Predictable Returns of Trade-Linked Countries: Evidence and. Explanations

Transcription

1 Predictable Returns of Trade-Linked Countries: Evidence and Explanations Savina Rizova Abstract Recent evidence shows that returns of trade-linked firms and industries are predictable due to the gradual diffusion of information. I test whether information diffuses gradually across the stock markets of trade-linked countries. I find that stock market returns of a country s major customers or suppliers predict the subsequent stock market return of that country. Strategies based on the country-level customer and supplier momentum yield monthly alphas of over 110 basis points. I also explore the economic channels behind the correlation of stock market returns of trading partners. I find that a positive shock to the stock market of a country s major customer predicts higher future exports to that customer. Similarly, the stock market return of a country s major supplier predicts future imports from that supplier. University of Chicago Booth School of Business; srizova@chicagobooth.edu. I would like to thank the members of my dissertation committee Eugene Fama, Tarek Hassan, Tobias Moskowitz and Lubos Pastor for their invaluable support and comments. I also thank Timothy Dore, Marina Niessner, Eduardo Repetto, Alexi Savov, Jhe Yun, participants at the University of Chicago PhD Student Finance Brownbag and participants at the London Business School Transatlantic International Conference for insightful discussions. 1

2 1 Introduction This paper investigates whether information diffuses gradually across the stock markets of trade-linked countries. I identify trade links by using annual bilateral exports and imports of goods. I find that stock market returns of a country s major customers or suppliers predict the subsequent stock market return of that country. Both the customer and supplier momentum effects yield large profits. For the period from July 1981 to March 2009, a monthly strategy of buying indices of countries whose equally weighted portfolio of customers had the highest returns in the previous month and selling short indices of countries whose equally weighted portfolio of customers had the lowest returns yields an average abnormal return of 117 basis points per month, or an annualized return of 14.0% per year. The same strategy applied to suppliers generates an average abnormal return of 81 basis points per month, or 9.7% per year. The customer and supplier momentum effects are robust to different portfolio specifications and are not driven by own country momentum. If the predictability in stock market returns is caused by the gradual diffusion of information, then the predictability should be weaker for countries that are widely followed in financial markets. Countries with the largest stock markets and/or largest gross domestic products attract a lot of investor attention and capital flows. Hence, market participants are more likely to collect information on these countries major trading partners. Indeed, both the customer and supplier momentum effects are significantly weaker for the largest sample countries. Moreover, the effects are weaker for countries with more trade-related articles in the Wall Street Journal. These findings are consistent with the hypothesis that the observed return predictability is driven by slow diffusion of information about smaller countries. I acknowledge that my results are also consistent with the hypothesis that the return predictability is driven by market frictions. Market frictions such as capital controls, trading costs and nonsynchronous trading could delay the reaction of stock markets to information about trading partners and generate crosscountry predictability. In addition, capital controls and trading costs are more likely to be binding for smaller countries. To address this alternative explanation, I implement the customer and supplier momentum strategies using U.S.-based single-country exchange-traded funds (ETFs). Since these instruments are traded throughout the day on the U.S. stock exchanges, they give investors easy and immediate access to foreign markets. For the period from April 1996 to March 2009, the average returns on the implemented strategies (net of trading costs) are large and statistically indistinguishable from those on the benchmark strategies. These results are inconsistent with the market frictions hypothesis, but are consistent with the gradualinformation-diffusion hypothesis. As Hong et al (2007) point out, under the hypothesis of gradual information diffusion, predictability of 2

3 returns across assets should be based on predictability of fundamentals. This means that the ability of a country s major trading partners to forecast its stock market should be based on their ability to forecast its fundamentals. Hence, it is necessary to identify the economic channels behind the customer and supplier effects. I present the following hypothesis for the economic channel behind the customer effect. A positive shock to the stock market of a major customer implies that the purchasing power of equity holders in the customer country has gone up and that purchases of imported goods are likely to rise due to the wealth effect. In addition, the rise of the stock market is likely to lower the cost of capital for domestic firms and stimulate them to make more investments some of which would involve purchasing imported goods. The expectation of higher sales in the customer country pushes the producer country s stock market up and yields a correlation between the stock market returns of the two trading partners. The logic for negative shocks is analogous. This hypothesis implies that a positive shock to a major customer s stock market is likely to generate a rise in exports to this customer in the subsequent months and that a negative shock is likely to generate a fall in exports. To test this hypothesis I examine the predictive power of customer stock market returns for changes in exports over the next twelve months. I find that customer stock market returns help forecast changes in exports to that customer in the near future. This is direct evidence in support of the hypothesis that a shock to the stock market of a major customer country generates a shock to the stock market of a producer country because of expected changes in future exports to that customer. I also present and test the following hypothesis for the economic channel behind the supplier effect. A positive technology or macroeconomic shock to the stock market of a major supplier is likely to enable firms in the supplier country to produce goods more cheaply and efficiently. As a result, the positive shock is likely to spill over to the firms in the buying country through lower import prices. Similarly, a negative macroeconomic shock to the firms in the supplier country is likely to affect negatively the firms in the buying country by forcing them to either accept higher import prices or search for alternative suppliers. If the main channel of cross-country spillover is import prices, a positive shock is likely to stimulate higher imports of goods from the supplying country and a negative shock is likely to stimulate lower imports of goods. To test this hypothesis, I examine the predictive power of supplier stock market returns for changes in imports over the next few months. I find that a supplier s stock market return indeed helps forecast changes in imports from that supplier over the next three to twelve months. To my knowledge, this paper is the first to document the predictability of stock market returns of trading partners and to investigate empirically the economic channels behind this predictability. The paper is motivated by the growing literature on gradual information diffusion. On the theoretical side, Merton (1987), Hong and Stein (1999), Hirshleifer and Teoh (2003), Peng and Xiong (2006), Hong et al (2007), as well as Menzly and Ozbas (2009) develop models in which due to limited market participation and/or limited 3

4 investor attention information diffuses gradually and there is stock return predictability. On the empirical side, a number of recent studies find evidence that information diffuses slowly across asset markets. The papers that are most closely related to mine are those of Cohen and Frazzini (2008), Menzly and Ozbas (2009) and Shahrur et al (2009). These studies provide evidence that economically linked firms and industries exhibit return predictability due to slow diffusion of information. Cohen and Frazzini (2008) find that stock returns of important customers forecast the subsequent stock returns of supplier firms in the United States. Menzly and Ozbas (2009) document that equity returns of industries are predicted by past equity returns of upstream and downstream industries in the United States. Shahrur et al (2009) look at developed markets around the world and also find that stock returns on customer industries lead the stock returns on supplier industries. My analysis differs from the above papers in three main aspects. First, I provide novel evidence on country-level customer and supplier momentum. Second, I examine the robustness of customer and supplier momentum profits to trading costs. Third, I shed light on the economic channels that explain the forecasting power of stock market returns of trading partners. None of the above papers examines the economic channel behind supplier momentum. Cohen and Frazzini (2008) investigate the real effects of customer momentum but since producer firms do not provide detailed and frequent information on sales to individual customer firms, the authors cannot test directly the economic channel behind the customer momentum effect, namely, whether customer stock returns predict future sales to that specific customer. The availability of monthly country-to-country export and import data, however, allows me to test directly the economic channels behind the observed customer and supplier effects at the country level. The remainder of the paper is organized as follows. Section 2 describes the data. Section 3 discusses the portfolio construction methodology. Section 4 presents the main customer and supplier momentum results. Section 5 examines whether the cross-country predictability is consistent with the gradual diffusion of information hypothesis. Section 6 explores the economic channels behind the customer and supplier momentum effects. Section 7 concludes. 2 Data My sample of countries consists of the 23 developed countries and the 24 emerging countries included in the MSCI World Index and MSCI Emerging Markets Index as of December The developed markets in the sample are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sweden, Switzerland, the United Kingdom and the United States. The emerging markets in the sample are Argentina, Brazil, 4

5 Chile, China, Colombia, the Czech Republic, Egypt, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Morocco, Pakistan, Peru, Philippines, Poland, Russia, South Africa, Taiwan, Thailand and Turkey. For each of the sample countries I obtain annual data on the U.S. dollar value of exports and imports of goods by partner from the Directions of Trade Statistics database of the International Monetary Fund. This leads to the exclusion of Taiwan from the sample because it is not included in the Directions of Trade database. Annual exports and imports are available from 1980 onwards. The Fund states that information on monthly exports and imports by trading partner is reported on a regular basis (with a delay of four months or less) by approximately 49 countries, representing all advanced economies and about 19 developing and emerging economies. Hence, it is safe to assume that data on annual trade for the developed and emerging markets studied here are publicly available six months after the end of a calendar year. Thus, I use annual bilateral exports and imports with a minimum six-month lag to identify the major customers and suppliers of the 46 sample countries for the period July 1981 to March I define major customers and suppliers as countries that account for at least 5% of a sample country s exports and imports, respectively. Note that major customers and suppliers need not be confined to the sample of 46 countries. For example, Slovakia is considered a major customer of the Czech Republic (accounting for 17% of Czech exports in 1995) even though it is not in the sample of main countries, referred hereafter as producer countries. The customer and supplier set of countries are allowed to be broader than the producer set of countries because none of the proposed trading strategies involve buying or selling short the stock markets of major customers or suppliers. Hence, concerns about liquidity and accessibility of stock markets to foreign investors need not limit the set of customers and suppliers analyzed in this study. To guarantee that export and import links are an important source of information for stock market participants, I exclude from the analysis sample countries in years when their total exports or imports account for less than 20% of their gross domestic product (GDP). When international trade represents a small portion of a country s domestic production, stock market returns of trading partners are unlikely to affect the country s fundamentals and stock market returns. Countries GDP data (in millions of current U.S. dollars) are obtained from the World Development Indicators (WDI) database of the World Bank. I match the trade data for calendar year t - 1 ( ) with monthly stock market returns for July t to June t + 1. As in many other international asset-pricing studies, stock market returns are measured by the Morgan Stanley Capital International (MSCI) country indices for gross returns (capital gains plus gross dividends). The country indices are value-weighted and include the largest and most liquid stocks in each market. Each index represents 80% to 85% of the investable stock universe from the perspective of international institutional investors. MSCI starts providing returns for emerging markets in Hence, emerging market countries begin appearing in the analysis in January of For four countries (Czech 5

6 Republic, Hungary, Pakistan, and Slovakia) I supplement the beginning of their return history with the S&P/IFCI country indices. The final sample for the customer momentum analysis includes 32,445 producer-customer monthly observations, or a total of 216 unique producer-customer links for July 1981 to March The final sample for the supplier momentum analysis includes 34,030 producer-supplier monthly observations, or a total of 268 unique producer-supplier links for the sample period. Note that the average duration of a producer-customer link is 13 years and the average duration of a producer-supplier link is 11 years. Hence these trade relations are typically long lasting. For some parts of the analysis I use the stock market capitalization of the sample countries and their partners. I obtain year-end total market capitalizations for 1980 to 1987 from the MSCI Blue Books. For 1988 to 2008 stock market sizes are from WDI. Since I need market values for emerging market countries for 1987, but the MSCI Blue Books cover only developed countries, I estimate emerging market values by adjusting the 1988 values for the price return of the country index over Also, MSCI provides combined market values for Singapore and Malaysia for 1983 to 1987, so I estimate the market values for Singapore by using its market size for 1982 and adjusting it for the price returns of its index for 1983 to Ireland appears in WDI in 1995 so I use MSCI for and estimate its market value for using price returns. Korea starts in WDI in 1989 so I estimate its value for Czech Republic starts in 1994 so I estimate its value for Portfolio Formation Methodology To investigate whether there is stock return predictability among trading partners at the country level, I conduct the following portfolio analysis. At the beginning of each month t from July 1981 to March 2009, I identify the major customers (suppliers) of each sample country and form an equally weighted or salesweighted average of the customer (supplier) returns, measured in local currency, for month t - 1. Sales weights are based on exports for customer countries and imports for supplier countries. Then I sort sample countries on the prior-month average return of their major customers (suppliers). I assign the top 30% of the sorted countries to the Top 30 portfolio, the bottom 30% to the Bot 30 portfolio and the middle 40% to the Med 40 portfolio. Portfolios are held for a month and then rebalanced. Portfolio returns are measured in U.S. dollars. I use three alternative types of country weights for the portfolios: equal weights, GDP weights and market capitalization (value) weights. To ensure that GDP data are known before the portfolios are formed, I match GDP for year t - 1 with returns for July of year t to June of t + 1. For the value weights, I match market capitalizations for the end of year t - 1 with returns for January to December of year t. I 6

7 also construct customer and supplier momentum strategies (Top-Bot). These strategies buy the customer (supplier) Top 30 and short sell the customer (supplier) Bot 30. If stock markets underreact to news about customer and supplier countries and fundamentals of trading partners are positively correlated, the customer and supplier momentum strategies should earn positive average abnormal returns. To estimate abnormal portfolio returns I first use a global CAPM model: R i,t RF t = α i + β i (MKT t RF t ) + e i,t where R i,t is the month t return on portfolio i, MKT t is the return on the MSCI All Countries Index, and RF t is the one-month U.S. Treasury bill rate. (I use the U.S. risk-free rate to compute excess returns since all strategies are constructed from the standpoint of a U.S. investor.) I also estimate abnormal portfolio returns using the following five-factor model: R i,t RF t = α i + β i (MKT t RF t ) + s i (SMB t ) + h i (HML t ) + m i (MOM t ) + f i (MOMF X t ) + e i,t In this model all factors are global, value-weighted and constructed at the country level. SMB is a smallminus-big factor constructed as the monthly difference in U.S. dollar returns between the bottom and top half of all sample countries sorted on their stock market capitalization for the end of the previous year. HML is a high-minus-low book-to-market factor. To construct this factor, at the end of each December I sort all sample countries into three equal groups based on their book-to-market ratios as of that date. Portfolios are rebalanced every year. HML is the difference in monthly U.S. dollar returns between the top and bottom tertile. The book-to-market ratio of each country is the value-weighted average book-to-market equity of the stocks in that country (using as weights the stocks market capitalization for December end). The country book-to-market ratios are from MSCI for developed-market countries and from S&P/IFCI for emerging-market countries. MOM is a global stock market momentum factor. To construct this factor, at the beginning of each month t from July 1981 to March 2009 I sort all sample countries into tertiles based on their compounded local-currency return for month t - 12 to month t Portfolios are rebalanced monthly. MOM is the difference in monthly U.S. dollar returns between the top and bottom tertiles. MOMFX is a global currency momentum factor. It is constructed analogously to MOM but using currency returns relative to the U.S. dollar instead of stock market returns. The HML, MOM and MOMFX factors constructed here are similar to the basic value and momentum strategies for country stock indices and currencies of Asness et al (2009). Using HML and MOM factors is in line with the empirical evidence that country stock indices 1 Using cumulative returns for month t-12 to month t-1 or for month t-6 to t-1 to form MOM yields similar results. 7

8 exhibit value and momentum effects (Asness et al (1997), Bhojraj and Swaminathan (2006) and Asness et al (2009)). The inclusion of MOMFX is motivated by the fact that even though ranking-period returns are measured in local currency, the holding-period returns are measured in U.S. dollars. 2 It is important to examine how well the global CAPM and the global multifactor model work since bad model problems could affect my main results. Appendix A tests whether the two models capture the crosssection of average returns on the market portfolios of individual countries. As Table A1 shows, the average global CAPM alpha is 0.6% per month. Only 12 of the 46 single-factor alphas, however, are greater than 1% per month and only 8 are more than two standard errors away from zero. Turning to the multifactor model, we see that the average alpha drops three times to 0.2% per month. Only three multifactor intercepts are greater than 1% per month and only two are more than two standard errors away from zero. Hence, the multifactor model appears to explain better the cross-section of average country returns. Gibbons, Ross and Shanken (1989) F-tests also support this conclusion. Overall, Table A1 suggests that the global CAPM and, especially, the global multifactor model do an adequate job of explaining the cross-section of average country returns. 4 Main Results 4.1 Customer Momentum I present the main results of the paper in Tables 1 and 2. Table 1 reports the average monthly returns for July 1981 to March 2009 for portfolios of producer countries sorted on the equally weighted (Panel A) or sales-weighted (Panel B) average of the prior-month local currency returns of their major customers. Results are shown for producer portfolios using equal, GDP and value weights. Portfolios are rebalanced monthly. Portfolio returns are monthly U.S. dollar returns in percent. Excess returns are net of the U.S. risk-free rate. Global CAPM Alpha is the intercept from a time-series regression of excess monthly portfolio returns on MKT-RF. Global Multifactor Alpha is from a time-series regression of excess monthly portfolio returns on MKT-RF, SMB, HML, MOM and MOMFX. Sorting countries on the lagged returns of their customers yields large differences in subsequent monthly returns. As Panel A shows, the equally weighted customer momentum strategy that is long the top 30% of the producer countries and short the bottom 30% delivers an average excess return of 1.11% per month (with a t-statistic of 4.12) or 13.3% per year. Adjusting returns for global factors has a negligible effect on the results. The global multifactor alpha for the equally weighted customer momentum strategy is 1.17% 2 For evidence on currency momentum, see, for example, Bhojraj and Swaminathan (2006) and Asness et al (2009). 8

9 (with a t-statistic of 4.10). The GDP and value-weighted portfolios earn large abnormal returns as well. The multifactor alpha is 0.98% per month (with a t-statistic of 3.03) for the GDP-weighted customer momentum strategy and 0.76% per month (with a t-statistic of 2.49) for the value-weighted customer momentum strategy. Therefore, the results are not driven by the smallest and least liquid countries in the sample. Panel A of Table 1 also reports average abnormal portfolio returns for the two halves of the sample period. The results for the sub-periods are of similar magnitudes to the results for the total period. The t-statistics are a bit lower for some specifications and sub-periods but this is most likely due to the smaller sample sizes. Overall, the subsample analysis shows that the customer momentum strategy is profitable in both halves of the sample period. Panel B of Table 1 provides further evidence that sorting countries on the prior-month returns of their major customers yields large differences in subsequent returns. Panel B is constructed analogously to Panel A, except that here the customer portfolios are sales-weighted. The results are similar to those in Panel A. The multifactor alpha is 1.04% per month (with a t-statistic of 3.68) for the equally weighted customer momentum strategy, 1.10% (with a t-statistic of 3.46) for the GDP-weighted strategy and 0.83% (with a t-statistic of 2.95) for the value-weighted strategy. Note that in almost all specifications in Table 1 alphas grow steadily across the producer portfolios as the lagged customer returns go from low to high. 3 This is consistent with the slow diffusion of information hypothesis because it suggests that the observed stock market reaction of the producer country is related to the size of the initial stock market reaction of the customer country. Therefore, Table 1 says that there is a lead-lag effect in the stock market returns of customer and producer countries. Although I do not report factor loadings for the portfolios (for brevity), the comparison of the excess returns and the alphas shows that portfolio abnormal returns cannot be explained by exposures to the global factors. The loadings on MKT, MOMFX and MOM are statistically indistinguishable from zero at the 5% significance level in all specifications of the customer momentum strategy for the overall period. The coefficient on SMB is statistically different from zero in only one specification and the coefficient on HML in only two. Figure 1 provides further evidence that customer momentum returns are not a reward for bearing global market risk. The graph shows the annual returns on the customer momentum strategy (based on equal 3 The alphas of most portfolios are positive. As Fama and French (1998) point out, this puzzling fact is frequently observed when country portfolios and a value-weighted global market portfolio are used in a global asset-pricing model. My examination of this issue suggests that the positive intercepts are in large part due to the role of Japan in the MSCI All Country Index. For July 1981 to March 2009, the average monthly return for Japan is 0.65% compared with 0.88% for the United States and 0.90% for the United Kingdom. Not only did Japan underperform substantially the other major markets in the MSCI All Country Index, but it also had a large weight in the index during long parts of the sample period (exceeding 20% for most of 1984 to 1996 and reaching 40% in ). Hence, it seems that the combination of Japan s underperformance and large country weight explains most of the positive-alpha issue. 9

10 customer weights) along with the annual returns on the MKT-RF factor. It is evident that customer momentum does not tend to underperform sharply in bad states of the world, measured by sharp declines in the global market. Indeed, in 2008 when almost all strategies in financial markets yielded abysmal returns and the world market dropped by more than 40%, the equally weighted and value-weighted customer momentum strategies both lost less than 10%. The monthly correlation between the customer momentum and the global market return series ranges from to across the different strategy specifications. To further examine the properties of the customer momentum portfolio, I plot in Figure 2 the evolution from month t + 1 to month t + 24 of the cumulative payoff on the long-short portfolio based on customer returns for month t. The solid line represents the average cumulative difference between the buy-and-hold returns of the long and short positions. The dashed lines represent a two-standard-error confidence interval. Standard errors are Newey-West adjusted for serial correlation. As in Figure 1, I present results for equally weighted producer portfolios in Panel A and for value-weighted producer portfolios in Panel B. The figure shows that the producer stock market underreacts to news causing jumps in the stock markets of its major customers. The cumulative abnormal return on the long-short portfolio over the twenty four months after the sorting month is 3.7% in Panel A and 1.8% in Panel B. The cumulative abnormal return on the customer momentum strategy remains statistically reliable for nine months in Panel A and for only three months in Panel B. The higher average cumulative return and the more persistent drift for the equally weighted portfolios are not surprising since the diffusion of information is likely to be slower for smaller stock markets. To summarize, Table 1, as well as Figures 1 and 2, supports the hypothesis that information about major customers gradually diffuses into the stock market of the producer country, generating a price drift and a profitable customer momentum strategy. 4.2 Supplier Momentum In Table 2 I examine whether the stock market returns of major suppliers also help predict the stock market return of a producer country. The table is organized in the same way as Table 1. The results for supplier momentum are almost identical to those for customer momentum, suggesting that there are lead-lag effects on both sides of the trade chain. For example, using equal weights for producers and equal weights for suppliers, the multifactor alpha for supplier momentum is 0.81% per month (with a t-statistic of 3.11), or 9.7% per year. Similarly, the supplier momentum strategy using value weights for producers and equal weights for suppliers yields a multifactor monthly alpha of 0.83% (with a t-statistic of 3.26) for the overall period. Hence, the profitability of the supplier momentum strategy is not concentrated in small less liquid stock markets. 10

11 As in Table 1, I also present results for the two halves of the sample period. The average abnormal returns on the long-short portfolio tend to be a bit lower and less reliable for the second half than for the first half. The magnitudes of the effects, however, are still large and all multifactor alphas are significant at the 10% level. Overall, the subsample analysis shows that the supplier momentum strategy generates substantial average abnormal returns in both halves of the sample period. Note that in all but one specification reported in Table 2, alphas get larger across the producer portfolios as the lagged supplier returns go from low to high. Hence, the observed stock market reaction of a producer country is related to the size of the initial supplier stock market reaction (as it is related to the size of the initial customer stock market reaction). The similarities between the customer and supplier momentum results continue. Supplier momentum returns also cannot be explained by exposures to the global risk factors. The loadings on MKT-RF, SMB and MOMFX are not statistically different at the 5% level in any of the specifications for the overall period. The coefficient on HML is statistically reliable in only one specification and the coefficient on MOM in only two. 4 The fact that the customer and supplier momentum average returns cannot be explained by sensitivities to MKT-RF, SMB, HML, MOM and MOMFX implies that adding such strategies to an opportunity set consisting of the five factors raises the maximum ex-post Sharpe ratio. Indeed, Appendix B shows that customer and supplier momentum strategies enhance the mean-variance opportunity set. Figure 3 also offers evidence that supplier momentum returns do not reward investors for taking on global market risk. The graph shows the annual returns on the supplier momentum strategy along with the annual returns on the MKT-RF factor. As with customer momentum, supplier momentum does not tend to underperform sharply in bad states of the world. For example, in 2008 the equally weighted and valueweighted supplier momentum strategies delivered returns of -10.1% and 6.6%, respectively. The monthly correlation between the supplier momentum and the global market return series ranges from to 0.02 across the different strategy specifications. I investigate further the behavior of the supplier momentum strategy in Figure 4. It plots the average cumulative return for months t + 1 to t + 24 on the long-short portfolio based on supplier returns for month t. Figure 4 shows that the producer stock market underreacts to news causing jumps in the stock markets of its major suppliers. The average cumulative return on the long-short portfolio over the twenty four months is 2.8% for the equally weighted strategy in Panel A and 3.9% for the value-weighted strategy in Panel B. Even though the average cumulative return for the value-weighted supplier momentum strategy tends to be higher, it remains statistically reliable for only three months. In contrast, the average cumulative return 4 I also test whether the customer and supplier momentum returns are correlated with carry trade returns. Using the annual excess returns of currency portfolios constructed by Lustig and Verdelhan (2007), I form two long-short carry trade strategies. The pairwise correlations between each of these strategies and each of the customer and supplier momentum strategies presented here are all below 0.33 and do not differ statistically from zero at the 5% significance level. 11

12 on the equally weighted strategy is statistically different from zero for ten months. Hence, the drift is once again much more persistent for the smaller stock markets. Overall, the results for customer and supplier momentum lend support to the hypothesis that news about major customers and suppliers gradually diffuses into the stock market of the producer, causing cross-predictability of returns along the trade chain. 4.3 Combining the Customer and Supplier Momentum Given the evidence that both customer and supplier momentum generate large average abnormal returns, it is worth examining the benefits of a momentum strategy based on the lagged returns of both major customers and major suppliers. 5 To do that, I implement two alternative trade momentum strategies. The first one is constructed as follows. At the beginning of each month from July 1981 to March 2009, I sort producer countries whose exports or imports accounts for at least 20% of GDP into three groups based on the equally weighted average of the prior-month local currency returns of their major trading partners. Major trading partners are defined as countries that account for at least 5% of exports (imports) when exports (imports) represent at least 20% of GDP. The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the middle 40% to the Med 40 portfolio. Portfolios are held for a month and then rebalanced. The combined trade momentum strategy goes long the Top 30 portfolio and sells short the Bot 30 portfolio. The second trading strategy is constructed as follows. At the beginning of each month from July 1981 to March 2009, I sort producer countries whose total trade (exports and imports) accounts for at least 40% of GDP into three groups based on the equally weighted average of the prior-month local currency returns of their major trading partners (countries accounting for at least 5% of the producers total trade). The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the middle 40% to the Med 40 portfolio. Portfolios are held for a month and then rebalanced. The total trade momentum strategy goes long the Top 30 portfolio and sells short the Bot 30 portfolio. Table 3 shows the average monthly U.S. dollar returns for portfolios of producer countries sorted on 5 A special feature of the country-level trade-link dataset is that for many countries the major customers are also their major suppliers. Indeed, about two thirds of the analyzed producer-customer links and producer-supplier links overlap. Hence, one can argue that the observed customer and supplier momentum are mostly driven by a common component and that including them together in a global investment strategy is redundant. The customer and supplier momentum portfolios, however, are far from identical. The correlations between the two long-short portfolios range from 0.54 for equal producer-country weights and equal partner-country weights to 0.64 for value producer-country weights and sales partner-country weights. Moreover, even though about two thirds of the analyzed producer-customer links and producer-supplier links overlap, two-thirds of all major suppliers are net suppliers, whereas two-thirds of all major customers are net customers. Thus, the supplier dataset is dominated by net suppliers, whereas the customer dataset is dominated by net customers. Hence, constructing a strategy based on both customer and supplier past returns is unlikely to be redundant. A more detailed investigation of the common component in the customer and supplier effects, presented in Section 6 of the paper, also supports the importance of both effects. 12

13 the prior-month return of the equally weighted portfolio of their major customers and suppliers. Panel A reports the results for the combined trade momentum strategies, and Panel B reports the results for the total trade momentum strategies. All long-short momentum portfolios earn large positive abnormal returns. For example, in Panel A the multifactor alpha is 0.98% per month (with a t-statistic of 3.69) for the equally weighted strategy, 0.84% per month (with a t-statistic of 2.86) for the GDP-weighted strategy, and 0.73% per month (with a t-statistic of 2.67) for the value-weighted strategy. The results in Panel B are similar. The multifactor alphas for the total trade momentum strategies range from 0.56% per month (with a t-statistic of 1.98) for the value-weighted specification to 0.91% per month (with a t-statistic of 3.12) for the GDPweighted specification. Using sales weights rather than equal weights for the major trading partners yields similar (untabulated) results. In summary, Table 3 shows that combining information about the lagged stock market returns of a country s major customers and suppliers generates large subsequent abnormal returns on a trade momentum portfolio. The slow diffusion of information across stock markets of trading partners yields predictability in returns that cannot be explained by global asset pricing models. 4.4 Robustness All of the main empirical findings of the paper are robust to various alternative portfolio specifications which are reported in Appendix C. While the results presented above are robust and consistent with the hypothesis that the slow diffusion of news about trading partners generates predictability in country returns, there are two immediate alternative explanations of the data. First, the observed predictability could simply be due to systematic differences in the averages returns and/or alphas of the long and short ends of the customer and supplier momentum strategies. Second, the observed predictability could be due to own-country momentum as opposed to cross-country momentum. To address the first alternative explanation, I examine the turnover of the customer and supplier momentum strategies. The average monthly turnover of the Bot 30 and Top 30 portfolios is about 65%. Hence, the composition of the long and short ends of the customer and supplier strategies changes frequently. I also plot in Figure 5 the fraction of months that each sample country spends in the Bot 30, Med 40 and Top 30 portfolios (based on equal partner weights). The distributions for the customer momentum portfolios (Panel A) as well as for the supplier momentum portfolios (Panel B) show that most countries spend a similar number of months in the long and short ends of the strategies. As a final test, I also rerun all strategies replacing raw country returns with country returns in excess of their averages for the months when the countries are included in the portfolios. The results, available on request, indicate that this change has no effect on the 13

14 average portfolio returns. To summarize, systematic differences in average returns and/or alphas do not appear to drive the results. To address the second alternative explanation (that the observed return predictability is due to owncountry momentum) I use pooled cross-sectional time-series regressions of country stock market returns. The dependent variable is the local currency stock market return of the producer country for month t. The key explanatory variables are the equally weighted average of the local currency returns of the country s major customers or suppliers for month t - 1 and for the eleven-month period from month t - 2 to t I add the local currency returns on the producer country for month t - 1 and for months t - 2 to t - 12 in order to control for possible own momentum effects. I also include (but do not report) the log of the market capitalization of the producer country for December-end of the year of the trade links as an additional control. To account for the time effect in the data, I cluster standard errors by month in all regressions and use month fixed effects in some of the specifications. 6 Table 4 reports regression results for all sample countries in Panel A and for developed countries in Panel B. Each panel reports results for the customer and supplier momentum effects. I first analyze the results for customer momentum. Column 1 of Panel A confirms the finding that the lagged one-month customer return is positively correlated to the current producer return. Column 2 adds to the model the lagged cumulative return for the customer portfolio. The results suggest that cross-predictability of country stock returns is driven mainly by the previous-month performance of major customers. Adding the lagged returns for the producer country has a negligible effect on the coefficient for the lagged one-month customer return. Therefore, own momentum is not driving customer momentum. Past customer country returns do forecast subsequent producer country returns. The effect is both large and statistically reliable. Columns 6-10 of Panel A yield the same conclusions for supplier momentum. The coefficient on the prior-month supplier return remains strong and statistically different from zero when the producer own lagged returns are added. Hence, supplier momentum cannot be explained by own momentum, either. In Panel B I repeat the analysis using only producer countries in developed markets to ensure that the observed predictability is not driven by emerging markets which tend to be smaller and less liquid. When I restrict my sample to developed countries the results remain strong and statistically reliable. Both customer momentum and supplier momentum are present in developed markets. 6 I do not use the Fama-MacBeth (1978) regression approach here because the monthly cross-section of producer countries consists only of 11 to 12 countries between July 1981 and December 1988, of less than 20 countries until July of 1993 and of less than 30 countries until July of Since the Fama-MacBeth approach would require monthly cross-sectional regressions, I would have too few degrees of freedom to obtain any power. As Petersen (2008) shows, the use of standard errors clustered by time also produces unbiased standard errors when there is a sufficient number of clusters. Since I have 333 months, the use of clustered standard errors should correct for the time effect in the data. 14

15 5 Gradual Diffusion of Information Hypothesis This paper is motivated by the literature on gradual diffusion of information. This literature is based on the hypothesis that investor attention constraints, as well as investor specialization, lead to the slow diffusion of information and generate return predictability. How plausible is this hypothesis for the transmission of shocks across the stock markets of trading partners? It is hard to argue that country returns are predictable because of investor attention constraints. After all, stock markets are easy to follow and there are only so many of them. In addition, the customer and supplier links between trading partners are long-lasting and easy to obtain. It is not hard to argue, however, that country returns are predictable because of investor specialization. First, most institutional investors do not participate in all stock markets around the world, and those that do, tend to participate through value-weighted international portfolios. Second, it is well-known that individual investors exhibit home bias in their portfolios and tend to participate actively in a limited number of international stock markets. 7 Hence, the hypothesis that information diffuses slowly across the stock markets of trading partners due to investor specialization appears plausible. In this section I test a few implications of the gradual-information-diffusion hypothesis for the customer and supplier momentum. If investor specialization is the driving force behind customer and supplier momentum, then predictability should be weaker for countries that are widely followed in financial markets. Countries with the largest stock markets and/or largest gross domestic products attract a lot of investor attention and capital flows. Hence, more market participants are likely to collect information on these countries major trading partners and, as a result, customer and supplier momentum should be mitigated for these countries. Similarly, the effects should be weaker for countries whose international trade is followed frequently in the news. In Table 5 I test these predictions explicitly, controlling for other possible determinants of producer stock market returns. The regressions in Panel A of Table 5 are similar to those in Panel A of Table 4. The only difference is that I interact prior-month average customer or supplier return with the indicator HIGHMCAP in Columns 1 and 4, with the indicator HIGHGDP in Columns 2 and 5 and with the indicator HIGH- TRADEART in Columns 3 and 6. HIGHMCAP is a dummy variable equal to 1 if the producer country s market capitalization is greater than the 80th percentile of all sample countries for the year of trade links. 8 Similarly, HIGHGDP is a dummy variable equal to 1 if the producer country s GDP is greater than the 80th percentile of all sample countries for the year of trade links. Finally, HIGHTRADEART is an indicator equal to 1 if the number of trade-related articles in the Wall Street Journal about a producer country exceeds the 7 For evidence on home bias in portfolio allocation see, for example, French and Poterba (1991), Lewis (1999), Faruqee, Li and Yan (2004) and Heathcote and Perri (2008). 8 Using the 85th or 90th percentile as a threshold yields similar results. 15

16 80th percentile of all sample countries for the year of trade links. 9 I cluster standard errors by month and include month fixed effects in all regressions. Consistent with the prediction that larger countries should exhibit lower customer and supplier momentum, I find negative and statistically reliable coefficients on the interactions with HIGHMCAP and HIGHDP. For example, the loading on the interaction with HIGHMCAP is (with a t-statistic of -2.53) for customer momentum and (with a t-statistic of -2.31) for supplier momentum. The estimated slope coefficients suggest that the customer and supplier momentum effects for larger countries are only about half the size of the effects for smaller countries. I also find support for the prediction that customer and supplier momentum should be weaker for countries whose international trade is often followed in the news. The loading on the interaction with HIGHTRADEART is (with a t-statistic of -2.20) for customer momentum and (with a t-statistic of -2.50) for supplier momentum. Overall, Panel A of Table 5 suggests that information diffuses faster for countries that are widely followed in financial markets. One could argue that the evidence here is also consistent with liquidity being the driving force behind customer and supplier momentum since smaller countries tend to have less liquid stock markets and less liquid stock markets are likely to react more slowly to news about trading partners. The analysis in the previous sections, however, does not support this argument. First, the long-short momentum strategies based on value-weighted country portfolios earn large and statistically reliable abnormal returns. Second, the customer and supplier effects are present in developed markets, which tend to be more liquid. Third, as Table A2 shows, skipping a month does not eliminate the profitability of the strategies based on equally weighted country portfolios. Fourth, Figures 2 and 4 show that both customer and supplier momentum effects persist for several months. As an explicit test of the liquidity hypothesis, in Panel B of Table 5 I interact prior-month average customer and supplier return with the indicator variable HIGHTURN. HIGHTURN is a dummy variable equal to 1 if the producer country s stock market turnover is greater than the 80th percentile of all sample countries for the year of trade links. Stock market turnover for a given year is defined as the total value of stocks traded during the year divided by the average of the total market capitalization for the beginning and end of the year. The data on share trading are from WDI but they are available only from 1988 onward. Hence, the regressions in Panel B are for the period from July of 1989 to March of For comparison, I also rerun the regressions from Panel A using the same period. The estimated coefficients on the interaction with HIGHTURN are (with a t-statistic of -1.01) for customer momentum and (with a t-statistic of -0.87) for supplier momentum. The slope estimates are not statistically different from 0 even at the 10% 9 The number of trade-related Wall Street Journal articles for each country in each year is from ProQuest. I define traderelated articles as articles that mention the words trade, trades, export, exports, import or imports. 16

17 significance level. Thus, we cannot reject the null that predictability is the same for stock markets with high and low turnover. In contrast, the coefficients on the interactions with HIGHMCAP, HIGHGDP and HIGHTRADEART are large and statistically robust for the same period. Hence, the tests in Panel B suggest that liquidity is not the driving force behind the observed cross-country predictability. Liquidity, of course, does not capture the whole spectrum of market frictions or limits to arbitrage that could explain the underreaction of stock markets to information about trading partners. Market frictions such as capital controls, transaction costs and nonsynchronous trading (as mentioned earlier) could delay the reaction of stock markets to information about trading partners and generate cross-country predictability. In addition, capital controls and transaction costs are more likely to be binding for smaller countries. To address this whole spectrum of alternative explanations, I implement the customer and supplier momentum strategies with U.S.-based single-country exchange-traded funds (ETFs). Since these instruments are traded throughout the day on the U.S. stock exchanges, they give investors easy and immediate access to foreign markets. 10 Appendix Table A4 lists the single-country ETFs included in this analysis. Most of the ETFs are based on the same MSCI country indices that I use to measure stock market returns. The only exceptions are the ETFs for Russia and Indonesia due to the fact that MSCI-based ETFs for these countries are not available for the sample period. I obtain monthly returns (net of expenses) for all ETFs from the CRSP Mutual Fund database. The earliest month with ETF return data is April Since ETFs are traded as stocks, a realistic implementation of an ETF-based trading strategy needs to take into account the bid-ask spreads and the commissions that an investor would have paid trading those instruments. I have not been able to find historical monthly data on bid-ask spreads by ETF, so I use one-month trailing average bid-ask ratios as of July 21, 2010 from XTF ETF Experts. 11 The bid-ask spreads are presented in the last column of Table A4. As for commissions, I assume a commission of 4 basis points per roundtrip transaction which is equivalent to $40 for a transaction size of $100,000. Even though the customer and supplier momentum strategies involve short selling, I do not account for short selling costs because these are unlikely to exceed 1 basis point per month. The results of the ETF-based implementations of the customer and supplier momentum strategies are reported in Table 6. Panel A presents the results for the customer momentum specifications. Panel B presents the results for the supplier momentum specifications. For the period from April 1996 to March 2009, the average returns on the implemented strategies (net of trading costs) are large and statistically indistinguishable from those on the benchmark strategies. For instance, the long short customer strategy that uses equal weights for customers and equal weights for producers yields an average return of 0.72% per 10 Bekaert (1995) shows that the existence of country funds could effectively integrate markets despite the presence of severe restrictions on foreign direct equity investments. 11 The bid-ask spreads were obtained from 17

18 month (with a t-statistic of 1.39). The same strategy applied to suppliers earns on average 1.33% per month (with a t-statistic of 2.78). Similarly to the average long-short returns, the alphas from the global asset pricing models are all positive and large. About half of the abnormal returns are not statistically reliable at the 10% significance level but this is most likely due to the short sample period (only 156 months). Overall, the large abnormal returns on the ETF-based trading strategies show that customer and supplier momentum can generate substantial profits even when implemented through ETFs traded in the United States. These findings are inconsistent with the market frictions hypothesis, but are consistent with the gradual-information-diffusion hypothesis. 6 Economic Channels behind the Customer and Supplier Effects 6.1 Customer Effect So far this study has shown that customer and supplier returns have a large, positive and statistically reliable effect on the subsequent monthly return of a producer country. Tests in the previous section suggest that the cross-country predictability along the trade chain is due to the gradual diffusion of information. As Hong et al (2007) point out, under the hypothesis of gradual information diffusion, predictability of returns across assets should be based on predictability of fundamentals. This means that the ability of a country s major trading partners to forecast its stock market should be based on their ability to forecast its fundamentals. Hence, it is necessary to identify the economic channels behind the customer and supplier effects. I now turn to exploring those channels. An obvious explanation for the customer effect is that a positive shock to the stock market of a major customer implies that the purchasing power of equity holders in the customer country has gone up and that purchases of imported goods are likely to rise due to the wealth effect. In addition, the rise of the stock market is likely to lower the cost of capital for domestic firms and stimulate them to make more investments some of which would involve purchasing imported goods. The expectation of higher sales to the customer country pushes the producer country s stock market up and yields a correlation between the stock market returns of the two trading partners. The logic for negative shocks is analogous. This hypothesis offers a few testable predictions. First, a country s stock market movements should predict changes in consumption growth and investment growth in that country. I test this prediction by examining the effect of annual stock market returns on the subsequent annual real growth rate of household final consumption expenditure, gross capital formation and gross fixed capital formation. The annual real growth rates of household consumption and investment are 18

19 from WDI. Gross fixed capital formation consists of outlays on additions to domestic fixed assets. Gross capital formation also includes changes in the level of inventories. Table 7 documents the results from panel regressions of a country s annual real growth rate of consumption and investment on its lagged annual stock market return. 12 All regressions include country fixed effects. Standard errors are clustered by year. The regressions show that there is a positive and statistically reliable relation between the annual stock market return and the subsequent growth rates of domestic consumption and investment. For instance, the slope on the lagged annual stock market return in Column 1 is 0.01 (with a t-statistic of 2.68). This implies that one standard deviation increase in the local annual stock market return leads to a one-quarter standard deviation increase in the annual real growth rate of household consumption. The economic significance of the effect on gross capital formation growth is similar. A one standard deviation increase in the local annual stock market return forecasts about one-fifth standard deviation increase in the annual real growth rate of investment. Even when I control for the lagged values of the dependent variables, the coefficients on the lagged stock market return remain robust. Stock market movements help predict changes in consumption and investment growth. 13 Hence, Table 7 supports the first prediction of the customer momentum hypothesis. The second testable prediction of that hypothesis is that a positive shock to a major customer s stock market is likely to generate a rise in exports to this customer in the subsequent months and that a negative shock is likely to generate a fall in exports. To test this implication I examine the predictive power of customer stock market returns for changes in exports to that customer over the next few months (controlling for the stock market return of the producer). Table 8 presents the results from this direct test of the mechanism behind the customer effect. The table reports coefficients from regressions of the change in a country s exports to a major customer on the lagged returns of that country and its customer. In Panel A the dependent variable is the change in three-month exports to a major customer country (the ratio of exports for months t to t + 2 and exports for months t - 1 to t - 3 minus one). In Panel B the dependent variable is the change in six-month exports (the ratio between exports for months t to t + 5 and exports for months t - 1 to t - 6 minus one). In Panel C the dependent variable is the change in twelve-month exports (the ratio between exports for months t to t + 11 and exports for months t - 1 to t - 12 minus one). Exports are measured in the local currency of the producer country. The explanatory variables are the local currency returns of the producer country and of its major customer for month t - 1 and for months t - 2 to t I also include (but do not report) the log of the producer country s GDP for the year of the producer-customer link (in millions of current U.S. dollars). All export ratios are winsorized at the 1% and 12 The lagged annual stock market returns are winsorized at the 99% level in the table to avoid giving excess weight to eight outliers due to Brazil, Argentina and Poland. 13 These results expand earlier international evidence on the links between stock market movements and consumption growth. For example, Campbell (1996) finds a positive contemporaneous correlation between annual stock market returns and consumption growth for a sample of 12 developed countries. 19

20 99% level, while the cumulative lagged customer returns are winsorized at the 99% level. The results are not sensitive to winsorizing at other levels or logging. The estimates in columns 1-5 are from pooled time-series cross-sectional regressions. These five regressions also contain month-of-the-year indicators to adjust for seasonality in exports. Standard errors in columns 1-5 are clustered by trade pair. The estimates in columns 6-8 are time-series averages of the coefficients from month-by-month Fama-MacBeth (1978) regressions. The Fama-MacBeth standard errors are Newey-West adjusted for serial correlation. The results in Table 8 show that the stock market return of a major customer does help forecast changes in exports to that customer in the near future. The coefficients on the lagged customer returns are positive in all panels and statistically reliable for the cumulative customer return. Therefore, good news about a major customer results in more exports to that customer over the following three to twelve months. The effects are not only statistically but also economically significant. Looking at Column 5 of Panel C, a one-standard deviation increase in the cumulative lagged customer return leads to 2 percentage points or 0.15 standard deviations increase in the growth of twelve-month exports. Given that the average growth of twelve-month exports is 9%, this is an important effect. Hence, Table 8 provides strong and direct evidence in support of the hypothesis that a shock to the stock market of a major customer country generates a shock to the stock market of the producer country because of expected changes in future exports to that customer. Since the positive effect of a customer stock market on a producer stock market is based on expected changes in exports to that customer, variation in the expected changes is likely to cause variation in the strength of this effect. A positive stock market return in the customer country is more likely to raise the purchases of final goods than of intermediate goods because domestic firms might buy some intermediate goods to produce output sold to other countries. 14 That is, demand for inputs could depend not only on the local stock market but also on the stock markets of other countries down the trade chain. This suggests a further test of the customer momentum hypothesis. Customer momentum should be stronger for producer countries that export more final goods. I present tests of this implication in Table 9. For every producer country I obtain exports to the world by commodity for 1980 to 2007 from the United Nations Commodity Trade Statistics Database. I classify commodities as either final goods or intermediate goods based on whether these goods are used mainly as an input for further production. Appendix Table A5 presents the classification. For each sample country and each available year I compute final exported goods as a fraction of total exported goods. Every year I sort sample countries based on their final goods fraction. Countries above the 80th percentile for a given year are assigned to the HIGHFINAL group and countries below the 80th percentile are assigned to the 14 For example, a UBS study cited by Wall Street Journal on December 29, 2009 estimates that 45% of Japan s exports to China are processed and shipped to other countries. 20

21 LOWFINAL group. I then regress monthly producer country stock market returns on the lagged stock market returns of the producer country and the equally weighted portfolio of its major customers separately for the two groups. The regression results are reported in the first three columns of Panel A. HIGHFINAL countries are much more affected by the prior month return of their customers than LOWFINAL countries. The difference in the effect for the two groups is positive and statistically reliable. Does the difference in stock market reactions reflect a difference in export reactions? The last three columns of Panel A say yes. A positive stock market return of a customer country forecasts a larger growth in exports to that country when the producer country sells more final goods. The coefficient on the interaction between the lagged one-month customer return and HIGHFINAL is positive, large and statistically reliable in all three columns. The interactions with the lagged cumulative customer return are also positive but unreliable indicating that most of the differential effect comes from the reaction of exports to the prior-month stock market return of the customer. Since high and low final-good exporters could differ systematically in other characteristics related to the customer effect (such as market size and gross domestic product), I repeat all tests replacing the raw fraction of final-good exports to total exports with the residual from a pooled regression of the raw fraction on the logs of the producer country s market capitalization and gross domestic product. The results, presented in Panel B of Table 9, show that the stronger reactions of high final-good exporters to customer shocks are not driven by systematic differences in country size. To summarize, Tables 7, 8 and 9 provide consistent support for the hypothesis that the stock market of a major customer affects the stock market of a producer country through expected changes in exports to that customer. 6.2 Supplier Effect I now turn to examining the economic channel behind the supplier effect. To explain the positive reaction of producer stock markets to positive shocks in supplier stock markets, I offer the following hypothesis. A positive technology or macroeconomic shock to the stock market of a major supplier is likely to enable firms in the supplier country to produce goods more cheaply and efficiently. The positive shock is likely to spill over to the firms in the buying country through lower import prices. Similarly, a negative shock to the firms in the supplier country is likely to affect negatively the firms in the buying country by forcing them to either accept higher import prices or search for alternative suppliers. This hypothesis offers at least two testable predictions. First, the fundamentals of supplier and producer countries should move together. I test this prediction in Table 10. Panel A contains correlations between a country s annual GDP growth for year t, the average 21

22 annual GDP growth of its major suppliers for year t, the country s stock market return for year t - 1 and the average stock market return of its major suppliers for t - 1. Annual stock market returns are measured in local currency and are winsorized at the 99% level to avoid giving excess weight to outliers. I identify major suppliers based on trade data for year t - 1. GDP growth data are from WDI and are based on GDP in constant 2000 U.S. dollars. Panel B contains coefficients and t-statistics from pooled time-series cross-sectional regressions of a country s GDP growth on its own lagged GDP growth, on the contemporaneous average GDP growth of its major suppliers and on the contemporaneous world GDP growth. Panel C contains coefficients and t-statistics from pooled time-series cross-sectional regressions of a country s GDP growth on its own lagged annual stock market return and the lagged average stock market return of its major suppliers. All regressions include country fixed effects. Standard errors are clustered by year. Panel A shows that the GDP growth of a producer country is positively correlated with the average GDP growth of its suppliers for the same year. It is also positively correlated with the lagged average annual return of the supplier countries. Note that annual stock market returns seem more correlated with next-year GDP growth than with current-year GDP growth. Panel B provides further evidence that the average supplier GDP growth has a positive contemporaneous effect on the producer GDP growth, even after controlling for the contemporaneous world GDP growth and for the lagged GDP growth of the producer country. Panel C documents that lagged supplier returns have a positive effect on the current GDP growth of the producer country, even when I control for the lagged producer return. The empirical evidence in Table 10 is consistent with the hypothesis that the supplier effect occurs because a positive macroeconomic or technological shock to the supplier country benefits firms not only in the supplier country but also in the producer country. This hypothesis also suggests that if the main channel of spillover is lower import prices, one should observe more imports of goods from the supplying country. Hence, another testable implication of this hypothesis is that supplier stock market returns should forecast changes in imports from that supplier. Table 11 presents tests of this implication. The regression analysis here is similar to that in Table 8. There are only two differences. First, the dependent variables are changes in imports from a major supplier instead of changes in exports to a major customer. Second, the key explanatory variables are lagged supplier returns instead of lagged customer returns. The results in Table 11 support the second implication of the supplier hypothesis. A supplier s stock market return helps forecast changes in imports from that supplier over the next 3 to 12 months. The coefficients on the prior-month return of the major supplier are always positive and statistically reliable in Panels B and C. For example, the slope on the prior-month supplier return is 0.04 (with a t-statistic of 2.57) in Column 5 of Panel B and 0.05 (with a t-statistic of 2.35) in Column 5 of Panel C. The estimates from the Fama-MacBeth regressions are even larger and more reliable. 22

23 Therefore, a positive shock to the stock market of a major supplier predicts higher imports from that supplier over the next twelve months. It is often the case that a country s major suppliers are also its major customers. Therefore, an alternative hypothesis for the observed supplier effect is that it is driven by the customer effect. That is, a positive stock market shock to a major trading partner signals higher expected sales to that partner and pushes the producer country s stock market up but it also prompts more imports of production inputs from that trading partner. Hence, this hypothesis also implies the predictive power of supplier stock market returns for producer stock market returns and imports. To disentangle the two hypotheses, I test whether the supplier effect in returns and imports is driven by major suppliers that are also major customers. The tests are presented in Table 12. Panel A of Table 12 reports results from pooled pair-level regressions of monthly stock market returns of a producer country on the individual lagged stock market returns of a major supplier and on its own lagged returns. 15 I cluster standard errors by month and by producer country and include month fixed effects in all regressions. Column 1 confirms the basic result about supplier momentum. The new result is in Column 2. In this specification I interact prior-month supplier return with a dummy variable, CUSTOMER, equal to one if the major supplier is also a major customer for the year of the producer-supplier link. The coefficient on the interaction measures the effect of being also a major customer on the ability of supplier stock market returns to predict subsequent producer returns. The results in Column 2 of Panel A imply that the supplier momentum effect is not primarily driven by suppliers that are also major customers. The loading on the interaction is indistinguishable from zero. For comparison purposes, Panel B reports the same test for the customer momentum effect. In contrast to Panel A, Panel B shows that the returns of pure major customers have a stronger effect on the future producer return than the returns of major customer-suppliers. This result is due to the fact that pure major customers tend to be much smaller (both in terms of market size and GDP) than major customersuppliers. The same is not true for pure major suppliers. They are similar in size to suppliers that appear on both sides of the trade chain. Hence, the results in Panel B provide additional support to the gradualinformation-diffusion hypothesis by showing that information about smaller partners diffuses more slowly in the producer stock market. In order to disentangle the two hypotheses about the supplier effect, I also examine whether the forecasting power of supplier returns for future imports is due only to major suppliers that are also major customers. The tests are reported in Panel C of Table 12. It shows results from pooled time-series crosssectional regressions of changes in imports from a given supplier on the lagged returns of the supplier and producer country. The regressions are specified similarly to those in Column 5 of Table 11. In Column 1 15 The lagged cumulative partner returns are winsorized at the 99% level to avoid giving excess weight to outliers. 23

24 the dependent variable is the change in 3-month imports, in Column 2 - the change in 6-month imports and in Column 3 - the change in 12-month imports. In all columns the key parameters are the coefficients on the interactions between lagged supplier returns and the dummy variable CUSTOMER, defined above. Note that the interaction between the lagged one-month supplier return and CUSTOMER is statistically unreliable in all three columns. Similarly, the interaction between the lagged cumulative supplier return and CUSTOMER is statistically indistinguishable from zero for 3-month and 6-month imports. Hence, Panel C says that the effect of supplier returns on imports is not driven only by two-way major trading partners. For comparison, Panel D presents the same tests for the effect of customer returns on future exports. The results show that being also a major supplier has little effect on the ability of customer returns to forecast subsequent changes in exports. To summarize, Table 12 offers evidence that the supplier effect is not driven by the customer effect. This lends support to the conclusion that the supplier effect is most likely due to technology or macroeconomic shocks in the supplier country that also affect firms in the producer country. 7 Conclusion This paper extends the evidence on return predictability along the trade chain by showing that the monthly stock market returns of a country s major customers or suppliers forecast the stock market return of that country for the subsequent month. Customer and supplier momentum strategies based on this cross-country return predictability yield average returns of 8% to 14% per year. Adjusting returns for global systematic factors leaves them largely unaffected both in magnitude and statistical reliability. 16 Since the customer and supplier links at the country level are long lasting and publicly available, these results lend support to the hypothesis that the observed predictability in returns is due to slow diffusion of information along the trade chain. Consistent with this hypothesis, I find that the return predictability is weaker for countries that attract a lot of investor attention and capital flows. I also show that implementing the customer and supplier strategies with U.S.-based ETFs yields large profits. These results suggest that the cross-country predictability cannot be attributed to market frictions and provide strong support for the gradual-information-diffusion hypothesis. The paper also explores the economic channels behind the correlation in stock market returns of trading partners. I present the following hypothesis for the economic channel behind the customer effect. A positive shock to the stock market of a major customer implies that the purchasing power of equity holders in the 16 It is possible that the abnormal returns documented here would disappear under an alternative asset-pricing model. Any test of market efficiency is a test of the joint hypothesis that markets are efficient and that the employed asset-pricing model is the right one. As a result, one can never reject market efficiency. 24

25 customer country has gone up and that purchases of imported goods are likely to rise. The expectation of higher sales to the customer country pushes the producer country s stock market up and yields a correlation between the stock market returns of the two trading partners. This hypothesis implies that a positive shock to a major customer s stock market is likely to generate a rise in exports to this customer in the subsequent months and that a negative shock is likely to generate a fall in exports. I find that customer stock market returns indeed help forecast changes in exports to that customer over the subsequent months. This is direct evidence in support of the hypothesis that a shock to the stock market of a major customer country generates a shock to the stock market of the producer country because of expected changes in future exports to that customer. As for the supplier effect, I offer and test the following hypothesis. A positive technology or macroeconomic shock to the stock market of a major supplier is likely to enable firms in the supplier country to produce goods more cheaply and efficiently. As a result, the positive shock is likely to spill over to the firms in the buying country through lower import prices. Similarly, a negative macroeconomic shock to the firms in the supplier country is likely to affect negatively the firms in the buying country by forcing them to either accept higher import prices or search for alternative suppliers. Thus, a positive shock to the stock market of the supplying country is likely to stimulate higher imports of goods and a negative shock is likely to stimulate lower imports. Consistent with this hypothesis, I find that a supplier s stock market return does help forecast changes in imports from that supplier over the next three to twelve months. A possible alternative explanation for the supplier effect is that it is driven by the customer effect because a country s major suppliers are often also its major customers. A positive stock market shock to a major trading partner signals higher future sales to that partner and pushes the producer country s stock market up but it also prompts more purchases of production inputs from that trading partner. Hence, this hypothesis also implies the predictive power of supplier stock market returns for producer stock market returns and imports. To disentangle the two hypotheses I test whether the supplier effects in returns and imports are mostly due to major suppliers that are also major customers. I find that the supplier effect is not driven by the customer effect. The country-level customer and supplier momentum documented in this study offer many directions for future research. One could further test the gradual-information-diffusion hypothesis by examining whether there is return predictability across currency markets of trading partners. Preliminary results reported in Appendix E suggest that information diffuses slowly not only across stock markets but also across currency markets of economically linked countries. Similarly, one could explore the lead-lag effect in bond market returns along the trade chain. Finally, cross-predictability across different asset markets of trading partners could be examined as well. 25

26 References Asness, Clifford S., John Liew, and Ross Stevens, 1997, Parallels between the cross-sectional predictability of stock and country returns, The Journal of Portfolio Management 23, Asness, Clifford S., Tobias J. Moskowitz, and Lasse H. Pedersen, 2009, Value and momentum everywhere, Unpublished working paper, AQR Capital Management, University of Chicago and New York University. Bekaert, Geert, 1995, Market Integration and investment barriers in emerging equity markets, World Bank Economic Review 9(1), Bhojraj, Sanjeev, and Bhaskaran Swaminathan, 2006, Macromomentum: returns predictability in international equity indices, The Journal of Business 79, Campbell, John Y., 1996, Consumption and the stock market: interpreting international evidence, NBER Working paper Cohen, Lauren, and Andrea Frazzini, 2008, Economic links and predictable returns, Journal of Finance 63, Fama, Eugene F., and James D. MacBeth, 1973, Risk, return and equilibrium: empirical tests, Journal of Political Economy 81, Fama, Eugene F., and Kenneth R. French, Value versus growth: the international evidence, Journal of Finance 53, Faruqee, Hamid, Shujing Li, and Isabel K. Yan, 2004, The determinants of international portfolio holdings and home bias, IMF Working Paper. French, Kenneth R., and James Poterba, 1991, Investor diversification and international equity markets, American Economic Review 81, Gibbons, Michael R., Stephen A. Ross, and Jay Shanken, 1989, A test of the efficiency of a given portfolio, 26

27 Econometrica 57 (5), Heathcote, Jonathan, and Fabrizio Perri, 2009, The international diversification puzzle is not as bad as you think, Working paper, Federal Reserve Bank of Minneapolis. Hirshleifer, David, Sonya Lim, and Siew Hong Teoh, 2007, Driven to distraction: extraneous events and underreaction to earnings news, Journal of Finance, forthcoming. Hong, Harrison and Jeremy Stein, 1999, A unified theory of underreaction, momentum trading, and overreaction in asset markets, Journal of Finance 54, Hong, Harrison, Walter Torous, and Rossen Valkanov, 2007, Do industries lead the stock market? Journal of Financial Economics 83, Lewis, Karen K., 1999, Trying to explain home bias in equities and consumption, Journal of Economic Literature 37, Lustig, Hanno, and Adrien Verdelhan, 2007, The cross-section of foreign currency risk premia and US consumption growth risk, American Economic Review 97(1), Menzly, Lior, and Oguzhan Ozbas, 2009, Market segmentation and cross-predictability of returns, Journal of Finance, forthcoming. Merton, Robert, 1987, A simple model of capital market equilibrium with incomplete information, Journal of Finance 42, Peng, Lin, and Wei Xiong, 2006, Investor attention, overconfidence, and category learning, Journal of Financial Economics 80, Petersen, Mitchell, 2009, Estimating standard errors in finance panel data sets: comparing approaches, Review of Financial Studies 22 (1), Shahrur, Husayn, Ying L. Becker, and Didier Rosenfeld, 2009, Return predictability along the supply chain: 27

28 the international evidence, Unpublished working paper, Bentley University and State Street Global Advisors. Simms, James, 2009, Chasing growth, Japan Inc. bullish on China s shop, Wall Street Journal, December 29, C10. 28

29 Table 1: Customer Momentum Strategy, 7/1981-3/2009 The table presents abnormal returns for portfolios of countries sorted on the equally weighted (Panel A) or sales-weighted (Panel B) average of the prior-month stock market returns of their major customers (customer countries account for at least 5% of exports, exports at least 20% of GDP). At the beginning of each month from July 1981 to March 2009, countries are sorted into three groups based on the previous-month local currency returns of their major customers. The top 30% are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Top-Bot is a zero-cost strategy that goes long the Top 30 portfolio and sells short the Bot 30 portfolio. Portfolios are rebalanced monthly. Portfolio returns are monthly U.S. dollar returns in percent. Excess returns are net of the US risk-free rate. Global CAPM Alpha is from a regression of excess monthly portfolio returns on a global market factor, MKT-RF. Global Multifactor Alpha is from a regression of excess monthly returns on MKT-RF, a global size factor SMB, a global value factor HML, a global stock market momentum factor MOM and a global currency momentum factor MOMFX. T-statistics are reported below the coefficient estimates. Panel A: Equal Weights for Customers Portfolio Country Weights Equal Weights GDP Weights Value Weights 7/1981-3/2009 (333 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.51] [2.78] [3.82] [4.12] [0.59] [2.66] [3.19] [2.82] [0.31] [2.46] [2.49] [2.42] Global CAPM Alpha [-1.08] [2.60] [3.92] [4.19] [-0.90] [2.32] [3.00] [2.93] [-1.40] [2.04] [1.99] [2.50] Global Multifactor Alpha [-3.04] [2.03] [3.28] [4.10] [-2.43] [2.05] [2.13] [3.03] [-2.77] [1.54] [0.96] [2.49] 7/1981-4/1995 (166 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.82] [3.05] [3.62] [3.05] [0.94] [2.78] [3.18] [2.14] [0.89] [2.80] [2.24] [1.31] Global CAPM Alpha [-0.78] [2.28] [2.97] [3.19] [-0.56] [1.88] [2.42] [2.25] [-0.66] [1.91] [1.20] [1.37] Global Multifactor Alpha [-2.20] [1.42] [2.10] [2.81] [-1.77] [1.23] [1.32] [2.06] [-1.75] [1.36] [0.05] [1.14] 5/1995-3/2009 (167 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.01] [1.05] [2.00] [2.78] [0.06] [1.10] [1.61] [1.86] [-0.28] [0.69] [1.38] [2.13] Global CAPM Alpha [-0.37] [1.73] [3.30] [2.77] [-0.27] [1.74] [2.48] [1.88] [-0.90] [1.02] [2.30] [2.15] Global Multifactor Alpha [-1.71] [0.90] [2.95] [2.86] [-1.37] [1.58] [1.92] [2.05] [-1.76] [0.50] [1.56] [2.17] 29

30 Table 1: Customer Momentum Strategy, 7/1981-3/2009 (continued) Panel B: Sales Weights for Customers Portfolio Country Weights Equal Weights GDP Weights Value Weights 7/1981-3/2009 (333 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.79] [2.50] [3.73] [3.73] [0.46] [2.18] [3.39] [3.28] [0.15] [2.19] [2.58] [2.90] Global CAPM Alpha [-0.64] [2.10] [3.78] [3.76] [-1.13] [1.50] [3.27] [3.39] [-1.78] [1.54] [2.13] [3.01] Global Multifactor Alpha [-2.39] [1.15] [3.20] [3.68] [-2.35] [0.46] [2.75] [3.46] [-2.94] [0.61] [1.36] [2.95] 7/1981-4/1995 (166 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [1.01] [2.89] [3.48] [2.93] [0.47] [2.44] [3.16] [2.69] [0.60] [2.69] [2.21] [1.72] Global CAPM Alpha [-0.56] [2.04] [2.80] [3.04] [-1.30] [1.47] [2.40] [2.91] [-1.27] [1.77] [1.16] [1.92] Global Multifactor Alpha [-1.90] [0.81] [2.20] [2.82] [-2.20] [0.13] [1.92] [2.90] [-2.12] [0.69] [0.55] [1.85] 5/1995-3/2009 (167 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.22] [0.81] [1.99] [2.37] [0.22] [0.81] [1.83] [1.95] [-0.29] [0.52] [1.50] [2.35] Global CAPM Alpha [0.01] [1.24] [3.21] [2.36] [0.01] [1.19] [2.89] [1.96] [-0.94] [0.65] [2.53] [2.36] Global Multifactor Alpha [-1.34] [0.52] [2.71] [2.48] [-1.07] [0.61] [2.35] [2.14] [-1.88] [0.12] [1.82] [2.42] 30

31 Table 2: Supplier Momentum Strategy, 7/1981-3/2009 The table presents abnormal returns for portfolios of countries sorted on the equally weighted (Panel A) or sales-weighted (Panel B) average of the prior-month stock market returns of their major suppliers (supplier countries account for at least 5% of imports, imports at least 20% of GDP). At the beginning of each month from July 1981 to March 2009, countries are sorted into three groups based on the previous-month local currency returns of their major suppliers. The top 30% are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Top-Bot is a zero-cost strategy that goes long the Top 30 portfolio and sells short the Bot 30 portfolio. Portfolios are rebalanced monthly. Portfolio returns are monthly U.S. dollar returns in percent. Excess returns are net of the US risk-free rate. Global CAPM Alpha is from a regression of excess monthly portfolio returns on a global market factor, MKT-RF. Global Multifactor Alpha is from a regression of excess monthly returns on MKT-RF, a global size factor SMB, a global value factor HML, a global stock market momentum factor MOM and a global currency momentum factor MOMFX. T-statistics are reported below the coefficient estimates. Panel A: Equal Weights for Suppliers Portfolio Country Weights Equal Weights GDP Weights Value Weights 7/1981-3/2009 (333 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.96] [2.15] [3.70] [3.41] [0.41] [1.64] [3.35] [3.64] [0.16] [1.52] [2.89] [3.60] Global CAPM Alpha [-0.48] [1.47] [3.92] [3.49] [-1.32] [0.62] [3.52] [3.61] [-1.87] [0.43] [2.75] [3.55] Global Multifactor Alpha [-1.77] [-0.18] [3.14] [3.11] [-2.15] [-1.17] [2.61] [3.17] [-2.84] [-1.11] [1.90] [3.26] 7/1981-4/1995 (166 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [1.38] [2.10] [3.78] [2.50] [0.85] [1.59] [3.33] [2.86] [0.97] [1.57] [2.74] [2.09] Global CAPM Alpha [-0.14] [0.93] [3.23] [2.75] [-0.94] [0.28] [2.65] [2.87] [-0.81] [0.26] [1.83] [2.12] Global Multifactor Alpha [-1.15] [-0.78] [2.36] [2.29] [-1.47] [-1.89] [1.56] [2.15] [-1.40] [-1.54] [1.03] [1.71] 5/1995-3/2009 (167 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.05] [0.99] [1.67] [2.31] [-0.14] [0.76] [1.63] [2.32] [-0.62] [0.59] [1.45] [3.03] Global CAPM Alpha [-0.32] [1.60] [2.81] [2.30] [-0.63] [1.24] [2.99] [2.31] [-1.60] [0.81] [2.71] [3.01] Global Multifactor Alpha [-1.53] [0.39] [2.08] [2.21] [-1.74] [0.54] [2.36] [2.49] [-2.69] [0.13] [1.99] [3.08] 31

32 Table 2: Supplier Momentum Strategy, 7/1981-3/2009 (continued) Panel B: Sales Weights for Suppliers Portfolio Country Weights Equal Weights GDP Weights Value Weights 7/1981-3/2009 (333 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.88] [2.35] [3.71] [3.64] [0.43] [2.31] [3.10] [3.23] [0.31] [1.92] [2.61] [2.93] Global CAPM Alpha [-0.52] [1.86] [3.97] [3.63] [-1.23] [1.81] [2.98] [3.24] [-1.51] [1.10] [2.22] [2.92] Global Multifactor Alpha [-1.79] [0.48] [3.23] [3.13] [-2.03] [0.41] [2.14] [2.82] [-2.32] [-0.13] [1.42] [2.61] 7/1981-4/1995 (166 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [1.08] [2.46] [3.84] [2.99] [0.57] [2.49] [3.09] [2.57] [0.87] [2.05] [2.65] [2.01] Global CAPM Alpha [-0.41] [1.41] [3.31] [3.08] [-1.16] [1.45] [2.30] [2.73] [-0.79] [0.76] [1.73] [2.09] Global Multifactor Alpha [-1.49] [0.28] [2.37] [2.50] [-1.96] [0.12] [1.35] [2.34] [-1.39] [-0.39] [1.16] [1.82] 5/1995-3/2009 (167 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.21] [0.89] [1.66] [2.13] [0.08] [0.85] [1.52] [1.99] [-0.34] [0.67] [1.15] [2.13] Global CAPM Alpha [-0.02] [1.46] [2.84] [2.11] [-0.25] [1.53] [2.68] [1.97] [-1.06] [1.11] [2.04] [2.12] Global Multifactor Alpha [-1.14] [0.22] [2.05] [1.89] [-1.03] [0.54] [1.92] [1.80] [-1.78] [0.18] [1.26] [2.00] 32

33 Table 3: Combining Customer and Supplier Momentum Strategies, 7/1981-3/2009 Panel A shows abnormal portfolio returns from a combined trade momentum strategy. At the beginning of each month from July 1981 to March 2009, I sort producer countries whose exports or imports account for at least 20% of GDP into three groups based on the equally weighted average of the prior-month local currency returns of their major trading partners. Major trading partners are defined as countries that account for at least 5% of exports (imports) when exports (imports) represent at least 20% of GDP. The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the middle 40% to the Med 40 portfolio. Portfolios are held for a month and then rebalanced. The combined trade momentum strategy goes long the Top 30 portfolio and sells short the Bot 30 portfolio. Panel B shows abnormal portfolio returns from a total trade momentum strategy. At the beginning of each month from July 1981 to March 2009, countries for which total trade (exports and imports) accounts for at least 40% of GDP are sorted on the prior-month stock market return of an equally weighted portfolio of their major trading partners (countries accounting for at least 5% of total trade). The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the middle 40% to the Med 40 portfolio. The total trade momentum strategy is a zero-cost strategy that goes long the Top 30 portfolio and sells short the Bot 30 portfolio. Portfolios are rebalanced monthly. All portfolio returns are monthly U.S. dollar returns in percent. Excess returns are net of the US risk-free rate. Global CAPM Alpha is from a regression of excess monthly portfolio returns on a global market factor, MKT-RF. Global Multifactor Alpha is from a regression of excess monthly returns on MKT-RF, a global size factor SMB, a global value factor HML, a global stock market momentum factor MOM and a global currency momentum factor MOMFX. T-statistics are reported below the coefficient estimates. Panel A: Combined Trade Momentum Strategy Portfolio Country Weights Equal Weights GDP Weights Value Weights 7/1981-3/2009 (333 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [0.59] [2.90] [3.57] [3.92] [0.58] [2.31] [3.04] [2.94] [0.13] [2.23] [2.46] [3.00] Global CAPM Alpha [-1.00] [2.76] [3.63] [3.95] [-0.99] [1.76] [2.89] [2.93] [-1.93] [1.63] [1.99] [2.93] Global Multifactor Alpha [-2.91] [1.81] [2.76] [3.69] [-2.55] [0.67] [1.86] [2.86] [-3.21] [0.36] [0.99] [2.67] Panel B: Total Trade Momentum Strategy Portfolio Country Weights Equal Weights GDP Weights Value Weights 7/1981-3/2009 (333 months) Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Bot 30 Med 40 Top 30 Top-Bot Excess Ret [1.08] [2.48] [3.68] [3.27] [0.62] [2.36] [3.45] [3.33] [0.66] [2.14] [2.62] [2.36] Global CAPM Alpha [-0.26] [2.06] [3.86] [3.35] [-0.93] [1.82] [3.55] [3.31] [-0.96] [1.45] [2.21] [2.39] Global Multifactor Alpha [-1.71] [0.74] [3.20] [3.11] [-1.90] [0.64] [2.84] [3.12] [-1.79] [0.35] [1.18] [1.98] 33

34 Table 4: Cross-Sectional Time-Series Regressions on Customer and Supplier Momentum, 7/1981-3/2009 The table reports pooled cross-sectional time-series regressions of country stock market returns, measured in local currency. The dependent variable is a country s stock market return for month t. The explanatory variables are the lagged equally weighted average return on the portfolio of the country s major customers (AVE CRET) or major suppliers (AVE SRET), the country s own lagged returns (RET), and the log of the country s total market capitalization for the year of the trade links (in millions of current U.S. dollars). To include countries in the customer (supplier) regressions, I require that their exports (imports) account for at least 20% of GDP for the year of the trade links. Major customers (suppliers) are countries that account for at least 5% of total exports (imports). Standard errors are clustered by month. T-statistics are reported below the coefficient estimates. Panel A: All Countries Customer Momentum Supplier Momentum (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) AVE CRETt AVE SRETt [3.12] [3.08] [2.84] [2.95] [1.99] [2.72] [2.66] [2.18] [2.51] [2.53] AVE CRETt 2:t AVE SRETt 2:t [1.11] [1.10] [0.54] [1.04] [1.42] [1.44] [0.80] [1.46] RETt RETt [1.13] [0.87] [0.60] [1.21] [0.68] [-0.10] RETt 2:t RETt 2:t [1.06] [1.85] [1.75] [2.65] Time Fixed Effects No No No No Yes Time Fixed Effects No No No No Yes Observations Observations Adjusted R Adjusted R Panel B: Developed Countries Customer Momentum Supplier Momentum (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) AVE CRETt AVE SRETt [2.90] [2.76] [2.01] [2.02] [2.39] [2.92] [2.79] [2.12] [2.11] [2.84] AVE CRETt 2:t AVE SRETt 2:t [1.62] [1.59] [0.80] [-0.03] [1.75] [1.71] [1.30] [0.17] RETt RETt [1.96] [1.90] [2.13] [1.81] [1.80] [2.37] RETt 2:t RETt 2:t [1.66] [2.32] [0.95] [1.64] Time Fixed Effects No No No No Yes Time Fixed Effects No No No No Yes Observations Observations Adjusted R Adjusted R

35 Table 5: Gradual Diffusion of Information Hypothesis Tests, 7/1981-3/2009 Panel A reports pooled time-series cross-sectional regressions of country stock market returns, measured in local currency. The dependent variable is the stock market return for a sample country for month t. The explanatory variables are the lagged returns on the equally-weighted portfolio of the country s major customers (CRET) or suppliers (SRET), the country s own lagged returns (RET), and the log of the country s total market capitalization for the year of trade (in millions of current U.S. dollars). HIGHMCAP is a dummy variable equal to 1 if the sample country s market capitalization is greater than the 80th percentile of all sample countries in that year. HIGHGDP is defined analogously. HIGHTRADEART is an indicator equal to 1 if the number of trade-related articles in the Wall Street Journal about that country exceeds the 80th percentile of all sample countries for the year of trade link. Panel B reports the same regressions but for a shorter period. HIGHTURN is a dummy variable equal to 1 if the stock market turnover of a sample country exceeds the 80th percentile of all sample countries for the year of the trade link. To include countries in the customer momentum regressions, I require that their exports account for at least 20% of GDP for the year of the trade links. Major customers are countries that account for at least 5% of total exports. To include countries in the supplier momentum regressions, I require that their imports account for at least 20% of GDP for the year of the trade links. Major suppliers are countries that account for at least 5% of total imports. All regressions include month fixed effects. The standard errors are clustered by month. T-statistics are reported below the coefficients. Panel A: MCAP, GDP and Trade Articles (7/1981-3/2009) Customer Momentum Supplier Momentum (1) (2) (3) (4) (5) (6) AVE CRETt AVE SRETt [2.14] [2.05] [2.18] [2.63] [2.61] [2.69] AVE CRETt 1 * HIGHMCAP AVE SRETt 1 * HIGHMCAP [-2.53] [-2.31] HIGHMCAP 0.01 HIGHMCAP 0.01 [2.75] [2.58] AVE CRETt 1 * HIGHGDP AVE SRETt 1 * HIGHGDP [-2.38] [-2.91] HIGHGDP 0.00 HIGHGDP 0.00 [1.55] [1.32] AVE CRETt 1 * HIGHTRADEART AVE SRETt 1 * HIGHTRADEART [-2.20] [-2.50] HIGHTRADEART 0.01 HIGHTRADEART 0.01 [2.58] [2.61] AVE CRETt 2:t AVE SRETt 2:t [1.05] [1.03] [1.09] [1.48] [1.44] [1.48] RETt RETt [0.58] [0.61] [0.61] [-0.11] [-0.09] [-0.09] RETt 2:t RETt 2:t [1.76] [1.83] [1.74] [2.58] [2.62] [2.53] Observations Observations Adjusted R Adjusted R

36 Table 5: Gradual Diffusion of Information Hypothesis Tests, 7/1981-3/2009 (continued) Panel B: Stock Market Turnover (7/1989-3/2009) Customer Momentum Supplier Momentum (1) (2) (3) (4) (5) (6) (7) (8) AVE CRETt AVE SRETt [1.69] [1.60] [1.72] [1.68] [1.98] [1.96] [2.05] [2.09] AVE CRETt 1 * HIGHMCAP AVE SRETt 1 * HIGHMCAP [-2.35] [-1.80] HIGHMCAP 0.01 HIGHMCAP 0.01 [2.31] [2.08] AVE CRETt 1 * HIGHGDP AVE SRETt 1 * HIGHGDP [-2.46] [-2.74] HIGHGDP 0.00 HIGHGDP 0.00 [1.42] [1.21] AVE CRETt 1 * HIGHTRADEART AVE SRETt 1 * HIGHTRADEART [-2.06] [-2.52] HIGHTRADEART 0.01 HIGHTRADEART 0.01 [3.14] [3.32] AVE CRETt 1 * HIGHTURN AVE SRETt 1 * HIGHTURN [-1.01] [-0.87] HIGHTURN 0.00 HIGHTURN 0.00 [0.88] [0.98] AVE CRETt 2:t AVE SRETt 2:t [0.86] [0.85] [0.94] [0.86] [1.50] [1.46] [1.54] [1.48] RETt RETt [0.41] [0.44] [0.42] [0.43] [-0.45] [-0.44] [-0.46] [-0.42] RETt 2:t RETt 2:t [1.53] [1.59] [1.45] [1.61] [2.31] [2.35] [2.17] [2.35] Observations Observations Adjusted R Adjusted R

37 Table 6: Implementing Customer and Supplier Momentum Strategies with Exchange-Traded Funds, 4/1996-3/2009 The table presents abnormal returns for long-short customer and supplier momentum strategies implemented with U.S.-based single-country exchange-traded funds (ETFs). Panel A reports average net monthly returns for customer momentum. At the beginning of each month from July 1981 to March 2009, countries are sorted into three groups based on the equally or sales weighted average of the prior-month stock market returns of their major customers (customer countries account for at least 5% of exports, exports at least 20% of GDP). The top 30% are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Customer momentum strategies go long the available ETFs for the Top 30 portfolio and sell short the available ETFs for the Bot 30 portfolio. Portfolios are rebalanced monthly taking into account brokerage commissions and bid-ask spreads. Panel B reports average net monthly returns for supplier momentum. Supplier momentum strategies are constructed analogously to customer momentum strategies. ETF returns are net of expenses. Portfolio returns are monthly U.S. dollar returns (net of trading costs) in percent. Excess returns are net of the US risk-free rate. Global CAPM Alpha is from a regression of excess monthly portfolio returns on a global market factor, MKT-RF. Global Multifactor Alpha is from a regression of excess monthly returns on MKT-RF, a global size factor SMB, a global value factor HML, a global stock market momentum factor MOM and a global currency momentum factor MOMFX. T-statistics are reported below the coefficient estimates. Panel A: Average Net Monthly Returns for Customer Momentum Partner Weight EW Sales Country Weight EW GDP VW EW GDP VW Excess Ret [1.39] [1.01] [1.38] [1.42] [1.39] [1.78] Global CAPM Alpha [1.39] [1.01] [1.38] [1.42] [1.39] [1.78] Global Multifactor Alpha [1.42] [1.04] [1.42] [1.39] [1.37] [1.79] Panel B: Average Net Monthly Returns for Supplier Momentum Partner Weight EW Sales Country Weight EW GDP VW EW GDP VW Excess Ret [2.78] [2.45] [2.61] [1.89] [1.85] [1.64] Global CAPM Alpha [2.78] [2.45] [2.60] [1.87] [1.83] [1.63] Global Multifactor Alpha [2.85] [2.50] [2.74] [1.97] [1.89] [1.60] 37

38 Table 7: Real Domestic Effects of Stock Market Movements, The table contains coefficients and t-statistics from pooled annual regressions of a country s consumption or investment growth on its own lagged stock market return, measured in local currency. Annual real growth rates of household consumption and investment are from WDI. Gross fixed capital formation (FIXED INVESTMENT GR) consists of outlays on additions to domestic fixed assets. Gross capital formation (INVESTMENT GR) also includes changes in the level of inventories. All regressions include country fixed effects. Standard errors are clustered by year. T-statistics are shown in brackets. Dependent Variable HH CONS GROWTH t INVESTMENT GR t FIXED INVESTMENT GR t (1) (2) (3) (4) (5) (6) RET t [2.68] [2.69] [1.97] [1.94] [1.86] [2.17] Dep Var t [4.85] [2.48] [5.09] Observations Adjusted R

39 Table 8: Regressions of Changes in Exports on Lagged Customer Returns, 7/ /2008 The table contains coefficients and t-statistics from regressions of the change in a country s exports to a major customer on the lagged local currency stock market returns of that country and its customer. In Panel A, the dependent variable is the change in 3-month exports to a country s major customer ( Exports t:t+2 Exports t 1:t 3 1). In Panel B, the dependent variable is the change in 6-month exports ( Exports t:t+5 1). In Panel C, the dependent variable is the change in 12-month exports Exports t 1:t 6 ( Exports t:t+11 1). Exports are measured in the local currency of the producer country. The explanatory variables are Exports t 1:t 12 the lagged returns on the country s major customer (CRET), the country s own lagged returns (RET), and the log of the country s GDP for the year of the trade link (in millions of current U.S. dollars). All export ratios are winsorized at the top and bottom 1% level, whereas the cumulative lagged customer return is winsorized at the top 1% level. The results are not sensitive to winsorizing at other levels or logging. To include countries in the regressions, I require that their exports account for at least 20% of GDP for the year of the trade link. Major customers are countries that account for at least 5% of total exports. The estimates in columns 1-5 are from pooled time-series cross-sectional regressions. These five regressions also contain month-of-the-year indicators. Standard errors in columns 1-5 are clustered by trade pair. The coefficient estimates in columns 6-8 are time-series averages of the slopes from month-by-month Fama-MacBeth regressions. The Fama-MacBeth standard errors are Newey-West adjusted allowing for 2, 5 and 11 lags of serial correlation in Panel A, B and C, respectively. T-statistics are reported below the coefficient estimates. 39

40 Table 8: Regressions of Changes in Exports on Lagged Customer Returns, 7/ /2008 (continued) Panel A: Change in Three-Month Exports Pooled Time-Series Cross-Sectional Regs Fama-MacBeth Regs (1) (2) (3) (4) (5) (6) (7) (8) CRET t [2.70] [1.56] [1.80] [0.44] [0.30] [0.09] CRET t 2:t [7.58] [7.55] [7.16] [4.74] [3.41] [3.47] RET t [-0.70] [-1.38] [-2.01] [-2.27] [0.84] [0.25] RET t 2:t [3.18] [3.28] [1.19] [-2.02] [0.85] [0.66] Time Fixed Effects No No No No Yes Pair Fixed Effects No No No Yes Yes Observations Adjusted R Panel B: Change in Six-Month Exports Pooled Time-Series Cross-Sectional Regs Fama-MacBeth Regs (1) (2) (3) (4) (5) (6) (7) (8) CRET t [3.44] [1.37] [2.26] [0.53] [1.05] [0.56] CRET t 2:t [6.65] [6.66] [6.62] [4.78] [3.54] [3.84] RET t [1.28] [0.47] [-0.58] [-1.79] [0.97] [0.74] RET t 2:t [3.23] [3.23] [1.32] [-1.48] [1.31] [1.25] Time Fixed Effects No No No No Yes Pair Fixed Effects No No No Yes Yes Observations Adjusted R Panel C: Change in Twelve-Month Exports Pooled Time-Series Cross-Sectional Regs Fama-MacBeth Regs (1) (2) (3) (4) (5) (6) (7) (8) CRET t [5.79] [3.03] [4.78] [1.42] [1.55] [0.94] CRET t 2:t [5.87] [5.86] [5.98] [4.02] [2.77] [2.90] RET t [3.45] [2.82] [1.22] [-0.77] [2.05] [1.95] RET t 2:t [3.48] [3.45] [1.63] [-0.79] [1.57] [1.44] Time Fixed Effects No No No No Yes Pair Fixed Effects No No No Yes Yes Observations Adjusted R

41 Table 9: The Customer Effect and Final Goods as a Fraction of Exports, 7/1981-3/2009 For each sample country and each year I compute final exported goods as a fraction of total exported goods. Every year I sort sample countries based on the final goods fraction. Countries above the 80th percentile for a given year are assigned to the HIGHFINAL group and countries below the 80th percentile are assigned to the LOWFINAL group. The first two columns of Panel A present regressions of monthly producer country stock market returns on the lagged stock market returns of the producer country and the equally weighted portfolio of its major customers separately for the two groups. The regressions are specified as in Column 5 of Table 4. The last three columns of Panel A present pooled regressions of changes in exports on the lagged stock market returns of the producer and customer country. Regressions are specified similarly to those in Column 5 of Table 8. Panel B repeats these tests replacing the raw fraction of final goods with the residual from a pooled regression of the raw fraction on the logs of market capitalization and gross domestic product. Panel A: Using Final-Good Exports as a Fraction of Total Exports Dependent Variable Returns Changes in Exports LOWFINAL HIGHFINAL Diff 3-month 6-month 12-month AVE CRET t CRET t [0.72] [2.80] [2.67] [-0.33] [-0.29] [0.46] AVE CRET t 2:t CRET t 1 * HIGHFINAL [1.56] [-0.37] [2.63] [2.96] [2.99] RET t CRET t 2:t [0.81] [-0.41] [4.63] [4.57] [3.50] RET t 2:t CRET t 2:t 12 * HIGHFINAL [1.72] [1.27] [0.45] [0.86] [1.65] HIGHFINAL [2.05] [1.06] [0.83] RET t [-2.44] [-1.88] [-0.90] RET t 2:t [-2.12] [-1.57] [-0.89] Observations Observations Adjusted R Adjusted R Panel B: Using the Residual from a Regression of Final-Good Fraction of Exports on Market Size and GDP Dependent Variable Returns Changes in Exports LOWFINAL HIGHFINAL Diff 3-month 6-month 12-month AVE CRET t CRET t [0.94] [2.18] [1.67] [-0.11] [-0.12] [0.67] AVE CRET t 2:t CRET t 1 * HIGHFINAL [1.61] [-0.03] [1.70] [2.28] [2.41] RETt CRET t 2:t [1.04] [-0.96] [4.61] [4.60] [3.58] RET t 2:t CRET t 2:t 12 * HIGHFINAL [1.68] [1.31] [1.01] [1.21] [1.58] HIGHFINAL [2.02] [1.70] [1.18] RET t [-2.46] [-1.93] [-0.98] RET t 2:t [-2.20] [-1.68] [-0.98] Observations Observations Adjusted R Adjusted R

42 Table 10: Supplier Effects on GDP Growth, Panel A contains correlations between a country s annual GDP growth (GDPGR), the average annual GDP growth of its major suppliers (AVE SGDPGR), the country s lagged annual stock market return (RET) and the average lagged stock market return of its major suppliers (AVE SRET). Panel B contains coefficients and t-statistics from pooled time-series cross-sectional regressions of a producer country s GDP growth on its own lagged GDP growth, on the contemporaneous average GDP growth of its major suppliers and on the contemporaneous world GDP growth. Panel C contains coefficients and t-statistics from pooled time-series cross-sectional regressions of a producer country s GDP growth on its own lagged annual stock market return and the lagged average stock market return of its major suppliers. To include countries in the analysis, I require that their imports account for at least 20% of GDP for the year of the trade links. Major suppliers are countries that account for at least 5% of total imports. Stock market returns are in local currency and are winsorized at the top 1% level. All regressions include country fixed effects. Standard errors are clustered by year. T-statistics are shown in brackets. Panel A: Correlations of GDP Growth and Returns GDPGR t GDPGR t 1 AVE SGDPGR t AVE SGDPGR t 1 RET t 1 AVE SRET t 1 GDPGR t 1.00 GDPGR t AVE SGDPGR t AVE SGDPGR t RET t AVE SRET t Panel B: Regressions of GDP Growth on Contemporaneous Average Supplier GDP Growth GDPGR t (1) (2) (3) AVE SGDPGR t [4.18] [3.89] [2.19] GDPGR t [2.89] [2.91] WORLDGDPGR t 0.15 [0.53] Observations Adjusted R Panel C: Regressions of GDP Growth on Lagged Average Supplier Return GDPGR t (1) (2) AVE SRET t [5.73] [1.87] RET t [2.96] Observations Adjusted R

43 Table 11: Regressions of Changes in Imports on Lagged Supplier Returns, 7/ /2008 The table contains coefficients and t-statistics from regressions of the change in a country s imports from a major supplier on the lagged local currency stock market returns of that country and its supplier. In Panel A, the dependent variable is the change in 3-month imports from a country s major supplier ( Imports t:t+2 1). In Panel B, the dependent variable is Imports t 1:t 3 the change in 6-month imports ( Imports t:t+5 1). In Panel C, the dependent variable is the change in 12-month imports Imports t 1:t 6 ( Imports t:t+11 1). Imports are measured in the local currency of the producer country. The explanatory variables are the Imports t 1:t 12 lagged returns on the country s major supplier (SRET), the country s own lagged returns (RET), and the log of the country s GDP for the year of the trade link (in millions of current U.S. dollars). All import ratios are winsorized at the top and bottom 1%, whereas the cumulative lagged supplier return is winsorized at the top 1%. The results are not sensitive to winsorizing at other levels or logging. To include countries in the regressions, I require that their imports account for at least 20% of GDP for the year of the trade link. Major suppliers are countries that account for at least 5% of total imports. The estimates in columns 1-5 are from pooled time-series cross-sectional regressions. These five regressions also contain month-of-the-year indicators. Standard errors in columns 1-5 are clustered by trade pair. The coefficient estimates in columns 6-8 are time-series averages of the slopes from month-by-month Fama-MacBeth regressions. The Fama-MacBeth standard errors are Newey-West adjusted allowing for 2, 5 and 11 lags of serial correlation in Panel A, B and C, respectively. T-statistics are reported below the coefficient estimates. 43

44 Table 11: Regressions of Changes in Imports on Lagged Supplier Returns, 7/ /2008 (continued) Panel A: Change in Three-Month Imports Pooled Time-Series Cross-Sectional Regs Fama-MacBeth Regs (1) (2) (3) (4) (5) (6) (7) (8) SRET t [3.97] [3.19] [2.99] [1.76] [1.52] [0.81] SRET t 2:t [3.88] [3.67] [2.51] [1.74] [1.43] [1.27] RET t [1.09] [-0.46] [-0.48] [-0.06] [2.02] [1.49] RET t 2:t [7.05] [7.00] [4.69] [3.32] [2.82] [2.52] Time Fixed Effects No No No No Yes Pair Fixed Effects No No No Yes Yes Observations Adjusted R Panel B: Change in Six-Month Imports Pooled Time-Series Cross-Sectional Regs Fama-MacBeth Regs (1) (2) (3) (4) (5) (6) (7) (8) SRET t [4.79] [3.49] [2.75] [2.57] [3.13] [2.65] SRET t 2:t [3.71] [3.62] [2.57] [1.97] [1.92] [1.70] RET t [4.37] [2.40] [2.14] [2.54] [2.22] [1.75] RET t 2:t [7.73] [7.67] [5.26] [3.81] [4.18] [3.88] Time Fixed Effects No No No No Yes Pair Fixed Effects No No No Yes Yes Observations Adjusted R Panel C: Change in Twelve-Month Imports Pooled Time-Series Cross-Sectional Regs Fama-MacBeth Regs (1) (2) (3) (4) (5) (6) (7) (8) SRET t [6.85] [4.92] [3.58] [2.35] [2.76] [2.53] SRET t 2:t [3.63] [3.51] [2.73] [1.60] [2.18] [1.92] RET t [7.44] [6.08] [5.39] [3.89] [3.03] [2.06] RET t 2:t [8.67] [8.59] [6.29] [4.36] [3.83] [3.72] Time Fixed Effects No No No No Yes Pair Fixed Effects No No No Yes Yes Observations Adjusted R

45 Table 12: Disentangling the Customer and Supplier Momentum Effects: 7/1981-3/2009 Panel A contains coefficients and t-statistics from pair-level pooled time-series cross-sectional regressions of country stock market returns, measured in local currency. The dependent variable is a country s stock market return for month t. The explanatory variables are the lagged stock market returns on a country s major supplier (SRET), the country s own lagged stock market return (RET), and the log of the country s total market capitalization for December end of the year of trade. Panel B contains coefficients and t-statistics from analogous regressions of monthly country returns on the lagged returns of that country and its major customers. SUPPLIER is a dummy variable equal to 1 if a country s major customer is also a major supplier. CUSTOMER is a dummy variable equal to 1 if a country s major supplier is also a major customer. To include sample countries in the supplier momentum regressions, I require that their imports account for at least 20% of GDP for the year of the trade links. Major suppliers are countries that account for at least 5% of total imports. To include sample countries in the customer momentum regressions, I require that their exports account for at least 20% of GDP for the year of the trade links. Major customers are countries that account for at least 5% of total exports. All regressions in Panels A and B include month fixed effects. The standard errors in Panels A and B are clustered by month and by country. Panel C shows regressions of the change in a country s imports from a major supplier on the lagged local currency stock market returns of that country and its supplier. The dependent variables are the change in 3-month imports (column 1), the change in 6-month imports (Column 2) and the change in 12-month imports (Column 3).The explanatory variables are the lagged supplier returns (SRET), the country s own lagged returns (RET), and the log of the country s GDP for the year of the trade link (in millions of current U.S. dollars). Panel D contains the analogous results for changes in exports. The reported estimates in Panels C and D are from pooled time-series cross-sectional regressions. These regressions also contain monthof-the-year indicators and month fixed effects. Standard errors in Panels C and D are clustered by trade pair. In all panels T-statistics are reported below the coefficient estimates. Panel A: Supplier Momentum Effect Panel B: Customer Momentum Effect (1) (2) (1) (2) SRETt CRETt [2.32] [1.45] [2.19] [2.40] SRETt 1 * CUSTOMER 0.01 CRETt 1 * SUPPLIER [1.02] [-1.53] CUSTOMER 0.00 SUPPLIER 0.00 [-0.61] [-0.01] SRETt 2:t CRETt 2:t [-0.14] [-0.15] [0.38] [0.37] RETt RETt [0.14] [0.13] [1.16] [1.14] RETt 2:t RETt 2:t [3.88] [3.87] [3.02] [3.02] Panel C: Changes in Imports Panel D: Changes in Exports 3-month 6-month 12-month 3-month 6-month 12-month SRETt CRETt [1.38] [1.58] [1.69] [-0.51] [-0.54] [0.94] SRETt 1 * CUSTOMER CRETt 1 * SUPPLIER [0.54] [1.61] [1.33] [1.33] [1.42] [0.18] SRETt 2:t CRETt 2:t [2.02] [2.03] [1.39] [4.99] [4.99] [3.49] SRETt 2:t 12 * CUSTOMER CRETt 2:t 12 * SUPPLIER [1.57] [1.53] [1.88] [-1.72] [-1.42] [-0.57] CUSTOMER SUPPLIER [-4.09] [-3.39] [-3.20] [-2.75] [-2.25] [-2.12] RETt RETt [0.15] [3.09] [4.66] [-1.52] [-0.49] [1.04] RETt 2:t RETt 2:t [5.46] [5.92] [6.37] [-0.25] [0.34] [1.06] 45

46 Figure 1: Annual Returns of Customer Momentum At the beginning of each month from July 1981 to March 2009, sample countries (whose exports represent at least 20% of their GDP) are sorted into three groups based on the equally weighted average of the local currency returns of their major customers for the previous month. Major customers are countries that account for at least 5% of total exports. The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Portfolios are equally (value) weighted and rebalanced monthly. The customer momentum portfolio is a zero-cost strategy that goes long the Top 30 portfolio and sells short the Bot 30 portfolio. This figure shows the annual U.S. dollar returns on the zero-cost customer momentum portfolio (Top-Bot), along with the annual returns on the MSCI All Countries Index in excess of the U.S. risk-free rate (MKT-RF). The returns for 1981 are for July to December. Panel A: Equal Weights Panel B: Value Weights 46

47 Figure 2: Customer Momentum, Event-Time Average Cumulative Returns At the beginning of each month from July 1981 to March 2009, sample countries (whose exports represent at least 20% of their GDP) are sorted into three groups based on the equally weighted average of the local currency returns of their major customers for the previous month (t). Major customers are countries that account for at least 5% of total exports. The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Portfolios are equally (value) weighted and rebalanced monthly. The customer momentum portfolio is a zero-cost strategy that goes long the Top 30 portfolio and sells short the Bot 30 portfolio. The solid line in this figure shows the average cumulative U.S. dollar return on the zero-cost customer momentum portfolio for month t + 1 to t The dashed lines represent a two-standard-error confidence interval. Standard errors are Newey-West adjusted for serial correlation. Panel A: Equal Weights Panel B: Value Weights 47

48 Figure 3: Annual Returns of Supplier Momentum At the beginning of each month from July 1981 to March 2009, sample countries (whose imports represent at least 20% of their GDP) are sorted into three groups based on the equally weighted average of the local currency returns of their major suppliers for the previous month. Major suppliers are countries that account for at least 5% of total imports. The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Portfolios are equally (value) weighted and rebalanced monthly. The supplier momentum portfolio is a zero-cost strategy that goes long the Top 30 portfolio and sells short the Bot 30 portfolio. This figure shows the annual U.S. dollar returns on the zero-cost supplier momentum portfolio (Top-Bot), along with the annual returns on the MSCI All Countries Index in excess of the U.S. risk-free rate (MKT-RF). The returns for 1981 are for July to December. Panel A: Equal Weights Panel B: Value Weights 48

49 Figure 4: Supplier Momentum, Event-Time Average Cumulative Returns At the beginning of each month from July 1981 to March 2009, sample countries (whose imports represent at least 20% of their GDP) are sorted into three groups based on the equally weighted average of the local currency returns of their major suppliers for the previous month (t). Major suppliers are countries that account for at least 5% of total imports. The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Portfolios are equally (value) weighted and rebalanced monthly. The supplier momentum portfolio is a zero-cost strategy that goes long the Top 30 portfolio and sells short the Bot 30 portfolio. The solid line in this figure shows the average cumulative U.S. dollar return on the zero-cost supplier momentum portfolio in for month t + 1 to t The dashed lines represent a two-standard-error confidence interval. Standard errors are Newey-West adjusted for serial correlation. Panel A: Equal Weights Panel B: Value Weights 49

50 Figure 5: Country Presence in the Bot 30, Med 40 and Top 30 Portfolios Panel A of Figure 5 plots the fraction of months that each sample country spends in each of the three customer momenutm portfolios. The portfolios are constructed as follows. At the beginning of each month from July 1981 to March 2009, sample countries (whose exports represent at least 20% of their GDP) are sorted into three groups based on the equally weighted average of the local currency returns of their major customers for the previous month. Major customers are countries that account for at least 5% of total exports. The top 30% of the sorted countries are assigned to the Top 30 portfolio, the bottom 30% are assigned to the Bot 30 portfolio, and the remainder to the Med 40 portfolio. Panel B plots the analogous results for supplier momentum portfolios. Panel A: Customer Momentum Portfolios Panel B: Supplier Momentum Portfolios 50