Third-Market Effects of Exchange Rates: A Study of the Renminbi

Transcription

1 PRELIMINARY DRAFT. NOT FOR QUOTATION Third-Market Effects of Exchange Rates: A Study of the Renminbi Aaditya Mattoo (Develoment Research Grou, World Bank), Prachi Mishra (Research Deartment, International Monetary Fund), and Arvind Subramanian (Peterson Institute for International Economics and Center for Global Develoment) December 17, 2011 Abstract This aer estimates the effects of China s exchange rate changes on exorts of cometitor countries in third markets. We derive an econometric secification which includes roduct and destination-secific indices of cometition between China and its develoing country cometitors. Our emirical strategy allows us to exloit the variation--afforded by disaggregated trade data--across exorters, imorters, roduct, and time to estimate the third-market effect. We find robust evidence for a statistically and quantitatively significant effect of China s exchange rate on cometitor country exorts. In articular, a country s exorts of roducts that comete more closely with China tend to increase significantly more when the Chinese exchange rate areciates. Our estimates suggest that a 10 ercent areciation of China s real exchange rate can boost on average a develoing country s exorts of a tyical 4-digit HS roduct category to third markets by 1-2 ercent. The magnitude of the third-market effect is consistent with the underlying theoretical model. Keywords: exchange rates, trade, China JEL codes: F13, F14, O53 Authors contact: amattoo@worldbank.org, mishra@imf.org, asubramanian@iie.com. The views exressed in this aer are those of the authors and do not necessarily reresent those of the International Monetary Fund (IMF) or World Bank (WB) or their olicies. We are grateful to George Akerlof, Andy Berg, Maurice Obstfeld, Hui Tong as well as seminar articiants at the IMF and World Bank for helful comments and discussions. We are articularly grateful to Rob Feenstra for very detailed suggestions and advice. Martin Kessler and Ujjal Basu Roy rovided excellent research assistance. All remaining errors are our own. 1

2 I. Introduction Exchange rates are a erennial source of discussion and analysis, and one of the oldest subjects in international macroeconomics. But the vanes to taste do veer, and do so cyclically in terms of which asects of exchange rates elicit attention and scrutiny. In the 1930s, under conditions of idle resources and weak demand, the cometitive consequences of exchange rates were rominent, esecially as many advanced countries successively went off the gold standard. In the 1970s and 1980s, as high inflation became one of the central issues in macroeconomic olicy making, exchange rates were increasingly used as anchors for inflation. Periodic financial crises, however, highlighted the role of fixed exchange rates in contributing to such crises articularly in the context of free caital mobility; with many countries starting to move towards flexible exchange rates. In the 2000s, the cycle seems comlete with the cometitive consequences of exchange rates back in the limelight. For examle, since the Asian financial crisis of the late 1990s, a number of emerging market countries have used exchange rates to self-insure against financial crises and to romote economic growth. Mercantilism, based on undervalued exchange rates, has come back into vogue, described by some as the hallmark of the new era of Bretton Woods II (Dooley et. al.; 2003). Simultaneously, contributions emerged about the ossible long run effects of undervalued exchange rates, notably Rodrik (2010) and Johnson et al. (2010). 1 But these and other 1 Note that the recent contributions on the growth effects of mercantilism have a historical counterart in the exerience of the 1930s described in Robinson (1947). But the difference is that the 1930s exerience related to macroeconomic mercantilism the short-run use of exchange rate during eriods of slack resources whereas the more recent contributions stress the medium-run benefits of undervalued exchange rates. 2

3 contributions have largely focused on the imacts of exchange rate olicies on the countries themselves. 2 There aears to be less evidence and quantification of the effect of exchange rates on the exorts of other countries, the sillover effect that is sometimes called the beggar-thy-neighbor effect. In this aer, we take a first ste toward estimating this sillover effect. We focus on the imact of movements in China s exchange rate on other develoing countries. The case of China resents an excellent oortunity to study this question. China, by virtue of being the world s largest exorter of goods, is guaranteed to have quantitatively more significant cometitive consequences for other countries than nearly any other exorter. This is reinforced by the fact that China is also a highly diversified exorter so that it otentially cometes with a broad range of countries and across the roduct sectrum. 3 In art, reflecting China s dominant size and encomassing scoe, its exchange rate olicy has been one of the most controversial asects of international macroeconomics during the 2000s. Ever since the global savings glut hyothesis (Bernanke 2005) gained credence as an exlanation of global imbalances and as a contributor to the global financial crisis that began in 2008, China s exchange rate olicy is being seen as one of the drivers of the global savings glut. 4 2 This is generally true of the older, voluminous literature on the trade consequences of exchange rates (Goldstein and Khan (1985) rovides a survey and other contributions include Deardorff (1984); Hooer, Johnson and Marquez (2000); Thursby and Thursby (1987)). It is also true of the more recent micro-literature (Dekle and Royoo (2002); Das, Roberts and Tybout (2001); Forbes (2002); Berman, Martin and Mayer (2011)). 3 So diversified is China s exort basket that one might even say it is the only country that has comarative advantage in all roducts. 4 The Global Saving Glut and the U.S. Current Account Deficit, Federalreserve.gov. March 10, 2005, Sandridge Lecture, Virginia Association of Economists, Richmond, Virginia,htt:// 3

4 Also, over the last two years, with considerable slack in industrial countries, reflected in high unemloyment rates and low caacity utilization, the Chinese exchange rate has been criticized for aggravating the demand shortfall in these countries. One estimate, for examle, by the Economic Policy Institute, which has been cited by Paul Krugman among others, suggests that if the renminbi were revalued to its equilibrium level, U.S. gross domestic roduct would increase by nearly 2 er cent, creating u to 2.25 million U.S. jobs (Scott, 2011). Viewing China's exchange rate olicy solely through the rism of global imbalances or industrial country difficulties has obscured an imortant issue, namely whether it has an effect on other develoing countries which comete with China. The trade consequences of China s exchange rate olicy is likely to be greater for develoing countries than for industrial ones because they comete more closely with China than the United States and Euroe, whose areas of comarative advantage are very different from China's. 5 But while there is a burgeoning literature on China s trade and exchange rates, there is very little evidence on the effect of China s exchange rate on the exorts of other develoing countries. The existing literature has mostly used a gravity framework augmented with China s exorts (e.g. Eichengreen, Ree and Tong (2004), and Ahearne et. al. (2003)) and finds some evidence that Chinese exorts crowd out other Asian exorts. Recently, Eichengreen and Tong (2011) have estimated the effect of renminbi revaluation on stock market valuations of foreign firms. They 5 As we show below in Section IV, on average, if a develoing country exorts a articular good to a articular destination, there is close to 90 ercent robability that China will do the same. 4

5 find that renminbi areciation has ositive effects on firms cometing with China, which is consistent with the findings in this aer. 6 Our aer contributes to this literature by quantifying the sillover effects of changes in the renminbi on other develoing countries. To be more secific: if China s exchange rate dereciates, it affects another country, say Brazil in its own market, in China s market and in the market of third countries to which both Brazil and China exort. First, a dereciation of the renminbi will increase Chinese exorts to Brazil because it is akin to a subsidy to Chinese exorters. Brazil s consumers might benefit from cheaer Chinese goods but Brazilian roducers of goods that comete with China will lose (effect on Brazilian roducers in their own market). Second, a renminbi dereciation will reduce Brazilian exorts to China because the renminbi dereciation will serve as an imort tariff in the Chinese market shielding Chinese roducers from Brazilian goods (effect on Brazilian roducers in China s markets). 7 Finally, a renminbi dereciation will increase Chinese exorts to third markets, dislacing Brazilian exorts, because a dereciation offers a subsidy to Chinese exorters which advantages them vis-à-vis Brazilian exorters (effect on Brazilian roducers in third markets). We call this latter the third-market effect and it is this effect of exchange rate movements on which this aer focuses. We estimate this third-market effect using disaggregated trade data at the 6-digit level sanning 124 develoing country exorters and 57 large imorters over the eriod , that allows us to exloit variation across imorters, exorters, roducts, and time. Our emirical aroach is 6 Eichengreen and Tong (2011) also find that the renminbi areciation has ositive effect on firms exorting to China, negative effect on firms roviding inuts for China s rocessing exorts and negative effects for firms imorting Chinese roducts. 7 Brazilian exorts of intermediate goods might ossibly benefit if greater Chinese final good roduction increases demand for imorted intermediates. 5

6 motivated by an analytical framework based on Feenstra, Obstfeld, and Russ (2011). We develo an identification strategy that relies on the following reasoning: the more a country cometes with China in a third market, the more a given dereciation of the renminbi is likely to hurt its exorts in that market. We develo indices of cometition with China at the exorterimorter-roduct level to imlement this strategy. The emirical secification with a battery of fixed effects hels us overcome to a large extent the roblems of endogeneity and omitted variables that lague estimation of trade-exchange rate equations using aggregated data. 8 We find robust evidence for the existence of a statistically and economically significant thirdmarket effect. In articular, exorts to third markets of countries with a greater degree of cometition with China tend to rise/fall significantly more as the renminbi areciates/dereciates. Overall our estimates suggest that a 10 ercent areciation of the renminbi increases a develoing country s exorts at the roduct-level on average by 1-2 ercent. For high indices of cometition, we find that the increase could be as large as 5 ercent. The magnitude of the estimates is consistent with the redictions from the analytical framework. The results imly that going forward, if the renminbi were to areciate, this could rovide a substantial boost to develoing country exorts. The rest of the aer is organized as follows. In Section II, we set out the analytical framework. Section III elaborates the estimation strategy and Section IV describes the data. The results are resented in Section V and Section VI concludes. 8 See Engel (2009), who argues how hard it is econometrically to searate out the effect of exchange rates on trade. 6

7 II. Analytical Framework As discussed above, the third-market effect that we seek to estimate is concetually straightforward. Take three countries: China as the exorter, the US as imorter, and Brazil as a cometitor to China in the US market. If the Chinese exchange rate dereciates relative to the dollar, its exorts will ceteris aribus become more cometitive in the US market (because Chinese exorters will now charge a lower dollar rice in the US). As a result, US consumers substitute away from Brazilian roducts towards Chinese ones, resulting in reduced Brazilian exorts to the US. We are interested in the size of this effect. In order to develo an analytical framework for our emirical exercise, we use the model in Feenstra, Obstfeld and Russ (2011). The setting is as follows. There are countries, different goods. Each country roduces a range of distinct varieties of each good. There is a constant elasticity of substitution () consumtion index for the reresentative consumer in country. Goods are differentiated not only by their characteristics, also by their country of origin (Armington assumtion), with a constant elasticity of substitution between domestically roduced and foreign varieties of good ( ), and a constant elasticity of substitution between different varieties of good originating in different exorters ( ). The same elasticity alies to different varieties of good roduced domestically. Feenstra, Obstfeld and Russ (2011) show that we can exress country s imorts from country of a articular good, defined at the HS 6-digit level,, as follows (equation 11 in their aer). 7

8 (1) 1 That is, the roortion imort demand ( ) of total consumtion in,, deends on three sets of comonents: the reference weight consumers in attach to imorts of good g from country, ; the rice of g imorts by from,, relative to the rice index of all g imorts, ; and the elasticity of substitution between imorted varieties of, ; the reference weight consumers in j attach to domestically roduced units of good, ; the rice index of all imorts by,, relative to the domestic rice of good, ; and the elasticity of substitution between the home and foreign varieties of good, ; the reference weight consumers in attach to consumtion of the good, ; the rice index of the good,, relative to the rice index of all goods in, ; and the elasticity of substitution between different goods,. We first establish the effect of a change in China s exchange rate changes vis-a-vis country,, on country s imorts of a articular good from country,. We can write this effect 8

9 as a chain effect, consisting of the effects of: the change in the Chinese exchange rate on the rice of the Chinese good, the change in the rice of the Chinese good on the foreign rice index, and the change in the foreign rice index on demand for good from country : (2) Now consider each term in the chain starting from the third term. Taking logs of Equation (1) and differentiating with resect to under the assumtion that a change in the rice index of imorted good has a negligible effect on the aggregate US rice index for good, we get: 9 (3) This imlies that the elasticity of demand for imorts of good from country with resect to the foreign rice index is simly the difference between the elasticity of substitution between imorted varieties of,, and the elasticity of substitution between home and foreign varieties, This is an innocuous assumtion from the emirical ersective because any additional terms for examle aggregate destination-secific rices will be absorbed in the very general fixed effects. 10 Note that in Broda and Weinstein (2006),, i.e. the elasticities of substitution between imorted varieties equals the elasticity of substitution between home and foreign varieties. In our framework, if, if the renminbi dereciates, consumers in country reduce their demand for varieties of good roduced at home and hence there is no third-market effect. 9

10 From Feenstra, Obstfeld and Russ (2011), we have the rice index for imorted goods,, (their equation 5): (4) Taking logs, differentiating with resect to the rice of the Chinese good in the market,, and simlifying, we get: (5) This imlies, as exected, that the elasticity of the foreign rice index for good with resect to the rice of the Chinese good is equal to the exenditure on the Chinese good as a share of exenditure on all imorts of,. We assume that the rice of the Chinese good in the US market,, deends on the rice in China,, the exchange rate, (defined in renminbi/imorter currency), and an exonent which catures the extent of roduct-secific exchange rate ass-through from China to,. (6) 1/ 10

11 Differentiating with resect to the exchange rate,, we have: (7) Substituting from Equations (3), (5) and (7) in Equation (2), we get: (8) Equation (8) imlies that a change in the Chinese exchange rate will have a non-zero effect on imort demand for good only if (i) elasticities of substitution across imorted varieties is different from that between imorted and domestic varieties, (ii) Chinese share in total imorts of that good is strictly ositive, and (iii) the exchange rate ass-through is non-zero. Given our assumtion regarding the symmetric elasticity of substitution between imorted varieties,, the effect of a change in China s exchange rate changes vis-a-vis country,, on country s imorts of a good from country,, does not deend on any exorter attribute. This makes Equation (8) less amenable to emirical analysis. For examle, if in order to test the rediction in Equation (8), we were to regress the imort demand at the exorterimorter-roduct level on the share of China in imorts at the imorter-roduct level, the effect of latter would not be estimated recisely as it would be absorbed by imorter-roduct fixed effects. 11

12 In order to introduce meaningful variation in the imact of China s exchange rate across exorter countries, so as to make it amenable to econometric analysis, we consider country s imorts,, from country of a articular bundle of goods, defined at a higher level of aggregation. In our emirical analysis, we use trade data at the Harmonized System (HS) 6-digit. Therefore is defined at the HS 6-digit level. Country s imorts of (at say the HS 4-digit level) can be exressed as: (9) G are the number of HS 6-digit lines in the roduct category. Taking logs and differentiating with resect to the exchange rate,, we get (10) This equation is intuitive: the elasticity of say Brazil's exorts to at the HS 4-digit category with resect to China s exchange rate vis-à-vis is related to the weighted average of China's share in total imorts in each constituent 6 digit category which Brazil exorts, where the weights are Brazil s exorts in the corresonding 6 digit category as a share of its total exorts in the 4 digit category. Thus, at this higher level of aggregation, the effect of China s exchange rate on exorts of a articular country to another country deends on the interlay between the relative imortance of secific 6-digit roduct lines,, in s exorts and the relative imortance of China as a source of imorts by of those 6-digit roduct lines. 12

13 Further, we also assume that the elasticities of substitution and the ass-through are constant for Cj Cj all 6-digit lines within the relevant four digit category. i.e., g g. Then Equation (10) can be rewritten as g (11) V Iij *[ Cj * ( )] G ij V Vg Cj where Iij [( ) * sg ] ij V g 1 is what we call the value-based index of cometition with China for good g exorted from i to j. For examle, if the HS 4-digit category, shirts, consisted of only two items, cotton shirts and non-cotton shirts, then our measure is simly the share of China in country j s imorts of each tye of shirt, weighted by the imortance of each tye of shirt in country i s shirt exorts to j. is what we define as the third-market effect. Equation (11) rovides the basis for the first secification we estimate. Under some additional symmetry assumtions, we can also estimate an alternative secification where we rely on the overla between China s exorts and those of country, at the extensive margin rather than measures of cometition at the intensive margin, as in Equation (11) above. We first assume that for each 6 digit category that exorts to within a 4-digit category, it exorts the same amount. If exorts ij N 6-digit categories in the relevant 4 digit category to, ij then the first term in Equation (10) simlifies to 1 /. Next assume that in each 6 digit N 13

14 category within the relevant 4-digit category where exorts to, China exorts either a fixed Cj Cj Cj share, s or nothing. sg s for N, lines or zero otherwise. Then summing the second ratio ij Ch over the relevant 6 digit lines gives us *. As above, we also assume that the elasticities Cj ij s N, Ch of substitution and the ass-through are constant for all 6-digit lines within the relevant four digit category. i.e., g g. Cj g Cj So that in this secial case, Equation (11) can be written as: (12) C Cj Cj I s * * ( )] ij [ where I C ij N is what we call the count-based index of cometition. This equation forms N ij, Ch ij the basis for our alternative emirical secification. In this secification the elasticity of 's exorts to with resect to China s exchange rate vis-à-vis for any HS 4-digit category is related, first of all, to the number of HS 6-digit categories within the 4-digit category that both and China exort, ij N, Ch, divided by the number of 6-digit categories exorts to within the relevant 4-digit category, Cj digit line with the relevant 4-digit, s ij N. Secondly, it is related to China's assumed constant share in each 6, which becomes art of the estimated coefficient in this secification - along with the difference in the two elasticities of substitution,, and the extent of ass-through,. 14

15 III. Estimation Strategy The analytical model above is useful in two imortant resects. First, it rovides an intuitive basis for caturing the cometition between an exorter and the country changing its exchange rate in third markets. Second, it allows us to identify clearly the key determinants of third-market effects. In articular, it gives us three redictions, relating to the sign and magnitude of the thirdmarket effect that we can take to the data. Secifically, Equations (11) and (12) would suggest the following: 11 (i) The value of exorts from country to country is negatively related to China s exchange rate vis-à-vis the imorter currency (measured in renminbi/imorter currency); i.e. 0 (ii) The magnitude of the third-market effect deends on the index of cometition with China; i.e. where I ij ) 0 I ij ) 11 We assume that the equilibrium value of bilateral trade is determined by imort demand, which would certainly be true in the extreme case of an infinitely elastic suly curve, but is also likely to be true with less strong assumtions. 15

16 In other words, greater is the degree of cometition with China, higher is the magnitude of the third-market effect. (iii) The magnitude of the third-market is higher, higher the elasticity of substitution between different imorted varieties ( ), the lower the elasticity of substitution between domestic and imorted varieties ( ) and higher the exchange rate ass-through ( ). 12 Cj It is not, however, straightforward to take rediction (i) above to the data. The basic roblem relates to the fact that the exchange rate (the exlanatory variable) and imorts (the deendent variable) have different dimensionality. Imorts vary by imorter, exorter, roduct and time. China s exchange rate on the other hand varies only by imorter and time. To identify the exchange rate effect would entail having to ignore a number of imorter-time varying factors that might be correlated with exchange rates and also affect imorts. In other words, in a correctly secified estimating equation, China s exchange rate would be absorbed by imorter-time fixed effects. To overcome this roblem, we exloit rediction (ii) to develo an identification strategy. The idea underlying our strategy is as follows. Take two countries, Malawi and Brazil. Assume that Brazil faces a greater degree of cometition with China in the US market. When the renminbi dereciates vis-à-vis the US$, exorts from Brazil to the US will fall more relative than exorts from Malawi to the US (Figure 1). 12 In the case of the count-based index, the magnitude of the third-market effect also deends on the share of China in the imorts ( s ). Cj 16

17 Figure1. Identification Strategy China Renminbi/$ Exorts Z (US) X (Malawi) Exorts Exorts Y (Brazil) Translating the same logic to the country-roduct-level, and focusing on variation within a country and roduct, we osit that a articular country s exorts of a articular roduct category to a destination country will be more affected by a change in China s exchange rate vis-à-vis country, the more in cometition that country s exorts of roduct category are with Chinese goods in that destination market. For examle, if China s exchange rate dereciates against the dollar, we osit that country s (Brazil s) exorts of a articular roduct category to a destination country (United States) will be more affected than Malawi s, the more in cometition Brazil s exorts of roduct category are with Chinese goods in US market. To this end, we create an index of cometition ( ) between Chinese exorts and those of its cometitors as described in the analytical section. Note that the index does not have a time subscrit which we exlain below. 17

18 With this index, we can define our estimating equation: (13) ln ln where is the value of exorts of HS 4-digit roduct from country to country. is the Chinese exchange rate vis-à-vis measured in renminbi er unit of s currency. This equation is similar to the difference-in-difference aroach due to the interaction term on the right hand side. The interaction term combines the exchange rate between China and the imorting country (say the renminbi-dollar exchange rate) and the index of cometition between the exorter and China in the imorting country. We exect that the coefficient in the equation will be negative: an increase in (a dereciation) will reduce s exorts more, the larger is the index of cometition. Econometrically, an advantage of this formulation is that we can control for a wide range of other effects on exorts through a set of very general fixed effects. In fact, in our core estimations, we emloy all three-way combinations of imorter, exorter, roduct and time fixed effects. For examle, and cature resectively any imorter and exorter countryroduct and time varying characteristics (e.g. country and roduct secific technological change or demand or olicy shock). Note that these fixed effects also encomass all country-time shocks both on the imorter and exorter side such as the state of the business cycle in both countries. catures any bilateral time-varying determinants of exorts (e.g. referential arrangements, 18

19 currency unions etc.). The existence of roduct-secific referential tariffs will not confound our identification because they will be absorbed in the fixed effects. The only factors that might not be controlled for are olicies of imorting country that vary by source country and roduct and time (for examle, changes over time in roduct-secific referential tariffs). Another virtue of this secification is that whereas the left hand side variable is imorterroduct- and destination-secific, the exchange rate is an aggregate variable which is neither roduct--nor destination-secific, and so there is less concern about reverse-causality flowing from exorts to the exchange rates. In other words, while exorts can affect exchange rates, it is less likely that exorts to certain destinations and of certain roducts affects the aggregate exchange rates. Moreover, note that unlike most existing macro and micro studies, we are not looking at the effect of a country s exchange rate on its own trade. The fact that we are trying to estimate cross-effects of Chinese exchange rates on exorts of other countries by itself lessens concerns about endogeneity of the exchange rate. For examle, it is less likely for the Chinese exchange rate olicy to be influenced by exorts of secific roducts to articular destinations. In sum, our secification minimizes the omitted variables roblem through a rich inclusion of fixed effects, while also addressing the endogeneity from trade to exchange rates. We believe that our secification makes a significant advance over the rior literature in disentangling the effects of exchange rates. Thus, one way of interreting our secification is that we exloit a reasonably exogenous macroeconomic shock exchange rate changes to identify the effect of exorts of cometitor countries through the variation afforded by the index of cometition. 19

20 The analytical section suggests that there are two ways of measuring the index of cometition. The value-based index (VBI) discussed above is defined as: (14) is defined at the 4-digit as in our baseline secifications above and denotes all the 6-digit categories within the 4-digit. As discussed above, this index is simly the weighted average of China's share in total imorts in each constituent 6 digit category which country exorts, where the weights are country s exorts in the corresonding 6 digit categories as a share of its total exorts in the 4 digit category. A count-based index (CBI) is the second way of measuring the index of cometition. This can be exressed relatively simly as: (15) ij N, C ij N This index is simly the number of 6-digit lines in which China s exorts overla with countryi s exorts in market j divided by the total number of 6-digit lines that country i exorts in the relevant 4-digit category in market j. The VBI is otentially vulnerable to the endogeneity roblem because it is exressed in values, like the deendent variable. The CBI, while derived from theory under symmetry assumtions, 20

21 has the emirical virtue of not being measured in value terms and hence being less related to the left hand side variable. In a sense, the count index is a measure of cometition on the extensive margin whereas the deendent variable relates more to the intensive margin. To minimize the endogeneity roblems we comute both indices based on the initial eriod observations (i.e. for the year 2000). IV. Data We focus on the eriod , during which concerns about China s exchange rate olicy have been most debated. For this eriod, we comile disaggregated data on bilateral exorts from the UN Comtrade database. The data are for roughly 6000 non-oil HS 6-digit lines covering digit roducts. We cover the 57 major imorting countries (making sure that we include all countries that together accounted for over 95 ercent of total exorts of develoing countries) and 124 develoing country exorters which are otentially in cometition with China (summary statistics are rovided in Aendix Table 1 and the list of imorting and exorting countries covered in Aendix Table 2). Trade data are deflated by the US CPI. We recognize that ideally we would use different rice indices to deflate trade between different country airs but this is not currently feasible. However, the resence of the very general fixed effects has the consequence also of imlicitly deflating the trade data. The data are imlicitly deflated by rices that vary by imorter, roduct and time, by imorter, roduct and exorter and by exorter, roduct and time. They are not deflated by rices that vary along all four dimensions (imorter, exorter, roduct, and time). So, 21

22 in some ways the left hand side variable is a hybrid, somewhere in between a ure quantity variable and a value variable. Exchange rate data are from the IMF s International Financial Statistics (IFS) database. Consistent with Equation (6), the bilateral exchange rate is deflated by China s CPI. Before we resent the econometric results, it is worth looking at some basic data. Figure 2 lots China s average index of cometition (where the average is over all exorters and roducts). The index is measured in two ways consistent with the discussion in the analytical section. Both the VBI and the CBI rise over time, consistent with China becoming a bigger and more diverse exorter. The CBI shows in articular that China occuies nearly all the roduct sace of all other develoing country exorters. Figures 3a and 3b lot the same indices but disaggregated by region. These charts show that China s overla with all regions has risen steadily over time, with the level of the overla greatest with other exorters in Asia (over 95 ercent in 2008 for the CBI) and least with Euroe and Central Asia. V. Results Main Findings All results are resented for both variations of our cometition index. In Table 1, we resent the baseline results. Our core samle has nearly 3.6 million observations. Columns [1]-[4] use the value-based index (VBI) while columns [5]-[8] use the count-based index (CBI). In both cases, the secifications rogressively increase the number of fixed effects, with a comrehensive set of 22

23 fixed effects in columns [4] and [8], making the secification a very demanding one. These will constitute our core secifications. All regressions are clustered at the imorter-exorter-roduct level. We find that the coefficient on the interaction term between the Chinese exchange rate and the index of cometition is consistently negative and significant at the 1 ercent confidence level. In other words, the more say Brazil is in cometition with China in a articular roduct, a dereciation of the Chinese exchange rate vis-à-vis say the dollar is associated with a greater reduction in Brazil s exorts of that roduct to the United States. We subject this core secification to a series of robustness checks in Tables 2-6. In Table 2, column [1], we dro outliers, defined as the to and bottom 1 ercentile of observations. The key coefficient is negative and statistically significant with the magnitudes close to those for the larger, core samle. In columns [2]-[4], we cluster the standard errors at the exorter-imorteryear, exorter-roduct-year and imorter-roduct year levels, and the statistical significance of the coefficients remain unchanged. Our core secification uses annual data. To test whether the results hold for the medium run, we use a long difference aroach suggested by Acemoglu and Johnson (2007). Thus, in column [5], we use observations only for 2000 and 2008 and find that the results remain similar to the baseline, with the magnitude of the interaction coefficient increasing by a little. In columns [6] 23

24 and [7], to make sure that the results are not driven by the choice of year for measuring the index of cometition, we measure the index for the years 2001 and 2002, resectively. 13 In column [8], we use an alternative measure of cometition the exort similarity index due to Finger and Kreinin. 14 Thus, for a wide range of robustness tests, the core results remain unaltered, both in the sense that the coefficients are stable and consistently significant at the 1 ercent confidence level. In Table 3, we test for robustness to alternative measures of the exchange rate variable. In our analytical framework, we assumed that the rice of Chinese goods in the imorting country market is determined by a simle relationshi between domestic rice in China and an exchange rate ass-through. Based on the framework, in our core secifications, we deflate the nominal 13 The estimated coefficient on the interaction between the index and exchange rate continues to be negative and statistically significant if we use a contemoraneous index of cometition, which would of course raise serious concerns about endogeneity. X 14 ijgt X Cjgt The Finger-Kreinin index can be exressed as: FK ijt min [, ] where g X X g g X X ijgt X Cjgt X ijgt Cjgt Share of roduct g Share of roduct g in total exorts from i to j in total exorts from China g ijgt at the 4 - digit level to j g Cjgt at the 4 - digit level. The results are also robust to using the alternative formulation of the Finger-Kreinin Index defined as and the weighted Finger-Krenin Index defined as

25 bilateral (between China and the imorting country) exchange rate by Chinese rices. The imlicit assumtion here is that Chinese roducers take account of changes in the bilateral exchange rate and average domestic inflation to determine exort rices. However, there could be alternative ways Chinese roducers and exorters determine their destination-secific exort rices. Chinese roducers could be influenced just by the nominal bilateral exchange rate ( ) or by the real bilateral exchange rate ( / ), with and denoting rices in imorting country and China resectively. The secifications corresonding to these two ways of measuring the exchange rate are in columns [1] and [2] (for the VBI) and columns [5] and [6] (for the CBI). In both cases, the results are robust although the magnitudes decline relative to the core secification. Yet other models of ricing behavior could involve Chinese roducers looking at changes in their multilateral cometitiveness in determining destination-secific exort rices. In this case, the relevant exchange rate is not destination secific but a multilateral one that is identical across all imorters ( ) where stands for China s multilateral exchange rate and hence without a subscrit). 15 We re-estimate the core regression to cater to these ossibilities by using the IMF s effective exchange rate as the relevant measure with the nominal rate in columns [3] and [7], and the real rate in columns [4] and [8]. Again, the coefficients are correctly signed and significant at the 1 ercent confidence level. Interestingly, these coefficients are substantially greater than for the core secification. 15 Note that in this case, the exchange rate varies across time and the index varies across imorter, exorter, and roduct so that the interaction term exloits the variation across all four dimensions. 25

26 In Tables 4 and 5, we test for robustness across exorters, defined in geograhic terms (Table 4) and in terms of income levels (Table 5). The results are robust in both cases. The results are also consistent between the two tables. For examle, the coefficients are lowest for African exorters and in Table 5, the coefficients are lowest for low income countries and greatest for uer middle income countries. 16 In Table 6, we check if the results are robust to the degree of roduct disaggregation. In the core secification, the data are at the HS 4-digit level. In Table 6, we use data at the HS 2-digit level. The indices of cometition are measured by aggregating across 6-digit lines within the 2-digit category. The samle size shrinks from over 3.6 million to about 860,000 observations. But the interaction term remains negative and significant. Overall, the results in Table 1-6 confirm the redictions from the analytical framework. The elasticity of develoing country exorts with resect to Chinese exchange rate deends significantly on the index of cometition with China. Further, higher the index of cometition, a given dereciation of the renminbi is associated with a bigger reduction in develoing country exorts. Discussion of Magnitudes 16 We also tested for robustness across imorters, defined in terms of advanced and other countries, and the results hold for each category of imorters. 26

27 How can we interret the magnitude of the estimated coefficients on the interaction between the exchange rate and the index of cometition? Evaluated at the average value of the indices, the results from our core secification (columns [4] and [8] of Table 1) yield a third-market effect of [i.e. (0.4)*(0.352)] and [i.e. (0.9)*(0.22)] for the value and count-based indices resectively. The estimates imly that a 10 ercent dereciation/areciation of the renminbi is associated with a reduction/increase in develoing country exorts at the roduct level on average by 1-2 ercent. If we use the range of coefficients from Tables 1-3, the effects are in the range of 1-4 ercent. Imortantly, for high indices of cometition (in the 90 th ercentile), the effect can be as high as 5 ercent. It is imortant to recognize that our estimates do not measure the overall effect of the exchange rate movements but rather a difference-in-difference estimate. Recall that the third-market effect we estimate in equation (13) is given by: (18) Our estimations identify which we multily by the relevant value of the index of cometition to obtain the average third-market effect. This estimate is akin to a difference-in-difference or marginal estimate which is to be distinguished from the overall exchange rate effect. From Equation (13), the estimated overall effect would be given by: 27

28 (19) Where is that comonent of the effect of the exchange rate on exorts which we cannot identify because it is absorbed in the fixed effects. One way of interreting the difference-in-difference estimate is to comare two countries, say Vietnam and Russia that vary in their average index of cometition with China. The two countries value-based indices are 0.5 and 0.3, resectively. In this case, using the estimates from Column [4] in Table 1 suggests that a 10 ercent dereciation of the Chinese exchange rate will reduce Vietnam s exorts of a tyical HS 4-digit category to third markets by 0.7 [i.e. ( )*0.352)] ercent more than Russia s. What does our analytical framework redict about the magnitude of the third-market effect? Recall from Equations (11) and (12) that the third-market effect redicted by theory based on the value and count-based indices are the following, resectively: (16) 28

29 (17) From the existing literature, we can obtain estimated values for each of the arameters. Of course, there is wide variation in each of these, but some ball-ark estimates are the following: 3, 1, 0.4 and s The estimates of (the elasticity of substitution between imorted goods, or the micro-armington elasticity) and (the elasticity of substitution between domestic and imorted goods, or the macro-armington elasticity) are based on Feenstra, Obstfeld and Russ (2011). The ass-through coefficient ( ) is an average of the estimates from Cama and Goldberg (2006) for industrial countries and the estimates of Goinath et. al. (2011) for the United States. s refers to the average share of China in the markets of each of the imorting countries which we obtain from our data. 17 Combining these estimates with the average value of the index of cometition for the value ( and count-based ( indices from our data (of 0.4 and 0.9, resectively), yield a 17 Xing (2010) looks secifically at ass-through of Chinese exchange rates to imort rices in US and Jaan, and estimates ass-through coefficients of 0.23 and 0.56 for the US and Jaan resectively. 29

30 magnitude of the third-market effect of [i.e. (0.4)*{(0.4)*(3-1)}] for the value-based index and [i.e. (0.9)*{(0.4)*(0.4)*(3-1)) for the count-based index. However, the values of the elasticity of substitution that we use to derive the theoretical rediction are based on Feenstra et. al. (2011), who estimate the elasticity for goods at a level of disaggregation close to the HS 10-digit level. 18 Our data on the other hand are at HS 4-digit so that the relevant elasticity for our urose could be well below the value of 3 that we use here. 19 Thus, third-market effect of 0.3 must be viewed as the uer bound redicted from theory. Therefore based on theory, we should exect that a 10 ercent dereciation of the renminbi against the dollar would lead at most to an average reduction of exorts of cometitor countries of 3 ercent. How do our estimates comare with those redicted from theory? Imortantly, our estimated third-market effects of 1-2 ercent from the baseline results are not far from the uer bounds suggested by theory. We should, however, be careful in comaring the estimates redicted from theory and those based on our emirical secification since while the former measures the total third- market effect, the latter measures only the marginal effect. The main thrust (and virtue) of our estimation rocedure is to show that the channel of transmission from exchange rates to cometitor country exorts can be detected in the data. The cost of being able to show this 18 Feenstra et al. (2011) actually use a unique data source called Current Industrial Reorts (CIR) which is ublished by the US Bureau of the Census. Their dataset covers 191 goods, of which 80 are based on a single 10-digit HS commodity, and another 42 are based on two or three 10-digit HS commodities. So the majority of the dataset is at a highly disaggregate level. 19 Broda and Weinstein (2006) argue that with more disaggregated data, one is likely to find higher estimates of the elasticity of substitution. 30

31 transmission mechanism recisely (by exloiting all the rich sources of variation in the data) is the difficulty in inning down the magnitude of the overall effect of exchange rates. In order to get an idea of the magnitude of the overall third-market effect, we estimate an alternative regression which is secified as follows: 20 ln ln ln Note the three differences relative to our core secification in Equation [13]. First, we have collased the data from four dimensions (imorter, exorter, roduct, and time) to three dimensions, eliminating the roduct-secific variation. The second difference relative to the core is that we have the exchange rate term on its own in addition to the interaction term (the latter being also a feature of the core secification). 20 Now, we can measure the total effect of an exchange rate change which is given by:. The secification thus allows us to identify both and. The third change is that because we have an exchange rate term on its own which varies by time and imorting country, we cannot include a fixed effect that varies along these two dimensions. Thus our ability to estimate the overall exchange rate effect comes at the rice of sacrificing both roduct-secific variation and key fixed effects, which is why we regard this secification as inferior to our core secification. 20 Note that the index of cometition is now defined for a country relative to China in the imorting market but averaged across all roducts. 31

32 The results from estimating Equation [20] are shown in columns [1] and [3] of Table 7 for the two indices. A slightly richer version that includes some controls that vary by imorter, exorter and time are resented in columns [2] and [4]. Because of the severe reduction in the variation in the data (the samle is about 13 ercent of that in the core secification), the coefficients are less recisely estimated. The exchange rate term is consistently negative and significant. Taking the most general secification reorted in column [2], we can ascertain the magnitude of the overall third-market effect. At the median value of the index of cometition of 0.4, this overall effect is about That is, a 10 ercent dereciation in the Chinese exchange rate relative to the dollar results in a nearly 2 ercent reduction in the average exorts of a cometitor country in the US market. This overall effect is very close to the more recisely estimated difference-in-difference magnitude from the core secification both of which are close to and consistent with the theoretical redictions. Overall, the estimates in this aer suggest that a 10 ercent dereciation/areciation in the renminbi exchange rate vis-à-vis an imorting country decreases/increases on average develoing country exorts by 1-2 ercent. Given the 30 ercent areciation of China s real exchange rate over (Figure 4), our findings suggest that this could have been associated with about a 2-6 ercent increase in the tyical develoing country s exorts to third markets. 32

33 Finally, we evaluate the third theoretical rediction discussed above which relates the magnitude of the exchange rate effect to the tye of good. Recall that equations [16] and [17] suggests that, the magnitude of the third-market effect is higher, higher the elasticity of substitution between different imorted varieties ( ) and the lower the elasticity of substitution between domestic and imorted varieties ( ), and higher the exchange rate ass-through ( ). Note that we do Cj not have information on, and at the roduct-level to take the theoretical redictions literally to the data. Cj However, in order to make a first ass, we artition the data into homogenous versus differentiated roducts based on Rauch (1999) classification. 21 As shown in Table 8, columns [1], [2], [6] and [7], we find that the coefficients on the interaction between the index of cometition and exchange rates are higher in magnitude for homogenous roducts vis-à-vis differentiated ones. However, we not find a large difference in magnitude of the estimated coefficients. There could be two ossible exlanations for this. First, although ossibilities for substitution across imorted varieties are likely to be higher for homogeneous roducts, the substitution between home and imorted varieties is also likely to be higher for these roducts; damening the third-market effect. Second, the ass-through coefficient is also likely to be lower for homogenous roducts. For examle, Yang (1997) shows both theoretically and emirically 21 Note that Rauch s classification is available at the SITC 4-digit; we concord it to HS 6-digit level using standard concordance tables, and then artition the data into homogenous and differentiated using Rauch s liberal classification (reference riced goods are included in the homogenous category). We then aggregate the data to the HS 4-digit level. 33

34 that the degree of exchange rate ass-through is negatively related to the degree of substitution (or ositively related to the degree of roduct differentiation). 22 We also emirically exlore whether the magnitude of the third-market effect varies by slitting the roducts along other dimensions likely to be correlated with the degree of roduct differentiation. Secifically, we look at variations by skill-intensity of the roduct low, medium and high-skill intensive. The classification of roducts by skill intensity is due to Peneder (2001). 23 Based on our roduct classification, the roortion of differentiated goods increases with skill-intensity (whereby more than 95 ercent of high-skill intensive are also differentiated by Rauch s definition). The results shown in Table 8, columns [3]-[5] and [8]-[10] suggest that our finding of a negative and statistically significant coefficient on the interaction between the index of cometition and exchange rate continues to hold across different slits by roduct-tye. Consistent with the redictions from our analytical framework, the magnitude of estimated coefficients on the interaction term declines with skill-intensity, with the magnitude of the effect being the highest for low-skill intensive roducts When roducts are highly substitutable, a rice increase is more likely to lead consumers to switch to other variants. Thus, foreign firms are more likely to kee their rices in line with the domestic rice and absorb exchange rate shocks rather than assing them on to rices. On the other hand, when roducts are highly differentiated (so that they are less substitutable), firms are less worried about losing customers in case of a rice increase and will be able to ass cost shocks to rices. 23 The classification of roducts into skill-intensity is at the ISIC 2-digit level. The classification is matched to the HS classification using concordance tables from the UN Statistics deartment. 24 We look at slit by end-use of the roduct: consumer versus caital and intermediate. Information on roduct tyes is taken from the UN s Broad Economic Classification (Pula, Gabor, and Peltonen, 2009). We did not find the interaction coefficient to differ significantly between these roduct categories. 34