Industry Concentration and Stock Returns: Australian Evidence

Transcription

1 Industry Concentration and Stock Returns: Australian Evidence David Gallagher Katja Ignatieva This version: June 3, 2010 Abstract This paper examines economic determinants of the cross-sectional stock returns on the Australian stock market. We argue that not only the standard risk factors such as the size of the company, or the book-to-market value affect the average stock returns, but also the structure of the product market itself. Motivated by the study of Hou and Robinson (2006), we address the issue of competition on the Australian stock market, comparing it to the US stock market. Given country-specific geographical and political features, as well as the Australian government policies, the structure of the Australian market is very distinct from the US market structure. In contrast to the large open US economy with a lot of competition, Australia is a relatively small open-economy, its market structure resembles many firms which either singularly or jointly dominate their own industries. The differences in the market structure lead to the different results when explaining the cross-section of stock returns. We find that the average stock returns on the Australian market are positively related to the size of the firm and negatively related to the book-to-market ratio. These are the opposite effects to those reported in Fama and French (1992) for the US market. Australian concentrated industries are dominated by the large companies with a high market power. In contrast to the US stock market, we find a significant evidence that the companies operating in higher concentrated industries generate higher risk-adjusted returns than the companies operating in lower concentrated industries. In addition, we find an interesting evidence regarding interaction between the size premium and the concentration premium: While increasing concentration indicates lower average stock returns for the large companies, it leads to an increase in average returns for the small companies. Keywords: Industry concentration, average stock return, Herfindahl index, size, book-tomarket, momentum, concentrated, competitive industry. JEL: AMS Subject Classification: School of Finance and Economics, University of Technology, Sydney, PO Box 123, Broadway, NSW 2007, Australia. David.Gallagher@uts.edu.au House of Finance, Goethe University, D Frankfurt am Main, Germany AND Faculty of Business and Economics, Department of Economics, Macquarie University, Sydney, Building E4B, Eastern Road North Ryde, NSW 2113 Australia. ignatieva@finance.uni-frankfurt.de/katja.ignatieva@mq.edu.au

2 1 Introduction Explaining average stock returns is a complicated process. Starting with an asset pricing models of Sharpe (1964), Lintner (1965) and Black (1972), the typical approach has been to consider various firm-specific characteristics which proxy various risks embodied in the expected returns. The most established sources of risk include the size of the company, or its market value first documented in Stattman (1980) and the book-to-market ratio, which is the ratio of book value of equity (total assets minus total liabilities) reported in the balance sheets to the market value of the company (share price times the number of shares outstanding). Further risk factors include the leverage and the earning-to-price E/P ratio. Using monthly US data of the CRSP NYSE returns, several studies document a significant relation between security returns and risk factors. For example, Banz (1981) and Rosenberg, Reid and Lanstein (1985) document a strong negative relation between the average return and the firm size. Fama and French (1992) point out a significant relation between firm size, book-to-market and stock returns. Bhandari (1988) finds positive relation between the leverage and the average stock returns. Ball (1978), Basu (1983) and Keim (1988) report positive relation of stock returns to the E/P ratio. Other firm attributes such as a dividend yield, exhibit significant correlation with average stock returns as reported in Litzenberger and Ramaswamy (1979). The idea that these risk factors can capture different types of risks such as distress of a firm, financial risk and cash flow riskiness is documented in Fama and French (1992), Daniel and Titman (1997), Davis, Fama and French (2000) and Griffin and Lemon (2000). In addition to the risk factors mentioned above, there are a number of other reasons related to the structure of the product market which may affect the average stock returns. For example, Schumpeter (1912) argues that innovation as a source of risk is more likely to occur in competitive industries, and it leads to the higher stock returns. Distress as reported in e.g. Hou and Robinson (2006) is another form of risk. Industries with high barriers to entry insulate some firms from the demand shocks while exposing others, and thus, will vary with the market structure. Thereby, the national competition policy in markets for products and services remains a significantly important area of economic policy. Competition policy relies on active monitoring and enforcement, and exhibits much public debate amongst regulators, consumers, unions, business and government. Competition policy is critical to a national economy, as it is considered an important driver in achieving the economic goals of productivity and efficiency, and serving as a catalyst for sustained economic growth and development. Competition policy is also important as a mechanism for promoting innovation in both product and services markets. Many studies have considered the linkage between stock returns and the extent to which competition (or concentration) in a market is related. Companies that dominate their industries through high industry concentration might be expected to have greater market power, higher pricing power, and exhibit greater barriers to entry for other firms. Schumpeter (1912) promotes the view that innovation in markets alters the structure of markets, in the sense that competitive pressures within an industry could lead to higher returns. However, a counterview is that firms operating in industries with higher barriers to entry are insulated from such competitive pressures, leading to lower levels of innovation, and therefore lower future stock returns. An important paper based on US data by Hou and Robinson (2006) considers an empirical nexus 1

3 between industry concentration and average stock returns, and their work identifies that firms operating in higher concentrated industries earn significantly lower risk-adjusted returns (controlling for book-to-market, size and momentum). Indeed, their results show that the economic magnitude between high and low concentrated industries is approximately 4% per annum. The differences in the average stock returns reported is Hou and Robinson (2006) are due to the risks which industry concentration proxies. Hou and Robinson (2006) show that concentration premium contains independent information about the cross section of average stock returns which is not embodied in the Fama-French factors such as size and book-to-market ratio. In this study, we provide the first out-of-sample evidence from the US applied to the Australian market. The Australian market is interesting in a number of respects. First, Australia is a relatively small open-economy, which in many industries exhibits market structures that are very different to the US and other developed economies. Given Australia s geographical position, small population, and large island-based continent, the structure of the market resembles many firms which either singularly or jointly dominate their own industries (for example, media, transport, banking, resources, and retail). In other words, there are many instances where one or a few firms either operate, or appear to operate, as monopolies or duopolies within their industry. Second, successive governments since the early 1990s have been exceedingly pro-active in their engagement and pursuit of robust competition policy outcomes. These commenced with the 1993 Hilmer Report 1, and led to a number of market-based reforms and changes in the Trade Practices Act (1974) (governing competition) as well as an overhaul in the late 1990s of the regulator (now known as the Australian Competition and Consumer Commission (ACCC)). These government policy changes have largely been aimed at improving competition within the Australian market. Third, there exists much debate by market participants concerning the extent of competition reform and the ability of the ACCC to effectively promote competition in a manner that leads to genuine market-based outcomes in the best interests of consumers. Consistent with our hypotheses concerning differences between the Australian and US markets in terms of market structure and industry concentration, we find significant evidence that companies operating in higher concentrated industries generate higher risk-adjusted returns. The spread between the most concentrated quintile and the most competitive quintile is 0.49% per month corresponding to the economic magnitude between high and low concentrated industries of approximately 5.8% per annum. Furthermore, we find that the average stock returns are positively related to the size of the company and negatively related to the book-to-market ratio. In addition, we find an interesting evidence regarding interaction between the size premium and the concentration. More precisely, we observe differences in stock returns depending on the market structure: (1) for the small companies, average stock returns increase with increasing concentration whereas (2) concentration tends to decrease the average stock returns with increasing size. In other words, increasing concentration indicates lower average stock returns for the larger companies, but it leads to an increase in average returns for the smaller companies. 1 The National Competition Policy Report released in 1993 was commissioned by the prime minister and undertaken by a committee chaired by professor Fred Hilmer - argues for greater competition among governmentowned entities, the removal of interstate barriers to trade in energies such as electricity and gas, the elimination of duopoly, e.g., in telecommunications, and the abolition of monopoly practices, e.g., in Australia Post over the delivery of mail, as well as among legal, medical and other sectors. 2

4 The paper is structured as follows. In Section 2 we provide a brief review on measuring concentration (our proxy for competition), as well as descriptive statistics concerning the structure of the Australian Stock Exchange and the extent to which one or a few firms dominate their industries. This is followed by our research design in measuring concentration. Section 3 relates industry concentration to the cross-section of average stock returns. Section 4 analyzes the factors which drive the differences in average returns across concentration quintiles. In particular, we study the differences in average returns occurring due to the cash-flow shocks. Further, Section 5 relates concentration premium to the various risk factors and business cycle indicators. We study the relationship between cross-sectional size and book-to-market factor premia and concentration premium in Section 6. The final Section 7 concludes and suggests areas for future research. 2 Data and Measures of Industry Concentration 2.1 Measuring Industry Concentration Herfindahl Index is applied in Industrial Organization to measure concentration of companies in an industry. The Herfindahl index relates the size of the company to the size of the industry which this company belongs, and thus, can be thought of as a measure of industry concentration, or an indicator of competitiveness among the companies. For each industry j the Herfindahl index is computed as: I H j = s 2 ij, (1) i=1 where s ij denotes the market share of company i in the industry j, calculated based on net-sales, total-assets, or book value of equity. This leads to three different types of the index denoted respectively as, H(Assets) and H(Equity). Clearly, small values of H j indicate that the market is shared by many competing companies, whereas higher values imply that several large firms operate in the industry j. Following the approach by Hou and Robinson (2006), we perform calculations of the concentration measures every calendar year for each industry, and then average these over the past three years to make the results more robust against large fluctuations in Herfindahl. 2.2 Data For our analysis we use historical Australian Stock Exchange (ASX) data, the Australian School of Business Share Price & Price Relative (SPPR) database, which contains historical records of share prices and calculated price relatives of all Australian listed companies with fully paid shares. The SPPR covers the period from December 1973 to December For each company, the share price, the number of shares outstanding and the price relatives are provided by SPPR on a monthly basis, together with company code, industry classification code and trading information. Industry classification is based on the current (as at December 2008) Standard & Poors Global Industry Classification System (GICS) which has replaced the complete ASX 3

5 Table 1: Standard & Poors GICS classifications. GICS Sector Industry Energy Energy Materials Materials, Metals and Mining Industrials Capital Goods, Commerical Services and Supplies, Transportation Consumer Discretionary Automobile and Components, Consumer Durables & Apparel, Consumer Services, Media, Retailing Consumer Staple Food and Staples Retailing, Food Beverage and Tobacco Health Care Healthcare Equipment and Services, Pharmaceuticals and Biotechnology and Life Sciences Financials Banks, Diversified Financials, Insurance, Real Estate (excl. REITs), Real Estate Investment Trusts Information Technology Software and Services, Technology Hardware and Equipment Telecommunications Telecommunication Services Utilities Utilities Other Miscellaneous Industrials Industry Classifications at the end of We consider 25 major industries classified into 11 GICS industry sectors summarized in Table 1. The accounting information is sourced from the Aspect Huntley (ASPECT) database. It includes raw and aggregate data items from final reports, annual reports, interim half yearly reports and quarterly cash flow reports. The data is collected directly from company financial statements and the notes include data on earnings, sales, book equity, market equity and total assets. The coverage of annual data goes back to 1989, and interim and preliminary final data is from In order to obtain comprehensive coverage of financial reporting data, for our analysis we merge equity data from SPPR with accounting data from ASPECT for the period from 1991 to 2007 since not enough accounting data is available prior to To assure that accounting information is available prior to equity data and thus, is reflected in the stock prices, we match SPPR stock return data of July of year t to June of year t+1 with the ASPECT accounting information of year t 1, as in Fama and French (1992) and Hou and Robinson (2006). We use a firm s market equity at the end of December of year t 1 to compute its book-to-market (B/M) and leverage ratios for year t 1, and we use its market equity for June of year t to measure its size. Therefore, to be included in the return tests for July of year t, a firm must have a stock price for December of year t 1 and June of year t. All together, the accounting ratios used in the analysis include: E/A ratio which is earnings before interest divided by assets; E/S is earnings divided by sales; B/M ratio is calculated by dividing book equity by market equity, which is calculated as SPPR stock prices times shares outstanding; V/A is a market value of firm divided by total assets; D/B is the ratio of dividends to book equity; R&D/A is the ratio of R&D expenditure to total assets; leverage is the ratio of book liabilities (total assets-book equity) to total market value of a firm; beta is a post-ranking β as in Fama and French (1992). Calculation of post-ranking βs is discussed in the Appendix 8. For its calculation we have to compute pre-ranking betas first, which requires monthly return data for at least 36 month preceding July of year t. Since the Herfindahl index is averaged over the past three years to enhance robustness, the final data sample includes period from 1993 to

6 Industry Concentration Measures H(Assets) H(Equity) Figure 1: Box-plots for the summary statistics of industry concentration measures. H(Sales/Asset/Equity) is the Herfindahl index calculated as the sum of squared sales/total assets/book equity-based market shares of all firms in each industry in a given year and then averaging over the past three years. The extreme of the lower whisker (2.5%-quantile), low quartile, median, upper quartile and the extreme of the upper whisker (97.5%-quantile) are plotted. 2.3 Measuring Industry Concentration We calculate Herfindahl index (1) based on sales, (2) total assets, or (3) book value of equity of a firm. denotes the Herfindahl index calculated using equation (1) as the sum of squared sales-based market shares of all firms in each industry in a given year and then averaging over the past three years. H(Asset) and H(Equity) use total assets and book equity, respectively, to compute market share of a firm Summary Statistics and Characteristics of Concentration-Sorted Portfolios Panel A of Table 2 reports the summary statistics of industry concentration measures for the 25 industries listed in Table 1. In addition, Figure 1 represents the box-plots for the summary statistics of industry concentration measures. The box contains 50% of observations belonging to the range between the low and the upper quartile (interquartile range). The extremes of the lower and the upper whisker correspond to 2.5%-quantile and 97.5%-quantile, respectively. From the Table 1 and the box-plots we observe that is on average higher than H(Asset) and H(Equity) and experiences higher standard deviation. The ranges between 0.09 (indicating high competition among small companies) and 0.90 (indicating that concentration drives the industry), whereas 60% of observations fall in the range between 0.15 and The three far right hand side columns represent correlations between the concentration measures: the Pearson correlation is presented above the main diagonal, whereas the Spearman 5

7 Energy Materials Energy Materials Metals and Mining Time Time Industrials Consumer Discretionary Capital Goods Commerical Services and Supplies Transportation Automobile and Components Consumer Durables & Apparel Consumer Services Media Retail Time Time Consumer Staple Health Care Food and Staples Retailing Food Beverage and Tobacco Healthcare Equipment and Services Pharm. and Biotech. and Life Scienc Time Time Figure 2: Industry concentration measures for sectors Energy, Materials, Industrials, Consumer Discretionary, Consumer Staple and Health Care plotted yearly from 1993 to is calculated as the sum of squared sales-based market shares of all firms in each industry in a given year and then averaging over the past three years. 6

8 Financials Information Technology Diversified Financials Banks Insurance Real Estate (excl. REITs) Real Estate Investment Trusts Software and Services Technology Hardware and Equipment Time Time Telecommunications Utilities Telecommunication Serv Utilities Time Time Miscellaneous Industrials Misc. Industrials Time Figure 3: Industry concentration measures for sectors Financials, Information Technology, Telecommunications, Utilities and Miscellaneous Industrials plotted yearly from 1993 to is calculated as the sum of squared sales-based market shares of all firms in each industry in a given year and then averaging over the past three years. 7

9 Table 2: Summary statistics of industry concentration measures. H(Sales/Asset/Equity) is the Herfindahl index calculated as the sum of squared sales/total assets/book equity-based market shares of all firms in each industry in a given year and then averaging over the past three years. Last three columns represent the Pearson correlation (above the main diagonal) and the Spearman correlation (below the main diagonal). Panel B and C represent mean respectively median of the industry characteristics within -sorted quintile portfolios. Q1 (low) corresponds to the 20% of industries with the lowest concentration whereas Q5 (high) corresponds to the 20% of industries with the highest concentration. Panel A: Summary of Industry Concentration Measures Spearman-Pearson Correlation mean median s.d. max min 20% 40% 60% 80% H(Assets) H(Equity) H(Assets) H(Equity) Panel B: Mean Characteristics of Sorted Quintile Portfolios Rank Newlist Delist Size Asset Sales E/A E/S V/A D/B R&D R&D/A Lev. B/M Beta Low Q Q Q High (rank) correlation is given below the main diagonal. We observe that the concentration measures H(Assets) and H(Equity) are highly correlated with correlation larger than 0.9, correlation for with two other measures decreases to approximately Further, Figure 2 and Figure 3 show the evolution of over the period 1993 to We observe a decrease in over the period for the following GICS sectors, indicating that these become more competitive over time: Industrials (rapid decrease for Transportation to moderate decrease for Commercial Services, and nearly constant low concentration for Supplies and Capital Goods); Health Care (Healthcare Equipment and Services concentration declines steadily over the whole period whereas Pharmaceuticals and Biotechnology and Life Sciences concentration decreases over the period and increases afterwards). indices for Telecommunication and Utilities experience similar patterns: they decrease rapidly from 1993 to 2000 (increase in competition over this period), slightly increase afterwards towards the peak in 2004, with a further minor decrease from 2005 to The Financial industry appears to be the most volatile regarding the concentration measure: decreases in period from 1995 to 1999 pointing out towards competition increase, concentration increases from 1999 to 2002 and decreases slightly afterwards for all Financials except Diversified Financials where decreases slightly after Note that Banks data is available only from 2001 and thus, the Herfindahl index can only be computed starting from It decreases from 2003 to 2007 indicating less concentration. For the remaining industries, all except Miscellaneous Industrials, we observe either no variation or a slight decrease in over the period Note however, firm data used to compute H(Sales/Assets/Equity) might not overlap, that is, e.g., total assets might not be available for some firms who report sales and vice versa. Excluding these data clearly produces smaller sample for which all concentration measures become highly correlated. 8

10 In the following, we group industries into quintiles based on their values into quintile portfolios. Average industry characteristics for each quintile portfolio are computed using accounting information from ASPECT for the quintile portfolios. Industry characteristics include Newlist (Delist) computed the average number of newly listed (de-listed) firms per year in each quintile; Size (market equity) computed as the price of a share times shares outstanding (in millions of dollars); Asset denotes total assets of the company; Sales are the net sales of the company; E/A and E/S denoting earnings before interest divided by total assets (sales); V/A gives the market value of a firm (market equity + total assets - book equity) divided by the total assets; D/B which is a ratio of dividends to book equity; R&D/A is a ratio of R&D expenditure to total assets; leverage is a ratio of book liabilities (total assets - book equity) to total market value of firm; B/M is a ratio of book equity to market equity; beta is a post-ranking beta as in Fama and French (1992). Each of these characteristic is calculated at the firm level and then averaged within each concentration quintile. The results are reported in Panel B of Table 2. We observe that the number of new-listings and de-listings is the highest for the second least concentrated quintile, corresponding to 21.6 and 22.1 respectively, and it drops to respectively 4.8 and 8.2 for the most concentrated industries, indicating higher barriers to entry. The average size and total assets increase towards more concentrated quintiles. Profitability measured through ratios earnings to assets (E/A) and earnings to sales (E/S) varies across quintiles. On average, earnings are negative, which leads to the negative E/A and E/S ratios. The E/A ratio appears to be lower ( 10.5%) for the most concentrated (high) quintile compared to the E/A of 1.1% for the most competitive (low) quintile. This indicates that highly concentrated industries have larger asset bases and lower unit profitability. V/A is a performance measure which gives the ratio of the market value to the book value of a firm (market value of firm divided by total assets). It is larger than one indicating that firms earn on average a rate of return higher than that justified by the cost of its assets. Such a return could not persist in the absence of long-run barriers to entry. V/A increases slightly when concentration increases. While smaller E/A ratio for the more concentrated industries indicates lower current profitability, an increase in V/A points out that concentrated industries expect higher profitability in the future. Dividend payout given by the ratio of dividend to book equity is nearly constant across quintiles Q2, Q3 and Q3. It increases slightly for the most competitive (Q1) and the most concentrated (Q5) quintiles. Fama and French (2000) and Hou and Robinson (2006) relate dividend policy to expected profitability showing that firms in more concentrated industries are more profitable. R&D expenditure increases from AUD 1.40 Mio for the most competitive quintile to AUD 1.72 Mio for the most concentrated quintile indicating that firms in the highly concentrated industries get more involved in risky innovations. Scaling by the total assets leads to the similar pattern: R&D/A increases with increasing concentration. The average ratio of book equity to market equity B/M decreases with increasing concentration, that is, for more concentrated industries the market is valuing equity relatively expensively compared to the book value. Since less risky investments are (all else equal) likely to have higher market value given that the associated companies have similar book values (and therefore, lower book to market ratio), the more concentrated industries appear to be less risky than the more competitive. The average betas decrease slightly with increasing concentration. Finally, leverage is higher for the industries falling into extreme 9

11 Table 3: Fama-MacBeth regression of on industry average characteristics. Crosssectional regressions are estimated every year and averaged over the past three years afterwards. Panel A reports the results for the estimates from the simple regression, Panel B reports the results of multivariate regression of on a combination of industry characteristics; t- statistics appear below the coefficient estimates. ln(size) ln(assets) ln(sales) E/A E/S V/A D/B R&D/A Leverage ln(b/m) beta Panel A: Simple Regression Panel B: Multiple Regression quintiles (Q1 and Q5) rather than for the middle-concentrated industries in Q3. Comparing the extreme quintiles, leverage appears to be slightly higher for the most competitive (Q1) quintile compared to the most concentrated quintile (Q5) Regression of on Industry Average Characteristics In the following, we apply regressions as in Fama and MacBeth (1973) of the cross-section of on industry average characteristics. Therefore, we estimate coefficients of the following regression equation: N jt = γ t + δ nt X jt + ε jt, (2) n=1 where X jt represents j th industry average characteristics in year t. Cross-sectional regressions are run every year from 1993 to 2007 and the means of annual regression coefficients together with the time-series t-statistics are reported in Table 3. Panel A reports the results for the estimates from the simple regression, Panel B reports the results of multivariate regression of on a combination of industry characteristics. Combining estimates from Tables 3 with the descriptive statistics from Table 2, we observe that the size and total assets are positively correlated with industry concentration with insignificant t-statistics in a simple regression. Positive size and total assets effect become significant when controlling for other average industry characteristics. Sales are negatively correlated with industry concentration with a sales effect being highly significant, even after controlling for other 10

12 Table 4: Industry Concentration and the Cross-Section of Average Stock Returns: we report cross-section of average stock returns (in percent) based on index. We calculate average monthly returns of the quintile portfolios, as well as the difference between the most concentrated quintile Q5 and the least concentrated quintile Q1; t-statistics are reported below the average returns. Firm-level returns are the average returns across firms within the same concentration quintile. Industry-level returns are calculated by first, averaging the returns within each industry, and then, across industries within the same concentration quintile. Adjusted returns are calculated by subtracting the return on a characteristics-based benchmark constructed as in Daniel and Titman (1997) and Hou and Robinson (2006). Firm-Level Returns Industry-Level Returns Quintile Quintile Q1 Q2 Q3 Q4 Q5 Q5 Q1 Q1 Q2 Q3 Q4 Q5 Q5 Q1 Panel A: Raw Returns Av.Return t-stats Panel B: Adjusted Returns adj. Av.Return adj. t-stats variable in the regression. The ratio E/A which measure unit profitability of a company, as well as ratios V/A and D/B, appear to be positively (but insignificant t-statistics) correlated with industry concentration. The effect of B/M on concentration is negative, but remain insignificant even after controlling for other variables in the regression. Further, R&D/A is significantly and positively related to concentration. Higher R&D to assets ratios with higher asset based for more concentrated industries indicate that these companies engage in risky innovations. On the other side, beta has a negative and significant effect on concentration indicating that concentrated industries are less risky than the competitive industries. Depending on the control variables, the effect of betas remain negative and statistically significant, or disappears (e.g. when controlling for size). Finally, leverage has significantly negative t-statistics indicating that the usage of debt to supplement investment decreases with concentration. 3 Cross-Section of Returns 3.1 Concentration and Average Cross-Sectional Stock Returns In Table 4 we report the cross-section of average stock returns based on index. Average monthly returns of the quintile portfolios are calculated, as well as the difference between the most concentrated quintile Q5 and the least concentrated quintile Q1; t-statistics are reported below the corresponding average return. Quintile returns are measured by either equally weighting the firm returns in each concentration quintile (firm-level returns), or by first forming industry portfolios (i.e. averaging the returns within each industry), and then by equally weighting industry returns in each concentration quintile (industry-level returns). In Panel A the first row represents on a firm level the average return in percent within the 11

13 quintile, calculated by equally weighting firms within each concentration portfolio: the least concentrated (most competitive) industries have an average return of 0.87% per month. This increases to 1.36% for the most concentrated industries of quintile Q5. The spread between the most concentrated quintile Q5 and the most competitive quintile Q1 is 0.49% per month with a significant t-statistic of This corresponds to the economic magnitude between high and low concentrated industries of approximately 5.8% per annum. For the average return calculated on the industry-level we observe nearly a double increase from 0.95% for the most competitive quintile Q1 to 1.99% for the most concentrated quintile Q5. The spread accounts to 1.04% and carries statistically significant t-statistic of In Panel B we perform the same calculations on the firm- and industry-level for the adjusted returns obtained by subtracting the return on a characteristics-based benchmark (constructed as in Daniel and Titman (1997) and Hou and Robinson (2006)) from the individual firm s return, and then averaging within the same concentration quintile. Subtracting the return on a characteristics-based benchmark allows to adjust individual returns for the risk factors such as size, book-to-market and momentum. Therefore, a triple-sorting is performed every month on all firms in our sample: we first sort firms into size quintiles, and then within each size quintile into book-to-market quintiles. Within each of these 25 portfolios we sort firms again based on the 12-month performance (12-month average return is build ignoring the most recent month). We subtract from each individual stock s return the benchmark return calculated by averaging individual return within each of these 125 portfolios. Similarly to Panel A, we observe an increase in the adjusted average stock returns with an increase in concentration. The spread in the average adjusted returns calculated on the firm (industry-) level for the most concentrated (Q5) and the most competitive (Q1) industries decreases from 0.49 (1.04) to 0.14 (0.85), compared to the spread in the average raw returns. While the spread Q5-Q1 remains significant on the industry-level, it carries insignificant t-statistic on the firm-level. Together, it suggests that on the industry-level the average return premium associated with industry concentration is independent from those of size, book-to-market and momentum. However, it is affected by these risk factors on the firm-level. 3.2 Fama MacBeth Cross-Sectional Regression Table 5 shows the results from regressing monthly average stock returns on the concentration measure and other characteristics. In Panel A, industry average returns are regressed on industry, ln(size), ln(b/m), momentum (the past one-year return on the industry portfolio), beta and leverage. In Panel B, individual company returns are regressed on of the industry which the company belongs to, and the firm-level characteristics ln(size), ln(b/m), momentum, beta and leverage. Cross-sectional regressions are estimated monthly and timeseries average estimates are reported together with the time-series t-statistics. From the estimates reported in Panel A, when regressing industry-level returns on only, we observe a positive and highly significant relationship between concentration and average stock returns. This indicates that on average the more concentrated industries earn higher average returns than the more competitive industries, which confirms the earlier results for the quintile portfolios. Further, we observe that the average stock returns are positively related to the 12

14 Table 5: Fama-MacBeth regression of average industry-level (Panel A) and firm-level (Panel B) cross-sectional returns on and industry-level (firm-level) average characteristics. In Panel A, industry average returns are regressed on industry, ln(size), ln(b/m), momentum (the past one-year return on the industry portfolio), beta and leverage. In Panel B, individual company returns are regressed on of the industry which the company belongs to, and the firm-level characteristics ln(size), ln(b/m), momentum, beta and leverage. Cross-sectional regressions are estimated monthly and time-series average estimates are reported together with the time-series t-statistics. ln(size) ln(b/m) Momentum Beta Leverage Panel A: Industry-Level Regression Panel B: Firm-Level Regression

15 firm size, and negatively related to the book-to-market ratio. Both carry significant t-statistics. In addition, there is a negative relationship between the average stock returns and beta in a single regression, which becomes insignificant after controlling for size, book-to-market and momentum. Finally, average stock returns are only insignificantly related to momentum and leverage, even after controlling for other variables in the regression. Combining these results with the regression estimates of on industry average characteristics, we conclude, that concentrated industries with a high market power are dominated by the large companies with the high market values (and therefore, the low B/M ratios), and have higher returns than the competitive industries shared by many small companies. From the estimates reported in Panel B, when running the regressions on a firm-level, we observe similar results to the industry-level regression: average stock returns increase with concentration. Average stock returns are also positively related to size and negatively related to the book-to-market ratio. Compared to the industry-level regressions, average returns appear to be positively related to betas (in a multiple regression when controlling for size, B/M and momentum) indicating higher (systematic) risk for the more concentrated industries. Furthermore, leverage has significantly negative impact on the average stock return in a single regression, the effect disappears when controlling for other variables in the regression. Momentum remains only insignificantly related to the average stock returns. 4 Concentration and Profitability This section analyzes factors which drive the differences in average returns across concentration quintiles. In particular, according to Campbell (1988), the returns can be decomposed into the sum of expected returns and shocks to cash flows and to discount rates. We are interested in whether these shocks across industries are responsible for different market structures when dealing with concentrated and competitive industries, and thus, for the differences in the returns. We assume the profitability model with a lagged profitability term, implemented in Vuolteenaho (2002) and Hou and Robinson (2006), which as an extension of the Fama and French (2000) profitability model: E t A t = α 0 + α 1 V t A t + α 2 DD t + α 3 D t B t + α 4 E t 1 A t 1 + ε t. (3) In equation (3) E/A denotes the earnings to assets ratio, V/A is the ratio of market value of assets to book value, DD is a dummy variable for non-dividend-paying firms and D/B is a ratio of dividend payments to book equity. The unexpected profitability U P is measured through the regression error ε t. The regression (3) is estimated yearly on a firm and industry-level, whereas industry total E/A ratio is calculated as a total sum of earnings within an industry divided by the total sum of assets within an industry, and then regressed on the four independent variables (total values) constructed in a similar way. Industry average E/A refers to the average industry earnings to assets ratio, which is regressed on industry average values of regressors. Panel A of Table 6 reports average coefficients of the cross-sectional regressions estimated yearly, with the time-series t-statistics reported below. Similarly to Fama and French (2000) and Hou and Robinson (2006), we observe significantly positive regression coefficients for D/B on the 14

16 Table 6: Relation between concentration and profitability. In Panel A firm/industry total/industry average earnings to assets ratio E/A is regressed on the ratio of market value of assets to book value of assets V/A, a dummy variable for non-dividend-paying firms DD, a ratio of dividend payments to book equity D/B and a lagged E/A ratio: E t A t = α 0 + α 1 V t A t + D α 2 DD t + α t E 3 B t + α t 1 4 A t 1 + ε t. The cross-sectional regressions are estimated yearly and the average coefficients together with the time-series t-statistics are reported below the coefficient estimates. The unexpected profitability is measured through the regression error ε t. Panel B reports in-sample unexpected profitability UP t measured through in-sample regression error, as well as the one-period-ahead unexpected profitability UP t+1 computed as a regression error from forecasting profitability in year t using regression coefficients estimated in year t 1. F-test and t-test compare profitability across quintiles and the last FM column shows estimated regression coefficients and t-statistics from regressing U P on the concentration measure. Profitability measure α 0 V/A DD D/B ROA t 1 Avg.R 2 Panel A: Expected Profitability Regression Firm-level E/A Industry total E/A Industry average E/A Panel B: Unexpected Profitability by Concentration Quintile Quintiles Profitability Q1 Q2 Q3 Q4 Q5 F(Q1=...=Q5) t(q1-q5) FM Firm-level Industry total Industry average UP t Tests UP t UP t UP t UP t UP t

17 firm-level and industry average-level, and significantly negative regression coefficients for DD on the firm-level indicating that the dividend-paying companies have higher unit profitability. Furthermore, profitability loads negatively (and significant) on V/A when running the regression on a firm-level, and positively (and significant) when running the regression on the industry-level. This shows that V/A captures differences across firms and industries in expected profitabilities, which are not captured by the dividend regressors DD and DB. On the firm-level, firms with a higher market value of assets to the book value of assets V/A tend to have lower unit profitability measured through E/A. Since the market value of assets of the less risky investments is likely to be higher than the market value of the more risky investments, more risky companies tend to have lower V/A ratios (given constant book value) and therefore, higher unit profitability. However, on the industry-level, industry s total (average) profitability increases with increasing industry s total (average) ratio of market value of assets to book value. Finally, we observe twice as higher regression R 2 for the industry-level regressions compared to the firm-level regression. In Panel B of Table 6 we compute in-sample unexpected profitability UP t measured through in-sample regression error, as well as the one-period-ahead unexpected profitability UP t+1 computed as a regression error from forecasting profitability in year t using regression coefficients estimated in year t 1. Quintiles Q1 (most competitive) to Q5 (most concentrated) report UP t and UP t+1 by concentration quintile. Hou and Robinson (2006) argue that if the differences in the results across quintiles were driven by cash flow shocks, we would expect large (in absolute values) unexpected profitability for Q1 and Q5. In our case we observe relatively low (and mostly insignificant) unexpected profitability, compared to e.g. middle-concentrated quintile Q3. The most competitive industries tend to have negative UP t indicating that the companies in these industries experience lower than expected profitability. The most concentrated industries have (on a firm-level) positive UP t values, that is, the associated companies have higher than expected profitability. Further, F-test always rejects the hypothesis on the equality of unit profitability across quintiles in-sample. For the one-period-ahead prediction UP t+1 there is an evidence on differences in unit profitability across quintiles on the firm-level, but not the industry-level. When we compare the least concentrated quintile Q1 and the most concentrated quintile Q5 by means of the t- test, we cannot reject the hypothesis that the profitability is the same in Q1 and Q5. Finally, the last column in Panel B reports the Fama-MacBeth regression of UP on with the t-statistics reported below the regression coefficient. We observe positive (but insignificant) relationship between concentration and unexpected profitability on the firm-level and negative (but insignificant) relationship on the industry-level. 5 Concentration and Time-Series Variation in Returns 5.1 Time-Series Variation of the Concentration Premium In this section we are interested in whether changes in the concentration premium are associated with the various risk factors and business cycle indicators. In order to examine whether concentration premium is associated with the size, book-to-market and momentum, and whether it 16

18 Table 7: Time-series variation of the concentration premium. Time series regression of the monthly concentration premia (obtained from the FM regression of cross-sectional stock returns on and other industry characteristics, see last row in Panel B of Table 5) on the risk factors: RMRF (market excess return), SMB (small market capitalization minus big), HML (high book-to-market minus low), and MOM (momentum); as well as business-cycle indicators: INF (monthly rate of inflation), the Term (term spread between 10-year and 3-month interest rate yields), the T-Bill (30-day T-Bill rate), g t and g t+1 denoting the current and the next year s growth rate of the GDP. t-statistics are reported below the regression coefficient estimates. Alpha RMRF SMB HML MOM INF Term T-Bill g t g t+1 Adj.R