Return Predictability along the Supply Chain F.AJ inancial Analysts Journal Volume 66 . Number 3 @2010 CFA Institute these two studies were based on the
Return Predictability along the Supply Chain F.AJ inancial Analysts Journal Volume 66 . Number 3 @2010 CFA Institute these two studies were based on the idea that the 2007, the last month for which data were available economic performance of economically linked at the start of the study Return Predictability along the Supply Chain: companies should be correlated. Thus, news about We also excluded a company if (1) it was a the operating performance of a customer should financial firm (primary SIC codes between 6000 and The International Evidence affect the stock price of both the customer and its 6999) because capturing the customers of such firms suppliers (and vice versa). Assuming a slow diffu- is notoriously difficult, (2) its total assets or total sion of information in the stock market, one should sales were under $10 million, (3) its monthly return Husayn Shahrur, Ying L. Becker, and Didier Rosenfeld, CFA observe a lead-lag effect in the stock returns of both was greater than 100 percent, or (4) the ratio of customers and suppliers. market value of assets to book value of assets was Using company-level customer-supplier data, Restrictions 2, 3, and 4 were In this study of a sample of equities listed on the exchanges of 22 developed countries, equity returns greater than 50. Restriction of customer industries led the returns of supplier industries. This customer-supplier/lead-lag effect Cohen and Frazzini (2008) found that returns of intended to omit very small companies, as well as observations with potential data entry errors. We exhibits characteristics consistent with the view that the effect results from a slow diffusion of customer companies in one month predict suppli- ers' returns in the subsequent month. They also found qualitatively similar results after relaxing value-relevant information. documented that this effect is economically signifi- these three restrictions. The resulting sample con- cant. Using the Fama-French (1993) three-factor sisted of 1,202,332 company-month observations for model to estimate abnormal returns, they found that 14,407 unique companies ("overall sample"). We used this sample to compute customer returns. To R cent research in finance suggests that capacity; in efficient markets, investors are able to a monthly long-short trading strategy based on this nvestors' inattention to information can process all publicly available information in a lead-lag effect can generate annual abnormal form the sample of supplier companies, we required result in the predictability of security timely fashion and thus are able, as a group, to price returns of up to 18 percent. Menzly and Ozbas (2007) that customer returns (described later in the article) returns. For instance, Cohen and Frazzini financial securities efficiently (see, e.g., Fama 1991). found a similar phenomenon by using industry- be constructible. This restriction resulted in a sample (2008) argued that if investors have limited ability Motivated by a large number of empirical anoma- of 478,727 company-month observations for 6,174 level linkages collected from the U.S. IO accounts. to gather and process all publicly available infor- lies, recent asset pricing models have assumed the In summary, evidence in the finance literature unique companies ("supplier sample"). mation, value-relevant information that is incor- opposite-namely, that investors have limited abil- is consistent with the view that stock prices do not porated in the stock prices of customer companies ity to gather and process information (see, e.g., Customer-Supplier Linkages. In our study, fully reflect publicly available information about may not be fully reflected in the stock prices of Hong and Stein 1999). A key finding derived from a primary challenge was to capture the customer- their suppliers. Using company-level data on U.S. these models is that some changes in security prices economically linked companies. Our study supplier linkages of companies that operated in a customer-supplier relationships, they found sup- builds on this evidence by investigating whether large number of countries. Studies that have used are predictable. port for this proposition: Lagged equity returns of equities along the same supply chain in developed Consistent with the predictions of such mod- customer-supplier company-level data (see, e.g., customer companies are positively correlated with markets exhibit lead-lag effects. In particular, it Kale and Shahrur 2007; Cohen and Frazzini 2008) els, a large body of empirical research shows evi- suppliers' contemporaneous equity returns. extends the work of Cohen and Frazzini (2008) and have relied on Statement of Financial Accounting dence that security prices do not incorporate all In our study, weexamined whether a customer- Menzly and Ozbas (2007) to international equity Standards Nos. 14 and 131, which require U.S. pub- supplier/lead-lag effect exists in international publicly available information. Of particular rele- markets by examining whether equity returns of lic companies to disclose the identity of any cus- vance to our study is the strand of literature that equity markets. One of our main challenges con- customer companies help predict the returns of tomer whose purchases represent more than 10 provides evidence of cross-correlation in equity cerned the difficulty in identifying customer- supplier companies. percent of the company's total revenues. We are returns for companies in the same industry. This supplier linkages in a large number of international unaware of such a disclosure requirement for the anomaly was first documented by Lo and Mackin- markets. To circumvent this problem at a poten- lay (1990), who found that equity returns on small Data countries in our study. An alternative approach to tially low cost, we used the benchmark input- identifying supply chain linkages, followed by We formed our sample in light of our main chal- output (IO) accounts for the U.S. economy to stocks tend to lag returns on large stocks. Badri- Shahrur (2005) and Menzly and Ozbas (2007), is to identify industry-level customer-supplier linkages nath, Kale, and Noe (1995) documented evidence lenge, which was to capture a large number of identify industry-level linkages by using country- in other developed markets. This approach is sup- suggesting that Lo and Mackinlay's result was international customer-supplier linkages, as well specific IO accounts. Doing so for 22 countries, how- as in view of the need to determine a customer ported by empirical research that provides strong driven by limited information gathering with ever, would require manually collecting data on 22 evidence of significant similarities among the pro- respect to small stocks. More recently, Hou (2007) return index. different IO accounts, which would be a monument duction structures of developed economies. found that Lo and Mackinlay's lead-lag effect is al task and perhaps an impractical approach predominantly an intra-industry phenomenon: Sample. To form our sample, we first identi- fied all stocks that were in the IDC/Exshare and To circumvent this problem at a potentially The Literature Returns on big companies lead returns on small low cost, we identified customer-supplier link- companies only within the same industry. This Worldscope databases and listed on the exchanges One of the assumptions of the efficient market par- ages by using the benchmark IO accounts for the evidence supports the conclusion that Lo and of developed countries in the MSCI World Index U.S. economy. We based this choice on evidence adigm is that investors have unlimited processing Mackinlay's lead-lag effect is driven by the slow (excluding the United States). Thus, our sample that the production structures of developed coun- diffusion of same-industry information, which is comprised equities from the following 22 markets: tries are very similar to one another. For instance, Husayn Shahrur is senior equity strategist at NBK incorporated in the equity prices of large compa- Australia, Austria, Belgium, Canada, Denmark, using IO accounts, Chenery and Watanabe (1958) Capital, Kuwait, and associate professor of finance at nies, to the equity prices of smaller companies. Finland, France, Germany, Greece, Hong Kong, found that the production structures of four devel- Bentley University, Waltham, Massachusetts. Ying L. Cohen and Frazzini (2008) and Menzly and Ireland, Italy, Japan, the Netherlands, New oped countries (the United States, Japan, Italy, and Becker is head of global active equity research and Didier Ozbas (2007) extended the literature on intra- Zealand, Norway, Portugal, Singapore, Spain, Norway) are highly similar. Other studies have Rosenfeld, CFA, is managing director and head of EAFE industry return predictability by examining Sweden, Switzerland, and the United Kingdom. shown that such similarities extend to other devel- and global active equity strategies at State Street Global whether a lead-lag effect exists among equities Because pre-1995 data were sparse, we began the oped countries and to such developing economies Advisors, Boston. along the same supply chain. The hypotheses of sample period in January 1995 and ended it in July as India (see Santhanam and Patil 1972; Song 1977).Financial Analyss journal The approach of using the ID accounts of one country to identify supply chain linkages in another has been followed by other researchers. For example, to estimate the Swiss [0 tables for 2001, the Centre for Energy Policy and Economics of ETH Zurich used data from the IQ accounts of other European countries.5 One drawback of using US. 10 accounts to capture supply chain linkages in other developed markets is that the actual linkages may not be captured accurately. This outcome could bias our tests against finding any signicant return cross- predictability along the supply chain. Thus, one could view our ndings as conservative estimates of the actual relationships because the analyzed factor is measured with noise. Moreover, variables constructed under this approach could capture fac- tors (e.g., industry momentum) that affect supplier returns but that are unrelated to suppermtomer linkagesl To mitigate this concern, we conducted a cross-sectional analysis in which we controlled for factors known to be correlated with contemporane- ous returns (eg., the own-industry lagged return). Therefore, to capture customerrsupplier industrylevel linkages, we relied on the \"use table\" of the benchmark IO accounts for the U.S, economy (for recent studies that used thisdataset, see Fan and Lang 2000; Shahrur 2005; and Kale and Shahmr 2007).6 For any pair of supplier/customer indus- tries, the use table reports estimates of the dollar value of the supplier industry's output that is used as an input in the production of the customer indus- try's output. For each supplier industry, the use table of the IO accounts enabled us to identify the customer industries and the importance of each customer industry to the supplier industry. We identified customer industries as follows. We rst classified our sample companies into indus- tries on the basis of their primary SIC code; we called these industries the supplier industries. (Note that Worldscope analysts assign SIC codes to companies on the basis of the business descriptions in the companies' annual reports) Next, given that the use table is constructed on the basis of the 10 six- digit coding system, we used an IO*SIC code con version table to identify the supplier industry's IO code (for more details, see Sluahrur 2005). We then used the supplier industry's IO code to identify its Customer industries from the use table. To avoid look-ahead bias, we relied on the 1987 use table, which was published in 1992 (before the rst year of our sample). Later in the article, we discuss the results from analyses of more recent 10 accounts. In determining our nal sample of supplier industries for which we had customer return data, we imposed restrictions. Some supplier industries sell their output to final users and have no identi- able customer companies that are publicly traded.7 Although other industries sell to interme- diate companies and not to final users, their cus- tomer industries may have no publicly traded companies. For such supplier industries, con structing a customer return factor on the basis of publicly traded customers is impossible. There fore, we included a supplier industry in the sample if at least 25 percent of its output was sold to customer industries with publicly u'aded compa- nies for which we had return data. Appendix A provides a list of selected supplier industries, together with the percentage of each industry's output sold to nal users. For example, the industry Motor Homes (SIC code 3716) sells all its output to nal users (mainly individuals); thus, we did not include in our supplier sample compa- nies in this industry. But we did include companies in the industry Primary Production of Aluminum (SIC code 3334) so long as there were publicly traded companies with return data in the industry Aluminum Rolling and Drawing or other customer industries that bought at least 25 percent of the supplier industry's output.a Customer Return Index. Ourcentralresearch question concerned whether lagged customer stock returns help predict the stock returns of supplier companies. As previously described, we identified the customer industries for each supplier industry in our sample by using [0 accounts. Next, for the ith supplier industry, we constructed the lagged cus- tomer return as follows: Curiamermiumw : )3 CR I, H (Indurlry percenmge :0qu ), F. (1) ,s. where r1 is the number of customer industries, CR]: is the value-weighted return on the portfolio of all companies in the jth customer industry, and Indus try percentage soldii is the percentage of the output of the ith supplier industry that is sold to the jth cus~ tomer industry. Intuitively, Industry percentage soldv- measures the importance of the ith customer indus- try as a buyer of the ith supplier industry's output. Simply put, Customer return is a weighted average of the returns on the customer industries' portfolios, where the weight is the percentage of the supplier industry's output that is sold to each customer industry. By definition, Customer return will be high if important customer industries have high returns. Note that in computing Industry percentage sold, we considered all customer industries, including those with no publicly traded companies, such as final usersl Because this approach is equivalent to assum- ing that the returns on customer industries with no publicly traded companies are equal to zero, it penalizes the returns on customer industries if they are insignificant buyersl In addition, given that the effect of new information about customers on sup- pliers may be country specic, we constructed Cus- tomer return at the country level, As a result, for each supplier industry in each country, we computed Customer return by using all publicly traded cus- tomer companies in that particular country. Descriptive Statistics. Table 1 presents descriptive statistics for July 2007, the last month in our sample period. Panel A shows statistics for the overall sample,- Panel B shows statistics for the sup- plier sample. As can be seen in Panel A, the overall sample contained 9,437 companies The median (average) company had a market value of equity (MVE) of $162 million ($2 billion) Japanese stocks accounted for the largest percentage of the sample in terms of market capitalization (about 20 percent) and number of companies (37 percent). In addition, the sample contained 402 unique 10 industries. Panel B shows that of these industries, 223 had available customer return data; these industries included 3,942 companies, or 42 percent of the total number of companies in the overall sample. The supplier sample constituted 31 percent of the overall sample in terms of market capitaliza- tion. For the supplier sample, the median (average) MVE was $146 million ($1.49 billion)smaller than the corresponding value in the overall sample. This nding suggests that supplier companies are, on average, smaller than their customers. The statistics in Table 1 also suggest that our sample of supplier companies was somewhat similar to the overall sample in terms of market composition. For instance, we found that across the two samples, the correlation among the various market MVEs (as a percentage of total MVE) was about 93 percent. Methodology and Results: LongShort Portfolio Approach We constructed supplier portfolios to test for the existence of a customersupplier/leadlag effect in our sample. Portfolio Formation. To examine whether the returns of supplier and customer industries exhibit cross-predictability, we tested whether a trading strategy based on buying or selling sup- plier companies selected on the basis of lagged customer returns generates abnormal returns. Thus, for each month t of the sample period, we Return Predictability along the Supply Chain ranked supplier industries in ascending order on the basis of the customer return for the previous month (Customer returnH) and formed ve quin- tile portfolios of supplier industries (Q1 to Q5), where Q1 (Q5) included supplier industries with the lowest (highest) Customer returnHl We also constructed a longshort (LS) portfolio that held the top quintile (Q5) and sold short the bottom quintile (Q1). The central hypothesis of our study is that if the returns of suppliers and customers exhibit cross-predictability, the LS trading strat- egy should generate positive abnormal returns. To estimate portfolio abnormal returns, we rst used a market model based on the return of a global market portfolio that included all the stocks in our overall sample. We also estimated abnormal returns by using a four-factor model with all the factors constructed at the universe level. Our four- factor model was the FamarFrench (1993) three- factor model augmented by Carhart's (1997) momentum factorl We followed the methodology described on Kenneth French's website? to con- struct the four factors, namely, Market, HML, SML, and MOM. Therefore, for each portfolio, we esti- mated the following model: Supplier portfolio excess return, = a + lMarkel, + a, HML, + 535MB, + B4MOM, , (2) where Supplier portfolio excess return = the equally (or value) weighted monthly excess return on the supplier industry portfolio Market : the value-weighted excess return on a portfolio thatincludes all the stocks in our sample HML : the return difference between high- and low-bookvtomarket stocks 5MB 2 the return difference between small- and large~cap stocks MOM = the return difference between stocks with high and low retums for the period t 12 to t 2 We defined excess portfolio returns as portfolio returns minus the risk-free rate. We used the one- month US. Treasury bill rate as a proxy for the risk- free rate.10 As is common in the literature, we lagged all nancial statement data to account for the delayed availability of dam to market partici- pants. Thus, for a month t, we used financial state- ment data for the scal year ended at least at t 3. Financial Analysts Journal Return Predictability along the Supply Chain Table 1. Descriptive Statistics, July 2007 Although we constructed Customer return at the Panel A suggest that the returns of supplier and No. of Mean MVE Median MV Market MVE % Of No. of Unique country level, we conducted our main tests at the customer industries exhibit cross-predictability. Marke Companies ($ billions) ($ billions) ($ billions Total MVE IO Industries universe level (i.e., not at the country level) because We also examined whether the L-S strategy is A. Overall sample we were interested in examining whether profitable after accounting for transaction costs. On Japan 3,494 1.104 0.115 3,858.166 20.38 329 customer-supplier return predictability existed in the basis of the findings in Domowitz, Glen, and United Kingdom 1,001 3.160 0.184 3,163.373 16.71 194 the whole universe of stocks in developed markets. Madhavan (2001), we computed the weighted aver- France 518 4.10 .200 2,125.613 1.23 157 Thus, we formed the quintiles and L-S portfolios age of round-trip trading costs for our sample, in Germany 502 2.868 0.117 1,439.621 7.60 146 and constructed the four risk factors by using data for all the countries in our study. We also conducted which the weight is based on the number of compa- Canada 719 1.606 0.221 1,154.379 6.10 140 nies in each market." We found that the weighted Netherlands 137 7.902 0.839 ,082.584 5.72 71 our analyses of Japan and the United Kingdom average is 91 bps. We also found that the total Switzerland 170 5.590 0.664 950.294 5.02 94 separately because those two markets combined 1.447 0.164 represented about 50 percent of our sample monthly turnover for our long-short trading strat- Hong Kong 593 857.998 4.53 147 egy is about 71 percent. This finding implies that the Australia 469 1.769 0.176 829.676 4.38 135 Supplier Portfolios' Abnormal Returns. estimate of total transaction costs per month is 65 Spain 100 6.465 1.307 646.520 3.42 59 Table 2 reports the results from a market model for bps. Together with the results for the equally Italy 206 2.95 0.445 609.257 3.22 103 quintile portfolios of stocks formed on the basis of weighted L-S portfolio, this result suggests that the Sweden 191 2.471 0.264 472.030 2.49 92 the lagged Customer return. Panel A reports the net return on our L-S trading strategy is about 7.2 Norway 136 2.23 0.402 303.234 1.60 52 results for equally weighted portfolio returns, which percent a year, which is economically significant. Finland 107 2.784 0.292 297.896 1.57 64 we computed by first calculating the average return Panel B of Table 2 shows the results for value- Singapore 486 0.535 0.080 260.208 1.37 149 or each supplier industry and then calculating the average return for all industries in the quintile. Sup- weighted quintile portfolios, in which the weights Belgium 86 2.430 0.322 208.971 1.10 53 Denmark 87 represent the stocks' market capitalizations at the 2.240 0.290 194.856 1.03 51 plier industries with the worst lagged customer per- beginning of the month. Although the results Greece 226 0.60 0.115 36.420 0.72 formance have an intercept (monthly alpha) of -0.30 exhibit some return predictability, the intercepts Austria 60 1.884 0.321 113.022 0.60 percent, which is not statistically significant at con- are largely statistically insignificant at conven- Ireland 43 2.133 .361 91.733 0.48 ventional levels (t-statistic is -1.26). Conversely, the Portugal 43 2.072 0.197 39.09 tional levels. The L-S portfolio has a monthly alpha 0.47 quintile that includes supplier industries with the New Zealand 63 0.720 0.213 45.391 0. 24 largest lagged customer return (Q5) has a positive of 0.35 percent and a t-statistic of 0.92. Together 9.437 with the results for equally weighted returns, these Total 2.005 0.162 18,930.335 100.00 alpha of 0.98 percent (statistically significant at the 1 percent level). A strategy of buying stocks in Q5 results suggest that small stocks exhibit more cross- B. Supplier sample and shorting stocks in Q1 yields a statistically signif predictability than do large stocks. This finding is United Kingdom 532 2.594 0.160 1,380.168 23.49 109 icant monthly alpha of 1.28 percent. This result not surprising because one would expect the diffu- Japan 1,404 0.942 0.126 1,322.263 22.51 163 translates into an annual abnormal return of 15.36 sion of information to be slower for small compa- Canada 436 1.500 0.232 653.947 11.13 percent. Further, the results for the five quintiles nies for several reasons, including less analyst Australia 230 2.092 0.198 481.203 8.19 show that alpha increases monotonically with following and less institutional holding than for Netherlands 40 10.29 0.589 411.744 7.01 lagged customer returns. Overall, the results in large companies. France 222 1.578 0.138 350.401 5.96 Germany 227 1.535 0.090 348.486 5.93 Switzerland 55 3.086 0.825 169.754 2.89 Table 2. Supplier Portfolios' Abnormal Returns: Market Model, Hong Kong 173 0.886 0.150 153.260 2.61 January 1995-July 2007 Sweden 85 1.349 0.193 114.636 1.95 (t-statistics in parentheses) Spain 28 3.939 1.050 110.300 1.88 Q1 (low) Q2 03 04 Q5 (high) Long-Short Norway 51 1.646 0.346 83.943 1.43 A. Equally weighted monthly returns Singapore 187 0.372 0.077 59.51 1.18 Intercept (%) -0.30 -0.17 0.24 0.55*** 0.98*#* 1.28** Italy 62 1.024 0.318 63.518 1.08 (-1.26) (-0.97) (1.38) 3.02 (4.79) 5.16 Greece 77 0.420 0.100 32.335 0.55 Market 0.77*** 0.70*** 0.73*** 0.73*** 0.80*** 0.03*# Finland 33 0.97 0.167 32.238 0.55 12.33 15.8 (16.06) (15.56) 15.10) (0.50) Austria 1.411 0.170 31.044 0.53 Adjusted R2 51.36% 63.73% 64.25% 62.77% 61.35% -0.53% Belgium 26 1.103 0.260 28.677 0.49 Denmark 19 0.729 0.100 13.856 0.24 B. Value-weighted monthly returns Portugal 1.504 0.108 13.540 0.23 Intercept (%) -0.C -0.39* 0.04 -0.04 0.27 0.35 Ireland 10 0.633 0.112 6.327 0.11 -0.29 -2.14) (-0.24) (-0.22) (1.09) (0.92) New Zealand 14 0.99*#* 0.96* ** 0.263 0.164 3.686 0.06 10 Market 1.00** 1.07*** 1.02*** 0.03* * * Total 3.942 1.490 0.146 5,874.837 100.0 223 13.2 20.14 25.03) (20.26) (16.06) 0.27) 73.89% Notes: This table reports descriptive statistics for a sample of companies traded on the stock exchanges of 22 markets in the MSC Adjusted R2 54.82% 31.39% 74.12% 64.24% 0.65% World Index (excluding the United States). Stock market data are for 1 July 2007. Panel A includes all companies that satisfy certain Notes: The sample includes 478,727 company-month observations for 6,174 unique companies. The size and other data restrictions. Panel B includes only companies for which a customer return can be computed. market model includes the value-weighted excess return on a market index. *Significant at the 10 percent level. #*Significant at the 1 percent level.Financial Analym Journal Table 3 shows estimates from the four-factor model. The results are almost identical to those of Table 2, which suggests that adding the other three risk factors has little effect on our results. For instance, using equally weighted returns (Panel A), we found that the LeS strategy has a fourrfactor monthly alpha of 1.23 percent. As shown in Panel B, using valueweighted returns, we found that the Les strategy yields an insignicant alpha. Because the results from both the market model and the four-factor model are similar, we report the results from only the four-factor model for all our subse- quent analyses. Panel A of Table 3 also shows that the quintile portfolios have some exposure to the model factors. For example, all five quintiles have significant exposure to the size factor (5MB), which is not surprising given thatthe subset of companies in the supplier sample is smaller than the subset in the overall sample, the latter being used in the factor construction. The LS portfolio, however, has less factor exposure than do the individual quintiles. The Les portfolio has insignificant loading on the market and size factors and modest exposure to the value and momentum factors. To examine whether the results in Tables 2 and 3 are driven by very small and highly illiquid com panies' stocks, we repeated our analysis after imposing several liquidity constraints. Thus, we excluded from the supplier sample companies below the 10th percentile of both MVE and trading turnover, which is defined as the average daily trading volume over the previous three months divided by the total number of shares outstanding. We also deleted stocks with a beginning-ofmonth price of less than $5. The results from our analysis of the trimmed sample are reported in Table 4. We found that the liquidity constraints did not have a material effect on our results. As shown in Table 4, the results pertaining to the equally weighted returns are very similar to those reported in Panel Aof Table 3. The monthly alpha of the LS portfolio Table 3. Supplier Portfolios' Abnormal Returns: FourAFactor Model, Return Predictability along the Supply Chain Table 4. Supplier Portfolios' Abnormal Returns: Trimmed Supplier Sample, January 1995July 2007 (lstatistics in parentheses) January 1995-July 2007 (t-statistics in parentheses) Q1 (low) Q2 Q3 Q4 05 (high) Long-Short A. Eryn/lily weighted monthly mums Intercept (7..) 0.27 41.22 0.13 0.39" 0.95m 1.23m (1.40) (1.59) (0.93) (2.56) (5.51) (4.70) Market 090m 034*\" 0.39"- 0.39m 0.94m 0.04 (17.43) (24.53) (25.49) (22.03) (20.32) (0.61) 5MB 0.57m 0.54m 0.61M 0.57M 0.53m 0.11 (6 39) (9.99) (11.00) (3.30) (9.34) (0.96) HML 0.37m 0.26m 0.24m 0.23m 017*\" 41.2 3* (5.65) (5.99) (5.54) (4.43) (3.01) (2.24) MOM 0.19m 41.07:. 41.01 0.03 0.02 0.17m (4.75) (2.59) (0.23) (1.00) (0.67) (3.09) Adjusted 31 73.53% 33.03% 33.49% 73.44% 77.65% 7.23% B. Value-weighted monthly mums intercept (9/5) 0 10 0.40" 0.10 41.15 0.24 0.34 (41.35) (72.05) (41.60) (059) (0.39) (0.35) Markcl 1.00m 1.02m 1.04m 1.13m 1.06m 0.01 (13.35) (19.70) (23.51) (20.10) (14.93) (41.13) 5MB 0 25" 0.25m 0.09 0.30' 0.22" 0.03 (2,06) (3.19) (1,29) (3.30) (1.93) (41.19) HML 0.35m 0.09 0.12" -0.04 41.03 -0.37m (3.52) (1.30) (2.09) (41.52) (0.30) (2.50) MOM 0.17* 41.02 0.02 0.12m 0.05 0.2.3m (72.32) (70.62) (41.43) (2.75) (1.03) (2.77) Adjusted 112 51.70% 75.75% 31.91% 76.37% 54.53% 0.02% Note: See the notes to Table 2. 'Significant at the 10 percent level. "Significant at the 5 percent level. \"Significant at the 1 percent level. Q1 (low) QZ OS 04 Q5 (high) Long-Short A. Equally weighted manlhly returns Intercept (%) 41.32 41.16 0.12 0.3 " 0.9 m 1.2 m (4.53) (1.14) (0.77) (2.42) (4.54) (5.00) Market 0 92m 0.92m 0.95m 0.97m 0.07m 0.05 (17.34) (24.24) (23.49) (22.95) (17.65) (0.73) SME 0.42m 0.55m 0.44m 0.53m 0.55m 0.13 (5.07) (3.94) (5.78) (3.59) (5.20) (1.13) Hill. 0 25m 0.15m 0.14m 0.10' 0.00 0.25m (3.91) (3.16) (2.57) (1.91) (0.00) (72.97) MOM 0.10" 41.03 0.05 0.07" 0.05 0.15m (2 49) (1.13) (1.44) (2.11) (1.25) (2.91) Adjusted R2 72 021u 32.35% 30.54% 30.00% 71.13% 9.33% 3 Valllt-wevghlcd monthly mums intercept (7..) 0.12 0.51m 0.21 0.03 0.20 0.31 (0 33) (2.55) (1.05) (0.35) (0.71) (0.77) Market 1.09m 1.02m 1.07m 1.011m 1.09m 0.01 (13.60) (19.32) (20.07) (17.59) (14.74) (0.06) 5MB 0.05 0.23m 0.05 0.19n 0.211 0.15 (0 43) (2.59) (0.57) (1.90) (1.77) (0.90) HML 0 27m 7003 0.05 41.17" 41.10 4133-" (2 65) (41.49) (0.73) (2.21) (1.12) (2.74) MOM 0 11* 0.05 0.04 0.15m 0.09 0.19" (1 70) (1.33) (1.05) (3.21) (1.55) (2.33) Adjusted R2 60.54% 75.37% 75.90% 72.92% 64.64% 6.11% Note: Restrictions placcdon the supplier sample resulted inatrimmed sample of 194.048 company-month observations for 3,930 unique companies. 'Significant at the 10 percent level. "Siylificant at the 5 percent level. "'Significant at the 1 percent level. is 1.28 percent, which is statistically significant at the 1 percent level. The results in Panel B of Table 4, which are based on value-weighted portfolios, also show some signs of crosSvpredictability, although the results are not statistically significant at conventional levels. Given the previous results, we explored the relationship between company size and customersupplier cross-predictability. Thus, we conducted similar analyses for subsamples of stocks formed on the basis of companies' market capital izations. We classified suppliers into four groups: micro-cap stocks (MVE less than $250 million), smallcap stocks (MVE between $250 million and $1 billion), midscap stocks (MVE between $1 billion and $5 billion), and largefap stocks (MVE greater than $5 billion).12 We used the trimmed supplier sample for this part of the analysis and report the results in Table 5. The results suggest that the customersupplier/leadlag effect is present in all size portfolios except the largecap portfolio. We also found that the abnormal return of the LS portfolio increases monotonically with company size. Moreover, the results in Table 5 suggest that the leadlag effect is not driven by microstocks. For instance, the LS strategy yields a monthly alpha of 0.87 percent even when it involves only stocks that are classied as mid-cap (Panel C). This finding is consistent with our earlier finding that dropping the most illiquid stocks from our sample does not sig- nificantly affect our results. Thus, this finding fur the:- addresses the concern that our results may be driven by the smallest and most illiquid stocks. Because about 50 percent of the sample com- panies were traded in Japan and the United King dom, we conducted separate analyses for each of those two markets. Thus, we reconstructed the risk factors separately for Japan and the United King- dom by using stocks traded in those markets. The results are reported in Table 6. We found that for Financial Analysts Journal Return Predictability along the Supply Chain Table 5. Supplier Portfolios' Abnormal Returns: Subsamples Based on Size, January 1995-July 2007 Table 6. Supplier Portfolios' Abnormal Returns: Japan and the f-statistics in parentheses United Kingdom, January 1995-July 2007 21 (low) 3 t-statistics in parentheses 24 Q5 (high) Long-Short A. Stocks with MVE less than $250 million (micro-cap); 46 percent of trimmed sample Q1 (low) Q3 Q4 25 (high) Long-Short Intercept (%) -0.60*** -0.38** 0.06 0.62*** 1.31* 1.91**# A. Japan; equally weighted monthly returns (-2.80) (-2.29) 0.32) 3.13 5.16) 5.79 Intercept (%) -0.14 -0.1 -0.06 0.27 0.52*** 0.66*** Market 0.87** 0.88** 0.86* * 0.95*** 0.93*+ 0.06 (-0.67) (-0.57) (-0.29) (1.28) (2.65 (15.29) (3.05 20.23 16.33 17.93) 13.79 (0.71) Market 0.89*#x 0.87*** 0.98** SMB 1.02** 0.63** 0.97*** 0.77*** 0.67*** 0.86*** 0.08 0.80*** 0.18 (19.40 20.17) 23.75 (11.00) (24.82) (6.87) 21.80 (7.92) (10.16) (1.54) (7.40) (1.25) SMB 0.61** * HML 0.24*4* 0.17* * * 0.61*** 0.53** 0.61** 0.55+* 0.10 -0.05 0.09 0.01 -0.22** (3.26) (8.87) 9.57) (8.56) (10.01) 3.00) (1.56) (1.38) (8.30) -0.65 MOM 0.13 (-2.01) HML 0.27*** 0.37** -0.08* 0.27*** 0.30** -0.01 0.0 0.07* 0.34*** 0.07 0.05 0.14+* (-1.88) (-0.40) (3.45 (0.51) (1.04) (4.98 (1.69) 3.86 (4.30) (4.50) (2.01) (0.79) Adjusted R2 66.03% MOM 0.06 77.28% 74.98% 72.41% 0.07 0.03 61.10% 3.62% 0.09** 0.07 0.01 (1.18) (1.54 0.80 (2.08) B. Stocks with MVE between $250 million and $1 billion (small-cap); 29 percent of trimmed sample Adjusted R (1.54) 0.27) 77.05% 78.80% 82.52% Intercept (%) -0.25 -0.26 83.88% 0.3 79.96% 0.63% 0.17 1.02*** 1.28*** (-1.01) (-1.52) (1.59) 0.85) (3.82) Market (3.75) B. United Kingdom; equally weighted monthly returns 1.02*** 1.02** 1.03** 1.09*** 1.00* -0.02 Intercept (%) -0.53 -0.20 -0.0 -0.21 0.18 0.71* (15.20) 22.10 (19.51) (20.97) (13.96) (-0.26) (-1.72) -0.64 0.24 -0.77) 0.57) SMB (2.09) 0.53** * 0.65*# 0.52** 0.73*** 0.66** 0.13 Market 1.06* * 0.90*#* 1.03*** 0.93*** 1.12*** 0.0 (4.94) 8.83 (6.13) 8.74) 5.77 (0.88) (12.34) (10.31) (15.54) (12.10) HML 12.93) 0.32** 0.20** (0.55) 0.07 0.10 -0.09 -0.41*** SMB 0.44*** 0.51**4 0.67*** 0.56** 0.53** 0.09 (3.78) 3.37 1.04) (1.52) -1.01 (-3.59) (5.03) 9.94 MOM (5.66 0.13** (7.17) 5.97 -0.01 0.0 0.10* (0.75) 0.10 0.23*** HML 0.13 .10 0 17* (-2.59) 0.20* 0.20* (-0.30) 0.08 (1.47) (2.47) 1.82 (3.34) Adjusted R2 (1.05) (0.79) 1.85 65.61% 79.20% 74.45% (1.84) 7.05% 1.66 (0.47) 61.41% 12.50% MOM -0.05 -0.04 0.02 0.04 0.0 0.04 C. Stocks with MVE between $1 billion and $5 billion (mid-cap); 18 percent of trimmed sample (-0.72) (-0.51) 0.39 (0.69) Intercept (%) (-0.09) -0.33 -0.23 (0.47) 0.08 0.39* 0.55** 0.87*** Adjusted R2 57.87% 50.11% 69.54% 56.15% 59.13% (-1.27) -2.33% (-1.33 (0.40) (2.11) Market (2.47 (2.60 1.06** 1.06*# 1.08*** Note: This table shows supplier portfolios' abnormal returns estimated by using a four-factor model. 1.04* 1.00*** -0.07 (15.44) (19.49) (21.24) *Significant at the 10 percent level 23.26 (16.89) (-0.73) SMB 0.21 **Significant at the 5 percent level. 0.38**# 0.37*** 0.31** * 0.26*** 0.04 4**Significant at the 1 percent level. (1.95) (4.36 (4.92 (3.97) (2.72) (0.29) HML 0.40** 0.06 0.10* 0.00 -0.06 -0.46*** (4.58) (0.82) (1.65) (0.07) -0.86) MOM (-4.09) -0.13** -0.12** * Japan (Panel A) and the United Kingdom (Panel B), 0.02 0.01 0.06 0.19* * * (-2.45) he L-S trading strategy yields monthly alphas of 10), which is defined as the variance of country (-2.86 (0.62) (0.16) 1.24 (2.70) Adjusted R2 65.96% 0.66 percent and 0.71 percent, respectively. We also returns that is explained by a global factor divided 76.73% 80.79% 78.65% 70.23% 12.56% found that the increase in alpha across the five by the variance explained by both global and local D. Stocks with MVE greater than $5 billion (large-cap); 7 percent of trimmed sample factors. We found that the correlation between the quintiles is largely monotonic in both markets. Intercept (%) 0.19 0.49** -0.2 two variables is negative and statistically signifi- 0.06 -0.17 -0.36 More generally, the results in Table 6 are qualita- (0.67) (-2.26) (-1.24) (0.25) cant at the 10 percent level. This result suggests that Market -0.64) (-0.90) 1.05** 1.10**# tively similar to those of our overall analysis. 1.09*** 1.15*** 1.03*** -0.02 he customer-supplier/ lead-lag effect is weaker in 13.66) 19.07 (20.22) We next examined whether the lead-lag effect (18.91) (15.07) (-0.21) countries whose financial markets are more inte- SMB 0.04 -0.08 -0.21* depends on the degree to which a particular market 0.13 -0.06 -0.10 grated with the world market. (0.33) (-0.88) (-2.42) is financially integrated with the world market. For (1.35) (-0.52) (-0.58) In sum, our results are consistent with the HML 0.13 0.06 0.12 example, an increase in financial integration with -0.02 -0.01 -0.14 hypothesis that the stock returns of customer (1.33) 0.88 (1.82) (-0.30) ( 0.10) other markets can result in more investors exploit- (-1.04) companies lead suppliers' returns. This cross- MOM -0.13** 0.03 -0.08* 0.12** 0.04 0.16** ing the customer-supplier/lead-lag effect, which predictability in returns is not driven by the -2.10 0.67 -1.85) (2.47) Adjusted R2 (0.71) should make that effect weaker. Toward that end, (1.99) smallest and most illiquid stocks, is less pro- 62.20% 75.74% 79.12% 74.57% 66.50% 1.39% we estimated the alphas from a long-short strategy for all the countries in our sample following the nounced for the largest stocks, is significant in the Notes: This table shows supplier portfolios' abnormal returns estimated by using a four-factor model for largest two markets in our sample, and is weaker subsamples formed on the basis of the company's market value of equity. All factors are computed by specification in Table 3 (using the specification in using all the companies in the overall sample. Table 2 yields similar results). We then estimated in markets that are more financially integrated the correlation between the alpha estimates and a with the world market. These findings add to the *Significant at the 10 percent level **Significant at the 5 percent level. country-specific measure of capital market integra- recent literature on customer momentum for U.S. #**Significant at the 1 percent level. tion. We took our measure for integration from the equities (Cohen and Frazzini 2008; Menzly and Ozbas 2007) and suggest that this phenomenon study by Bekaert, Hodrick, and Zhang (2005, Table extends to other developed markets.Financial Analysts Journal Return Predictability along the Supply Chain Methodology and Results: included them in the regressions to control Table 7. Fama-MacBeth Regression Analysis of Supplier's Monthly Regression Approach for the intra-industry lead-lag effect found Stock Return, January 1995-July 2007 by Hou (2007). Following Hou, we classified t-statistics in parentheses We conducted a regression analysis to examine our four-digit SIC code industries into 12 indus- Model 1 Model 2 Model 3 findings from the L-S methodology after control- tries. We also repeated our analyses by Intercept (%) 0.520* -0.053 0.07 ling for other variables related to contemporaneous using equally weighted portfolio returns (1.68) (-0.18) -0.25) stock returns and found similar results. Customer return-1 0.157*** 0.123*** 0.043*** Regression Model. We used a regression 4. MVE is the market value of the company's (5.42) (5.91) (3.01) approach to test whether the findings from the L-S equity. Company return-1 -0.040*** -0.041*** methodology hold after controlling for other vari- 5. BTM is the ratio of the company's market (-6.44) (-7.30) value of equity to its book value of equity. Company return-12,t-2 0.009*** 0.009*** are shown in the literature to be correlated with contemporaneous stock returns. Thus, using Unlike the L-S methodology, the regression (3.97 (3.84 0.040*** the Fama-MacBeth (1973) regression methodology, model in Equation 3 is estimated at the supplier- Industry return-1 we estimated the following model: company level and not at the supplier-industry (5.35) level. This approach enables us to control for Large-size industry return-1 0.027 Supplier firm return, = a + B, Customer return,-1 company-specific determinants of contemporane- (1.48) ous returns. The model (Equation 3) is estimated Medium-size industry return-1 0.098*** + B2 Return,-1 (4.90) + B3 Return,-12, 1-2 for every calendar month, and the averages of the coefficient estimates are reported. Following Fama Small-size industry return-1 0.042*** + BA Industry return,-1 (3) and MacBeth (1973), we determined the statistical (3.30) or By Large-size industry return,-1 significance of these averages by using the stan- Industry return-12, t-1 0.010*** 0.011* * * + Bo Medium-size industry return,-1 dard deviation of monthly estimates. The main (4.42) (5.37 + B, Small-size industryreturn,-1 coefficient of interest in Equation 3 is B1. The central MVEt-1 0.030 0.033 + Bg Industry return,-12,1-2 prediction hypothesis of this part of the analysis is (1.23) (1.28 + B,MVE,_1 + BloBTM,_ + H, as follows: If the returns of suppliers and customers BTM -1 0.506*** 0.484*** exhibit cross-predictability, we should expect Bj to (6.07) (6.22) where be positive and statistically significant. Adjusted R2 1.06% 4.79% 5.33% 1. Return is the supplier company's return. No. of months 151 151 151 We included Return,_1 to control for the Regression Results. The results obtained Notes: This table reports the results of a Fama-MacBeth (1973) regression analysis of the supplier- reversal effect of Jegadeesh (1990) and from the L-S methodology may be driven by the company monthly stock return on the customer lagged monthly return and control variables. The Return-12,t-2 to control for the momentum correlation between the customer returns and sample includes 478,727 company-month observations for 6,174 unique companies. The dependent effect of Jegadeesh and Titman (1993). other factors that are known to be correlated with variable is the stock return for the supplier company for month t. Customer return,_1 is the value- 2. Industry return is the value-weighted return contemporaneous returns. In particular, the lagged weighted return for month t - 1 on a portfolio of companies in customer industries. Company return-1 (Company return-12,t-2) is the supplier's return for t-1 (t -12 to t -2). Industry return_1 (Industry on a portfolio of the supplier company's customer return may be highly correlated with the return-12,t-1) is the value-weighted return on the supplier's industry portfolio for t -1 (t - 12 to f - 1); four-digit SIC code industry. We included lagged own-industry return given the nature of the t-statistics are computed by using the standard deviation of the time series of monthly estimates for Industry return_1 and Industry return-12, t-2 relationships between supplier and customer each coefficient. The average of the monthly adjusted R2 is reported. to control for the momentum effect at the industries. To examine this issue, we conducted a *Significant at the 10 percent leve industry level (see Moskowitz and Grinblatt Fama-MacBeth (1973) regression analysis, in *Significant at the 1 percent level. 1999). This control variable is especially which we regressed the stock returns of supplier important because the lagged customer companies on the customer returns and several return is likely to be correlated with the control variables. Recall that we conducted the 3, we included the lagged returns on size-based high future returns. The coefficients on the size lagged supplier return given that suppliers regression analysis at the supplier-company level portfolios (Hou 2007). The coefficient on Customer variable (MVE), however, are insignificant (using and customers operate in related industries. (i.e., not at the industry level) to control for return_1 remains positive and statistically signifi- the natural log of MVE also yields statistically Using industry portfolios at the two-digit company-specific determinants of contemporane- cant, although the magnitude of the coefficient insignificant coefficients). SIC code level or equally weighted portfo- ous returns. The results of this analysis for the estimate is smaller. lios yields qualitatively similar results. supplier sample are reported in Table 7 (the results The results pertaining to the control variables Conditional Regression Analysis. Next, we 3. Large-, medium-, and small-size industry of this analysis for the trimmed supplier sample are are generally consistent with those reported in the examined whether the effect of customer-supplier returns are the value-weighted returns on very similar and are untabulated here) literature. For example, evidence of short-term linkages on the predictability of supplier returns portfolios of large-industry companies The results of Model 1 in Table 7 are consistent reversal in stock returns is seen in the negative and varies with the characteristics of suppliers. Thus, (70th to 100th percentile of MVE), medium- with our main finding: The lagged customer statistically significant coefficients on Company we examined whether the documented customer- industry companies (30th to 70th percentile return is positively related to the contemporane- return_1. The results pertaining to the long-term supplier/lead-lag relationship is stronger (or of MVE), and small-industry companies (0 ous supplier return. In Model 2, we included the own and industry returns are consistent with the weaker) for (1) smaller suppliers, (2) those whose to 30th percentile of MVE), respectively. main control variables. The results suggest that the momentum effect (see Jegadeesh and Titman 1993; sales to their customers are concentrated, and (3) Following Cohen and Frazzini (2008), we customer-supplier/lead-lag relationship is Moskowitz and Grinblatt 1999). We also found those with higher relationship-specific investments constructed the size-based portfolios and robust to the inclusion of these controls. In Model that high-book-to-market companies tend to have with their customers.FinancialAna/ystsloumal Return cross-predictability and supplier size. As discussed earlier, under the hypothesis of slow diffusion of information, smaller suppliers are expected to exhibit stronger return predictability (we have already reported results consistent with this prediction). We examined whether this size effect persists after controlling for other determi- nants of contemporaneous return. Thus, we included in our regression model Customer returnH, the interaction term Customer returnH (M VE H), and the control variables. The results for this specification are reported in the first column of Table 8. Consistent with our prediction, we found a negative and significant coefficient on the interaction term, which suggests that smaller sup pliers exhibit stronger leadlag effects. Return cross-predictability and sales concen- tration. Next, we proposed that the concentration of the suppliers' sales can also affect the magnitude of the cross-predictability of returns. On the one hand, a supplier company that sells its output to many customers may be able to mitigate the effect of a particular shock that affects one particular cus- tomer by selling to other companies that do not Table a. Conditional FamaMacaeth Regression Analysis 0! Supplier's Monthly stock Return, January 1995-July 2007 (tstatistics in parentheses) Model 1 Model 2 Model 3 Intercept (%) 43.055 411120 -o.019 (0.22) (0.37) (0.07) Customer returnH 11.13 m 0157"- 0.091m (6.35) (5.114) (4.77) Customer mum... (Mvc,_.) 4021-" (-6.29) Customer returnH (Supplier snles 41197"- concentration) (73.04) Salts concenirrilon 0.535 (1 59) Customer retllrnH (Supplier lac/D inlensily) 5.624'" (2.81) iron inrensriy 4.955 (0.60) Company retllrnH 411140"- mulm 41042" (5.45) (4.60) (7703) Company rclurnHZH nomm arms-v omen" (3.97) (3.95) (4.05) Industry relur-nH 0.04m\" warm 0.037m (5.32) (5.37) (5.33) lnrusmy returnHz H noiom arms-v 0009*" (4.44) (4 311) (4.20) MVEH 0042* (1.031 0.1133 (1.83) (1126) (1.35) BTMH (3.507m 0493'" 0.5mm (a 09) (am) (6.41) Adlusted R2 4.32% 5.37% 5.55% No. of months 151 151 151 Notes: See the notes to Table 7. This table reports the results of a Fama7MacBeth (1973) regression analysis of the supplier-company monthly stock return on the customer lagged monthly return and control variables. Returns are expressed m percentage terms Supplier sales concentration is a measure that takes high (low) values when the output of the supplier's industry is sold to few (many) customer industries. Supplier asp intensity is total supplier industry Roi) divided by total industry assets; t-statistics are computed by using the standard deviation of the time series of monthly estimates for each coefficient. The average of the monthly adiusted K2 is reported. 'Significant at the 10 percent level. "'Slgnificant at die 1 percent level. Return Predictability along the Supply Chain suffer the same shock. Thus, a supplier company that deals with many customers should have weaker return crosspredictability with its custom- ers. On the other hand, if a company sells most of its output to just one customer, the effect of a shock to that customer on the supplier's operations will probably be easier for investment professionals to understand. In such a scenario, the stock price of supplier companies should adjust more quickly to information pertaining to their customers. Under this view, one would expect stronger return cross~ predictability if the supplier deals with a larger number of customers. With no companylevel data on supplier sales, weused an industry-level variable as an alternative because we expected industry sales concentration to be positively correlated with company-level activity. Thus, to test the relationship between return crosspredictability and the concentration of supplier sales, we constructed the following vari- able by using data from the IQ accounts: Supplier soles conceniroiion, = (4) n 2 (Indurrry percentage sold? Fl yer where n is the number of customer industries and Industry percentage soldiy is, as defined earlier, the percentage of the ith industry's output that is sold to the jth customer industry Intuitively, Supplier sales concentration will be large for supplier indus- tries that sell a large percentage of their output to just one customer industry. As reported in the second column of Table 8, we found a negative and statistically signicant coefficient on the interaction term Customer returnH (Supplier sales concentration), which suggests that suppliers that have more dispersed sales experi- ence stronger retum crosspredictability. This nd- ing is consistent with the notion that the diffusion of information from customers to suppliers is slower when investment professionals must assess the effect of news about multiple customers on the supplier companies. Return predictability and relationship- specific investments. Companies that undertake relationship-specic investments with their cus tomers are more likely to be affected by the operat- ing performance of the customer companies; 3 the prospects of such companies are largely tied to the performance of their customers. For instance, if a company has made significant investments that are specic to its customer, the value of such assets will be much lower (higher) if the customer's business is expected to deteriorate (improve). Thus, we expect such companies to exhibit more return cross-predictability because the economic link with their customers is stronger. To capture the intensity of relationshipspecific investments between the company and its customers, we used the R&D inmnsity of the supplier's industry.\"1 The use of R&D intensity to proxy for asset specicity is abundant irl the empirical literature on transaction cost economics. For example, Armour and Teece (1980) suggested that RkDintensive vertical chains are likely to have complex interstage interdepen- dencies, Levy (1985) argued that research-intensive industries tend to involve specialized inputs that require transaction-specific investments by suppliers (see also Allen and Phillips 2000; Kale and Shahrur 2007). Moreover, many R&D-intensive industries have tight vertical coordination, which tends to involve highly specialized asses.15 Thus, our proxy for the intensity of relationship- spedfic investments is Supplier R'D intensity, which is defined as total industry R&D divided by total industry assets.16 To test our hypothesis, we inter- acted Supplier RfrD intensity with Customer returnH and report the results in the third column of Table 8. We found that the coefficient on this interaction variable is positive and signicant at the 1 percent level. This finding is consistent with our hypothesis and suggests that suppliers and customers with strong economic ties are associated with strong return crosspredictability. Robustness Analysis To check for robustness, we repeated our analysis with more recent 10 accounts, as well as with equally weighted portfolios for Customer return. IO Accounts. Our earlier ndings are based on the 1987 10 accounts for the whole sample period Because customersupplier linkages can change over time, we repeated our analysis with more recent 10 accounts. Thus, we conducted our main analysis with the 1987, 1992, and 1997 accounts for 199571997, 199872002, and 200372007, respectively. In choosing which 10 accounts to use for which period, we ensured that the 10 accounts were published before the start of the particular period, thereby avoiding look-ahead bias. For example, the 1992 10 accounts were published in 1997 and were thus used for 19982002. We found that the results from this analysis are qualimtively similar to those reported here and are thus not tabulated. This finding is not surprising given the high degree of correlation among the 10 accounts. For instance, we found that the mean pairwise correlation for the variable Industry percentage sold; (defined earlier)7constructed by using the 1987/, 1992, and 1997 accountsis around 0.95. Financial Analysts Journal Return Predictability along the Supply Chain Equally Weighted Customer Return. We and higher relationship-specific investments with Notes also repeated our main analysis after constructing their customers. Overall, the lead-lag effect exhib- Customer return by using equally (instead of value-) its characteristics that are consistent with the view 1. For related theoretical research, see, for example, Merton comprises consumption expenditures and investments by weighted portfolios. In untabulated results, we that it is the result of the slow diffusion of value- 1987); Hirshleifer and Teoh (2003); and Peng and Xiong federal, state, and local governments. found that the customer-supplier / lead-lag effect relevant information. (2006) 8. Note that the use table shows the commodities as moving is weaker when using this specification. This find- 2. Related to this literature, some studies have found that directly to consuming industries or final users-that is, ing is unsurprising given that one would expect equity prices do not appear to reflect other value-relevant even if they reach customers by means of wholesalers or the stock prices of large customers to quickly This article benefited tremendously from the support of publicly available information. Hong, Torous, and Val- retailers, which are treated as service industries whose Mark Hooker at State Street Global Advisors. We also primary service is the distribution of goods. If trade were incorporate any new information about their oper- kanov (2007) found that for a significant number of U.S. thank James F. Bryson, Otgontsetseg Erhemjamts, Atul industries (including retail, service, commercial real estate, shown as the buying and reselling of commodities, many ating performance. Gupta, Henry Marigliano, Kartik Raman, Anya Suvo- metal, and petroleum), industry returns forecast the U.S. industries would have wholesalers and retailers as their Conclusion rov, Anand Venkateswaran, Roy Wiggins, Mike Yang, stock market by up to two months. Moreover, the propen- suppliers and customers. This method of treating trade was and seminar participants at State Street Global Advisors sity of an industry to predict the market is correlated with crucial to our study because it enabled us to capture the Recent research in finance has documented a actual consuming industries or final users even if a com- for helpful comments and suggestions. its propensity to forecast various indicators of economic return lead-lag effect along the supply chain activity. Further, Della Vigna and Pollet (2007) used lagged modity was sold through retailers and wholesalers. This article qualifies for 1 CE credit. consumption and demographic data to forecast future con- 9. Using U.S. equities, Cohen and Frazzini (2008) Available at http:/ /mba.tuck.dartmouth.edu/pages/faculty/ sumption demand growth induced by changes in age struc- ken.french/. found that equity returns of customer companies 10. We collected the rates from Kenneth French's website. Grif- lead the stock returns of supplier companies (see ture. They found that forecasted-demand changes 5-10 Appendix A. Percentage of years in the future predict annual industry stock returns. fin (2002) found that using the country-specific estimates also Menzly and Ozbas 2007). Our study contribu For other related empirical research, see Bartov and Bodnar for the risk-free rate results in little difference in the esti- utes to this recent literature by examining whether such a customer-supplier/lead-lag effect is Output Sold to Final Users (1994); Hong, Lim, and Stein (2000); Huberman and Regev mates of abnormal returns. (2001); Hirshleifer, Lim, and Teoh (2004); Hou and 11. Although the L-S portfolio is equally weighted, using a present in other developed markets. Using a sam- For a number of selected industries, Table Al Moskowitz (2005); Della Vigna and Pollet (2005); Hirshleifer weighted average of the transaction costs is appropriate ple of equities listed on the exchanges of 22 devel- shows the percentage of each industry's output that and Teoh (2006); Hou, Peng, and Xiong (2006). because the L-S portfolio has more companies from larger oped countries, we found that equity returns on is used as intermediate input in the production of 3. Of course, the interpretation of return predictability as evi- markets. Thus, we based our weight on the number of customer industries lead the returns of supplier other goods and services, as well as the percentage dence inconsistent with the efficient market hypothesis is companies and not on the total capitalization of each mar- ket. In addition, Domowitz, Glen, and Madhavan (2001) industries. We also showed that this customer- sold to final users. Final users are of three types: subject to many critiques. For instance, some scholars argue Personal Consumption Expenditures, Government that the abnormal returns documented in the literature are estimated transaction costs by using data for September supplier/lead-lag effect is economically signifi- an artifact of the asset pricing model used; the reported 1996-December 1998. Assuming that transaction costs have cant and close in magnitude to that found in U.S. Consumption Expenditures and Gross Invest- abnormal returns should disappear once one uses the declined over time owing to technological advances and equities. In cross-sectional analyses, we found that ments, and Gross Private Fixed Investments. The "right" asset pricing model that fully controls for risk expo- more competitive markets, one can consider our estimate this effect is more pronounced for small suppliers data were taken from the 1987 use table of the sure. Further, some of the anomalies may be the result of of the average of transaction costs to be the upper bound of and for supplier industries with dispersed sales benchmark IO accounts. data mining; researchers consistently search for factors that such costs for our sample period. help predict returns, and by the law of large numbers, some 12. Although our size classifications follow the conventions of of these factors will be correlated with future returns even some investment firms, they remain somewhat ad hoc. Table A1. Selected List of Supplier Industries if no meaningful economic relationship exists. Finally, some Therefore, to check the robustness of our results, we % Used as repeated our analysis after using different size categoriza- Another Significant researchers question the economic significance of some of these anomalies and suggest that they affect only a tiny tion criteria. We found a similar pattern in abnormal returns Intermediate % Sold to Main Customer Customer Industry / Industry SIC Code Final User Industry / Final User Final User segment of the market, which renders them irrelevant. For when we classified stocks on the basis of the median MVE Input ($304 million for the trimmed sample). We also found sim r Homes 3716 10 Personal Consumption na a detailed discussion of these and other issues, see, inter ilar results when we sorted stocks into size quartiles on the Expenditures alia, Fama (1991, 1998) and Fama and French (2007) Small Arms Ammunition 99 4. These markets were included in the MSCI World Index as basis of MVE. 3482 Federal Government na 13. As noted by Williamson (1975) and others, relationship- Motor Vehicles and 3711 Personal Consumption na of June 2006. Owing to the incompleteness of census data on the Swiss specific assets are those that support specific transactions Passenger Car Bodies Expenditures 5. between the company and its stakeholders. The assets are Petroleum Refining 2911 51 Personal Consumption Air Transportation economy, the practice of using the IO accounts of other coun- termed relationship-specific because the value derived Expenditures tries has been followed in Switzerland since the 1970s (for from their use outside the relationship is less than that Aircraft Engines and 3724 62 38 Aircraft Federal Government more details, see www.input-output.ethz.ch/project/index). within the relationship. Engine Parts 6. The Bureau of Economic Analysis of the U.S. Department of Motors and Generators 3621 81 10 14. We used industry-level rather than company-level R&D Gross Private Investments Refrigeration and Heating Commerce publishes the benchmark IO accounts for the U.S. intensity to reduce the effect of data problems on our Equipment economy every five years. The accounts are primarily based Motor Vehicle Parts and 3714 93 7 results. For instance, the R&D value is missing for a signif- Motor Vehicles and Passenger Automotive Repair Shops on data collected from economic ce uses conducted by the icant number of companies in our sample. Further, we Accessories Car Bodies and Services U.S. Census Bureau. The economic censuses provide com- Sheet Metal Work 3444 95 5 believe that the intensity in relationship-specific invest- Motor Vehicle Parts and Industrial and Commercial prehensive data-including information on industry and ments is largely an industry characteristic rather than a Accessories Building Construction commodity production, materials consumed, and operating company one. Therefore, we expect companies that oper- Screw Machine Products 3451 2 Motor Vehicles and Passenger Motor Vehicle Parts and expenses-that are not available on a more frequent basis. ate in R&D-intensive industries to ha
Step by Step Solution
There are 3 Steps involved in it
Step: 1
Get Instant Access with AI-Powered Solutions
See step-by-step solutions with expert insights and AI powered tools for academic success
Step: 2
Step: 3
Ace Your Homework with AI
Get the answers you need in no time with our AI-driven, step-by-step assistance