Home-Biased Gravity: The Role of Migrant Tastes in International Trade

Transcription

1 Home-Biased Gravity: The Role of Migrant Tastes in International Trade Penglong Zhang December, 2018 Abstract Immigrants tend to buy products from their home countries. As a result, the more immigrants of a given ethnicity a country has, the more it will tend to import from those immigrants country of origin. This effect of migrant heterogeneity is ignored by the standard gravity literature that assumes homogeneous preferences among resident national consumers. This paper embeds that observed regularity into a structural gravity model. Gravity derived from the Almost Ideal Demand System generates bilateral trade shares with three distinct components: ethnic composition of resident population, bilateral trade cost, and per capita income. Using international trade and transnational migration data among 40 countries, this paper estimates the home bias of each ethnic group in tastes. The results show that consumers tastes for products from their country of origin deviate from unbiased levels by 35 percent on average. Ethnic taste bias is found to explain half of the home bias in trade. Counterfactuals suggest that anti-immigration policy significantly impedes trade with countries of origin. Keywords: Home bias, Tastes, Migration. JEL Classification: D10, F11, F14, F22. School of Public Policy and Management, Tsinghua University, Beijing ( zhangpenglong@tsinghua.edu.cn). I am deeply grateful to James Anderson for his guidance throughout my graduate study at Boston College, and to my committee Arthur Lewbel, Zhijie Xiao, and Ben Li for their invaluable feedback, which greatly improved the paper. Pol Antràs, James Rauch, David Atkin, Roberto Rigobon, Felix Tintelnot, Pablo Fajgelbaum, and Amit Khandelwal provided very helpful comments. I also benefited from discussions with Susanto Basu, Fabio Schiantarelli, Rosen Valchev, Jaromir Nosal, Michael Grubb, Maurizio Zanardi, Juanjuan Zhang, and the seminar participants at Boston College, MIT Sloan School of Management, Tsinghua University, the meetings of the Midwest International Economics Group, as well as the European Trade Study Group. All errors are mine.

2 1 Introduction The world is still far from flat today. An extensive literature finds that international trade flows are too small relative to intra-national trade to be explained by observable trade costs the border puzzle or the home-bias-in-trade puzzle. 1 Figure 1 plots the shares of inter- and intra-national trade in total expenditures on manufacturing goods across countries, as well as their native population shares in First, the home expenditure shares (blue bars) are much larger than the levels under hypothetical frictionless trades (dash bars) which are equal to countries production shares. 2 This implies a large home bias in trade (the difference between blue and dash bars). Second, the import (home) share is positively correlated to the immigrant (native) population share, implying a important role of migrant heterogeneity in international trade. Consumers have home-biased preferences. 3 Consumers bring their idiosyncratic preferences with them when they migrate. Thus immigrants tend to buy products from their home countries. This effect of migrant heterogeneity is ignored by the standard trade literature that assumes homogeneous preferences among resident national consumers. 4 The home bias in taste on home goods is believed to contribute to the home bias in trade, but are empirically indistinguishable from trade costs in gravity model (see Anderson, 2011). Ignoring home-biased taste would lead us to overestimate the extent to which trade liberalization can foster market integration. Failure to account for preference heterogeneity causes problems in welfare gain across consumers (see Lewbel and Pendakur, 2017). This paper relaxes the assumption of representative consumer to heterogeneous ethnic consumers in terms of taste biases. More specifically, any given consumer has a taste biased towards the good produced by her country of origin wherever she currently resides 1 The border puzzle (home-bias-in-trade puzzle) is a widely discussed problem in macroeconomics and international finance, first documented by McCallum (1995). He finds that the U.S.-Canadian border led to 1988 trade between Canadian provinces that is a factor 22 times trade between U.S. states and Canadian provinces. Obstfeld and Rogoff (2001) identify this as one of the six major puzzles in international economics. Anderson and van Wincoop (2003) develop the structural gravity model to explain the puzzle with multilateral resistances that are the average trade resistance between a country and its all trading partners. But even after controlling for multilateral resistances, there is still a border barrier that is implausibly high to be explicable by directly measured trade cost tariffs and freight costs (See Anderson and van Wincoop, 2004). 2 In a standard benchmark with homogeneous consumer preference, if foreign products were just as accessible and desirable as domestic ones, then under complete specialization each country would consume its expenditure share of every other country s production. 3 Empirical work using micro-level data provides much evidence. See rice in Morey (2016) and car in Coşar et al. (2018). 4 Atkin (2013) and Bronnenberg et al. (2012) use domestic migration to explore the regional taste differences. Di Comite et al. (2014) propose a new model where consumer preferences are heterogeneous across countries but there are no migrations. 1

3 in the world. Thus consumers are heterogeneous in terms of how large their taste biases are. I extend the structural gravity model by building and estimating a structural component of home-biased preferences. The gravity model generates bilateral trade shares with three distinct components: ethnic composition of resident population, bilateral trade cost, and per capita income. Market taste depends in part on the ethnic origin of consumers. When ethnicities are home-biased in tastes, migration promotes trade with countries of origin. Using international trade and transnational migration data among 40 countries, this paper estimates the home bias of each ethnic group in tastes. The results show that consumers tastes for products from their country of origin deviate from unbiased levels by 35 percent on average. Ethnic taste bias is found to explain half of the home bias in trade. Similar to Atkin (2013), I define the tastes for goods as pure budget share shifters in expenditure-share equations, which measure how much a consumer likes the goods when the qualities and prices are exactly the same. In the Armington setup, 5 a consumer with unbiased tastes (a hypothetical metropolitan) spends her income equally on goods from different countries, 6 while a consumer with fully biased tastes (a hypothetical parochial) spends all her income only on home good given all else equal. Then any arbitrary consumer is in the middle between those two. Then I define the taste bias of a consumer as relative distance between the unbiased and fully-biased tastes. The taste bias is zero for unbiased consumers and one for fully-biased consumers. Thus consumers are heterogeneous in terms of how large their taste biases are. In order to distinguish the taste differences from income differences acting through non-homotheticity, I adopt the Almost Ideal Demand System (AIDS) constructed by Deaton and Muellbauer (1980). 7 AIDS is a second-order approximation to any demand system and it is flexible enough to impose any taste, price or income elasticity structures. The key assumption in this paper is that a consumer s tastes only depend on her country of origin, unrelated to where she currently lives. Thus the consumers from the same ethnic group share the common tastes regardless wherever they are geographically located. 8 One viable explanation is the habit formation in tastes proposed by Atkin (2013), 5 Each country specializes in the production of a distinct variety. 6 Actually I allow for asymmetric tastes for the hypothetical metropolitan in both model and empirical part. The simplification here is just for the intuition discussion. 7 Researchers typically regard products as differentiated by taste (horizontal differentiation) and quality (vertical differentiation). Recent literature usually interprets the high income-elastic goods in the model as high-quality goods and the low income-elastic goods as low-quality goods. Trade papers by Hallak (2006), Fieler (2011), Caron et al. (2014), Feenstra and Romalis (2014), and Fajgelbaum and Khandelwal (2016) find that richer countries export goods with higher income elasticities. 8 I also allow a taste difference between ethnic consumers at home and abroad in the extension part. 2

4 which argues that adult taste favors the foods consumed as a child and the preferences gained in childhood persist into adulthood. 9 Thus, consumers are biased towards the goods with the same national origin of themselves all else equal regardless wherever they are located. For example, a Chinese immigrant in Canada is biased towards Chinese food products, an Italian immigrant in U.K. has a biased taste towards Italian clothing products, and a Japanese immigrant in the U.S. prefers Japanese electronic products, although they have left their home countries for years. This assumption on the preference yields an aggregate demand system that is feasible to estimate. With the taste structure and the ethnic-preference assumption, I derive the home-biased gravity equation by aggregating expenditure shares across ethnic groups within each country. In addition to trade costs and income effects, the gravity also captures the impact of the trade promoter that stimulates bilateral trade share via immigration. The intuition is that the biased immigrants spend larger shares on their home products than the natives do, increasing the import shares from their home countries to their host countries. For example, the higher the Mexican immigrant share in the U.S. population, the higher import share of Mexican products in the U.S. market. The marginal effect of this promoter is stronger when ethnic Mexicans taste bias is larger. In additional to multilateral resistances, the home-biased gravity also gives multilateral promoters which are the average terms of bilateral promoters (immigrant shares) across all trade partners. Next, I obtain the symmetric home-biased gravity by assuming that both products and taste biases are symmetric. By comparing this simplified gravity with CES structural gravity, I provide structural components to the unobservable bilateral market tastes that are usually taken as error terms in literature. The key implication of the home-biased gravity equation is that the bilateral taste is determined by the bilateral migration, and its marginal effect is determined by the taste bias. The home-biased gravity equation gives three implications on intra-national trade. First, non-immigration countries or those with more biased consumers have a higher home bias in trade. Second, countries with lower domestic trade barriers, or producing more price-elastic goods have a higher home bias in trade. Third, rich countries producing high-quality goods, or poor countries producing low-quality goods have a higher home bias in trade than the others. The last two implications are straightforward. The intuition of the first one is that immigration countries, like the U.S., import more than non-immigration ones, like Japan, because large shares of immigrants spend more on products imported from their homes, resulting in a smaller domestic share of the host 9 He terms this process habit formation and provides ample evidence in the psychology and nutrition literature. See Bronnenberg et al. (2012) for more discussion. 3

5 country. The magnitude of this effect is determined by the native population share and their taste bias for home good relative to global ones. In the empirical part of the paper, I start with estimating the home-bias gravity model. The taste shifters are the ethnicity-specific coefficients of immigrant shares. With international manufacturing trade and transnational migration data among 40 countries, I find that the import share is significantly increasing in immigrant share for most ethnic groups with other standard gravity variables as controls. The four ethnic groups with the highest taste shifters are Japanese, Chinese, Brazilians, and Americans, while the four with lowest taste shifters are Luxembourgers, Maltese, Belgians, and Dutchmen. On average, the taste shifters is This paper also estimates sectoral gravity in agricultural and service sectors, final and intermediate goods, as well as 14 manufacturing industries. It is found that consumers have larger taste bias in agricultural products, while almost unbiased on services. Import share is more sensitive to immigrant share for final goods than for intermediate goods. Sectoral estimation makes my results applicable to the micro level literature on home preference. I also address two potential econometric issues by estimating the symmetric homebiased gravity equation. First, this trade-creating effect by immigration could also work through the trade cost channel in which migration network reduces information barriers and enhances contract enforcement. I add two variables measuring network effect into the regression. First one is the probability that, if we select an individual at random from each country, they will have a connection defined as common national origin. This measure extends the one in Rauch and Trindade (2002) that only captures the ethnic- Chinese Network. The second one is emigration share that also contributes to the network but has no effect on trade through preference channel. The results show that the network effect channel is not significant to aggregate manufacturing import share. Second, the estimated coefficient is subject to potential endogeneity issues. In order to address the reverse causality from migration to trade, I instrument the stock of immigrants in 2010 by the stock in 2000, similarly to Card (2001). I also estimate the gravity equation by taking first-order differences between two cross-sections to eliminate any potential timeinvariant omitted variables. Both strategies leave the main results unchanged. In robustness checks, I examine asymmetry on the unbiased tastes and the price elasticities. By combining results of unbiased tastes and taste biases, I find that both world and American consumers favor American products. Germans and British are less biased in their taste but world consumers favor their products. By combining results of price and income elasticities, I find that American, German, and Italian products are high quality and more price-elastic, while Indonesian, Indian, and Chinese products are low quality 4

6 and less price-elastic. Using the estimated parameters, I conduct two counterfactual experiments. First, I decompose the contribution of the taste effect, trade cost effect, and income effect to the home bias in trade. About half of the trade bias is explained by the trade cost, the other half is explained by the taste bias, very little is explained by the income effect. Second, I check how migration policy affects home bias in trade. The U.S. will decrease its openness (import share) by 6 percent with an immigration ban. Lastly, I extent the model by introducing the local bias in addition to the home bias. On one hand, a consumer has a biased taste towards the good produced by her home country (where she is originally from). On the other hand, her taste is also biased toward to the good produced locally (where she is living). To estimate the local bias parameters, I impose the constraint that local bias is proportional to native population share which implies that immigrants are more likely to assimilate in a country with more natives. The results show that consumers located at countries with higher native share are more likely to locally bias in their tastes. After controlling for local bias in gravity, the home taste shifters estimated are mostly smaller but close to the baseline results. This suggests that the immigrants taste bias is robust even with assimilation. This paper contributes to the gravity literature in international trade in two aspects. First, it introduces consumer preference heterogeneity into gravity models, which is additional to the gravity with firm heterogeneity (Chaney, 2008; Helpman et al., 2008), the non-homothetic gravity with consumer income heterogeneity (Fajgelbaum and Khandelwal, 2016), and the gravity with consumer geography heterogeneity (?). Second, it allows home-biased tastes when discussing trade with AIDS preference, which is supplementary to translog gravity (Novy, 2013), trade includes entry of new goods (Feenstra, 2010), price and income of international comparison (Neary, 2004), and so on. My paper helps explain the border puzzle or home-bias-in-trade puzzle by homebiased preferences, which is supplementary to other explanations, e.g., trade cost (Obstfeld and Rogoff, 2001), multilateral resistance (Anderson and van Wincoop, 2003), nonhomothetic preference (Caron et al., 2014), multi-stage production (Yi, 2010), contractual enforcement (Anderson and Marcouiller, 2002), and scale effect (Ramondo et al., 2016). Chaney (2014, 2016) and Allen (2014) discuss the resistances from information barrier. Head and Mayer (2013) summarize the sources of resistance as imperfect information, very localized tastes, and distribution networks. There is a large literature that examines the effect of international migration on international trade. Some papers have explored the effect of immigration on bilateral trade flows, typically finding that immigration and trade are complements. In particular, im- 5

7 migrants are found to reduce barriers to exports by facilitating communication between firms and reducing setup costs in the destination country, which is called network effect by Rauch and Trindade (2002, 2003). This trade cost channel of migration on trade is explored also in Combes et al. (2005), Iranzo and Peri (2009), and Felbermayr et al. (2015). This paper is along this line by providing a different perspective on preference channel. More broadly, my work is also related to the rich literature on the effect of migration on economic outcomes, as well as work examining the importance of ethnicity and ethnic diversity. Hanson (2010) discusses more consequences of international migration in additional to trade. Literature focuses on migration effects on labor market (Ottaviano et al., 2013; Foged and Peri, 2016), on GDP per worker (Fulford et al., 2017), on invest (Burchardi et al., 2017), on income (Ortega and Peri, 2014), and on wealth redistribution (Luttmer and Singhal, 2011). The rest of this paper is organized as follows. In Section 2, I present the model and derive the home-biased gravity. In Section 3, I estimate the gravity to get the estimates of the taste biases and all the other theoretical parameters. The robustness checks are conducted in Section 4. Section 5 gives counterfactual experiments and Section 6 extends the model to include local bias. I conclude and discuss future directions in Section 7. 2 Model I present how to incorporate home-biased preferences into the international trade framework in Section 2.1, and derive the home-biased gravity equation in Section 2.2. Section 2.3 discusses the gravity s implications for the home-biased trade. 2.1 Preferences The world consists of N countries, indexed by i as exporter and n as importer. Each country specializes in the production of a distinct variety. Let p ni be the price in country n of the good imported from country i. The iceberg trade cost of exporting from i to n is t ni. No-arbitrage condition implies that p ni = p ii t ni. There are N ethnic groups of consumers, indexed by h as the national origin. Each ethnic group is globally distributed and thus each country is populated by N ethnic groups. Note that n, i, h = 1, 2,..., N. Let s nh be the immigrant population share in country n from country h if h = n, and s nn be the native population share in country n. Then h=1 N s nh = 1 holds for all n. Consumers have the Almost Ideal Demand System (AIDS) preference introduced by Deaton and Muellbauer (1980), which can be rationalized as a non-homothetic second- 6

8 order approximation to an arbitrary expenditure system. Specifically, any consumer who is from country h and resides at country n has expenditure functions given in logarithmic form as ln e h n = ln Q h n + u h n N i=1 (p ni ) φ i, (1) where e h n is the minimum expenditure at which the consumer can obtain utility u h n given prices p ni. The price index ln Q h n is given in logarithmic form as ln Q h n = N i=1 α h i ln p ni N i=1 N i =1 γ ii ln p ni ln p ni. (2) Applying the Shephard s lemma and differentiating the expenditure function with respect to log price p ni generates the expenditure share in good (produced by country) i for consumer (of ethnic group) h at country n equal to w h ni = αh i + for i = 1,..., N. The shares must be non-negative. ( ) N en γ ii ln p ni + h φ i ln i Q =1 n h, (3) These expenditure shares have some nice features. First, αi h is a taste parameter of consumer h for the good i which shifts the expenditure share independently from the prices and income. Second, γ ii is the cross-price elasticity for good i when i = i and the own-price elasticity for good i when i = i. Third, φ i is the income elasticity which captures the non-homothetic component of the preference. Positive φ i implies luxury goods (with high quality) while negative φ i implies necessary goods (with low quality). 10 I refer to e h n/q h n as adjusted real income (expenditure) by individual price index which varies by the individual consumption basket compositions. To satisfy homogeneity of degree one, the parameters are constrained by α h i [0, 1], N i=1 αh i = 1 for all h, N i=1 γ ii = 0 for all i, and N i=1 φ i = 0. Symmetry is imposed to satisfy the Young s Theorem, γ ii = γ i i. Concavity is imposed by the requirement that {γ ii } is negative semi-definite. When φ i = 0 for all i, AIDS becomes the homothetic translog preference. When γ ii = 0 for all i, i and φ i = 0 for all i, AIDS becomes the Cobb-Douglas preference. 10 Note that γ ii and φ i are semi-elasticities since they relate expenditure shares to logs of prices and income, but I refer to them as elasticities to save notation. Actually the price elasticities are 1 + γ ii /w h ni + w h ni, and the income elasticities are 1 + φ i/w h ni. 7

9 Following Anderson and Zhang (2018), 11 I specialize the general γ ii to γ ii = γβ i ( δ ii + β i ), (4) where the Kronecker delta δ ii is 1 when i = i and 0 otherwise, β i [0, 1], and i=1 N β i = 1. Specialization (4) satisfies all the general restrictions but imposes a tight restriction on the cross-effects. In particular, complementarity is ruled out, all off-diagonal terms of the substitution effects matrix are non-negative. β i = 1/N is the special case proposed by Feenstra (2003). 12 The variation of β i allows for asymmetric demand responses to price changes. This gives AIDS preference CES-like components because the price terms in equation (3) become γβ i ln (p ni / p n ) which captures cross-effects in substitution with the log of a ratio of own price to an average price ln p n = i N =1 β i ln p ni. Home-Biased Tastes In equation (3), α h i is a taste parameter for good i but heterogeneous across consumers. It is unvarying to where consumers consume but varying to where they are originally from. One viable explanation is the habit formation in tastes proposed by Atkin (2013), which argues that adult taste favors the foods consumed as a child and the preferences gained in childhood persist into adulthood. 13 Therefore, consumers from the same ethnic group share the common tastes regardless wherever they live. Different to Atkin (2013), the tastes in this paper are assumed to be exogenous. 14 Another important feature of the differences in consumer tastes is home bias. A large amount of empirical work from disaggregate level data provides evidence for its existence. For any consumer, she has a biased taste towards the good produced by her home country (where she is originally from) whatever her host country (where she resides and consumes) is. In other word, consumers are biased towards the goods with same national origin of themselves all else equal regardless wherever they are located. For example, an Italian immigrant in the U.S. has a biased taste towards Italian products even though he has resided in the U.S. for years. An American living in the U.S. is biased towards American products since they have the same national origin. More structure on the home bias yields an aggregate demand system that is feasible to estimate. First, I assume a hypothetical world unbiased consumer with tastes for good 1, 2,..., N 11 They use AIDS preference to obtain the Almost Ideal gravity to study international zero trade flows. 12 This special case is followed by Novy (2013) and Fajgelbaum and Khandelwal (2016). 13 He terms this process habit formation and provides ample evidence in the psychology and nutrition literature. 14 Lan and Li (2015) discuss how economic openness affects nationalism. 8

10 as {α 1, α 2,..., α N }, and N i=1 α i = 1. Then I specialize ethnic consumer h s tastes as α h i = (1 θ h )α i + θ h δ h i, (5) where the Kronecker delta δi h is 1 when i = h meaning home goods, and 0 otherwise, and θ h [0, 1] is the taste shifter which measures how biased ethnic consumer h s taste is towards her home good h. This specialization satisfies conditions N i=1 αh i = 1 for all h. For any consumer h, she raises her taste for home good h by a shifter θ h, while reduces those for all the other goods by 1 θ h (as well as the home good to make sure taste aggregates equal to 1). When θ h = 0, ethnic consumer h are unbiased, exactly like the hypothetical world consumer. When θ h = 1, the consumers from ethnic group h have fully biased tastes, i.e. 1 for home good while 0 for all the others. Thus the set of taste shifters {θ h } N h=1 is the measurement of the taste biases across ethnicities, and it is the key set of parameters in this paper. Now I rewrite the expenditure equations (3) as w h ni = (1 θ h)α i + θ h δ h i γβ i ln ) ( + φ p i ln n ( pni e h n Q h n ), (6) where average price in market n is ln p n = i=1 N β i ln p ni. And γβ i are the semi price elasticities and φ i are the semi income elasticities. A different way to interpret the taste shifter θ h is the fraction of fully biased consumers in ethnic group h. In this case, I assume there are two types of consumers in each ethnic group. One is a cosmopolitan who is unbiased exactly like the hypothetical world consumer, and the other one is a parochial who spends all on home product while zero on others given all else equal. If their population shares are 1 θ h and θ h respectively, then the average taste of ethnic group h for home good is (1 θ h )α h + θ h 1, and tastes for the other goods is (1 θ h )α i + θ h 0, which are exactly the same as equation (5). Thus {θ h } i=1 N could be interpreted as the parochial population shares of each ethnic group, and they measure how ethnicities are biased. When θ h = 0, ethnic group h are all cosmopolitans. When θ h = 1, ethnic group h are all parochial. A byproduct of this taste structure is within-country consumer heterogeneity. Given good i, only consumer i raises her taste by a shifter θ i, while all the others reduce their tastes by different levels, 1 θ h, in terms of their own taste shifters. For any good, consumers in one market thus have heterogeneous tastes for it due to their heterogeneous taste shifters. When θ h = 0 for all h = 1, 2,..., N, all consumers anywhere in the world are homogeneous in preference, which is exactly the same to the case with a representative consumer. 9

11 2.2 Home-Biased Gravity In this part, I first aggregate individual expenditure shares across all ethnic groups within each market (country), and then derive the gravity equation for each country pair that depends on aggregate data and the demand parameters. I assume that ethnic groups within a country share the same income, i.e. e h n = e n. 15 Aggregating individual expenditure shares across ethnic groups gives the market import share of country n from country i by w ni = N h=1 s nhw h ni, i.e., w ni = (1 θ n )α i + θ i s ni γβ i ln where the average taste bias in market n θ n = ( pni p n ) ( ) en + φ i ln, (7) Q n N s nh θ h, (8) h=1 and the average price index in market n by aggregating individual price index ln Q h n, ln Q n = N s nh ln Qn. h (9) h=1 Equivalently, equation (7) could be taken as the expenditure shares of a representative consumer in market n who is a mixed-blood with all ethnic origins. Since ethnic compositions differ, the representative consumers in different markets have different demand. Next, I let X ni be the value of imports of importer n from exporter i, and let E n be the total expenditure of the importer. Then w ni = X ni E n (10) is the share of aggregate expenditures in country n devoted to goods from country i. Then equation (7) could be rewritten as X ni E n = [α i γβ i ln ( ) pii ] + (θ p i s ni α i θ n ) γβ i ln ( tni P n ) ( ) en + φ i ln, (11) Q n 15 On the absence of data, I skip within-country cross-ethnicity income difference, but keep cross-country income difference. Lewbel (1991) finds E[e h ln e h ]/Ee h = ln Ee h + Σ with U.S. data and later literature, like Fajgelbaum and Khandelwal (2016), follow his idea to control for the index Σ with country fixed effects. 10

12 where p = N i=1 β ip ii and ln P n = N i=1 β i ln t ni. Income of each exporter i equals the sum of sales to every country, Y i = N X ni, (12) n=1 where Y i measures exporter i s total income. Substitute equation (11) into equation (12) and divide through by Y which is the total world income. Subtract the ( resulting ) expression from equation (11) and simplify by canceling term [α i pii γβ i ln p ]. Denote ln r n = ln (e n / Q n ) as the real expenditure, then I term the resulting expression as the home-biased gravity X ni Y ( ) i E n Y = (θ is ni θ i Πi s α ipn) s tni ( rn ) γβ i ln + φ Π i P i ln, (13) n r where and and ln Π i = Π s i = N n=1 N n=1 ( ) En ln t Y ni, ln P n = ( ) En s Y ni, Pn s = ln r = N n=1 N β i ln i=1 ( tni Π i ), (14) N θ i (s ni Πi s ), (15) i=1 ( ) En ln r n. (16) Y On the left hand side, X ni E n Y i Y is the deviation of bilateral trade per unit of n s expenditure from its frictionless level Y i Y. There are three terms on the right hand side, which capture the taste effect, price effect, and income effect, respectively. The second term, γβ i ln ( tni Π i P n ), is the effect of relative bilateral trade resistance from origin i to destination n, where ln Π i and ln P n are the outward and inward multilateral resistances in log, respectively. This term is very similar to the CES structural gravity in Anderson and van Wincoop (2003). The last term, φ i ln ( r nr ), is the non-homothetic component of the gravity equation and captures the effect of relative income per capita of market n where ln r is the average world income per capita in log. This term is very similar to the non-homothetic component of the gravity equation in Fajgelbaum and Khandelwal (2016). The first term, θ i s ni θ i Π s i α ip s n, is the one of the most important novelties in this paper. It captures the effect of relative trade promoter that stimulates bilateral trade via bilateral migration from origin i to destination n. The intuition is that immigrants raise the market average taste for their home country s products. This is because immigrants with home-biased tastes spend larger shares on home-produced products, increasing the 11

13 import shares of the host country from their home countries. The magnitude of the promoter is increasing in how large the bilateral immigrant share, s ni, is, as well as how large their taste shifter, θ i, is. I refer to Πi s and Ps n respectively as the outward and inward multilateral promoters that summarize the average trade promoters between a country and its trading partners. When everybody is unbiased, i.e. θ i = 0 for all i, the trade promoters are shut down, and, together with a symmetry assumption β i = 1/N, this gravity equation corresponds to that in Fajgelbaum and Khandelwal (2016). Symmetric Home-Biased Gravity In order to compare the home-biased gravity with the CES gravity in Anderson and van Wincoop (2003), I impose symmetry conditions, i.e. θ i = θ, α i = 1/N, β i = 1/N and φ i = 0 for all i, denote γ = γ/n, then I get the symmetric home-biased gravity X ni Y ( ) i E n Y = θ(s ni Πi s Ps n) γ tni ln, (17) Π i P n where and ln Π i = Π s i = N n=1 N n=1 ( ) En ln t Y ni, ln P n = 1 N ( ) En s Y ni, Pn s = 1 N N ln i=1 ( tni Π i ), (18) N (s ni Πi s ). (19) i=1 When θ = 0, the taste effect is shut down and the symmetric home-biased gravity is identical to the translog gravity in Novy (2013). Recall the CES gravity is ln X ni ln Y ( ) i E n Y = Ω tni ni + (1 σ) ln Πi, (20) P n where (Π i )1 σ = N n=1 ( En Y ) ( tni P n ) 1 σ, (P n ) 1 σ = N i=1 ( ) ( ) 1 σ Yn tni Y Πi. (21) Comparing equation (17) and (20) gives the the difference of the functional form on the left hand side: a difference in shares vs. a difference in log shares. The own price effect in the two gravity equations is essentially the same while the cross price effects are captured by price index terms that differ, but would ordinarily be econometrically controlled for with origin and destination fixed effects. What s more important is that the unobservable bilateral taste effect Ω ni is offered a micro fundamental structure in this paper while taken 12

14 as part of the errors in traditional gravity literature. Since the multilateral promoters are absorbed by origin and destination fixed effects, econometrically the difference boils down to an additional term, the immigrant share. The coefficient of immigrant shares is just the taste shifter. 2.3 Home-Biased Trade Since we are interested in home bias in trade, I can write down the home share deviation based on the home-biased gravity equation (13) as HB n X nn E n Y ( ) n Y = (θ ns nn θ n Π s n α n Pn) s tnn ( rn ) γβ n ln + φ n ln, (22) Π n P n r where HB n is defined as country n s home bias in trade, or trade bias, that measures the deviation of home expenditure share from its frictionless level. Equation (22) shows that home bias in trade could be explained by three effects separately: taste effect, price (trade cost) effect, and income effect. Regarding the three effects, the following implications are obtained. Implication 1. Non-immigration countries have a larger trade bias than immigration ones. Home countries of more biased consumers have a larger trade bias. Implication 2. Countries with lower internal trade barriers have a larger trade bias. Countries producing more price-elastic goods have a larger trade bias. Implication 3. Rich countries producing luxury goods or poor countries producing necessary goods have a larger trade bias than the others. Implication 1 holds because HB nn s nn = θ n > 0 and the intuition is that immigration countries (with smaller s nn, like the U.S.) import more than non-immigration countries (with larger s nn, like Japan) since large amount of immigrants spend more on products produced by their home countries, resulting in less domestic expenditure share of the host country. The magnitude of this effect is determined by the natives taste shifter for home good relative to that for the global good, which are controlled by Π s n and P s n. Implication 2 is about the trade cost effect that higher internal cost reduces domestic consumption relative to imports, which is consistent with the structural gravity literature Anderson and van Wincoop (2003) and Anderson and Yotov (2010). 16 Implication 3 implies that the 16 In Anderson and Yotov (2010), they propose and estimate a similar index constructed home bias (CHB) by CES structural gravity: CHB n = ˆX ( ) ni /Y n Y i /Y = ˆt 1 σ. nn ˆΠ n ˆP n 13

15 income effect increases trade bias for rich countries above average (ln ( r nr ) > 0) which produce positively semi-income-elastic goods (φ i > 0, luxury goods), like Germany, or poor countries below average (ln ( r nr ) < 0) which produce negatively semi-income-elastic goods (φ i < 0, necessary goods), like China. This result of income effect is consistent with Caron et al. (2014). 3 Estimation In this section, I estimate the home-biased gravity derived in Section 2. Section 3.1 describes the data, and Section 3.2 presents the main estimation results. I also discuss some other econometric issues in Section Data To estimate the home-biased gravity, I merge datasets on bilateral migration, trade, and gravity respectively. First, the migration data are from United Nation International Migrant Stock Dataset that presents estimates of total immigrant stock at mid-year by origin and destination for year of 2000 and 2010, and are available for all countries and areas of the world. The estimates are based on official statistics on the foreign-born or the foreign population. I use the data in year of 2010 as the baseline, and do robustness checks with the data in year of Second, trade flows are from World Input-Output Database (WIOD) that records bilateral trade flows and production data for 40 countries (27 European countries and 13 other large countries) across 35 sectors that cover food, manufacturing, and service. The data also record total expenditures by sector and country of origin, as well as final consumption. Third, I obtain bilateral distance, common language, border, and tariff information from CEPII s Gravity database. Price levels, adjusted for cross-country quality variation, are obtained from Feenstra and Romalis (2014). GDP, GDP per capita, and population are from the Penn World Tables. Recall the home-biased gravity equation X ni E n = Y ( ) i Y + (θ is ni θ i Πi s α ipn) s tni ( rn ) γβ i ln + φ Π i P i ln, n r where there are a large number of parameters to estimate. Actually only the taste shifters {θ i } i=1 N are the focus of this paper. Thus I can reduce the number of estimated parameters by imposing some restrictions. First, let β i = 1/N for all i, which leads constant own price 14

16 elasticity following Feenstra (2003). 17 Second, income elasticities and exporter income are assumed to satisfy φ i = ψ c + ψ ln r i for all i, similar to Feenstra and Romalis (2014) and Fajgelbaum and Khandelwal (2016). The theoretical restriction N i=1 φ i = 0 implies ψ c = ψ(1/n) i=1 N ln r i, transforming this linear relationship to φ i = ψ(ln r i ln r) where ln r = (1/N) i N =1 ln r i, and reducing the number of income elasticity parameters to be estimated from N to one. For the tastes of the hypothetical unbiased world consumer, I proxy α i = EM i / i EM i where EM i is the extensive margin in global trade estimated by Hummels and Klenow (2005). 18 I will also go through a number of robustness checks to ensure that my results do not solely depend on this particular extensive margin measure. Then the home-biased gravity equation becomes X ni E n = θ i s ni γ ln t ni + ψ ln r ni + λ n α i + f i + g n + ɛ ni, (23) where ln r ni = ln r n ln r i, f i = Y i /Y θ i Π s i + γ ln Π i + ψ ln r i ln r, and g n = γ ln P n + ψ ln r n ln r + ψ ln r ln r. I proxy bilateral trade costs with bilateral observables. Specifically, ln t ni = ρ ln d ni + ρ b border ni + ρ l lang ni + ρ t tari f f ni + ρ i internal ni + ε ni, (24) where d ni is the bilateral distance between origin i and destination n. Parameter ρ reflects the elasticity between distance and trade costs, border ni and lang ni are common border and common language, tari f f ni is bilateral tariff level, and ε ni is idiosyncratic errors for trade costs. Importantly, internal ni is 0 for import and 1 for internal trade similar to Ramondo et al. (2016) and Anderson and Yotov (2017). Then the specification of the home-biased gravity is X ni E n = θ i s ni + b ln d ni + b b border ni + b l lang ni + b t tari f f ni + b i internal ni + ψ ln r ni + λ n EM i + f i + g n + ɛ ni. (25) And f i, g n could be controlled by fixed effects of origins and destinations. There are 86 parameters to be estimated {b, b b, b l, b t, b i, ψ, {θ i } 40 i=1, {λ i} 40 i=1 }. Actually {λ i} 40 i=1 are used to control the effects of multilateral promoters, and thus not the coefficients of interest. 17 I allow asymmetric price elasticities in robustness check. 18 They construct a measure of the extensive margin across countries based on shipments in more than 5000 six-digit product categories from 126 exporting countries to 59 importing countries for the year The extensive margin is measured by weighting categories of goods by their overall importance in exports. They document that the extensive margin tends to be larger for big countries. 15

17 Recall the real expenditure per capita is defined as ln r n = ln (e n / Q n ), where e n, nominal expenditures per capita, are observable too. Aggregate price index ln Q n can be proxied by a Stone index following literature, 19 that is ln Q n = N w ni ln(p ii d ρ ni ), (26) i=1 where p ii are the quality-adjusted prices estimated by Feenstra and Romalis (2014) and the value of ρ will be discussed later. This proxy of the aggregate index is identical to the aggregate proxy of the individual indices due to nice property of the Stone index. 20 Because we do not directly observe trade costs, I cannot separately identify ρ and γβ. Following Fajgelbaum and Khandelwal (2016) and Novy (2013), I set ρ = Then the parameters of interest {θ i } i=1 N could be estimated directly and other theoretical parameters {φ i } i=1 N and γ are identified by φ i = ψ(ln r i (1/N) i N =1 ln r i ), and γ = Nb/ρ. To investigate more variation of the taste biases, I also estimate sectoral (and industrial) gravity equations across 14 manufacturing industries, final and intermediate good sectors, as well as food and service sectors using more disaggregate level data. Disaggregate estimation makes my results able to speak to many micro level literature on home preferences. More specifically, I estimate Xni k En k = θ k i s ni + b k ln d ni + b kb border ni + b kl lang ni + b kt tari f f ni + b ki internal ni + ψ k ln r k ni + λk nem i + f k i + g k n + ɛ k ni, (27) where all variables with a superscript k is defined in the same way to those without any superscript but in good class k. 21 I run the regression separately with corresponding data and obtain the taste shifter estimates across all ethnicities in all good classes. 19 Deaton and Muellbauer (1980) first time use a Stone index to proxy the AIDS price index. The trade literature, like Atkin (2013) and Fajgelbaum and Khandelwal (2016), follow this approximation. 20 If I instead proxy individual price index ln Qn h = i=1 N wh ni ln(p iid ρ ni ), the aggregate of individual index, followed by equation (9), becomes ln Q n = h=1 N s nh ln Qn h = h=1 N i=1 N s nhwni h ln(p iid ρ ni ) = i=1 N w ni ln(p ii d ρ ni ) which is exactly the same to equation (26). 21 Good classes could be final and intermediate goods, or food, manufacturing and services, or different industrial goods. 16

18 3.2 Main Results The baseline results are reported in column (1) in Table 1. It reports the estimates of 40 ethnicity-specific coefficients on Immigrant share, and coefficients on variable Distance, Border, Language, Tariff, Internal, as well as that on income per capita product, while estimates of the fixed effects and other coefficients are dropped since they are not the parameters of interest. The estimate of distance elasticity is , significantly negative, suggesting that bilateral distance reduces trade flows between countries, which is consistent with literature. Under the assumption that ρ = 0.177, the estimate implies γ = The additional trade costs common border, common language, tariff, internal trade also have the intuitive signs. The estimate of the income elasticity parameter, ψ, is 0.006, significantly different from zero. This means that richer countries are more likely to spend on products from richer countries, conditional on trade costs and tastes, consistent with literature. The table also reports the estimates of the 40 θ i parameters, one corresponding to each ethnic group, in the subsequent rows. The four ethnic groups with highest θ s are Japanese, Chinese, Brazilians and Americans, with taste shifters above 0.5. The four ethnic groups with lowest θ s are Luxembourgers, Maltese, Belgians, and Dutchmen, with taste shifters not different from 0. This means the former ethnicities are highly homebiased consumers whereas the latter are almost unbiased. On average, the taste bias measured by the ethnicity-specific immigrant coefficients is My results above provide the evidence of taste bias existence and estimate the magnitude of the bias for every ethnic group. But what causes the biases and why they differ among ethnicities are very complicated. Thus they are not the key issues in this paper. To investigate more variation of the taste biases, I also estimate sectoral gravity in equation (27) across manufacturing, agricultural and service sectors, as well as final and intermediate goods using more disaggregate level data. First I estimate the taste bias in three sectors in Figure 2 (See full results in Table 2). Compared to manufacturing sector (the baseline estimates) in column (1), agricultural sector in column (2) tends to be more favored by home consumers, while service sector in column (3) is much less biased. In other word, consumers have larger taste bias (0.448 on average) when they are facing agricultural products, while almost unbiased (0.053 on average) on services. This makes sense based on the taste habit formation assumption that food tastes are even more persistent. Note Indians are the most biased ethnic group in agricultural sector. This is consistent with what we usually observe that Indian immigrants prefer Indian food although they tend to be far more highly educated and. Next I report the results for two types of goods in Table 3. Import share is more sensitive to immigrant share for final good in column 17

19 (2) than for intermediate good in column (3) on average. This is very consistent to the consumer taste theory. The industrial estimates are obtained in Table A.1-A.2 in Appendix. Table 4 displays the consumers average taste bias on different industrial goods. Their tastes are more biased for goods from Textiles, Leather products, transportation products, and food, beverages, and tobacco, while much less biased in non-metallic minerals, rubber and plastics, and electrical and optical equipment. Table 5 displays the most biased ethnic groups in different industries. For examples, Australians and Americans are biased when buying food, beverage, and tobacco, Japanese and Koreans are biased when buying transportation products, and Canadians are biased when buying wood products. 3.3 Other Results Symmetric Bias To compare the estimates with literature, I also estimate the symmetric home-biased gravity. The econometric specification of equation (17) is X ni E n = θs ni + b ln d ni + b b border ni + b l lang ni + f i + g n + v ni, (28) where f i = Y i Y θps + γ ln Π i and g n = γ ln P n θπ s i. And f i, g n could be controlled by fixed effects of origins and destinations. To make the estimates comparable with literature, I only include the foreign trade. Thus the tariff term is captured by the importer s fixed effect. Internal trade dummy is also dropped since there are no internal trade observations any more. The results are reported in Table 6. Column (1) reports the estimates of coefficients on variable Immigrant share, Distance, Common border, and Common language, while the estimates of fixed effects are dropped since they are not the parameters of interest. The estimate of distance elasticity is negatively significant suggesting that bilateral distance reduces trade flow between countries, consistent with literature. Coefficients on other gravity variables are also significant. Common border and common language increase bilateral trade flow, while tariff reduces bilateral trade flow significantly. The coefficient on immigrant share is estimated to be 0.405, significantly different from zero, which confirms the existence of consumer s taste bias. An interpretation of its implication is 41 percent of consumers in the world are parochial and they will spend all on home goods given all else equal, while the other 59 percent are cosmopolitans and they are indifferent among goods with different national origins. This result is also close to the average taste 18

20 shifter estimates, 0.352, in the baseline regression. I also estimate equation (28) for a different cross-section, the year of 2000, for which migration data are available. One of the reasons is to check that my results are not the artifact of a particular time period. The second reason is to check whether there is a change in the immigrant share coefficient. Column (2) reports the estimates. The taste bias estimate is with little change compared to that in the year of 2010, which confirms the robustness of the results. This is consistent with my basic assumption that the consumer tastes are persistent and the estimate is a good measure for consumers taste bias. Column (3) tests the gravity shutting down the taste term, i.e. translog gravity, and the results are very similar to Novy (2013). Together with column (1), it reveals that the coefficients of traditional gravity variables change very little with or without the additional migration variable. This implies that migration share is not correlated to traditional gravity variables, and it does explain a new part of the variance of trade shares that traditional gravity variables do not capture. Column (4) estimates the standard CES structural gravity. Since the dependent variable is in log, the elasticity estimated is different to that in other columns Endogeneity There are some potential concerns related to the estimation. The first concern is that immigrant share may be correlated with trade shocks v ni. When this is the case, the OLS estimation leads to inconsistent results. Such endogeneity bias can arise from two sources. The first is reverse causality. It is possible that some positive shock on the value of bilateral trade between two countries leads to more migration between the two places. For example, more people migrate internationally to do business associate to more international trade. The second is omitted variables. Migration may be correlated with unobserved factors that also affect trade, such as the trading partners cultural similarity or bilateral economic policies. In order to address the reverse causality from migration to trade, I instrument the stock of immigrant in 2010 by that in 2000 similarly to Hanson (2010). 22 Then I run a Two-Stage Least Square regression for equation (28) to test whether my estimation suffers from a big endogeneity problem. I also estimate the gravity equation by taking first differences between the above two years to eliminate any potential time-invariant omitted variables. 22 Card (2001) firstly suggest thats supply-push component of the immigrant inflows to a particular city, which is based on historical settlement patterns and the total number of newly arriving immigrants from different source countries, is a potential candidate for such an instrumental variable for migration. 19

21 More specifically, I estimate to test whether θ is significant or not. X ni E n = θ s ni + f i + g n + v ni, (29) The results are reported in Table 7. First, I check whether the lagged immigrant share in the year of 2000 is a valid instrument for current immigrant share in the year of column (2) shows the instrument s coefficient is significant and the F-test value is 41.2, which implies that the explanatory power of instruments is high. Then I do a 2SLS regression. The immigration s coefficient is still significant reported in column (3). It is 0.08 smaller than, very close to, the OLS estimate in column (1). The potential endogeneity problem causes an under-estimation of the taste bias but in a negligible magnitude. One of the reasons is that immigrant stock share is a state variable which is accumulative for a very long period while import share is a flow variable which is annual in my dataset. Thus there is very weak causality relationship on the opposite direction. Another reason is that a large portion of the international migration is caused by non-economic reasons, e.g., conflict, climate, education, religion, or economic reasons but unrelated to trade, e.g., job opportunity. In addition, I also do a long-difference estimation to equation (29) and obtain the result in column (4). The migration coefficient is 0.782, still significantly different from zero, and is robust with the gravity variables as controls in column (5). Therefore, I am confident that endogeneity issues is not big concern in my estimation Alternative Channel The second concern related to my estimation is the coefficient of immigrant shares may capture the stimulating effect more than just taste bias. The taste shifters may be overestimated because migration increases trade not only through the preference channel but through the trade cost channel as well. As Rauch (2001) argues, migration builds up business and social networks cross national borders can help to alleviate problems of contract enforcement and provide information about trading opportunities. Rauch and Trindade (2002) find that ethnic Chinese networks, proxied by the product of ethnic Chinese population shares, increased bilateral trade significantly, and more for differentiated than for homogeneous products. In order to check the network channel, I construct a new variable Network ni = N s nj s ij. (30) j=1 20

22 This variable gives the probability that, if we select an individual at random from each country, they will have a connection defined as common national origin. By construction, it is symmetric for country pairs. When j = CHN, it s an indirect link identical to that in Rauch and Trindade (2002), and when j = n, i, it s a summation of two direct links similar to that in Combes et al. (2005). But my measure extends them to a summation of all indirect and two direct links, and thus captures the network effect as much as possible. Then I re-specify bilateral trade cost by adding a new variable, Network, and rerun the regression of equation (28) to control the possible network channel. The results in Table 8 show that the coefficients on network is insignificant with migration in column (2) and without in (4). In column (3) and (5), Emigration share is added but its coefficient is not significant either. It is straightforward because emigrants contribute to bilateral network but not preference effect. column (6) checks the network effect in CES gravity and the coefficient on network is significantly positive, which is consistent with Rauch and Trindade (2002), which implies that my measure for network is valid. All those evidence tell that the network channel is negligible for import share of the aggregate manufacturing products. 4 Robustness In this section, I do robustness check with different specifications. Unbiased tastes are proxied by alternative measures in Section 4.1 and price elasticities are asymmetric in Section Alternative Measure of Unbiased Tastes In the baseline estimation, I proxy α i = EM i / i EM i where EM i is the extensive margin in global trade. Now I will replace it with other measures to ensure that my results do not solely depend on this particular extensive margin measure.. The first one is exporter s population in log, and the second is exporter s GDP in log. Table 9 reports the estimates in column (2) and column (3) respectively. The results are very close to the baseline estimation in column (1). I even also use 1/N to replace EM to impose symmetric unbiased tastes. It is reported in column (4). All results are robust. Figure 4 plots the unbiased tastes and taste biases. Because unbiased tastes α i measure to what level global consumers favor the good produced by country i, and θ i measures how ethnicity i s taste on their home product i deviates from that level. Then I can conclude from the figure that both world and home consumers favor American products, and 21

23 neither favor Luxembourg products. Also note that Germans and British are less biased in their taste but world consumers favor their products a lot. 4.2 Asymmetric Price Elasticity In the baseline estimation, it is assumed the consumers have symmetric price elasticity for each country s product. Now let s relax this symmetry restriction to more general case. 23 Specifically, {β i } i=1 N are asymmetric. Then the home-biased gravity is specified as where X ni E n = θ i s ni γβ i ln t ni + ψ ln r ni + λ n α i + f i + g n + γβ i ln P n + ɛ ni, (31) ln P n = N β i ln i=1 ( tni Π i ). (32) And ln r ni = ln r n ln r i, f i = Y i /Y θ i Π s i + γ ln Π i + ψ ln r i ln r, and g n = ψ ln r n ln r + ψ ln r ln r. As seen from equation (32), the inward multilateral resistances are unobservable because they have inside parameters {β i } i=1 N to be estimated. If I take unobservables as parameters to be estimated, the interactive terms will cause nonlinearity in regression. Bai (2009) extends the argument in Chamberlain (1984) to models with interactive effects, and shows that more consistent estimates are obtained with a projection of the interactive term onto an average of regressors when interest is centered on coefficients of non-interactive terms. I follow his idea to project the unobservables such that ln P n = η ln P n + ε n where ln P n = N µ i ln t ni, (33) i=1 and I let µ i = 1/N following Mundlak (1978). I also test the different sets of values of µ s. Therefore the econometric specification of the home-biased gravity equation becomes X ni E n = θ i s ni + b i ln d ni + b b border ni + b l lang ni + b t tari f f ni + b i internal ni + ψ ln r ni + λ n EM i + η i P n d + f i + g n + v ni. (34) Note that the coefficient on distance is exporter-specific. And f i, g n could be controlled by fixed effects of origins and destinations. v ni = ɛ ni + γβ i ε ni, and ln P n d are calculated by 23 Anderson and Zhang (2018) estimate the asymmetric price elasticities by projecting elasticities on exporter s export sophistication. 22

24 equation (33) but with ln d ni instead of ln t ni. 24 Then I can use the General Least Square method to estimate the equation. Table 10 reports the estimates in column (2). Exporter-specific distance elasticity is not reported since it is not the parameters of interest. The results are very close to the baseline estimation in column (1) and the average is I also try different weights, {Y i /Y} N i=1, to project the multilateral promoters in column (3). All results are robust. Then the theoretical parameters γ and {β i } are identified by γ = N i=1 b i/ρ, and β i = b i / N i =1 b i. Figure 5 plots the price elasticity ργβ i and income elasticity φ i. American, German, and Italian products are luxury goods because they have positive semi income elasticities, and thus their income elasticities are larger than one. These products are also price elastic which means consumers are very sensitive to their price changes. Indian, Indonesian, and Chinese products are necessary goods (with negative semi income elasticities), and price inelastic. Different to the estimates in Fajgelbaum and Khandelwal (2016), I allow {β i } N I=1 varying to exporters. This variation of price elasticities also reduces the variation of income elasticities estimated in their paper Counterfactuals In this section, I do counterfactual experiments with my baseline estimates. Section 5.1 studies to what extent the home bias in trade can be explained by taste effect, trade cost effect, and income effect, respectively. In Section 5.2, I conduct migration policy counterfactual experiment to test its impact on international trade. 5.1 Variation Decomposition First, I use these estimates to assess the role of each in contributing to the home bias in trade. Recall the formula implied by home-biased gravity equation HB n = (θ n s nn θ n Π s n α n Pn) s }{{} Gn s ( tnn γβ n ln Π n P n ) } {{ } Gn t ( rn + φ n ln r ). } {{ } Gn i Specifically, I term the taste effect, trade cost effect, and income effect as G s n, G t n, and G i n respectively. Then I calculate the contribution of each effect on trade bias of each country 24 Here I impose asymmetry only on distance elasticities. If all other standard gravity variables have heterogeneous coefficients, too many variables in regression would cause degrees of freedom problem. 25 Fajgelbaum and Khandelwal (2016) obtain the estimates of income elasticities with a range of to 0.052, whereas the range in this paper is to

25 by Contribution l n = G l n/hb n (35) where l = s, t, i. First column in Table 11 lists the trade biases for 40 countries, which are the deviations of home shares from their frictionless levels in manufacturing sector. The average bias is as high as 54%. The country with the largest bias, 86%, is Brazil and the one with the lowest, 10%, is Luxembourg. China and the U.S. both have modest level of home bias around 60% because they have large production shares in the world market which means high frictionless level of intra-national trade. Column (2)-(4) report the percentage contribution of the three effects to home bias in trade. On average, 43.2% of the market home bias could be explained by trade cost, 56.3% explained by home-biased taste, only 0.5% explained by income effect. In other word, trade cost and home-biased taste are the main causes of home bias in trade, while income effect is negligible. Taste effect larger than trade cost effect for most of the economies. China, Japan, and the U.S., have the largest taste effect above 70% because their ethnic consumers are most biased. Head and Mayer (2013) show that between 50% to 85% of the distance effects on trade flows are due to indirect trade costs (that they call as dark trade costs ). Another paper by Hou et al. (2017) find the contribution of distance effect through dyadic preferences is about 96%. Compared to their work, my results show that the contribution of taste effect (corresponding to dark trade costs in their paper), about 56.3%, is reasonable and consistent. 5.2 Policy Implications In this part, I discuss how migration policy affects the home bias in trade with the parameter values estimated in the baseline regression. If the bilateral immigrant share s nn changes, this has a direct effect on the corresponding home bias HB n. But the change in s nn also has an indirect effect on home bias through a change in multilateral promoters, which is similar to indirect trade cost effect on trade through multilateral resistances in Anderson and van Wincoop (2003). I refer to the direct and indirect effects together as a general equilibrium effect following Novy (2013). Specifically, HB n = θ n s nn }{{} direct effect θ n Π s n α n Pn s }{{} indirect effect (36) The taste bias θ n only captures the direct effect of a change in s nn on HB n. The indirect effect could be captured by multilateral promoters θ n Π s n and α n Pn. s Consider an extreme migration policy, like migration ban, for country n, to raise native 24

26 population share to 1. How does its home bias response to this shock? Since transnational migration promotes international trade, we expect an increase in the home market share. Using the equation above, I can calculate the percentage change of home bias HB n /HB n for each country. The results are reported in column (2) in Table 12. It is found that, the increase in home bias is a small number on average, only 3.7%, probably because international migration is at a very low level comparing to the total world population. But for immigrant countries, the increase is much larger. Australia will increase its home bias by 12.8 percent and Canada will increase by 10.5 percent. The U.S. will increase intra-national trade by 6.6 percent with an immigration ban. On the opposite, if we remove all the migration barrier at country n, and allow consumers freely migrant across its border, the country will end up with an ethnic composition exactly the same as the world population. Instead we expect a decrease in home market share as well as the home bias. column (3) reports the results. The home bias decreases by 56 percent on average, much larger than that in the case of the migrant ban. China s home biases will go down by the most because they have very small immigrant population and their consumers are highly biased in taste. 6 Extension In this section, I introduce the local bias in addition to home bias into the model to allow immigrants to assimilate. I redefine the taste parameter in equation (3). On one hand, a consumer has a biased taste towards the good produced by her home country (where she is originally from). On the other hand, her taste is also biased toward to the good produced locally (where she is living). 26 Then I specialize ethnic consumer (ethnic group) h s tastes at location (country) n as α h ni = (1 θ h Θ n )α i + θ h δ h i + Θ n δ ni, (37) where the Kronecker delta δ ni is 1 when i = n meaning local goods, and 0 otherwise, and Θ n [0, 1] is the local shifter which measures how consumers at location n s taste is towards to local good n. This specialization still satisfies conditions N i=1 αh ni = 1 for all h and all n. For any consumer h at location n, she raises her taste for home good h by a shifter θ h, for local good n by a shifter Θ n, while reduces those for all the other goods by 26 Bronnenberg et al. (2012) use micro-level data to infer that approximately 40 percent of the geographic variation in market shares is attributable to persistent brand preferences. They also show that their data also strongly reject the hypothesis that all that matters is where consumers lived in childhood: consumers who move after age 25 still eventually converge to the consumption patterns of their new state of residence. 25

27 1 θ h Θ n. Thus the set of local shifters {Θ h } h=1 N is the measurement of the local biases across locations. Now I rewrite the expenditure equations (3) as w h ni = (1 θ h Θ n )α i + θ h δ h i + Θ n δ ni γβ i ln ) ( + φ p i ln n ( pni e h n Q h n ). (38) Aggregating individual expenditure shares across ethnic groups gives the market import share of country n from country i by w ni = N h=1 s nhw h ni, i.e., w ni = (1 θ n Θ n )α i + θ i s ni + Θ n δ ni γβ i ln Then equation (39) could be rewritten as X ni E n = [α i γβ i ln ( pni ( ) pii ] + (θ p i s ni α i ( θ n + Θ n )) + Θ n δ ni γβ i ln p n ) ( ) en + φ i ln. (39) Q n ( tni P n ) ( ) en + φ i ln. Q n Substitute equation (40) into equation (12) and divide through by Y which is the total world income. Subtract( the) resulting expression from equation (40) and simplify by canceling term [α i pii γβ i ln p ]. Then the home-biased gravity with local bias becomes X ni Y ( ) i E n Y = (θ is ni θ i Πi s α ipn) s tni ( rn ) ( ) E γβ i ln + φ Π i P i ln + Θ n δ n r ni Θ i i, (41) Y (40) where ln Π i, ln P n, and Π s i are the same as the baseline model, while P s n = On the left hand side, X ni E n N θ i (s ni Πi s) + (Θ n i=1 Y i Y N n =1 ( En Y ) Θ n ). (42) is the deviation of bilateral trade per unit of n s expenditure from its frictionless level Y i Y. There are a fourth term on the right hand side now, which capture the local effect, in additional to taste effect, price effect, and income effect in the baseline mode. It captures the effect of relative trade promoter that stimulates internal (local) trade due to the local bias in taste. When consumers everywhere has non local bias, i.e. Θ n = 0 for all n, the internal trade promoter is shut down, and this gravity equation corresponds to that in the baseline model. To estimate the gravity equation (41), I follow the methodology in the baseline estimation. Note that Θ i E i /Y is captured by exporter-specific fixed effect. δ ni is a dummy for internal trade. In addition, I assume local shifters to satisfy Θ n = µs nn for all n. This 26

28 is reasonable because immigrants are more likely to assimilate in a country with bigger native population share. Then the number of local bias parameters to be estimated is reduced from forty to one. Thus the specification of the gravity equation (41) is X ni E n = θ i s ni + b ln d ni + b b border ni + b l lang ni + b t tari f f ni + b i s nn internal ni + ψ ln r ni + λ n α i + f i + g n + ɛ ni. (43) In contrast to (25), the dummy internal now interacts with native population share. The idea behind this variable is to capture the variation of border barriers across locations. 27 The results are reported in column (2) in Table 13. The coefficient estimates of distance, common border, common language, tariff, and income are exactly the same as the baseline results in column (1). The coefficient of the interaction term of internal trade and native population share is positively significant, which suggests consumers located at countries with higher native share are more likely to locally bias in their tastes. The table also reports the estimates of the 40 θ i parameters, one corresponding to each ethnic group, in the subsequent rows. After controlling local bias in gravity, the home taste shifters estimated are mostly smaller. On average, the taste bias measured by the ethnicity-specific immigrant coefficients is 0.302, which is lower than in the symmetric estimation. Now the four ethnic groups with highest θ s are Japanese, Australian, Chinese, and Americans. The four ethnic groups with lowest θ s are Maltese, Belgians, Luxembourgers and Dutchmen, with taste shifters not different from 0. 7 Conclusion Trade models usually have identical preferences across consumers. Differences in demand across countries, such as home-biased tastes, are often understood by unobservable trade costs which are indistinguishable. It is important to quantitatively assess the two because declines in trade costs and homogenization of tastes have different implications for trade policies and welfare changes. This paper provides a structural component of the home-biased preferences which is feasible to estimate. The bilateral market taste is determined by bilateral migration and their taste bias. I estimate taste bias for each ethnic group and find that taste bias explains the trade bias more than trade cost does. The difference in taste bias explains the large variation of home bias across countries. 27 Anderson et al. (2015) first time use the interaction terms of internal trade and a series of countryspecific characteristics to estimate their impacts on international borders in the presence of a complete set of exporter-time and importer-time fixed effects within a structural gravity model. 27

29 The home-biased gravity model here suggests several extensions. (i) Endogenous migration can be added into the model to provide more policy implications. (ii) Another potential extension is to allow for within-country income differences to investigate more variations of taste biases. In empirical there are some other possible extensions. (i) More information on immigrants, like education, age, or religion, can be included to test whether the taste biases differ within ethnicity. (ii) By looking at different migration generations, check whether the taste bias decays over time. More challenging, perhaps leading outside the scope of the present model, is finding the deep causes of the taste bias formation. This is related to other research on how culture and institution persist or change across ethnic groups. I believe that these are all fruitful directions for future research. 28

30 References Allen, Treb, Information frictions in trade, Econometrica, 2014, 82 (6), Anderson, James, The Gravity Model, Annual Review of Economics, 2011, 3, and Douglas Marcouiller, Insecurity And The Pattern Of Trade: An Empirical Investigation, Review of Economics and Statistics, 2002, 84, and Eric van Wincoop, Gravity with Gravitas: A Solution to the Border Puzzle, American Economic Review, 2003, 93, and, Trade Costs, Journal of Economic Literature, 2004, XLII, and Penglong Zhang, AI Gravity, Latent Trade, and Zero Flows, working paper, and Yoto Yotov, Specialization: Pro- and Anti-Globalizing, , NBER Working Paper, and, Short Run Gravity, NBER Working Paper, 2017., Ingo Borchert, Aaditya Mattoo, and Yoto V Yotov, Dark costs, missing data: shedding some light on services trade, European Economic Review, 2015, 105, Atkin, David, Trade, tastes and nutrition in india, American Economic Review, 2013, 103, Bai, Jushan, Panel Data Models With Interactive Fixed Effects, Econometrica, 2009, 77, Bronnenberg, Bart, Jean-Pierre Dubé, and Matthew Gentzkow, The evolution of brand preferences: Evidence from consumer migration, American Economic Review, 2012, 102 (6), Burchardi, Konrad, Thomas Chaney, and Tarek Hassan, Migrants, Ancestors and Foreign Investments, Review of Economic Studies RR, Card, David, Immigrant inflows, native outflows, and the local labor market impacts of higher immigration, Journal of Labor Economics, 2001, 19 (1), Caron, Justin, Thibault Fally, and James Markusen, International trade puzzles: A solution linking production and preferences, Quarterly Journal of Economics, 2014, 129,

31 Chamberlain, Gary, Panel Data, Handbook of Econometrics, 1984, 1984, Chaney, Thomas, The Intensive and Extensive Margins of International Trade, American Economic Review, 2008, 98, , The network structure of international trade, American Economic Review, 2014, 104 (11), , Networks in international trade, in The Oxford Handbook of the Economics of Networks Combes, Pierre-Philippe, Miren Lafourcade, and Thierry Mayer, The trade-creating effects of business and social networks: evidence from France, Journal of International Economics, 2005, 66, Comite, Francesco Di, Jacques-François Thisse, and Hylke Vandenbussche, Vertizontal differentiation in export markets, Journal of International Economics, 2014, 93 (1), Coşar, A Kerem, Paul LE Grieco, Shengyu Li, and Felix Tintelnot, What drives home market advantage?, Journal of International Economics, 2018, 110, Deaton, Angus and John Muellbauer, An almost ideal demand system, American Economic Review, 1980, 70, Fajgelbaum, Pablo and Amit Khandelwal, Measuring the Unequal Gains from Trade, Quarterly Journal of Economics, 2016, 131, Feenstra, Robert and John Romalis, International prices and endogenous quality, Quarterly Journal of Economics, 2014, 129, Feenstra, Robert C, A Homothetic Utility Function for Monopolistic Competition Models, Economics Letters, 2003, 78, , New products with a symmetric aids expenditure function, Economics Letters, 2010, 106, Felbermayr, Gabriel, Volker Grossmann, and Wilhelm Kohler, Migration, International Trade, and Capital Formation: Cause or Effect?, Handbook of the Economics of International Migration, 2015, 1B, Fieler, Ana Cecilia, Nonhomotheticity and Bilateral Trade: Evidence and a Quantitative Explanation, Econometrica, 2011, 79,

32 Foged, Mette and Giovanni Peri, Immigrants effect on native workers: New analysis on longitudinal data, American Economic Journal: Applied Economics, 2016, 8 (2), Fulford, Scott, Ivan Petkov, and Fabio Schiantarelli, Does It Matter Where You Came From? Ancestry Composition and Economic Performance of US Counties, , Boston College working paper, Hallak, Juan Carlos, Product Quality and the Direction of Trade, Journal of International Economics, 2006, 68, Hanson, Gordon, International Migration and the Developing World, Handbook of Development Economics, 2010, 5, Head, Keith and Thierry Mayer, What separates us? Sources of resistance to globalization, Canadian Journal of Economics, 2013, 46, Helpman, Elhanan, Marc Melitz, and Yona Rubinstein, Estimating trade flows: trading partners and trading volumes, Quarterly Journal of Economics, 2008, 123, Hou, Yulin, Yun Wang, and Hakan Yilmazkuday, Gravity Channels in Trade, Federal Reserve Bank of Dallas working paper, Hummels, David and Peter Klenow, The Variety and Quality of a Nation s Exports, American Economic Review, 2005, 95, Iranzo, Susana and Giovanni Peri, Migration and trade: Theory with an application to the Eastern Western European integration, Journal of International Economics, 2009, 79 (1), Lan, Xiaohuan and Ben Li, The economics of nationalism, American Economic Journal: Economic Policy, 2015, 7 (2), Lewbel, Arthur, The Rank of Demand Systems: Theory and Nonparametric Estimation, Econometrica, 1991, 59, and Krishna Pendakur, Unobserved Preference Heterogeneity in Demand Using Generalized Random Coefficients, Journal of Political Economy, 2017, 125, Luttmer, Erzo and Monica Singhal, Culture, Context, and the Taste for Redistribution, American Economic Journal: Economic Policy, 2011, 3, McCallum, John, National borders matter: Canada-U.S. regional trade patterns, American Economic Review, 1995, 85,

33 Morey, Mitchell, Preferences and the home bias in trade, Journal of Development Economics, 2016, 121, Mundlak, Yair, On the Pooling of Time Series and Cross Section Data, Econometrica, 1978, 46, Neary, Peter, Rationalizing the penn world table: true multilateral indices for international comparisons of real income, American Economic Review, 2004, 94, Novy, Dennis, International trade without CES: Estimating translog gravity, Journal of International Economics, 2013, 89, Obstfeld, Maurice and Kenneth Rogoff, The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?, NBER Macroeconomics Annual, 2001, 15, Ortega, Francesc and Giovanni Peri, Openness and income: The roles of trade and migration, Journal of International Economics, 2014, 92 (2), Ottaviano, Gianmarco, Giovanni Peri, and Greg Wright, Immigration, Offshoring and American Jobs, American Economic Review, 2013, 103, Ramondo, Natalia, Andrés Rodríguez-Clare, and Milagro Saborío-Rodríguez, Trade, Domestic Frictions, and Scale Effects, American Economic Review, 2016, 106, Rauch, James, Business and Social Networks in International Trade, Journal of Economic Literature, 2001, XXXIX, and Vitor Trindade, Ethnic Chinese Networks in International Trade, Review of Economics and Statistics, 2002, 84, and, Information, International Substitutability, and Globalization, American Economic Review, 2003, 93, Yi, Kei-Mu, Can Multistage Production Explain the Home Bias in Trade?, American Economic Review, 2010, 100,

34 Tables and Figures Production Share Home Share Import Share Native Population Share 100% 1 90% % % % % % % % % % China Japan Brazil India Korea Russia United States Italy France Germany United Kingdom Estonia Belgium 0.8 Figure 1: Home Bias in Trade and Native Population Share Notes: Import (home) share is the share of a country s imports (domestic goods) in its total expenditures. Native population share is the share of a country s natives in its total population, which equals to one minus the foreign-born population share. Production share is the share of a country s production in world total production. Data source: WIOD and UNdata. 33

35 Table 1: Home-Biased Gravity Estimates: Baseline Import share Distance (0.001) Border (0.002) Language (0.002) Tariff (0.001) Income (0.001) Internal (0.071) JPN X Immigrant (0.075) CHN X Immigrant (0.073) BRA X Immigrant (0.074) USA X Immigrant (0.080) IDN X Immigrant (0.074) AUS X Immigrant (0.087) TUR X Immigrant (0.075) IND X Immigrant (0.074) KOR X Immigrant (0.074) ITA X Immigrant (0.076) RUS X Immigrant (0.073) ESP X Immigrant (0.078) CAN X Immigrant (0.085) FIN X Immigrant (0.076) IRL X Immigrant (0.082) FRA X Immigrant (0.077) ROM X Immigrant (0.074) GRC X Immigrant (0.075) BGR X Immigrant (0.074) PRT X Immigrant (0.075) SWE X Immigrant (0.079) POL X Immigrant (0.074) MEX X Immigrant (0.074) TWN X Immigrant (0.075) LTU X Immigrant (0.076) DEU X Immigrant (0.080) CZE X Immigrant (0.075) GBR X Immigrant (0.078) AUT X Immigrant (0.081) CYP X Immigrant (0.083) LVA X Immigrant (0.082) HUN X Immigrant (0.076) SVK X Immigrant (0.075) SVN X Immigrant (0.074) EST X Immigrant (0.085) DNK X Immigrant (0.077) NLD X Immigrant (0.078) BEL X Immigrant (0.080) MLT X Immigrant (0.079) LUX X Immigrant (0.104) Observations 1600 Adjusted R Mean Std Notes. Table reports the estimates of the homebiased gravity equation. Robust standard errors in parentheses. Significance 34 *.10, **.05, ***.01.

36 Table 2: Home-Biased Gravity Estimates: by Sector (1) (2) (3) Manufacturing (Baseline) Agricultural Service Distance (0.001) (0.001) (0.000) Border (0.002) (0.003) (0.001) Language (0.002) (0.003) (0.001) Tariff (0.001) (0.000) (0.007) Income (0.001) (0.000) (0.000) Internal (0.071) (0.104) (0.080) JPN X Immigrant (0.075) (0.109) (0.037) CHN X Immigrant (0.073) (0.108) (0.033) BRA X Immigrant (0.074) (0.108) (0.032) USA X Immigrant (0.080) (0.117) (0.060) IDN X Immigrant (0.074) (0.108) (0.030) AUS X Immigrant (0.087) (0.127) (0.074) TUR X Immigrant (0.075) (0.109) (0.100) IND X Immigrant (0.074) (0.108) (0.218) KOR X Immigrant (0.074) (0.109) (0.146) ITA X Immigrant (0.076) (0.111) (0.031) RUS X Immigrant (0.073) (0.106) (0.029) ESP X Immigrant (0.078) (0.114) (0.032) CAN X Immigrant (0.085) (0.124) (0.033) FIN X Immigrant (0.076) (0.111) (0.031) IRL X Immigrant (0.082) (0.120) (0.034) FRA X Immigrant (0.077) (0.112) (0.031) ROM X Immigrant (0.074) (0.108) (0.030) GRC X Immigrant (0.075) (0.109) (0.031) BGR X Immigrant (0.074) (0.108) (0.030) PRT X Immigrant (0.075) (0.110) (0.031) SWE X Immigrant (0.079) (0.115) (0.032) POL X Immigrant (0.074) (0.108) (0.030) MEX X Immigrant (0.074) (0.108) (0.029) TWN X Immigrant (0.075) (0.110) (0.029) LTU X Immigrant (0.076) (0.111) (0.031) DEU X Immigrant (0.080) (0.117) (0.033) CZE X Immigrant (0.075) (0.109) (0.031) GBR X Immigrant (0.078) (0.114) (0.032) AUT X Immigrant (0.081) (0.118) (0.033) CYP X Immigrant (0.083) (0.121) (0.034) LVA X Immigrant (0.082) (0.120) (0.034) HUN X Immigrant (0.076) (0.111) (0.031) SVK X Immigrant (0.075) (0.110) (0.031) SVN X Immigrant (0.074) (0.109) (0.031) EST X Immigrant (0.085) (0.124) (0.035) DNK X Immigrant (0.077) (0.112) (0.032) NLD X Immigrant (0.078) (0.114) (0.032) BEL X Immigrant (0.080) (0.117) (0.033) MLT X Immigrant (0.079) (0.114) (0.032) LUX X Immigrant (0.104) (0.151) (0.043) Observations Adjusted R Mean Std Notes: Table reports the estimates of home-biased gravity equation with sectoral trade flow data. Robust standard errors in parentheses. Significance 35 *.10, **.05, ***.01.

37 Table 3: Home-Biased Gravity Estimates: by Good (1) (2) (3) Aggregate (Baseline) Final Intermediate Distance (0.001) (0.001) (0.001) Border (0.002) (0.002) (0.002) Language (0.002) (0.002) (0.002) Tariff (0.001) (0.001) (0.001) Income (0.001) (0.001) (0.001) Internal (0.071) (0.075) (0.072) JPN X Immigrant (0.075) (0.079) (0.076) CHN X Immigrant (0.073) (0.078) (0.075) BRA X Immigrant (0.074) (0.078) (0.076) USA X Immigrant (0.080) (0.084) (0.082) IDN X Immigrant (0.074) (0.078) (0.075) AUS X Immigrant (0.087) (0.092) (0.089) TUR X Immigrant (0.075) (0.079) (0.076) IND X Immigrant (0.074) (0.079) (0.076) KOR X Immigrant (0.074) (0.079) (0.076) ITA X Immigrant (0.076) (0.080) (0.078) RUS X Immigrant (0.073) (0.077) (0.074) ESP X Immigrant (0.078) (0.083) (0.080) CAN X Immigrant (0.085) (0.090) (0.087) FIN X Immigrant (0.076) (0.080) (0.078) IRL X Immigrant (0.082) (0.087) (0.084) FRA X Immigrant (0.077) (0.081) (0.078) ROM X Immigrant (0.074) (0.078) (0.075) GRC X Immigrant (0.075) (0.079) (0.076) BGR X Immigrant (0.074) (0.078) (0.076) PRT X Immigrant (0.075) (0.080) (0.077) SWE X Immigrant (0.079) (0.083) (0.080) POL X Immigrant (0.074) (0.078) (0.076) MEX X Immigrant (0.074) (0.078) (0.075) TWN X Immigrant (0.075) (0.079) (0.077) LTU X Immigrant (0.076) (0.081) (0.078) DEU X Immigrant (0.080) (0.084) (0.082) CZE X Immigrant (0.075) (0.079) (0.076) GBR X Immigrant (0.078) (0.082) (0.080) AUT X Immigrant (0.081) (0.085) (0.082) CYP X Immigrant (0.083) (0.088) (0.085) LVA X Immigrant (0.082) (0.087) (0.084) HUN X Immigrant (0.076) (0.080) (0.078) SVK X Immigrant (0.075) (0.080) (0.077) SVN X Immigrant (0.074) (0.079) (0.076) EST X Immigrant (0.085) (0.089) (0.086) DNK X Immigrant (0.077) (0.081) (0.079) NLD X Immigrant (0.078) (0.083) (0.080) BEL X Immigrant (0.080) (0.084) (0.082) MLT X Immigrant (0.079) (0.083) (0.080) LUX X Immigrant (0.104) (0.110) (0.107) Observations Adjusted R Mean Std Notes: Table reports the estimates of home-biased gravity equation with intermediate and final trade flow data. Robust standard errors36 in parentheses. Significance *.10, **.05, ***.01.

38 Taste Bias (θi) Ethnicity ID Manufacturing (Baseline) Service Agricultural Figure 2: Taste Shifters θ i by Sector Figure plots ethnicity i taste bias toward home products in different sectors. Taste Bias (θi) Ethnicity ID Aggregate (Baseline) Intermediate Final Figure 3: Taste Shifters θ i by Good Figure plots ethnicity i taste bias toward home products on different types of goods. 37

39 Table 4: Home-Biased Gravity Estimates: by Industry Industry Mean Standard Deviation Aggregate Textiles Leather and footwear Transport equipment Food, beverages, and tobacco Machinery Chemical products Manufacturing, nec Basic metals and fabricated metal Wood products Printing and publishing Electrical and optical equipment Rubber and plastics Other non-metallic minerals Notes: Table reports the estimates of home-biased gravity equation with industrial trade flow data. Only mean and stardard deviation of taste bias estimates are reported. Robust standard errors in parentheses. Significance *.10, **.05, ***

40 Table 5: The Most Biased Ethnic Groups by Industry Notes: Table reports the most biased ethnic groups by industry. 39