Industry Concentration and Average Stock Returns

Transcription

1 Industry Concentration and Average Stock Returns Kewei Hou Fisher College of Business The Ohio State University David T. Robinson Fuqua School of Business Duke University This Draft: December 3, 2003 ABSTRACT This paper shows that differences in industry concentration help explain the crosssection of average stock returns. Firms in concentrated industries earn lower returns, even after controlling for size, book-to-market and momentum. The premium for industry concentration exhibits systematic business cycle variation. In addition, the premium on book-to-market is higher in more concentrated industries. Standard explanations based on measurement error, deregulation, capital structure, and correlation with known risk factors do not explain these findings. We hypothesize that this occurs because either (i) barriers to entry in highly concentrated industries insulate firms from aggregate shocks that lead to economic distress, or (ii) firms in highly competitive industries are riskier because they engage in more innovative activities and thus command higher expected returns. Additional tests support both these conjectures. JEL Classification Codes: G12, G33, L10 We thank Anne-Marie Knott and seminar participants at Ohio State for helpful comments. This paper is a significantly revised version of a paper entitled, Market Structure, Firm Size, and Expected Stock Returns. Judith Chevalier, Eugene Fama, Peter Hecht, Owen Lamont, Toby Moskowitz, Per Olsson, Per Strömberg, and seminar participants at Chicago and the Chicago Quantitative Alliance provided many helpful comments on this earlier draft. Any remaining errors are our own. Correspondence to: Kewei Hou, Fisher College of Business, Ohio State University, 2100 Neil Avenue, Columbus, OH hou.28@osu.edu.

2 Industry Concentration and Average Stock Returns ABSTRACT: This paper shows that differences in industry concentration help explain the cross-section of average stock returns. Firms in concentrated industries earn lower returns, even after controlling for size, book-to-market and momentum. The premium for industry concentration exhibits systematic business cycle variation. In addition, the premium on bookto-market is higher in more concentrated industries. Standard explanations based on measurement error, deregulation, capital structure, and correlation with known risk factors do not explain these findings. We hypothesize that this occurs because either (i) barriers to entry in highly concentrated industries insulate firms from aggregate shocks that lead to economic distress, or (ii) firms in highly competitive industries are riskier because they engage in more innovative activities and thus command higher expected returns. Additional tests support both these conjectures.

3 Firms generate cash flows through their actions in product markets. These risky cash flows are in turn priced in financial markets. Yet, the economic link between product markets and asset prices remains relatively unexplored. This paper explores the link between industry concentration and average stock returns, offering the first empirical evidence of the asset pricing implications of industry market structure. The main finding in this paper is that firms in highly concentrated industries earn lower returns, even after controlling for size, book-to-market, momentum, and other known risk characteristics. This is true both at the industry portfolio level, as well as at the individual firm level, and it is robust to alternative empirical specifications. Moreover, the economic magnitude of these effects is large. Our results indicate that firms in the quintile of most competitive industries earn annual returns that are nearly four percent higher than those of similar firms in the quintile of most concentrated industries. This difference is highly statistically significant. Existing asset-pricing theories do not offer explanations for this finding. Thus, our first step is to rule out likely candidate explanations for our results. One potential explanation is that industry characteristics are simply correlated with risk measures that are known to explain the cross-section of stock returns. If this were the case, industry concentration would covary with expected returns without industry concentration playing an independent role in determining asset prices. For example, this would be true if size and book-to-market ratio completely determine the cross-section of expected stock returns, and industry concentration simply described industry-level, cross-sectional differences in those characteristics. Industry concentration is indeed correlated with firm characteristics that are known to describe the cross section of average stock returns: firms in highly concentrated industries have larger market capitalizations and lower book-to-market ratios. However, controlling for these effects, as well as CAPM β and momentum, we still find strong evidence that firms in more concentrated industries earn lower stock returns. The fact that we find a significant 1

4 concentration effect even after controlling for those known determinants of average returns indicates that these correlations in characteristics are not driving our results. A second alternative is that industry concentration affects expected returns through capital structure choice, by increasing or decreasing the amount of leverage that firms in an industry can sustain. We account for this possibility by including leverage in our empirical analysis. The results show that firms in more concentrated industries tend to have lower leverage. But even after controlling for this effect, our industry concentration results still hold. Another potential explanation has to do with regulation. Regulated industries may face lower costs of capital either because they have lower operating risks (due to regulated entry and exit), or because their capital structure and/or capital charges are regulated. If regulation is correlated with industry concentration, then this could explain our findings. To explore this possibility, we have repeated our analysis after excluding regulated industries. Doing this has virtually no impact on our findings. Thus, it seems unlikely that our results are being driven by regulation. Yet another alternative mechanism through which industry concentration could potentially affect stock returns is through information quality. For example, investors might demand lower returns on firms in more concentrated industries because the cost of collecting and processing information is lower in those industries. However, if this were the case, we would expect information transmission to be faster in more concentrated industries. Hou (2003) documents the opposite result: the speed of information diffusion between large and small firms is slower in more concentrated industries, not faster. Thus, although we do not control for this directly, it seems unlikely that our findings are attributable to information quality. A final alternative is that the correlation between concentration and returns is spurious, an artifact of measurement error in the data. To control for the possibility that de-listing is correlated with concentration in a manner that affects returns, we employ the corrections described in Shumway (1997) to account for de-listing bias. Alternatively, international conglomerate firms could be distorting our measures of concentration. To guard against this, we delete such 2

5 firms. Finally, perhaps sales-based concentration figures are suspect, and asset-based concentration measures should be used instead. These specification choices have no effect on our findings. Since this is the first paper to demonstrate a role for industry concentration to affect the cross-section of asset returns, and given existing explanations cannot account for our findings, our next step is to hypothesize an economic explanation for our results. We offer two potential channels through which industry concentration amplifies or diminishes economy-wide risk as it is transmitted to individual firms. One hypothesis is that barriers to entry, which differ at the industry level, insulate firms from aggregate economic shocks. High barriers to entry also lead to high levels of industry concentration. Thus, highly concentrated industries naturally earn lower stock returns than highly competitive ones because their barriers to entry are higher. In other words, the industry concentration spread is correlated with a priced barriers to entry factor in an ICAPM sense. A second hypothesis is that firms in highly competitive industries are more innovative and growth-oriented, and hence more risky. This hypothesis draws on recent work in industrial organization suggesting that rates of innovation are lower in more concentrated industries (Knott and Posen 2003). Implicit in this explanation is the presumption that this innovation risk is not captured by existing asset pricing factors. Although the economic mechanisms differ, these hypotheses make similar predictions for returns data. Using these hypotheses, we explore two new predictions. First we examine the book-to-market spread in returns within industries. We show that the premium associated with the book-to-market ratio is larger in more concentrated industries. Under a barriers to entry interpretation, this suggests that high book-to-market firms are likely to be in greater distress if they are operating in industries dominated by a few firms with big market share, all else equal. Equivalently, this means that conditional on being in distress, distress risk is greater when exit costs are higher. Under an innovation risk interpretation, 3

6 this result suggests that innovative firms have lower expected returns when they are in more concentrated industries, which is consistent with the view that innovation opportunities (or incentives) are lower in concentrated industries. Second, we examine how the concentration premium varies with the business cycle. The premium associated with industry concentration co-moves positively with inflation and the t-bill rate and negatively with the term spread. Since the average value of the premium is negative, this means that the magnitude of the premium is highest at an economic trough, suggesting that the spread in returns between firms that are insulated from economic distress and those that are not is greatest when general economic conditions are weakest. The timeseries results also support an innovation risk interpretation, since the premium is highest when expected growth opportunities, and hence innovation risk, are greatest. Regardless of the precise mechanism linking market structure to expected returns, our findings are important for a number of reasons. First, different industries have experienced dramatically different average returns over time. For example, returns for the drug industry (SIC 283) average 24.3% per year over the period, while steelworks industry (SIC 337) averaged only 8.34% per annum over the same period. Indeed, Fama and French (1997) estimate annual costs of equity from the CAPM and 3-factor model that differ by as much as a factor of two across industries (note that their actual average returns differ by considerably more). Understanding how market structure affects stock returns may ultimately lead to a better understanding of the industry-level variation in expected returns. Second, a growing literature in asset pricing explores the importance of industries for explaining empirical regularities in stock returns. For example, Asness and Stevens (1996) decomposes size and book-to-market factors into inter- and intra-industry components and finds that intra-industry factors have greater explanatory power in the cross-section. Moskowitz and Grinblatt (1999) document industry portfolios exhibit significant momentum. In addition, Hou (2003) finds that the lead-lag effect in stock returns is due to an intra-industry component which also drives the industry momentum effect. However, these papers stop short of testing a 4

7 precise mechanism for why industry differences are important for understanding the behavior of stock returns. Our results offer insight into two such mechanisms. This paper is part of a larger literature that links industrial organization to issues in financial economics. Earlier work such as Titman (1984) studies how capital structure and product markets interact through the liquidation decision. A number of recent papers have examined the link between capital structure and industry characteristics; see, for example, Mackay and Phillips (2002) or Almazan and Molino (2001). To our knowledge, however, ours is the first paper to link expected stock returns to industry product-market characteristics through the channel we propose. The remainder of the paper is structured as follows. Section I describes the data and how we construct industry concentration measures. In addition, this section illustrates the relation between industry concentration and industry-level characteristics. Section II examines how industry concentration affects the cross-section of stock returns. Given the inability of existing explanations to account for our findings, we devote Section III to a discussion of our hypotheses linking industry concentration to stock returns. Section IV studies the interaction between industry concentration and firm characteristics, while Section V considers business cycle variation in concentration premia. Section VI concludes. I. Data and Measures of Industry Concentration A. Sample Selection Our sample includes all NYSE/AMEX/NASDAQ listed securities with sharecodes 10 or 11 (e.g., excluding ADR s, closed-end funds, REIT s) that are contained in the intersection of the CRSP monthly returns file and the COMPUSTAT industrial annual file between January, 1973 and December, Our main reason for focusing on the post-1973 period is to ensure 5

8 that we have the widest possible industry coverage. Prior to 1973, the CRSP sample includes NYSE and AMEX firms only; NASDAQ firms are not added to the sample until To ensure that accounting information is already impounded into stock prices, we match CRSP stock return data from July of year t to June of year t + 1 with accounting information for fiscal year ending in year t 1 as in Fama and French (1992a). In order to be included in our return tests, a firm must have CRSP stock price, shares outstanding and 3-digit SIC classification for June of year t. 1 It must also have returns data from the previous three years for market β estimates. In addition, it must have COMPUSTAT data on sales, book equity, market equity and total assets for its fiscal year ending in year t 1. This data requirement probably biases our sample towards larger firms, which in turn diminishes the overall variation in the concentration measures. Thus, if anything, the results in this paper probably understate the true relation between industry concentration and the source of risk that is impounded in stock returns. Book equity is stockholder s equity plus balance sheet deferred taxes and investment tax credit minus the book value of preferred stock. COMPUSTAT market equity is stock price times shares outstanding at fiscal year end. The book-to-market ratio is calculated by dividing book equity by COMPUSTAT market equity. Leverage is defined as the ratio of book liabilities (total assets minus book equity) to total market value of firm (COMPUSTAT market equity plus total assets minus book equity). Size (CRSP market equity) is measured by multiplying shares outstanding by stock price for June of year t. We follow Fama and French (1992a) to estimate market β by computing full-period βs for portfolios sorted by size and pre-ranking β and then assigning portfolios βs to stocks in those portfolios. The pre-ranking β is estimated as the sum of the coefficients of regressions of individual stock returns on contemporaneous and lagged market returns over the past three years. Finally, return series are adjusted for delisting bias following Shumway (1997). This addresses the possibility that our results are driven by 1 Kahle and Walkling (1996) report problems between CRSP and COMPUSTAT with regard to SIC industry classifications. To minimize any impact this may have on our results, and to maintain internal consistency with our variable construction, we disregard COMPUSTAT SIC classifications. 6

9 measurement error arising from a correlation between de-listing and industry concentration that has no impact on stock returns. Throughout the paper, we use three-digit SIC classifications to define industry membership. This choice reflects the desire to balance two offsetting concerns. On the one hand, we wish to use fine-grained industry classifications so that firms in unrelated lines of business are not grouped together. On the other hand, using too fine an industry classification results in portfolios that are statistically unreliable and firms being grouped into distinct industries arbitrarily. Choosing 3-digit classifications strikes a balance between these two concerns. Although all of the results in the paper are presented with 3-digit SIC classifications, we have also replicated our main findings for 2- and 4-digit SIC classifications. 2 Our main results are not sensitive to the choice of two-, three-, or four-digit SIC classification. B. Measuring Industry Concentration We measure industry concentration using sales-based and asset-based Herfindahl indexes. The Herfindahl index for industry j is defined as follows: Herfindahl j = I i=1 s 2 i j, (1) where s i j is the market share of firm i in industry j. Thus, the Herfindahl measure uses the entire distribution of industry market share information to obtain a complete picture of industry concentration. Small values of the Herfindahl index imply that the market is shared by many competing firms with none owning a very large chunk of the market, while large values imply that market share is concentrated in the hands of a few, large firms. To ensure that our results are robust to alternative specifications of industry concentration, we use both sales and assets to calculate market share. 3 To calculate Herfindahl ratios for the industries in our sample, we 2 These results are available from the authors upon request, but are omitted here for the sake of brevity. 3 In the appendix, we present further robustness checks using the 5-firm ratio, which is the fraction of total market share owned by the five largest firms in the industry. None of our findings hinges on our choice of 7

10 perform the above calculations each year for each industry, and then average the values over the past three years. This ensures that potential data errors in sales or assets do not have undue influence on our measures. It is important to point out that our use of the Herfindahl index differs from prior applications in corporate finance. In particular, many papers studying the diversification discount use the Herfindahl index to measure the degree to which internal investment opportunities of a single firm are spread across many projects (diversified firms) or only a few projects (focused firms). For example, see Berger and Ofek (1995), Rajan, Servaes, and Zingales (2000) or many others. Instead, our measure captures market shares across all firms in a given 3-digit industry. One concern with using Compustat data to generate measures of industry concentration is that we do not have information on privately held firms. In principal, privately held firms are included on Compustat if they have engaged in public debt issues, but in practice, the number of such firms is small. While we recognize that this may be a shortcoming of our measures, we think that this shortcoming only makes it more difficult to establish our results. To the extent that privately held firms tend to be smaller relative to their publicly traded peers, omitting them probably has only a small effect on the herfindahl index. The fact that our results are robust to different measures of industry concentration and different industry classifications suggests that this is unlikely to be a problem for our findings. Panel A of Table I reports summary statistics for the two industry concentration measures. H(Sales) is the sales-based Herfindahl index, while H(Assets) is the measure using asset shares. H(Sales) has a mean of 0.52, a median of 0.46 and a standard deviation of The asset-based Herfindahl measures are highly correlated with the sales-based figures. H(Assets) has a mean value of.53, with a standard deviation of.30; this is nearly identical to the distribution of H(Sales). industry concentration measure. However, Schmalensee (1977) argues that the Herfindahl statistic is the preferred measure of industry concentration in empirical settings, therefore we focus on it throughout the discussion. 8

11 C. Correlations In Panel B, we report industry characteristics averaged across concentration quintiles. The spread in H(Sales) is large: the most competitive quintile has an average H(Sales) of.145, while the most concentrated quintile has an average of.985. Industry median sales and median size is significantly larger for the most concentrated quintiles, but the average industry β is roughly flat across the quintiles. The number of new listings and the number of de-listings is significantly higher in the quintile of most competitive industries, which suggests that barriers to entry are higher in more concentrated industries. Finally, we examine how industry concentration covaries with other industry-level characteristics. This is presented in Panel C, where we report Fama and MacBeth (1973, henceforth FM) regressions of the cross-section of industry concentration measures on industry average characteristics. We estimate equations of the following form: H jt = α t + N n=1 λ nt X jt + ε jt (2) where the H jt are industry concentration measures and the X jt are industry average characteristics. Regressions are run every year from 1973 to 2001, and the time-series means of annual cross-sectional coefficient estimates are reported along with the time-series t-statistics. This procedure allows for multivariate correlation analysis, and it is robust to cross-correlated error terms. The resulting coefficients can then be interpreted as simple or conditional correlations between concentration and industry-average characteristics, and appropriate statistical inferences can be drawn about the magnitude of these relations. The row described as simple reports results from FM regressions of concentration on each characteristic in isolation. (Thus, there are eight separate sets of regressions reported.) The row described as multiple reports a single set of regressions in which all eight characteristics are included as independent variables. 9

12 Industry mean log assets is negatively correlated with industry concentration, which suggests that firms are smaller in more concentrated industries. The same holds true for average log market equity. This would seem to contradict the results of Panel B, but it largely reflects skewness in the size distribution of firms within an industry. The correlation between industry concentration and average minimum log assets, however, is.06 and highly statistically significant (not reported); this positive loading is consistent with the idea that high industry concentration reflects higher barriers to entry, which we explore in detail below. Measures of profitability are positively correlated with industry concentration. Earnings to sales ratios are highly positively correlated with industry concentration, both unconditionally, and conditional on other industry characteristics. Combine this with the fact that sales to asset ratios are lower for concentrated industries, and the picture that emerges is concentrated industries are ones with large asset bases and high unit profitability. This description is consistent with the IO literature that we draw on in Section III below. Our measure of Q is simply market value of assets over book value of assets. The positive correlation between Q and industry concentration suggests that firms high industry concentration is associated with positive expected future economic profits, which again is consistent with the view that industry concentration is an indirect measure of barriers to entry. Likewise, the book-to-market ratio is negatively correlated with industry concentration at the industry level, suggesting that discount rates are potentially lower for firms in concentrated industries. II. Industry Concentration and the Cross-Section of Returns A. The Concentration Spread Table II relates industry concentration to the cross-section of average stock returns, measured both at the industry and firm level. In June of each year, industries are sorted into quintiles based on their Herfindahl index. We then report average monthly returns and t-statistics for 10

13 these portfolios, as well as the difference between Quintile 5 (most concentrated) and 1 (least concentrated). Panel A reports returns for Herfindahl quintiles, while Panel B presents a variety of robustness checks. The first row in Panel A presents raw average returns computed by equally weighting firms within each concentration portfolio. Looking across Herfindahl quintiles, firms in the least concentrated (most competitive) industries earn an average return of 1.558% per month. This declines to 1.262% per month for firms in the most concentrated quintile. The spread between the two is % per month, which is statistically different from zero at the 5% level. Since Table I shows that industry concentration is associated with a number of known determinants of average returns, we also report characteristics-adjusted returns. We use the procedure in Daniel, Grinblatt, Titman, and Wermers (1997) to adjust individual stock returns for size, book-to-market, and momentum. All firms in our sample are first sorted each month into size (CRSP market capitalization) quintiles, and then within each size quintile, further sorted into book-to-market quintiles. Within each of these 25 portfolios, firms are again sorted into quintiles based on the firm s past 12-month return, skipping the most recent month. Stocks are averaged within each of these 125 portfolios to form a benchmark that is subtracted from each individual stock s return. The expected value of this excess return is zero if size, bookto-market and past one-year return completely described the cross-section of expected returns. The characteristic-adjusted average returns of the above quintile portfolios as well as the average spread between Quintile 5 and 1 are reported in the second row of Panel A. Even after adjusting for these known premia, we still see a significant spread in average returns across concentration quintiles. Interestingly, adjusted returns for the quintile of the most competitive industries (Q1) are positive and statistically significant, and they decrease to negative and statistically significant for the quintile of most concentrated industries (Q5). However, the adjustments have little or no effect on the total spread in returns. The spread for Herfindahl index is statistically significant at % per month, only three basis points lower than the 11

14 raw return figure. Together, this suggests that the return premium associated with industry concentration is independent from those of size, book-to-market and momentum, and that controlling for industry concentration is important for understanding the cross-section of stock returns. Since concentration is an attribute of an industry, not a firm, there is flexibility in how quintile returns are measured. The third and fourth rows in Panel A report raw and adjusted returns calculated by first forming industry portfolios, and then equally weighting industry returns within each concentration quintile. These industry-level returns mirror the firm-level results. In each case, we see a large and statistically significant spread between the most concentrated and the most competitive quintile. The spread ranges from % to % per month depending on whether raw or characteristics-adjusted returns are reported. In Panel B, we take a number of steps to control for a variety of potential explanations for our result. The first two sets of rows report average returns that have not been corrected for the Shumway (1997) de-listing bias. The first of these reports firm-level characteristics-adjusted returns; the second reports returns formed by the industry portfolio scheme described above. Neither of these adjustments affects our results, which suggests that our findings are not being driven by spurious correlation in the data arising from the way in which returns are recorded on CRSP. The next set of rows repeats the returns spread calculations but omits regulated industries. The industries are taken from Barclay and Smith (1995). If anything, this strengthens our results. Thus, it seems unlikely that our findings are being generated by regulation. The final set of rows reports returns spreads according to H(Assets) instead of H(Sales). Since the two variables are so highly correlated, it is no surprise that these spreads are also significant. But, given the 5-firm spreads that we report in the appendix, these H(Sales) spreads illustrate that our results are not sensitive to the choice of industry concentration measure. 12

15 B. Industry Concentration and Industry Portfolio Returns To further examine the relation between industry concentration and average stock returns, we conduct Fama and MacBeth (1973) regressions of monthly stock returns on industry concentration measures and other characteristics. In Table III, we report regressions of industry portfolio returns regressed on industry characteristics and industry concentration measures. More specifically, for each month t, we estimate cross-sectional regressions of the form: R jt = α t + N n=1 λ nt X jnt + ε jt (3) where R jt is the return of industry j in month t, X jnt (from 1 to N) are industry-level characteristics of industry j. The time series average of the cross-sectional regression loadings λ nt is reported along with its time-series t-statistic. The interpretation of the regression coefficient is that it is the return to a zero-cost portfolio with the weighted characteristic equal to one on the corresponding regressor and zero on all other regressors (Fama 1976). These regressions provide robustness check of the relationship between industry concentration and average returns without imposing quintile breakpoints and allow us to control for additional alternative explanations. The first two rows of Table III show that more concentrated industries earn lower average returns, consistent with our previous results from quintile portfolios. The cross-sectional regression coefficient on the H(Sales) index is negative and statistically significant at the 5% level. The next five rows demonstrate that industry average returns are positively related to industry mean book-to-market ratio, leverage and past one year s industry return, and insignificantly related to industry average size and market β. Finally, the last row shows that controlling for these variables does not drive out the significance of the industry concentration effect. Thus, while the results of Table I suggest that industry concentration is correlated with other industry characteristics that describe average returns, the results from Table III suggest 13

16 that those correlations are not the driving forces behind the inverse relationship between industry concentration and average stock returns. C. Industry Concentration and Individual Stock Returns Table IV repeats the analysis described above, but replaces industry portfolio returns with firm-level stock returns, and replaces industry characteristics with firm-level measures of size, book-to-market, leverage, and market β. The first two rows of Table IV show that the degree of concentration of the industries to which individual firms belong is important even for understanding the cross-section of firm-level returns. Fama and MacBeth (1973) regressions of individual stock returns on Herfindahl index alone produce an average slope coefficient of % with a t-statistic of The next four rows confirm the standard results found in the literature that average individual stock returns are negatively related to size, positively related to book-to-market ratio and past one year return, and once they are controlled for, leverage and market β are not priced in the cross-section. Accounting for the premia associated with these variables does not alter our results on industry concentration. On the contrary, it enhances the results. Introducing size, book-to-market ratio, past one year return, leverage and market β to the cross-sectional regressions raise both the point estimates as well as the t-statistics for industry concentration. The conclusion to take away from this section is that not only do industry returns vary with industry concentration, but individual stock returns do as well: firms in concentrated industries earn lower stock returns than firms in more competitive industries. The results hold under a variety of different empirical strategies, and are robust to whether or not we control for characteristics such as size, book-to-market, and past returns, both at the firm and industry levels. (As we show in the appendix, these findings also hold for alternative concentration measures.) These controls suggest that the industry concentration effect we have identified are not being driven by correlations with other determinants of expected returns, or through capital structure choice. 14

17 III. The Link Between Market Structure and Stock Returns The previous sections have shown both that industry concentration measure comove with characteristics known to explain stock returns, and more importantly that stock returns covary with industry concentration measures above and beyond that which can be explained by the correlation in these characteristics. The purpose of this section is to develop a hypothesis that not only is consistent with the previous findings, but that offers additional empirical tests that go beyond what we have presented thus far. While the industrial organization literature does not establish a direct connection between market structure and asset returns, it has identified two channels through which industry concentration could potentially affect stock returns. One is through the Schumpeterian hypothesis, linking growth and innovation (and hence risk and return) to the distribution of firms in an industry. The second is through barriers to entry. A. Innovation Risk and Stock Returns Schumpeter is associated with two influential views of the link between market structure and innovation. His later view, the Schumpeterian hypothesis (Schumpeter 1942), argues that monopolistic firms have stronger incentives to innovate than firms in competitive industries, since monopolistic firms can realize the economic profits arising from their innovation, rather than have their super-normal profits competed away. (They also can afford larger R&D outlays.) In terms of our analysis, this hypothesis predicts that firms in more concentrated industries should earn higher expected returns to compensate investors for these risky innovative activities. This is flatly contradicted by our findings, since we show that more competitive industries earn higher average returns. This hypothesis has also received criticism in recent industrial organization literature. Work by Geroski (1990) finds evidence against the hypothesis that competitive rivalry di- 15

18 minishes innovation. Reinganum (1985) models an industry with a single incumbent and multiple challengers and shows that the challengers have stronger innovation incentives, suggesting that the level of innovation varies non-monotonically with the number of firms in the industry. Knott and Posen (2003) show that innovation is increasing in the degree of industry competition. The evidence described above indicates that innovative activities cluster in competitive industries. This is consistent with the idea of creative destruction, the earlier of Schumpeter s two hypotheses (Schumpeter 1912), which argues that innovation occurs in small firms on the fringes of established industry, and that these small challengers ultimately overturn the existing status quo and usher in a new paradigm. 4 Inherent in this view is that innovation involves risktaking. Thus, one hypothesis is the innovation risk hypothesis. Under this interpretation, our main results suggest that the concentration spread reflects the higher returns for the additional risk associated with increased innovation that occurs in more competitive industries. If innovation risk is driving our results, then the concentration spread should be correlated with business conditions, since rates of innovation vary across the business cycle. In particular, the concentration spread should be highest when expected innovation risk is the greatest, which should occur at an economic trough, when expected future business conditions are improving. Likewise, if we interpret the book-to-market ratio as measuring value or growth, then the innovation risk story makes predictions for intra-industry spreads in book-to-market as well. If there are fewer innovative opportunities in highly concentrated industries, then we should expect growth firms (low book-to-market firms) to earn lower expected returns in these industries. In other words, value firms should outperform growth firms by an even wider margin in settings in which growth opportunities are lower, since here the expected returns 4 See also He, Morck, and Yeung (2003) for evidence relating turnover in firm dominance to economic growth on an international scale. 16

19 for growth firms should relatively lower. Under an innovation risk interpretation, these lower opportunities are found in more concentrated industries. B. Risk, Return, and Barriers to Entry Early work in industrial organization originating with Bain (1954) established an influential paradigm linking the exogenous production characteristics of an industry to firms pricing behavior, which in turn determined firm performance. This is often called the Structure/Conduct/Performance (S/C/P) paradigm. The observational starting point for the S/C/P paradigm is the nature of the production technology in an industry, which is taken to be exogenous. For example, the chip manufacturing industry has high fixed costs, since large, expensive plants must be built and customized to each new chip that is designed. The S/C/P paradigm would view this high fixed costs as a natural barrier that restricted competitive entry (structure). Since entry to this industry would be limited, the number of incumbent firms would be few, and each would be able to price significantly above marginal cost without fear of arousing entry (conduct). As a result, firms in this industry would earn super-normal economic profits (performance). Alternatively, consider the internet boom of the late 1990s. An S/C/P interpretation of the 90 s Internet technology boom would work as follows. Entry costs were low, as was witnessed by the rush of firms onto the internet. Since entry was virtually costless, it was necessary to price at or below marginal cost to attract business. As a result, most firms fared poorly, and were unable to meet long-run fixed costs, prompting exit. The S/C/P paradigm suggests that barriers to entry will affect expected returns whenever differences in the number of competitors in an industry, or in the pricing practices they observe, change the risk characteristics of firms in question. Since barriers to entry affect entry and exit, they can potentially affect stock returns through distress risk. This can have both a cross-sectional and a business cycle effect. 17

20 The cross-sectional effect of barriers to entry has implications for inter-industry as well as intra-industry comparisons of stock returns. Looking across industries, we would expect firms in high barriers to entry industries to have lower expected returns since the average distress risk would be lower in these industries. This predicts the negative correlation between industry concentration and stock returns, as we have already documented in Tables II and III. However, conditional on observing a firm in distress, we would expect the distress premium to be in higher in a high barriers to entry industry, since this is where the costs of entry and exit are the highest. The business cycle effect of barriers to entry comes into play when we consider how entry and exit vary over the business cycle. For instance, suppose industries with low barriers to entry can sustain high competition during times of economic prosperity, but in times of economic decline, this competition induces exit. Then, the underlying source of risk is economic fluctuation, but industry concentration determines how sensitive different firms are to this risk. By keeping out competitors, the same barriers to entry that lead to high industry concentration cushion that industry s firms from the effects of aggregate shocks by lowering the probability of economic distress in times of economic downturn. The business cycle effect predicts that the spread in returns between firms in high and low barriers to entry industries is greatest when expected business conditions are poor. The crosssectional effect predicts that returns covary negatively with barriers to entry across industries, but conditional on being in distress, a firm s distress premium is higher in high barriers to entry industries, since the costs of entry and exit are higher. To test these predictions, one empirical approach would be to measure barriers to entry directly and relate them with stock returns. However, recent work in industrial organization has focused on the fact that barriers to entry reflect the strategic choices of incumbent firms in addition to the inherent production characteristics of the industry. This is illustrated in a large body of work including Schmalensee (1978), Salop (1979), Schmalensee (1981), Sutton (1991), and Sutton (1998). The fact that barriers to entry reflect strategic choices of incumbent 18

21 firms as well as the primitives of industry production technology makes it impractical to stock returns directly to barriers to entry. In particular, the strategic nature of barriers to entry not only makes them difficult to measure, but introduces potential endogeneity with stock returns. Instead, we focus on industry concentration as a measure of barriers to entry, since it is a natural consequence of barriers to entry, regardless of the manner in which they came to be. As we discussed in Section I, the positive correlation between industry concentration and minimum firm size supports the use of industry concentration as a proxy for barriers to entry. We hypothesize that firms in highly concentrated industries earn lower returns because, all else equal, they are better insulated from undiversifiable, aggregate shocks. C. Predictions To summarize this section, we offer two new predictions linking concentration to stock returns. These are as follows: Prediction 1 (Intra-Industry) Firms in high concentration industries earn lower average stock returns. However, the spread in returns associated with book-to-market is highest in concentrated industries. Under an innovation risk interpretation, Prediction 1 stems from the fact that the spread between value and growth is larger in concentrated industries. This occurs because innovation opportunities are lower in concentrated industries, and therefore the required returns for growth firms are lower as a result. On the other hand, if industry concentration is related to distress, as is true under a barriers to entry interpretation, then we expect high book-to-market firms to carry a higher premium in more concentrated industries. This coincides with Fama and French (1992b), who interpret book-to-market ratio as a risk proxy related to relative distress in a multi-factor ICAPM or APT framework (see Merton (1973) or Ross (1976)). Chan and Chen (1991) also propose 19

22 the the existence of a risk factor in returns and expected returns that is related to relative distress. This interpretation is further supported by the evidence in Fama and French (1995) that that low book-to-market firms have persistently strong fundamental performance, while high book-to-market firms have persistently weak performance. Chen and Zhang (1998) show that high book-to-market firms are associated with poorer earnings performance, higher levels of earnings uncertainty, leverage, and dividend cut or omission. Using data from ten developed countries, Liew and Vassalou (2000) find that HML (a factor-mimicking portfolio that is long on high book-to-market stocks and short on low book-to-market stocks) contains important information regarding future GDP growth. Shumway (1996) also documents a positive premium for distressed firms, although he shows that book-to-market is only weakly correlated with his firm-specific proxy for default probability. Griffin and Lemmon (2002) find that the book-to-market premium is higher among firms with the highest level of distress risk proxied by Ohlson s O-score (Ohlson (1980), Dichev (1998)). Thus, under a barriers to entry interpretation, Prediction 1 reflects the fact that distress is more costly in concentrated industries. Conditional on the level of distress, firms in more concentrated industries have higher distress premia than firms in low concentration industries. Comparing the two interpretations, the innovation risk interpretation attributes the increase in the value spread to lower expected returns among growth firms in concentrated industries, while the barriers to entry interpretation attributes this to increasing required returns to distressed firms. Prediction 2 (Business Cycle) The concentration spread is highest at an economic trough, and lowest at an economic peak. Under a barriers-to-entry interpretation, Prediction 2 occurs because current distress risk is highest at an economic trough. This is also when expected future innovation risk is highest, which is consistent with an innovation risk interpretation. In the remainder of the paper, we offer empirical evidence on these two predictions. 20

23 IV. Industry Concentration and Firm Characteristics The results thus far have shown an economically important and statistically significant link between industry concentration and average stock returns. This section explores the interaction between concentration and book-to-market to test Prediction 1. We explore Prediction 1 by first including an interaction term between H(Sales) and the book-to-market ratio in the firm-level Fama and MacBeth (1973) cross-sectional regressions. The last row of Table IV report the estimation results. The coefficients on the interaction terms are positive and statistically significant, suggesting that the premium associated with being a high book-to-market firm grows as industry concentration increases. This lends support to the idea that book-to-market ratio is related to distress risk and industry concentration is a mechanism through which this risk is propagated through the economy. At the same time, it is consistent with the view that growth firms underperform value firms by relatively more in concentrated industries, which is consistent with the innovation risk interpretation. To get a sense of the economic magnitude involved here, next we examine returns to bookto-market portfolios for different levels of industry concentration. This also allows us to distinguish the two explanations, since they have different predictions for which type firm is responsible for the increasing book-to-market spread. The innovation risk story predicts that decreases in the returns to growth firms are responsible; the barriers to entry story predicts that increases in value firms are driving the result. In June of each year, we sort industries into concentration quintiles according to their H(Sales) value. Then firms within each concentration quintile are further sorted into five portfolios according to their book-to-market ratio. The equal-weighted returns on these doublesorted portfolios are calculated over the following years from July to June. Table V reports the average monthly returns of the five book-to-market portfolios as well as the difference in returns between quintile 5 and 1 for each H(Sales) group. Each row demonstrates the prevalence of the book-to-market effect within each concentration group. 21

24 As the table indicates, the spread in returns associated with book-to-market ratio is the largest among the most concentrated industries. For example, high book-to-market stocks outperform low book-to-market stocks by 1.072% per month in the lowest Herfindahl quintile, and this number grows to 1.814% per month for the highest Herfindahl quintile. These doublesorted portfolio results reinforce the findings from the cross-sectional regressions, which show that the book-to-market premium grows as industry concentration increases. This table also suggests that the innovation risk story may do a better job at explaining the relation between industry concentration and stock returns, since the increase in the spread is entirely attributable to the lower average returns of the first book-to-market quintile. V. Time Series Variation of Industry Concentration Premium This section links changes in the premium associated with industry concentration to various risk factors and business cycle indicators. This allows us to revisit the question of whether the concentration premium remains significant after controlling for existing risk factors, and also enables us to examine Prediction 2 by tracking the concentration premium throughout the business cycle. In Table VI, we report results from the following time-series regressions of monthly concentration premia on risk factors and economic indicators: λ H t = α + I i=1 β i F it + J γ j X jt + ε t, (4) j=1 where F it are returns to the factor-mimicking portfolios in month t, and X jt are month t values of the business cycle indicators. The dependent variable, λ H t, is the time-series of H(Sales) risk premia generated from the Fama Macbeth regressions reported in Table IV. 22