets within each state. I apply the same no-arbitrage condition given by equation (9) and calculate 30 rural–urban trade barriers. I find that the median rural–urban trade barrier is 1.48 and the 90–10 percentile is 2.38 and 1.24. These results imply that for the median state, prices in the urban area are 50% more expensive than in the rural area and that the rural–urban trade barrier in certain states is almost as large as the median cross–state tradebarrier.25 International Trade Barriers Finally, I estimate the international trade barriers from the internationaltradeflows. EachIndianstateistradingwiththeRoW,butnotallstateshaveaccesstoan international port. Hence, states without an international port ship goods to and from the lowest cost port in order to trade internationally. Given the computed cross–state trade barriers from the price data, the cost to the lowest cost port is given by the minimum trade barrier from the specific non-port state to each of the port states. Then, I assume that once the goods are at the port, internationalexportandimportcostsarethesameacrossportsbutnotacrossgoods. I estimate four international trade barriers: agricultural and manufacturing import and export barriersdenotedbyτ ,τ ,τ andτ ,respectively. Icalibratethosecostsbymatching a,imp m,imp a,exp m,exp the aggregate data moments given in the first row of Table 4. Let X be the state- and goodg,sRoW specific import value from the RoW for good g ∈ {a,m}. Similarly, define X as the stateg,RoWs andgood-specificexportvaluefromthestateinIndiatotherestoftheworldforgoodg ∈ {a,m}. Thefourtargetedmomentsaregivenby (cid:80) (cid:80) (cid:80) (cid:80) (cid:34) X X X X (cid:35) a,sRoW m,sRoW a,RoWs a,RoWs s , s , s , s , (cid:80) (cid:80) (cid:80) (cid:80) X X X X a,s m,s a,s m,s s s s s (cid:80) where X isthetotalIndiansectoralproduction. g,s s Table 6 shows the calibrated international trade barriers. Note that these are the border costs, that is, the trade barriers that the urban regions in port states face. Section 5 compares their mag- 25I find the computed rural–urban trade barriers are positively correlated with the density of the infrastructure network in each state. More specifically, I find a correlation coefficient of 0.58 between the estimated rural–urban tradebarriersandthemediandistancefromeachmarkettothenearestrailway. 23

nitudetothatoftheinternaltradebarriers. These calibrated border trade barriers are consistent with earlier work on international trade barriers for developing countries. First, the export barrier for manufacturing is higher than the import barrier. This is consistent with Waugh (2010). Second, the import barrier for agriculture is aboutthreetimeslargerthanformanufacturing,whichislinewithTombe(2015). Table6: InternationalBorderTradeBarriers Agriculture Manufacturing Import τ = 2.57 τ = 1.49 a,imp m,imp Export τ = 2.14 τ = 2.92 a,exp m,exp 4.3 Fit Model Given the model calibration, I test how well the model fits the international and domestic trade data. International Trade One of the model’s innovations is to allow for state heterogeneity in terms ofaccesstoaninternationalport. Thiswasmotivatedbytheinternationaltradeopennessresultsin the top panel of Table 4, which show that port states trade substantially more internationally than non-port states. To test the fit of the model regarding international trade, I perform two analyses. Thefirsttestshowwellthemodelcapturesthenon-targetedheterogeneityintradeopennessacross portandnon-portstates. Thesecondrecalibratesthebordertradebarriersforthesametrademodel with the exception that all states have an international port to quantitatively assess the importance ofdifferentialportaccess.26 Table 7 shows the results for agricultural and manufacturing trade. Column one for each good shows the data: the export and import openness as reported in the top panel in Table 4. Column two shows the results for the “new model” in this paper with heterogeneous international port 26SectionA.5intheonlineAppendixprovidesmoredetailonthecalibrationandtheinternationaltradebarriers. 24

access. Column three shows the export and import openness results for the “old model” without differentiatingportaccess. First,Icomparecolumnsoneandtwoforeachgoodtoevaluatethefitofthemodel. Themodel performsverywellintermsofbothcapturingthenon-targetedheterogeneityininternationaltrade openness across port and non-port states and matching the shares. For instance, for agriculture, the model captures that port states trade substantially more than non-port states. Furthermore, the model matches the import and trade openness almost perfectly: port states import 4% and export 10% of the total production, whereas for non-port states it is only 0.6% and 2%. With regard to manufacturing,themodelcanaccountforthelargedifferenceinimportandexporttradeopenness of port states compared to non-port states. Nevertheless, it slightly underestimates international trade for port states (2 percentage points) and overestimates trade for non-port states (4.5 percentagepoints). Table7: Non-TargetedModelMoments Agriculture Manufacturing Data New Old Data New Old Model Model Model Model Port Imports 4% 4% 3% 39% 36% 23% States Exports 9% 10% 6% 34% 33% 21% Non-Port Imports 0.7% 0.6% 2% 6% 11% 32% States Exports 3% 2% 6% 6% 10% 29% Second, I compare columns one and three for each good. As mentioned before, column three shows the results for a recalibrated model in which all states have access to an international port. Table7showsthatthis“oldmodel”doessignificantlyworsethanthe“newmodel”incapturingthe heterogeneityandmagnitudeininternationaltradeforportversusnon-portstates. Foragriculture, 25

non-portstatestradeasmuchasportstates,andformanufacturing,theytradeevenmorethanport states. Bothoftheseimplicationsarecounterfactual. Thus, taking into account differential port access is quantitatively important to explain statebasedinternationaltradingpatterns. Domestic Trade Next, I test how well the computed domestic trade barriers predict domestic trade flows in India for agricultural and manufacturing goods. Figure 3 shows the model- versus X the data-weighted trade flows z = g,sl for both agriculture and manufacturing. The model g,sl Xg,sX g,l predicts domestic trade flows well, especially for agriculture. The correlations are reasonably high and significant, with a correlation of 0.67 for agricultural and 0.55 for manufacturing trade. Nevertheless,themodeloverestimatesdomestictradeinmanufacturinggoods. (a)Agriculture: corr=0.67 (b)Manufacturing: corr=0.55 Figure3: IndianDomesticTrade 5 Results First, I compare the magnitude of internal versus external trade barriers and their heterogeneity across port and non-port states. Second, I perform several counterfactual experiments to assess the trade and welfare effects of internal versus external integration. Finally, I study the nature of internaltradebarriers. 26

5.1 Size of Internal versus External Trade Barriers Table 8 presents the total trade barriers Indian states face to import and export internationally.27 TheseareacombinationofthebordertradebarrierspresentedinTable6,togetherwiththeinternal trade barriers, which consist of a cross–state trade barrier for non-port states, and a rural–urban trade barrier for all states. Table 8 contains two blocks, one for imports and the other for exports, each consisting of four columns. The first column in each block reports the median populationweighted trade barrier faced by all states. The second column shows the internal trade barrier as a fractionofthetotaltradebarrier. Columnsthreeandfourshowthebreakdownoftheinternaltrade barrier into a cross–state and a rural–urban trade barrier, respectively. These percentages sum up tothetotalinternaltradebarrier. Inaddition,alltradebarriersarebrokendownforagriculturaland manufacturinggoodsandforportandnon-portstates. AfirstresultfromTable8isthatinternaltradebarriersaccountforasubstantialfractionofthe totaltradebarriers: 44%oftotalinternationalimportbarriersand29%oftotalinternationalexport barriers. In other words, for international imports, shipping from the RoW to the ports of India accounts for 56% of the total barrier, whereas shipping from the ports in India to the destination accounts for 44% of the total barrier. For exports the fraction of internal barriers is slightly less, 29%,duetotherelativehighborderexportcostsIcalibratedinTable6. The breakdown of import barriers by good shows that shipping agricultural goods from the Indian ports to the final destination in India represents one-third (34%) of the total import trade barrier. For manufacturing imports, it represents almost two-thirds (62%) of the total import barrier. Thereasonisthattheborderimportbarrierissmallerformanufacturing. AsecondresultfromTable8isthatthenon-portstatesfacethehighesttotaltradebarriers,that is, around three times higher than the port states, for both international imports and exports. This is mainly driven by the additional cross–state barrier non-port states incur when trading internationally. The fraction of internal barriers for non-port states is on average 62% of total import and 46% of total export barriers. This implies that moving goods from the port to the destination in IndiaissubstantiallymoreexpensivethanmovinggoodsfromtheRoWtotheIndianport. 27TablesA.4,A.5,A.6,andA.7intheonlineAppendixshowtheresultsforeachstate. 27

Table8: FractionandBreakdownInternalTradeBarriers InternationalImports InternationalExports Total Internal Cross- Rural- Total Internal Cross- Rural- Barrier State Urban Barrier State Urban AllStates Average 3.65 44% 19% 25% 4.15 29% 13% 16% Agriculture 4.62 34% 15% 19% 4.32 47% 20% 27% Manufacturing 2.67 62% 27% 35% 4.00 16% 7% 9% PortStates Average 2.42 16% 0% 16% 2.90 9% 0% 9% Agriculture 3.07 11% 0% 11% 2.88 23% 0% 23% Manufacturing 1.77 29% 0% 29% 2.92 0% 0% 0% Non-PortStates Average 5.48 62% 32% 30% 5.81 46% 24% 22% Agriculture 6.95 52% 27% 25% 6.62 65% 34% 31% Manufacturing 4.02 78% 41% 37% 5.00 27% 14% 13% Figure 4 shows internal barriers as a fraction of the total trade barriers for all states. This is an average across import and export barriers. As mentioned before, there is large heterogeneity in the fraction of internal trade barriers: the 90–10 percentile is 70–13%. Most of this heterogeneity is driven by port versus non-port states, where internal barriers account for 17% and 51%, respectively. Furthermore, as states are more removed from a major port, the fraction of internal trade barriersincrease,goingupto73%ofthetotaltradebarrierinHimachalPradesh. 28

Jammu and Kashmir Himachal Pradesh Chandigarh Punjab Uttarakhand Haryana Arunachal Delhi Pradesh Uttar Pradesh Rajasthan Assam Nagaland Bihar Meghalaya Manipur Jharkhand Madhya Pradesh West Tripura Mizoram Gujarat Bengal Chhattisgarh Orissa Maharashtra Andhra Pradesh Goa % Internal Trade Barrier Karnataka [0,15] Puducherry (15,30] Tamil (30,45] Kerala Nadu (45,60] (60,75] Figure4: FractionofInternalTradeBarriers AfinalresultfromTable8isthebreakdownofinternaltradebarriersintocross–stateandrural– urban trade barriers. On average, the rural–urban trade barrier represents a higher fraction of the total compared to the cross–state trade barrier: 25% versus 19% for imports and 16% versus 13% forexports. Thisisdrivenbythefactthathalfofthepopulationlivesinportstatesthatdonotface any additional cross–state trade barrier when trading internationally. This is clear from the port results, which shows that port states face no cross–state trade barriers. Nevertheless, even for the non-port states, the rural–urban trade barriers are almost as high as the cross–state trade barriers, which suggests that rural–urban trade barriers also impose a substantial barrier to international trade. 29

5.2 Counterfactuals Withthecalibratedmodel,Inowanalyzecounterfactualexperiments. Theprocedureisasfollows: given the same state population, L ; rural population shares, β ; parameter values σ ,σ ,θ ,θ ; s s a m a m and the productivity vectors T and T , I change specific trade barriers. Then, I recalculate a,s m,s wagesintheurbanandruralregionsofeachstatesuchthatlabormarketscleargivenbyequations (6)and(8). Tostudytheimpactofreducingtradebarriers,Iconsidertwomeasures. Thefirstisthe change in overall international imports as a fraction of production.28 The second is the change in welfare. Welfare is measured as a compensating variation: by how much do wages in each region need to change in order to receive the same utility in the baseline as in the counterfactual? Utility canbewrittenasfollows:29 (cid:20)(cid:18) (cid:19) (cid:18) (cid:19) (cid:21) σ σ U = λr a xr + m xr , (10) s s σ −1 s,a σ −1 s,m a m 1 (cid:82) wherexr = qr(j )pr(j )dj isdefinedasthestate-region-specificexpenditureongoodvariety s,g s g s g g 0 j . To aggregate each region- and state-level compensation variation into an aggregate welfare g measure,Iweighthembyregion-andstate-levelpopulation. First,Istudytwocounterfactualsaimedatinternationalintegration,afterwhichIexaminethree counterfactualsaimedatinternalintegration.30 5.2.1 RemovingInternationalImportBarriers Thefirstpolicyexperimentistheremovalofinternationalimportbarriers,thatis,τ = τ = a,imp m,imp 1. The impact on international imports is shown in column (1) of the top panel in Table 9. It first presents the results for all states and breaks down the results by port and non-port states. After a removal of international import barriers international trade would increase substantially; 28As trade is balanced, the impact on international exports is similar. There are some differences in exports and imports across goods for port states and non-port states, however, and these results are reported in Table A.8 in the onlineAppendix. 29SeeSectionA.7intheonlineAppendix. 30Note that I only study the potential benefits from international versus internal integration, and do not take into account the costs. One might be concerned that building better infrastructure in India is more costly than reducing internationalimporttariffs,forinstance.Nevertheless,Section5.3providesevidencethatinternaltradebarriersconsist ofphysicalandpolicybarriers,whichimpliesthatthereisscopeforremoving‘lowercost’policybarriersinIndia. 30

whereas India currently imports 17% of its total production, it would now import 37%. However, the international trade breakdown across port and non-port states shows that the total fraction of imported goods is still largest for the port states; they would import 50% of total production as opposedto18%inthenon-portstates. Thereasonforthesedifferentimpactsoninternationaltrade is that non-port states still face the additional cross–state trade barrier to move goods to and from the international port. As these trade barriers make up more than half of the total trade barriers, removinginternationalimportbarriersdoesnotincreaseimportstotheleveloftheport-states. Table9: InternationalTradeandWelfareEffects Baseline Import Cross-State Cross- Rural- Cross-State toPort State Urban toU.S. (1) (2) (3) (4) (5) InternationalImports Aggregate 17% 37% 26% 11% 21% 14% PortStates 24% 50% 24% 11% 29% 16% Non-PortStates 6% 18% 28% 10% 9% 11% Welfare Aggregate . 7% 2% 30% 18% 13% PortStates . 12% -0% 29% 19% 12% Non-PortStates . 2% 4% 31% 17% 14% These trade results are also reflected in the welfare impacts presented in the bottom panel of Table 9. Column two shows that after removing import barriers, total welfare increases by 7%. Separating the impact across port and non-port states shows that welfare gains are concentrated in theportstates. Theyexperiencewelfaregainsof12%,whereasthenon-portstatesonlygain2%. 31

The aggregate welfare gains from removing international import barriers primarily stem from increased international imports, which has two effects: prices decrease because of lower cost imports, but wages also decrease due to higher international competition. In the aggregate, I find that the price effect outweighs the wage effect. However, because the effect on international importsislowerfornon-portstates,thepriceeffectisdampened. Thus,theirwelfaregainsarelower comparedtotheportstates. Overall, this shows that policies focused on reducing international import barriers have very heterogeneous trade and welfare effects depending on international port access. In fact, Table A.10 of the online Appendix shows that the previously outlined model (Table 7), without differential international port access across Indian states, would overestimate the gains from removing internationalimportbarriersby6percentagepoints,asitwouldoverestimatethetradeandwelfare effectsfornon-portstates. 5.2.2 BuildingInternationalPortsinAllStates The previous section showed that transporting goods to and from an international port for nonport states is quantitatively important for the welfare gains from international trade liberalization. Therefore, the second counterfactual is to estimate the impact of building an international port in allnon-portstates,therebyincreasingmarketaccesstotheRoW.Inthemodel,onlythecross–state tradebarrierstoandfromtheinternationalportfornon-portstatesareremoved,whilekeepingthe cross–statetradebarrierswithinIndiaconstant. The trade results are shown in column (2) in the top panel of Table 9. Aggregate international imports increased to 26% from 17%, but not by as much as removing import barriers (37%). Distinguishing international imports to the port states from those to the non-port states, the aggregate increase is coming from the latter: international imports as a fraction of production increase from 6% to 28%. The intuition is that the cross–state trade barriers to move goods to and from international ports are on average larger than the international trade barriers. By removing them, international imports increase by more for non-port states compared to when import barriers are completelyremoved. Thewelfareeffectsincolumn(2)inthebottompanelshowthataggregatewelfareincreasesby 32

2%. Thisimpliesthatremovinginternationalimportbarriersincreaseswelfaremorethanremoving the barriers to access international ports. This is primarily driven by the small welfare losses experienced in the port states. They are marginally worse off due to increased port competition, which drives down wages more than prices, as international imports only slightly increase. Nonport states, however, benefit more from having direct access to an international port than from removinginternationalimportbarriers: 4%comparedto2%. Thisresultsuggeststhatremovingthe costsofmovinggoodsfromandtotheportsarequantitativelyimportantandshouldbeconsidered whendesigninginternationaltradepolicies. Thisconcludestheexaminationofthetwocounterfactualsaimedatinternationalintegration. I willnowconsiderthreecounterfactualsaimedatdomesticintegration. 5.2.3 RemovingCross–StateTradeBarriers The first counterfactual aimed at internal integration is the removal of cross–state trade barriers, that is, d = 1 ∀s,l = 1...S. The impact on international imports is shown in column (3) in the s,l top panel of Table 9. It shows that aggregate international imports would decrease 5 percentage pointsto11%oftotalproduction. ThisaggregatedeclineisdrivenbyportstatesastheyreduceinternationalimportsandincreaseimportsfromotherIndianstates. Fornon-portstates,international imports actually increase from 6% in the baseline to 10%. This is because removing cross–state barriers not only makes trade with other states in India cheaper, but it also decreases the cost of tradinginternationallyasitbecomescheapertomovegoodstoandfromthenearestport. Now consider the welfare effects in column (3) in the bottom panel of Table 9. First, the total welfare gains are substantial: 30% compared to 7% when removing import barriers, which suggests that India has more to gain from becoming more integrated internally than removing importbarriers. Second, both port and non-port states experience welfare gains around the same magnitude: 29 and 31%, respectively. Thus, the welfare gains are more equally distributed across states, which is in contrast to when import barriers are removed. The reason is that when international import barriers are removed, the non-port states do not experience large welfare gains compared to port states because they still incur large internal barriers to trade to and from the international 33

port. Thus, there are big heterogeneous welfare impacts, with the largest gains concentrated in the port states. When internal trade barriers are removed, however, all states benefit from cheaper Indian goods. In addition, non-port states experience an additional welfare gain from cheaper international goods. To summarize, removing cross–state trade barriers in India generates larger andmoreequallydistributedwelfaregainsthanremovingimportbarriers. 5.2.4 RemovingRural–UrbanTradeBarriers The second counterfactual aimed at internal integration is the removal of rural–urban trade barriers, that is, δ = 1 ∀s = 1...S. Column (4) in the top panel of Table 9 shows that aggregate s internationalimportswouldincrease4percentagepointsto21%oftotalproduction. Bothportand non-port states import more from the RoW, with the overall level still being higher in port states: 29% compared to 9% for non-port states. An interesting thing to note is that international trade is affected differently depending on the type of internal integration in India. When cross–state trade barriers are removed, international trade decreases because port states increase internal trade. Alternately, when rural–urban trade barriers are removed, international trade increases as port states trade more internationally because the barrier to international trade has decreased. Overall, this suggests that rural–urban trade barriers impose a significant barrier to international trade for all states. Column (4) in the bottom panel of Table 9 shows the respective welfare impacts. Aggregate welfare gains are 18%, which is larger than any aggregate welfare impact from policies aimed at internationalintegration. Thissuggeststhattherearelargepotentialwelfaregainsfromconnecting rural and urban regions. In addition, the gains are, again, more equally distributed across states compared to removing international import barriers. Nevertheless, if we compare the two types of internalintegrationinIndia,theresultsindicatethatIndiahasmoretogainfromconnectingstates inIndiabydecreasingcross–statetradebarriers. 5.2.5 ReducingCross–StateTradeBarrierstoU.S.Level The previous four counterfactuals have shown the welfare gains of removing barriers internationallyanddomestically. Asthesepoliciesareextreme,Iperformafinal,more‘realistic’counterfac- 34

tualexperimentinwhichIreducecross–statetradebarriersinIndiatothoseobservedintheUnited States. Figure2showedthatcross–statetradebarriersinIndiaareaboutfivetimeslargerthanthose intheU.S.Hence,Idivideallcross–statetradebarriersinIndia,d ,bythesamefractionsuchthat sl themediantradebarrierinthecounterfactualIndiandataisthesameasintheUnitedStates(1.32). Column(5)inTable9showstheeffectsoninternationaltradeandwelfare. Thesameintuition holds as in the counterfactual when cross–state barriers are removed. The primary result from this final counterfactual is that reducing cross–state trade barriers in India to the level of those in the United States leads to welfare gains of 13%, which is still larger than the welfare gains from removing international import barriers: 7%. This implies that India has more to gain from becomingmoreintegratedinternallythanfromremovinginternationalimportbarriers. Figure 5 graphically presents the welfare results across Indian states. Panel (a) presents the state-wise welfare changes when international import barriers are removed. A removal of them mostly benefits the port states, with welfare gains rising to 38% in Delhi. The non-port states, however,experiencesmallerwelfaregainsupto5%,andsomeevenexperiencewelfarelosses. Panel (b) presents the state-based welfare changes when internal barriers in India are reduced totheU.S.level. Inthisscenario,allstatesexperiencewelfaregainsandtheyareaslargeorlarger than those in Panel (a). Furthermore, the states that benefit most are the non-port states located in thenorthernandnortheasternregionsofIndia. Thus,Indiahasmoretogainfrombecomingmore integrateddomesticallyandthewelfaregainsaremoreequallydistributedacrossstates. The intuition is that large barriers to trade within India make port states trade more with the rest of the world because they have access to cheaper international goods. Non-port states, on the other hand, trade more within India as international trade is expensive due to high internal trade barriers. When internal trade barriers are reduced, trade across all states in India increases. This increaseswelfareforportstatesastheybenefitfromcheaperIndianproducts. Non-portstatesalso benefit from cheaper Indian products due to increased cross–state trade, but also from lower cost internationaltradeduetosmallertradebarrierstoandfrominternationalports. 35

Jammu and Kashmir Himachal Pradesh Punjab Chandigarh Uttarakhand Haryana Delhi Arunachal Pradesh Uttar Pradesh Rajasthan Assam Nagaland Bihar Meghalaya Manipur Jharkhand Madhya Pradesh West Tripura Mizoram Gujarat Bengal Chhattisgarh Orissa Maharashtra Andhra Pradesh Goa Karnataka Puducherry Tamil Kerala Nadu (a)RemoveInternationalImportBarriers Jammu and Kashmir Himachal Pradesh Punjab Chandigarh Uttarakhand Haryana Delhi Arunachal Pradesh Uttar Pradesh Rajasthan Assam Nagaland Bihar Meghalaya Manipur Jharkhand Madhya Pradesh West Tripura Mizoram Gujarat Bengal Chhattisgarh Orissa Maharashtra Andhra Pradesh Goa % Change Welfare Karnataka [-1,5] Puducherry (5,15] Tamil (15,25] Kerala Nadu (25,35] (35,50] (b)ReduceCross–stateTradeBarrierstoU.S. Figure5: WelfareComparisonExternalversusInternalIntegration 36

5.3 The Nature of Cross–State Trade Barriers The previous section showed that India has more to gain from reducing cross–state trade barriers, where cross–state barriers were computed from cross–state price dispersion. The advantage of using price dispersion to discipline trade barriers instead of direct measures such as distance is thatitencompassesmorethanjustinfrastructurebarriers. Nevertheless,todesignpoliciesfocused on domestic integration, it is important to decompose these cross–state trade barriers. Therefore, this section provides more insight into the nature of the estimated cross–state trade barriers by regressing them on measures of infrastructure as well as policy and culture. I run the following regression: log(d ) = α+βX +ε , (11) sl sl sl where d is the cross–state trade barrier between states s and l and X is the set of explanatory sl sl variables. Toproxytheinfrastructuralbarrier,Iusetwomeasures: thedistancebetweenstatecapitalsand the number of state borders between states. Because these two measures are highly correlated, I runthemainspecificationgivenbyequation(11)separatelyinPanelsIandIIofTable10. Forthepolicybarriers,Iusemeasuresofcorruption,thearduousnessofthetaxadministration, and tax rates themselves. To measure these variables, I use the World Bank Enterprise Survey data for India in 2014, which contains almost 10,000 firms spread across all Indian states except the union territories Puducherry and Chandigarh. The survey asks to what degree tax rates, tax administration, and corruption are an obstacle to the current operations of the establishment. The responsesrangefrom“NoObstacle”to“VerySevereObstacle”. Iusetheseresponsestoprovidea comparable measure of these factors across states. I convert these qualitative measures to dummy variablesbyfirsttakingthefractionoffirmsthatansweredeither“MajorObstacle”or“VerySevere Obstacle”inresponsetothethreevariablesofinterest. Ithencreateadummyvariablethatisequal to one if the fraction of firms is above the 90th percentile in either the origin or destination state, and zero otherwise.31 As a final non-infrastructural barrier, I include a cultural barrier to trade, 31Iarbitrarilychoosethe90thpercentiletocapturetheeffectofthosestateswhereeitherofthethreevariablesisa majorissue. 37

whichindicateswhetherstatessharethesamemainlanguage. PanelIofTable10showstheresultswheninfrastructureisproxiedbydistance. First,Iconsider the infrastructural barrier. Column (1) presents the results for distance as the only explanatory variable in equation (11). It shows that if the distance between two states increases by 1%, the tradebarrierincreasesby0.19%,andthiscoefficientishighlysignificant. Inaddition,theadjusted R2 is 0.12, which implies that distance in itself can explain 12% of the variation in cross–state tradebarriers. Second, I consider the impact of the three policy barriers in columns (2), (3), and (4). Column (2) shows that trade barrier between two states increases by 17% if corruption is high in either state,andthiscoefficientisalsohighlysignificant. TheadjustedR2 risesfrom0.12to0.18,which suggests that corruption plays an important role in the size of trade barriers. In column (3), I include a dummy that indicates a burdensome tax administration. The coefficient is again highly significantandeconomicallyimportant. Itshowsthattradebarrierswouldincreaseby23%ifeither theimportingorexportingstatefacesadifficulttaxadministration. TheadjustedR2 alsoincreased from 0.18 to 0.24. Column (4) includes the final policy barrier: the tax rate. The significant coefficientindicatesthatstateswithhighertaxratesfaceanadditionaltradebarrierof12%. Finally,Iincludeadummyvariablethatindicateswhetherthemainlanguageisthesameacross states. There are 14 different main languages across the 27 states and three union territories in the analysis. The results are presented in column (4). They show that if two states share their main language,theircross–statetradebarrierdecreasesby17%. Thisresulthighlightsthatevencultural barriersinIndiaareeconomicallyimportant. Panel II of Table 10 shows the same results as Panel I, but with the second proxy for infrastructure: the number of state borders between any two states in India. Column (1) shows that the coefficient is positive and highly significant. An interesting thing to note is that the adjusted R2 is higher compared to that when infrastructure was proxied by distance: 0.16 compared to 0.12, respectively. This implies that the actual number of state borders when trading across states is moreinformativeaboutthecross–statetradebarriersthandistance. Thisisexactlyinlinewiththe earlierresultwhichindicatedthatthelargestwelfaregainsfrominternalintegrationareassociated withconnectingstatesinIndia. 38

Table10: NatureofCross–StateTradeBarriers (1) (2) (3) (4) (5) PanelI Log(Distance) 0.19*** 0.22*** 0.22*** 0.22*** 0.18*** s.e. (0.02) (0.02) (0.02) (0.02) (0.02) HighCorruption 0.17*** 0.18*** 0.19*** 0.22*** s.e. (0.03) (0.03) (0.03) (0.03) HighTaxAdministration 0.23*** 0.24*** 0.25*** s.e. (0.03) (0.03) (0.03) HighTaxRates 0.12*** 0.12*** s.e. (0.03) (0.03) Language -0.17*** s.e. (0.03) Constant -0.38*** -0.61*** -0.62*** -0.64*** -0.32** s.e. (0.13) (0.13) (0.13) (0.13) (0.14) Observations 756 756 756 756 756 AdjustedR2 0.12 0.18 0.24 0.25 0.28 PanelII #ofBorders 0.09*** 0.12*** 0.12*** 0.12*** 0.10*** s.e. (0.01) (0.01) (0.01) (0.01) (0.01) HighCorruption 0.19*** 0.21*** 0.21*** 0.23*** s.e. (0.03) (0.03) (0.03) (0.03) HighTaxAdministration 0.22*** 0.23*** 0.24*** s.e. (0.03) (0.03) (0.03) HighTaxRates 0.09*** 0.09*** s.e. (0.03) (0.03) Language -0.11*** s.e. (0.03) Constant 0.71*** 0.64*** 0.61*** 0.60*** 0.65*** s.e. (0.03) (0.03) (0.03) (0.03) (0.03) Observations 756 756 756 756 756 AdjustedR2 0.16 0.30 0.35 0.36 0.37 ***, **, and * indicate statistical significance at the 1%, 5%, and 10% confidence levels, respectively. The results areestimatedcoefficientsforaregressionofcross–statetradebarriers(inlogs)onthestatedvariables. PanelIshows theresultswithdistanceastheinfrastructuralbarrierandPanelIIshowstheresultswiththenumberofbordersasthe infrastructuralbarrier. Thevariables”HighCorruption”,”HighTaxAdministration”,and”HighTaxRates”indicate whethereithertheimportingorexportingstateisinthetop10percentileofIndianstateswithrespecttothesevariables. 39

Columns (2), (3), and (4) in Panel II show the result with the corruption, tax administration andtaxratesdummies. TheresultsareverysimilartothoseinPanelII.Allcoefficientsarearound the same magnitude and are highly significant. An interesting difference with Panel I is that the adjusted R2 is higher when including the policy barriers: 0.36 compared to 0.25 in column four. This is partly due to the higher explanatory power of the border effect, but also due to the higher explanatory power from the policy variables. Finally, column (5) includes the language barrier dummyvariable. TheresultsaresimilartothoseinPanelI,althoughslightlysmaller. To summarize, I find that policy barriers to trade in the form of corruption, a burdensome tax administration, and tax rates are quantitatively as important as infrastructural barriers. Therefore, studyingthenatureandimpactofthesepolicybarriersisanimportantavenueforfutureresearch. 6 Robustness Inowperformtworobustnessexercises. Thefirstistoallowforlabormobilityacrossregions. The secondistorecalibratethemodelusingtradebarriersestimatedfromagravityequation. 6.1 Mobile Labor InthemainanalysisIassumelaborisimmobileacrossstatesandregionswithinstate. Inowrelax the second assumption and allow for labor mobility across regions within a state. Nevertheless, I assume that the wedge between agricultural and manufacturing wages, w s U , remains the same as wR s thebaselinecalibrationineachstate. Theprocedureisasfollows: giventhesamestatepopulation, L ; labor wedge, w s U ; parameter values σ ,σ ,θ ,θ ; and productivity vectors T and T , I s wR a m a m a,s m,s s change specific trade barriers. Then, I recalculate the fraction of the rural population β and rural s wages w such that (1) the urban wage in each is equal to the rural wage multiplied by the wage a,s wedgeand(2)labormarketsclear. Welfareisagainmeasuredasacompensatingvariation,butbecauselaborwithinastateisnow mobile,theweightsonthecompensatingvariationineachregionwillbedifferent. Table11shows theimpactoninternationaltradeandwelfare,whichisequivalenttoTable9,butwithmobilelabor across the rural and urban regions. In addition, it shows the fraction of the rural population and 40

howitchangedinresponsetoeachofthecounterfactualscomparedtothebaselinecalibration. Table11: InternationalTradeandWelfareImpactswithMobileLabor Baseline Import Cross-State Cross- Rural- Cross-State toPort State Urban toU.S. (1) (2) (3) (4) (5) InternationalImports Aggregate 17% 43% 28% 12% 20% 16% PortStates 24% 58% 24% 11% 29% 17% Non-PortStates 6% 20% 33% 14% 8% 14% Welfare Aggregate . 8% 2% 46% 17% 18% PortStates . 14% -0% 41% 19% 16% Non-PortStates . 3% 4% 51% 16% 20% RuralPopulationβ Aggregate 0.69 0.57 0.66 0.46 0.74 0.55 PortStates 0.60 0.44 0.60 0.41 0.69 0.49 Non-PortStates 0.77 0.69 0.72 0.51 0.78 0.60 ThemiddlepanelinTable11showsthatthemainresultholdsevenwhenlaborismobileacross regions: India has more to gain from becoming more integrated internally than from removing internationalimportbarriers. Aggregatewelfaregainsfromreducingcross–statesbarriersinIndia to the level of those in the United States amount to 18% (column (5)) compared to aggregate welfare gains of 8% from removing international import barriers (column (1)). This indicates that theeffectonwelfareisamplifiedunderbothcounterfactualscomparedtoTable9,butmoreforthe case of domestic integration. The intuition is given in the bottom panel in Table 11. It shows that as cross–state trade barriers are reduced to the U.S. level (column (5)), labor reallocates more to 41

the urban regions than when import barriers are completely removed. Hence, the welfare gains of removing internal barriers are amplified because workers in the urban regions face lower overall tradebarriers,asbothdomesticaswellasinternationaltradeoccursthroughtheurbanregions. 6.2 Trade Barriers from the Gravity Equation Instead of using price variation to measure trade barriers, I can also compute trade barriers by estimatingagravityequationforbothagriculturalandmanufacturingtradefollowingSimonovska and Waugh (2014). Section A.6 of the online Appendix provides more detail on how these trade barriers are estimated. One drawback of this method is that it identifies cross–state trade barriers astheonlyinternaltradebarrierinIndia,astradeflowsaredisaggregatedatthestatelevel. Table12: GravityEstimationResults Baseline Import Cross-State Cross- Cross-State toPort State toU.S. (1) (2) (3) (4) InternationalTrade Aggregate 17% 34% 29% 10% 15% PortStates 28% 57% 27% 10% 22% Non-PortStates 1% 2% 32% 9% 4% Welfare Aggregate . 7% 4% 66% 19% PortStates . 15% -0% 53% 16% Non-PortStates . -0% 7% 78% 23% Table 12 presents the results.32 The main result still holds: the welfare gains from reducing internal barriers to trade to the level in the United States generates larger welfare gains than re- 32Becauserural–urbantradebarrierscannotbeidentifiedusingthegravityestimationmethod,Icannotperformthe counterfactualofremovingthem. 42

moving international import barriers: 19% compared to 7%, respectively. In addition, the welfare gainsfromreducinginternalbarriersaremoreequallydistributedacrossIndianstates. 7 Conclusion Thispaperquantifiedthesizeofinternalversusexternaltradebarriersandassesstheeffectontrade and welfare. I developed a quantitative multi-sector international trade model featuring nonhomothetic preferences in which Indian states trade both domestically and internationally. I discipline the model using rich micro data on price dispersion as well as foreign and domestic trade flows at theIndianstatelevel. I find that internal trade barriers make up a substantial part of the total trade barrier: on average,theyaccountfor40%,withlargeheterogeneityacrossIndianstatesdependingonthedistance to the closest port. Second, I find that the welfare gains from reducing cross–state trade barriers to the level in the U.S. are larger than from removing international import barriers: 13% versus 7%, respectively. In other words, India has more to gain from becoming more integrated domestically than from removing international import barriers. In addition, the largest welfare gains from domesticintegrationareassociatedwithreducingtradebarriersbetweenIndianstates. The next step is to study the components of these cross–state trade barriers. I provided some evidencethatcross–statetradebarriersnotonlyconsistofinfrastructuralbarriersbutalsoofpolicy barriers. Giventhatpolicybarrierssuchascorruptionandaburdensometaxadministrationplague many developing countries, studying the size and nature of policy barriers and their interaction withinfrastructureisanimportantavenueforfutureresearch. References ADAMOPOULOS, T. (2011): “Transportation Costs, Agricultural Productivity and Cross-Country IncomeDifferences,”InternationalEconomicReview,52,489–521. ALDER, S. (2016): “Chinese Roads in India: The Effect of Transport Infrastructure on Economic Development,”Workingpaper. 43

ALLEN, T. (2014): “InformationFrictionsinTrade,”Econometrica,82,2041–2083. ALLEN, T. AND C. ARKOLAKIS (2014): “Trade and the Topography of the Spatial Economy,” QuarterlyJournalofEconomics,129,1085–1140. ALVAREZ, F. AND R. E. LUCAS (2007): “General Equilibrium Analysis of the Eaton-Kortum ModelofInternationalTrade,”JournalofMonetaryEconomics,54,1726–1768. ARKOLAKIS, C., A. COSTINOT, AND A. RODR´IGUEZ-CLARE (2011): “New Trade Models, SameOldGains?” AmericanEconomicReview,102,94–130. ASTURIAS, J., M. GARC´IA-SANTANA, AND R. ROBERTO(2016): “CompetitionandtheWelfare Gains from Transportation Infrastructure: Evidence from the Golden Quadrilateral of India,” Workingpaper. ATKIN, D.(2013): “Trade,TastesandNutritioninIndia,”AmericanEconomicReview,103,1629– 1663. ATKIN, D. AND D. DONALDSON (2016): “Who’s Getting Globalized? The Size and Impact of IntranationalTradeCosts,”NBERWorkingPaperNo.21439. BEHRENS, K., C. GAIGNE´, G. I. P. OTTAVIANO, AND T. JACQUES-FRANC¸OIS (2006): “Is remotenessaLocationalDisadvantage?” JournalofEconomicGeography,6,347–368. CALIENDO, L., F. PARRO, E. ROSSI-HANSBERG, AND P.-D. SARTE (2014): “The Impact of Regional and Sectoral Productivity Changes on the U.S. Economy,” NBER Working Paper No. 20168. CARON, J., T. FALLY, AND J. R. MARKUSEN (2014): “International Trade Puzzles: A Solution LinkingProductionandPreferences,”QuarterlyJournalofEconomics,129,1501–1552. COC¸AR, A. K. AND P. D. FAJGELBAUM (2016): “Internal Geography, International Trade, and RegionalOutcomes,”AmericanEconomicJournal: Microeconomics,8,24–56. COSTINOT, A. AND D. DONALDSON (2014): “How Large are the Gains from Economic Integration? TheoryandEvidencefromU.S.Agriculture,1880-1997,”Workingpaper. 44

CRUCINI, M. J. AND M. SHINTANI (2008): “Persistence in Law of One Price Deviations: EvidencefromMicro-Data,”JournalofMonetaryEconomics,55,629–644. DATA PORTAL OF INDIA (2012): “AgriculturalMarketingInformationNetwork,”. DIRECTORATE GENERAL OF COMMERCIAL INTELLIGENCE AND STATISTICS, MINISTRY OF COMMERCE AND INDUSTRY (2012a): “ForeignTradeStatisticsofIndia,”. ———(2012b): “TheInter-StateMovements/FlowsofGoodsbyRail,RiverandAir,”. DONALDSON, D. (2016): “Railroads of the Raj: Estimating the Impact of Transportation Infrastructure,”AmericanEconomicReview,forthcoming. DONALDSON, D. AND R. HORNBECK (2016): “Railroads and American Economic Growth: A MarketAccessApproach,”QuarterlyJournalofEconomics,131,799–858. DORNBUSCH, R., S.FISCHER, ANDP.A.SAMUELSON(1977): “ComparativeAdvantage,Trade, and Payments in a Ricardian Model with a Continuum of Goods,” American Economic Review, 67,823–839. EATON, J. AND S. KORTUM (2002): “Technology, Geography, and Trade,” Econometrica, 70, 1741–1779. ENGEL, C. AND J. H. ROGERS (1996): “HowWideIstheBorder?” AmericanEconomicReview, 86,1112–1125. FEENSTRA, R. C. AND D. E. WEINSTEIN (2016): “Globalization, Markups, and the U.S. Price Level,”JournalofPoliticalEconomy,forthcoming. FIELER, A. C. (2011): “Non-homotheticity and Bilateral Trade: Evidence and a Quantitative Explanation,”Econometrica,79,1069–1101. GOLLIN, D., D. LAGAKOS, AND M. WAUGH(2014): “TheAgriculturalProductivityGapinPoor Countries,”QuarterlyJournalofEconomics,129,939–993. 45

GOLLIN, D. AND R. ROGERSON (2014): “Productivity, Transport Costs, and Subsistence Agriculture,”JournalofDevelopmentEconomics,107,38–48. HERRENDORF, B., J. A. SCHMITZ, AND A. TEIXEIRA (2010): “The Role of Transportation in U.S.EconomicDevelopment: 1840-1860,”InternationalEconomicReview,53,693–716. HUMMELS, D. AND G. SCHAUR (2013): “TimeasaTradeBarrier,”AmericanEconomicReview, 103,1–27. JENSEN, R. (2007): “The Digital Provide: Information (Technology), Market Performance, and WelfareintheSouthIndianFisheriesSector,”QuarterlyJournalofEconomics,122,879–924. LIMA˜O, N.ANDA.J.VENABLES(1999): “Infrastructure,GeographicalDisadvantage,andTransportCosts,andTrade,”TheWorldBankEconomicReview,15,451–479. NAGY, D. (2016): “Citylocationandeconomicdevelopment,”Workingpaper. RAMONDO, N., A. RODR´IGUEZ-CLARE, AND M. SABOR´IO-RODR´IGUEZ (2016): “Trade, DomesticFrictions,andScaleEffects.”AmericanEconomicReview,forthcoming. RESTUCCIA, D., D. T. Y. YANG, AND X. ZHU (2008): “AgricultureandAggregateProductivity: AQuantitativeCross-CountryAnalysis,”JournalofMonetaryEconomics,55,234–250. ROSSI-HANSBERG, E. (2005): “A Spatial Theory of Trade,” American Economic Review, 95, 1464–1491. SIMONOVSKA, I. AND M. WAUGH (2014): “The Elasticity of Trade: Estimates and Evidence,” JournalofInternationalEconomics,92,34–50. SOTELO, S. (2016): “DomesticTradeFrictionsandAgriculture,”Workingpaper. STOREYGARD, A. (2016): “Farther on Down the Road: Transport Costs, Trade and Urban Growth,”ReviewofEconomicStudies,83,1263–1295. THE WORLD BANK (2014): “EnterpriseSurveys,”. 46

TOMBE, T. (2015): “TheMissingFoodProblem: Trade,Agriculture,andInternationalProductivityDifferences,”AmericanEconomicJournal: Macroeconomics,7,226–258. TOMBE, T. AND X. ZHU (2015): “Trade, Migration and Productivity: A Quantitative Analysis of China,”Workingpaper. UNITED NATIONS (2012): “IndustrialStatisticalDatabase,”. UNITED STATES DEPARTMENT OF AGRICULTURE (2012): “National Aggregate Statistics ServiceintheAgriculturalPrices,”. WAUGH, M. (2010): “International Trade and Income Differences,” American Economic Review, 99,2093–2124. 47

Appendix for Online Publication This online appendix contains additional empirical results, robustness tests, and a mathematical proof for the paper, “A Passage to India: Quantifying Internal and External Barriers to Trade” by Van Leemput. I have organized the results into the following sections: (A.1) additional data, (A.2) price data, (A.3) state-based barriers to international trade, (A.4) international trade results bysector,(A.5)modelwithoutportdifferentialportaccess,(A.6)gravityequationestimation,and (A.7)welfare. A.1 Additional Data FigureA.1showsamapwithall27Indianstatesandthreeunionterritoriesusedfortheanalysis. FigureA.1: IndianStates 1

TableA.1providesadditionalstate-baseddatausedinthemodelcalibration. TableA.1: DataModel TotalPop. ShareRuralPop. RuralWage Urbanwage L β wR wU s s s s AndhraPradesh 84,665,533 67% 0.28 0.67 ArunachalPradesh 1,382,611 77% 0.30 0.97 Assam 31,169,272 86% 0.12 0.82 Bihar 103,804,637 89% 0.06 0.60 Chandigarh 1,054,686 3% 0.51 0.65 Chattishgarh 25,540,196 77% 0.17 0.64 Delhi 16,753,235 3% 0.60 0.52 Goa 1,457,723 38% 0.35 1.14 Gujarat 60,383,628 57% 0.34 0.66 Haryana 25,353,081 65% 0.39 0.98 HimachalPradesh 6,856,509 90% 0.25 1.36 JammuandKashmir 12,548,926 73% 0.17 0.55 Jharkhand 32,966,238 76% 0.08 0.52 Karnataka 61,130,704 61% 0.23 0.56 Kerala 33,387,677 52% 0.25 0.55 MadhyaPradesh 72,597,565 72% 0.16 0.45 Maharashtra 112,372,972 55% 0.27 0.68 Manipur 2,721,756 70% 0.17 0.37 Meghalaya 2,964,007 80% 0.12 0.79 Mizoram 1,091,014 48% 0.22 0.37 Nagaland 1,980,602 71% 0.25 0.66 Orissa 41,947,358 83% 0.13 0.88 Puducherry 1,244,464 32% 0.24 0.49 Punjab 27,704,236 63% 0.46 0.71 Rajasthan 68,621,012 75% 0.22 0.68 TamilNadu 72,138,958 52% 0.25 0.55 Tripura 3,671,032 74% 0.15 0.63 UttarPradesh 199,581,477 78% 0.13 0.44 Uttarakhand 10,116,752 69% 0.19 0.91 WestBengal 91,347,736 68% 0.19 0.53 RestofWorld(RoW) 579,144,440,361,130,704 40% 1 1.88 2

A.2 Price Data Tables A.2 and A.3 show the list of agricultural goods used in the analysis in India and the United States,respectively. TableA.2: AgriculturalGoodsinIndia 1. Apple 28. Mango 2. Arhar(Tur) 29. MasurDal 3. Bajra(PearlMillet) 30. Mousambi 4. Banana 31. Onion 5. Beetroot 32. Orange 6. BengalGrams(Gram) 33. Paddy(Dhan) 7. Bhindi(LadiesFinger) 34. Papaya 8./9. Bittergourd(TwoVarieties) 27. Maize 10. BlackGrams(UrdBeans) 35. Peach 11. Bottlegourd 36. Pears 12./13. Cabbage(TwoVarieties) 37. Pomegranate 14./15. Carrot(TwoVarieties) 38. Potato 16. Cauliflower 39./40. Pumpkin(TwoVarieties) 17. Cucumber 41./42. Raddish(TwoVarieties) 18. FrenchBeans 43. RedGrams 19. Garlic 44. Rice 20. Ginger 45. Spinach 21./22. GreenChilly(TwoVarieties) 46. Tomato(Local) 23. GreenGrams(Moong) 47. Tomato 24. Greenginger 48. Turmeric 25. Gur(Jaggery) 49. WaterMelon 26. Lemon 50. Wheat 27. Maize 3

TableA.3: AgriculturalGoodsintheUnitedStates 1./4. Barley(FourVarieties) 18. Peanuts 5. Canola 19./21. Peas(ThreeVarieties) 6. Corn,Grain 22. Rice 7. Cotton,Cottonseed 23. Rye 8. Flaxseed 24. Safflower 9./11. Hay(ThreeVarieties) 25. Sorghum,grain 12. Hops 26. Soybeans 13. Lentils 27. Sugarbeets 14. Maplesyrup 28. Sugarcane 15. Millet,proso 29./31. Sunflower(ThreeVarieties) 16. Mint(TwoVarieties) 32./36. Tobacco(FiveVarieties) 17. Oats 37./40. Wheat(FourVarieties) A.3 State-based Barriers to International Trade Tables A.4, A.5, A.6, A.7 show the state-based (categorized under port versus non-port states) results for the internal trade barriers as a fraction of the total trade barrier for agricultural and manufacturingimportsandexports,respectively. 4

TableA.4: AgriculturalImportTradeBarriers PortStates TotalImportBarrier %Internal (1−β )∗τ + (1−β )+ s a,imp s β ∗δ ∗τ β ∗δ s s a,imp s s Median 3.07 11% AndhraPradesh 3.09 11% Delhi 2.57 0% Goa 2.92 8% Gujarat 3.05 11% Karnataka 3.14 12% Kerala 3.31 16% Maharashtra 2.95 9% Orissa 3.47 18% TamilNadu 2.89 7% WestBengal 4.68 34% Non-PortStates TotalImportBarrier %Internal (1−β )∗d ∗τ + (1−β )∗d + s s,l a,imp s s,l β ∗δ ∗d ∗τ β ∗δ ∗d s s s,l a,imp s s s,l Median 6.95 52% ArunachalPradesh 8.19 58% Assam 10.6 67% Bihar 3.98 26% Chandigarh 3.77 23% Chattishgarh 4.26 30% Haryana 4.95 37% HimachalPradesh 15.1 76% JammuandKashmir 12.8 72% Jharkhand 6.22 48% MadhyaPradesh 4.56 33% Manipur 7.96 57% Meghalaya 13.1 72% Mizoram 7.68 56% Nagaland 8.17 58% Puducherry 3.04 11% Punjab 4.68 34% Rajasthan 5.21 40% Tripura 9.01 62% UttarPradesh 3.87 24% Uttarakhand 10.3 66% 5

TableA.5: ManufacturingImportTradeBarriers PortStates TotalImportBarrier %Internal (1−β )∗τ + (1−β )+ s m,imp s β ∗δ ∗τ β ∗δ s s m,imp s s Median 1.77 29% AndhraPradesh 1.79 29% Delhi 1.49 0% Goa 1.69 22% Gujarat 1.76 28% Karnataka 1.81 31% Kerala 1.91 37% Maharashtra 1.70 23% Orissa 2.01 42% TamilNadu 1.67 21% WestBengal 2.71 63% Non-PortStates TotalImportBarrier %Internal (1−β )∗d ∗τ + (1−β )∗d + s s,l m,imp s s,l β ∗δ ∗d ∗τ β ∗δ ∗d s s s,l m,imp s s s,l Median 4.02 78% ArunachalPradesh 4.74 82% Assam 6.14 87% Bihar 2.30 53% Chandigarh 2.18 49% Chattishgarh 2.46 58% Haryana 2.86 66% HimachalPradesh 8.74 91% JammuandKashmir 7.40 89% Jharkhand 3.59 75% MadhyaPradesh 2.64 62% Manipur 4.60 81% Meghalaya 7.57 89% Mizoram 4.44 80% Nagaland 4.72 82% Puducherry 1.76 28% Punjab 2.70 63% Rajasthan 3.01 68% Tripura 5.21 84% UttarPradesh 2.24 51% Uttarakhand 5.94 86% 6

TableA.6: AgriculturalExportTradeBarriers PortStates TotalExportBarrier %Internal δ ∗τ δ s a,exp s Median 2.88 23% AndhraPradesh 2.79 21% Delhi 2.14 0% Goa 2.93 24% Gujarat 2.84 22% Karnataka 2.92 24% Kerala 3.32 32% Maharashtra 2.72 19% Orissa 3.05 27% TamilNadu 2.66 18% WestBengal 4.73 51% Non-PortStates TotalExportBarrier %Internal δ ∗d ∗τ δ ∗d s l,s a,exp s l,s Median 6.62 65% ArunachalPradesh 7.50 69% Assam 9.67 75% Bihar 3.41 34% Chandigarh 3.14 29% Chattishgarh 3.77 40% Haryana 4.51 49% HimachalPradesh 13.4 82% JammuandKashmir 12.6 81% Jharkhand 5.74 60% MadhyaPradesh 4.16 45% Manipur 7.53 69% Meghalaya 12.5 81% Mizoram 8.36 72% Nagaland 7.73 70% Puducherry 2.93 24% Punjab 4.46 49% Rajasthan 4.76 52% Tripura 8.57 72% UttarPradesh 3.41 34% Uttarakhand 9.84 76% 7

TableA.7: ManufacturingExportTradeBarriers PortStates TotalExportBarrier %Internal τ N/A m,exp Median 2.92 . AndhraPradesh 2.92 . Delhi 2.92 . Goa 2.92 . Gujarat 2.92 . Karnataka 2.92 . Kerala 2.92 . Maharashtra 2.92 . Orissa 2.92 . TamilNadu 2.92 . WestBengal 2.92 . Non-PortStates TotalExportBarrier %Internal d ∗τ d l,s m,exp l,s Median 5.00 27% ArunachalPradesh 6.22 37% Assam 5.33 30% Bihar 3.52 10% Chandigarh 4.28 20% Chattishgarh 3.82 14% Haryana 4.63 24% HimachalPradesh 7.16 43% JammuandKashmir 7.51 45% Jharkhand 4.66 24% MadhyaPradesh 3.92 15% Manipur 6.24 37% Meghalaya 6.56 39% Mizoram 6.23 37% Nagaland 6.22 37% Puducherry 3.21 5% Punjab 4.04 17% Rajasthan 4.22 19% Tripura 6.22 37% UttarPradesh 3.52 10% Uttarakhand 7.70 46% 8

A.4 International Trade Results by Sector TableA.8showstheimportandexporteffectsinresponsetoallcounterfactualsstudiedinSection 5.2. It breaks down the overall trade effects into the sectoral imports and exports for all states, brokendownbyportandnon-portstates. TableA.8: InternationalSectoralImportsandExportsImpact Baseline Import Cross-State Cross- Rural- Cross-State toPort State Urban toU.S. (1) (2) (3) (4) (5) Aggregate Agriculture Import 3% 33% 4% 1% 2% 1% Export 6% 22% 10% 8% 22% 8% Manufacturing Import 27% 39% 54% 23% 45% 23% Export 24% 47% 49% 16% 28% 16% PortStates Agriculture Import 4% 49% 4% 1% 3% 1% Export 10% 33% 10% 8% 31% 8% Manufacturing Import 36% 51% 36% 16% 45% 16% Export 33% 60% 33% 12% 29% 12% Non-PortStates Agriculture Import 1% 15% 4% 1% 1% 1% Export 2% 8% 10% 7% 9% 7% Manufacturing Import 11% 20% 49% 19% 16% 19% Export 10% 25% 44% 13% 8% 13% 9

A.5 Model without Differential Port Access ToquantitativelyassesstheimportanceofdifferentialinternationalportaccessacrossIndianstates, Iestimatetheentiremodelwithoutdistinguishingstateswithdifferentialaccesstoaninternational port, that is, all states have an international port. I recalibrate the four international trade costs: agriculturalandmanufacturingimportandexportcosts,denotedbyτ ,τ ,τ andτ , a,imp m,imp a,exp m,exp respectively,inordertomatchthesameaggregateimportandexportsharesasafractionofsectoral production for agriculture and manufacturing, as shown in Table 4. Table A.9 shows the international trade barriers in a recalibrated model without taking into account differential port access across Indian states. Compared to the border costs in Table 6, the estimated international import cost are almost 40% higher for agriculture and 60% for manufacturing, whereas the export trade barriers are similar. As shown in Table 7, the “old model” does significantly worse in capturing internationaltradeforportversusnon-portstatesinIndia. TableA.9: BorderTradeBarrierswithoutDifferentialPortAccess Agriculture Manufacturing Import τ = 3.13 τ = 1.79 a,imp m,imp Export τ = 2.12 τ = 2.98 a,exp m,exp Next, I perform the same counterfactual exercise of removing international import barriers in India,asgivenincolumnoneofTable9. Thecounterfactualresultsinamodelwithoutdifferential access to international ports across Indian states are presented in Table A.10. An interesting thing to note is that the welfare gains from removing international import barriers are higher: 13% on aggregate compared to 7% in Table 9. Thus, a model without heterogeneous port access, would overestimate the welfare gains by 6 percentage points. In addition, the welfare gains are equally distributed among port and non-port states, which is intuitive because they are no longer differentiated by whether they have access to an international port or not. Therefore, these results show that to assess the aggregate and distributional welfare gains from international integration, it is importanttoincludedifferentialportaccess. 10

TableA.10: InternationalTradeandWelfareImpactswithoutDifferentialPortAccess Baseline Import Cross- Rural- Cross-State State Urban toU.S. (1) (2) (3) (4) InternationalImports Aggregate 17% 48% 6% 22% 10% PortStates 16% 48% 7% 19% 10% Non-PortStates 18% 48% 6% 25% 10% Welfare . Aggregate 13% 29% 19% 12% . PortStates 14% 31% 18% 13% . Non-PortStates 13% 27% 20% 12% A.6 Gravity Equation Estimation As a robustness check, I estimate two separate gravity equations in order to compute the trade barriers directly from the trade flows. Following Simonovska and Waugh (2014), I assume that there is a representative consumer in each state in India that has a CES preference for each good (agricultureandmanufacturing): ˆ σ−1 1 σ σ−1 U s =  q s (j) σ dj . 0 11

Each state draws a good-specific productivity z from a state-specific distribution for each variety, s j. IassumethisdistributionisFréchet: (cid:0) (cid:1) z (j) ∼ F (z) = exp −T z−θ . s s s s If production in both sectors has constant returns to scale (CRS) and labor is immobile across states,thentheunitcostforaspecificvarietyj is w s . z (j) s Each Indian state trades with other Indian states at a per-unit iceberg trade cost τ > 1 to ship sl goodsfromstatel tostates. Therefore,thetotalpriceofanimportedgoodvarietyj fromstatel is τ w sl l p = . sl z (j) l Consumerspurchasefromthelowestcostproducer,whichimplies p (j) = min{p (j );l = 1...S}. s sl g TheCESexactpriceindexforeachdestinationsisgivenby (cid:20) (cid:18) θ+1−σ (cid:19)(cid:21) 1− 1 σ (cid:34) (cid:88) S (cid:104) (cid:105) (cid:35)− θ 1 P = Γ T (w τ )−θ . s l l sl θ l=1 Then, the share of expenditures that state s spends on goods from state l, π , can be written as sl follows X T (w τ )−θ sl l l sl π = = . sl X S (cid:104) (cid:105) s (cid:80) T (w τ )−θ l l sl l=1 Divide the expenditure share for state s on goods from state l by the expenditure share for state s onitsowngoods: X sl / Xs = T l w l −θ τ−θ. Xss / Xs T s w s −θ sl 12

Takelogstoderivethetheoreticalgravityequation: (cid:18) (cid:19) log X sl / Xs = S −S −θlogτ , l s sl Xss / Xs whereS = T w−θ andS = T w−θ. Theempiricalgravityequationcanbewrittenas l l l s s s (cid:18) (cid:19) log X sl / Xs = S −S −θlogτ +ν . l s sl sl Xss / Xs Iassumethattradecoststakethefollowingfunctionalform: logτ = d +b +ex , sl k sl l where d with k = 1,2,...,5 is the effect of distance between the capitals of state s and l lying in k the kth distance interval, and b is a dummy variable that indicates whether states share a border. sl Finally, ex is an exporter fixed effect that allows for asymmetric trade costs at the exporter level. l Byestimatingthegravityequation,Iobtainthetradebarrierslogτˆ uptoaconstantθ. Ipreviously sl estimated the trade elasticity for agriculture, using the methods in Simonovska and Waugh (2014) to be 5.6. For manufacturing I fix it at 5, following the literature. Hence, I then compute the cross–statetradebarriersinIndia. (a)Agriculture: corr=0.73 (b)Manufacturing: corr=0.67 FigureA.2: IndianDomesticTrade 13

A.7 Welfare Welfare is measured as a compensating differential: by how much does the wage in each region need to change in order to receive the same utility in the baseline as in the counterfactual? Total state utility is given by equation (10). Recall that a representative agent in region r in state s choosesthequantityconsumedofbothagriculturalandmanufacturinggoodsofvarietyj ,denoted g by{q(j )} , g ∈ {a,m}tomaximize g jg∈[0,1] ˆ ˆ 1 1 (cid:18) (cid:19) (cid:18) (cid:19) σ (cid:104) (cid:105) σ (cid:104) (cid:105) q s r(ja)≥ m 0, a q s r x (jm)≥0 σ a − a 1 q s r(j a ) σa σ − 1 1 dj a + σ m m −1 q s r(j m ) σm σm −1 dj m 0 0 subjectto ˆ ˆ 1 1 qr(j )pr(j )dj + qr(j )pr(j )dj = wr. s a s a a s m s m m s 0 0 Thefirst-ordernecessaryconditionsaregivenby q s r(j a ) σa σ − a 1−1 = λr s pr s (j a ) q s r(j m ) σm σm −1−1 = λr s pr s (j m ) ˆ ˆ 1 1 qr(j )pr(j )dj + qr(j )pr(j )dj = wr. s a s a a s m s m m s 0 0 Multiplyingthefirsttwofirst-ordernecessaryconditionsbyqr(j )implies s a q s r(j a ) σa σ − a 1 = λr s pr s (j a )q s r(j a ) q s r(j m ) σm σm −1 = λr s pr s (j m )q s r(j m ) ˆ ˆ 1 1 qr(j )pr(j )dj + qr(j )pr(j )dj = wr. s a s a a s m s m m s 0 0 14

Then,integratingthefirsttwoexpressionsoverallvarietiesimplies ˆ ˆ 1 1 q s r(j a ) σa σ − a 1 dj a = λr s pr s (j a )q s r(j a )dj a 0 0 ˆ ˆ 1 1 q s r(j m ) σm σm −1 dj m = λr s pr s (j m )q s r(j m )dj m 0 0 ˆ ˆ 1 1 qr(j )pr(j )dj + qr(j )pr(j )dj = wr. s a s a a s m s m m s 0 0 Substitute these expressions into the objective function in order to obtain the indirect utility at the statelevel:  ˆ ˆ  1 1 (cid:18) (cid:19) (cid:18) (cid:19) σ σ U s = λr s  σ − a 1 q s r(j a )pr s (j a )dj a + σ m −1 q s r(j m )pr s (j m )dj m a m 0 0 (cid:20)(cid:18) (cid:19) (cid:18) (cid:19) (cid:21) σ σ = λr a xr + m xr , s σ −1 s,a σ −1 s,m a m where ˆ 1 xr = qr(j )pr(j )dj s,g s g s g g 0 isdefinedasthestate-andregion-specificexpenditureongoodvarietyj . g 15