Jose´-Miguel Gaspar ESSEC Business School

Massimo Massa INSEAD

Idiosyncratic Volatility and Product Market Competition*

I.

Introduction

How important is the link between a firm’s competitive environment and its idiosyncratic volatility? This question, which the literature has so far surprisingly neglected, is the subject of this article. Campbell et al. (2001) have documented a substantial rise in the average stock’s idiosyncratic volatility over the past decades. It is possible that the tougher competitive conditions faced by firms in many industries—caused by increased deregulation, globalization, and the speed of technological change—could be at least partially to blame. After all, “the best of all monopoly profits is a quiet life” (Hicks 1935, 8). It is important to understand idiosyncratic volatility because of its direct implications on investors’ portfolio and hedging strategies. In addition, underdiversified investors demand a premium for holding a firm’s shares that is positively related to its idiosyncratic risk (Merton 1987). Consistent with this idea, there is evidence * We would like to thank Alexandre Baptista, Jean Dermine, Bernard Dumas, Miguel Ferreira, Nicolae Garleanu, Harald Hau, Pascal Maenhout, Pedro Matos, Urs Peyer, Ludovic Phallippou, Matti Suominen, and an anonymous referee for their valuable comments and Jean Cropper for editorial assistance. Gaspar gratefully acknowledges the financial support of Programa Praxis XXI of Fundac¸a˜o para a Cieˆncia e Tecnologia. All remaining errors are our own. Contact the corresponding author, Massimo Massa, at [email protected]. (Journal of Business, 2006, vol. 79, no. 6) 䉷 2006 by The University of Chicago. All rights reserved. 0021-9398/2006/7906-0017$10.00 3125

We investigate the link between a firm’s competitive environment and the idiosyncratic volatility of its stock returns. We find that firms enjoying high market power, or established in concentrated industries, have lower idiosyncratic volatility. We posit that competition affects volatility in two distinct ways. Market power works as a hedging instrument that smoothes out idiosyncratic fluctuations. Also, market power lowers information uncertainty for investors and therefore return volatility. We find strong support for both effects. Our results contribute to the understanding of recent trends of idiosyncratic volatility and confirm the link between stock performance and firm’s competitive environment.

3126

Journal of Business

that idiosyncratic volatility is priced and that it drives stock market forecastability (Malkiel and Xu 2001; Goyal and Santa-Clara 2003), although these effects seem to be mostly associated with small firms (Brown and Ferreira 2003; Bali et al. 2005). This article qualifies and extends the literature on volatility by proposing and testing hypotheses concerning the influence of product market conditions on idiosyncratic risk. We hypothesize two ways in which competition affects volatility. First, market power works as a natural hedge that smoothes out cash flow fluctuations resulting from idiosyncratic cost shocks. This hypothesis results from the standard view of volatility as a function of shocks to the economy’s discount factor and to the company’s underlying cash flows (LeRoy and Porter 1981; Campbell and Shiller 1988; Vuolteenaho 2002). Competitive positioning can influence the impact of company-specific shocks. A firm with monopoly power is able to pass on a bigger proportion of any idiosyncratic cost shocks to its consumers. In contrast, a firm acting in a highly competitive industry can be driven out of business entirely if costs get much out of line with those of its competitors. Second, the ability to exercise market power and avoid competition decreases uncertainty about the firm’s future performance. Pa´stor and Veronesi (2003) construct a model where investors learn about a firm’s average profitability over time. In their model, idiosyncratic volatility increases with uncertainty about a firm’s average profitability because (firm-specific) learning uncertainty is only weakly correlated with the (economy-wide) stochastic discount factor. As a result, all learning uncertainty is transmitted to returns in the form of idiosyncratic shocks. Our hypothesis claims that competition increases uncertainty about the firm’s average profitability. We should therefore observe higher idiosyncratic volatility by firms in competitive industries even when the volatility of their underlying profits are kept constant. Our hypotheses are tested using various measures of product market competition. To measure the market power of firms within an industry, we use the industry-adjusted price-cost margin, or the Lerner index (Lerner 1934). To measure market power between industries, we employ the HerfindahlHirschman index of concentration (Hirschman 1945; Herfindahl 1950). Using these and other measures, we show that a firm’s market power is negatively related to the absolute level of the firm’s idiosyncratic volatility. An increase of one standard deviation in the industry-adjusted price-cost margin of the average firm implies a 2% reduction in the absolute level of idiosyncratic volatility. These firm-level results hold at the industry level and are robust to using alternative definitions of market power. Additional analysis reveals that both the natural hedge channel and the uncertainty channel separately contribute to the impact of competition on firmspecific volatility. On the one hand, the natural hedge channel posits that competition increases return volatility by raising profit volatility. Using a simultaneous equations approach, we indeed find evidence of this effect. On

Volatility and Competition

3127

the other hand, the uncertainty channel implies that the market is more uncertain about competitive firms’ cash flows. Multivariate regressions confirm that the presence of market power is negatively associated with firm-level uncertainty, as measured by the observed dispersion in analysts’ earnings forecasts. Our results are particularly relevant for two reasons. First, they are important for understanding the upward trend in idiosyncratic volatility documented by Campbell et al. (2001). Pa´stor and Veronesi’s (2003) model contains two sources of volatility—profit volatility and uncertainty about average profitability—that are both seen as key for understanding the rise in volatility. Our article argues that competitive conduct strongly influences both these sources of volatility, providing a simple economic background to their story. Managers and commentators often insist that competition has increased in the past decades due to deregulation and globalization.1 Second, our results provide insights into an important indirect channel between business conditions and firm value. If idiosyncratic risk is priced, investors will require a higher rate of return to invest in firms acting in very competitive environments. We would therefore expect to find lower valuations for these firms, not only due to their possibly lower cash flow (relative to otherwise identical firms enjoying market power) but also due to their higher cost of capital. Other papers have investigated the influence of firm characteristics on idiosyncratic volatility. Wei and Zhang (2003) find a strong negative correlation between profitability and idiosyncratic risk, as well as strong positive correlation between profit volatility and idiosyncratic risk. Although the authors conclude that “changes in fundamental variables” are behind their results, they do not provide a theoretical explanation of what those variables might be. Malkiel and Xu (2001, 2003), in separate univariate analyses, suggest a link between idiosyncratic volatility and growth opportunities and also between idiosyncratic volatility and institutional investment. Brown and Ferreira (2003) show that the power of idiosyncratic volatility to predict returns originates mostly on the volatility of small firms. The remainder of this article is structured as follows. Section II presents our hypothesis development, and Section III describes the data. Section IV 1. Shepherd (1982, 613) documents a “striking increase in competition during 1958 to 1980, virtually throughout the U.S. economy,” and the Organization for Economic Cooperation and Development (OECD) reports that the movement toward regulatory reform has been stronger in the United States than in other countries and attributes a “sizable” portion of the good economic performance in the past decade to increased competition (Suppanz, Wise, and Kiley 2004). Concerning globalization, Bernard, Jensen, and Schott (2006) show that the average industry’s exposure to competition from low-wage countries increased substantially between 1972 and 1992 and that the impact of this competition is stronger in more labor-intensive industries. Ryan (1997) shows slightly negative trends in price-cost margin ratios. Using the Federal Trade Commission (FTC) guidelines, our own calculations show that the percentage of industries considered as highly concentrated (defined as industries exhibiting Hirschmann-Herfindahl indices bigger than 0.18) decreased from 53% in 1962 to about 34% in 1999.

3128

Journal of Business

reports our main regression results. Section V provides additional evidence on the pervasiveness of the two channels hypothesized above. A brief conclusion follows in Section VI.

II.

Main Hypothesis and Testable Predictions

A.

The Natural Hedge Channel

The first channel of impact of market power on idiosyncratic volatility is derived using standard economic assumptions. The value of a firm is the present value, adjusted for risk, of its stream of profits. Firms maximize profits subject to their residual demand. Information arrives to the market in the form of news shocks, some of which are of a systematic nature while others are specific to the firm. We associate firm-specific shocks with shocks to costs so as to differentiate them from shocks to demand that might affect all firms in a given industry and are thus of a systematic nature. To see how the impact of a cost shock varies with the firm’s degree of market power, consider two otherwise identical firms: firm R operates in a market with comparatively rigid demand, while firm E operates in a market where demand is relatively elastic. Firm R commands greater market power through its ability to extract a greater price from its customer base. Profits of firm R are also higher since its pricing power enables it to charge higher prices. If an idiosyncratic cost shock arrives, firm R suffers less of an impact on profits than firm E.2 There are two reasons for this. First, firm R has higher profits to start with, so for a given size shock it is affected proportionately less. Second, and more important, the firm facing a rigid demand has a greater ability to pass on a bigger proportion of the cost shock to its consumers. Relative to firm E, firm R can increase prices by a bigger amount in case of a positive cost shock since its customers are less able to resort to substitutes to escape the price rise. The reverse happens in the case of a negative cost shock: firm R smoothes the impact of the cost shock by not lowering prices and expanding output as much as a firm with more elastic demand would do. 2. For an illustration in a monopoly context, assume a constant-elasticity demand Q p AP w, where w is the elasticity of demand. Assume also that marginal cost c is constant. The profit maximizing solution is P ∗ p 1/ (FwF ⫺ 1) # c # Q. Suppose that a cost shock moves c to c1 1 c. Differentiating the modulus percent change in profit in order to w, we get ⭸ (FDP1/PF) /⭸w p (c1/c) 1⫺w # (ln c1 ⫺ ln c) 1 0; that is, the percent change in profit caused by a shock in c is greater for a firm facing a more elastic demand. The result is valid under other market structures. As an example, consider a heterogeneous Cournot duopoly with linear demand P p a ⫺ bQ and constant marginal costs ci, i p 1, 2. In this setting the elasticity of the demand curve faced by all firms is similar, so we model market power as the ability of a firm to produce at lower costs. We assume that firm 1 has a cost advantage, such that c1 ! c2. The Cournot solutions are qi∗ p (a ⫺ 2ci ⫹ cj) /3b and Pi (qi∗) p b # qi2, i p 1, 2. A few steps suffice to show that FDP2/P2F 1 FDP1/P1F if c1 ! c2, i.e., the percentage change in profit caused by a shock in marginal cost is greater for a firm that has lower market power.

Volatility and Competition

3129

The net result is that the profits of a firm with the more rigid demand changes by less than those of a firm with a more elastic one.3 Market power works as a natural hedging instrument that firms have at their disposal to smooth out idiosyncratic fluctuations. Idiosyncratic noise in the firm’s returns should therefore be lower: there should be a negative correlation between a firm’s level of idiosyncratic volatility and its ability to exert market power. B.

The Uncertainty Channel

The second effect links the degree of the firm’s market power to the level of information uncertainty faced by investors. Pa´stor and Veronesi (2003) develop a simple firm valuation model in which underlying average profitability (the drift of the firm’s stochastic earnings process) is unknown. Their model predicts that idiosyncratic volatility is increasing in uncertainty about the firm’s average profitability. The reason is that investors’ learning errors are only weakly correlated with the stochastic discount factor and hence do not affect expected returns. Learning errors show up in returns as idiosyncratic noise instead. Since uncertainty is highest for younger firms, the authors use firm age as a proxy for investor uncertainty. We argue that competition influences the amount of uncertainty faced by investors. It should be relatively easier for investors to learn about underlying profitability for firms that have dominant market positions in established, wellunderstood industries. In contrast, for firms operating in competitive, fastchanging industries, it is more difficult for analysts and the market in general to forecast future cash flows and profitability. If competition increases uncertainty about average profitability, it will therefore increase idiosyncratic volatility. This uncertainty channel is different from the “natural hedge” argument presented earlier, since it means that competition affects return volatility directly and not through volatility of profits. We nevertheless reach the same conclusion: there should be a negative correlation between a firm’s level of idiosyncratic volatility and its ability to exert market power. Although we present the natural hedge and the uncertainty arguments as separate for expositional purposes, we see them as a single joint hypothesis. The natural hedge argument claims that competition increases profit volatility, and the uncertainty argument claims that competition increases learning uncertainty. Both imply that higher competition increases return volatility.4 3. The limit case of perfect competition offers a good illustration of what we have in mind. In such a situation, demand is perfectly elastic. A firm that suffers an idiosyncratic negative shock in costs instantaneously grabs the whole market and has an infinite percentage profit increase. A firm that suffers a persistent positive shock in costs shuts down and exits the industry (so, in a way, its “change” in profits is also maximal). This line of reasoning, which is focused on cost shocks, extends to idiosyncratic shocks to demand in the form of a downward shift in each firm’s residual demand curve. In such a scenario, it is still the case that the firm facing the more rigid demand is less affected, since it will recover some of cost of the lost unit sales revenue by charging a higher price per unit. 4. We thank an anonymous referee for help in defining and refining our hypothesis.

3130

III.

Journal of Business

Data Issues and Sample Construction

Our base sample is the CRSP-Compustat Merged universe of stocks for the period 1962–2001. We keep only those securities with a CRSP share code equal to 10 or 11, we exclude financial companies and regulated utilities, and we winsorize variables at the 1% level to deal with extreme outliers in sales and profit measures.5 The resulting sample contains 85,122 firm-year observations with nonmissing values for all our main variables. A detailed description of the Compustat data items used in the construction of the data set can be found in the appendix. Our main dependent variable, idiosyncratic volatility, is defined as the sum of squared residuals with respect to a capital asset pricing model.6 Our empirical implementation follows Malkiel and Xu (2003). For each month and each stock in the sample, we run a time-series regression of the past 36-month returns on the market excess return. We fit the resulting beta estimates to the daily returns in the current month, obtaining daily residuals. Idiosyncratic volatility is the sum (across all days and all months) of the squares of these daily residuals. Our regressions use the log of idiosyncratic volatility as a dependent variable to address the concerns raised in the literature (Goyal and Santa-Clara 2003; Malkiel and Xu 2003) that the skewness of idiosyncratic volatility might affect statistical inferences. We use two main proxies for market power. The first proxy is the excess price-cost margin (henceforth, EPCM), defined as the difference between a firm’s operating profit margin (PCM) and the average operating profit margin of its industry. This proxy is based on the concept of the price-cost margin, or the Lerner index, used in the industrial organization literature to measure a firm’s ability to price above marginal cost (Lerner 1934). Our implementation of the price-cost margin (as equal to operating profits over sales) makes the assumption, usual in the literature, that average variable cost is a meaningful proxy for marginal cost (Carlton and Perloff 1989).7 We subtract the industry’s average price-cost margin due to the fact that different industries might have structurally different rates of profit for reasons unrelated to market power. Our variable is thus better able to capture intraindustry differences in pricing power. Our second proxy for market power is the Herfindahl-Hirschman index of concentration. The Herfindahl-Hirschman index is equal to the sum of the squared market shares (sales over total industry sales) of firms in the industry. The index is widely used to measure market power in antitrust policy (e.g., Federal Trade Commission 1992). In our regressions, each firm is assigned 5. We found cases in Compustat of firms showing sales thousands of times greater than their reported costs. 6. We also calculated idiosyncratic volatility relative to a Fama and French (1993) three-factor model. The correlation between the two measures of idiosyncratic volatility is superior to 0.98. 7. The criticisms made to the price-cost margin is that it does not take into account the cost of capital and that it is an approximation valid only for companies that operate in a single line of business. In unreported regressions, we find results similar to ours using other definitions of the price-cost margin that address these issues.

Volatility and Competition TABLE 1

3131

Summary Statistics

Idiosyncratic volatility Share of systematic volatility Price-cost margin (PCM) Excess price-cost margin (EPCM) EPCM, equally weighted Herfindahl-Hirschman index C4 concentration index C10 concentration index

N

Mean

Standard Deviation

Q1

Median

Q3

85,122 83,767 84,426 84,556 84,556 84,559 84,559 84,559

.425 .189 .085 ⫺.083 .002 .108 .493 .690

.560 .154 .161 .169 .156 .097 .171 .158

.110 .056 .044 ⫺.126 ⫺.038 .054 .364 .582

.230 .157 .095 ⫺.058 .011 .085 .473 .678

.494 .292 .151 ⫺.005 .067 .124 .606 .799

Note.—This table presents summary statistics for some of the variables used in this study (see the appendix for a detailed description of the variables’ construction, sources, and references). Idiosyncratic volatility is the yearly sum of squares of daily residuals with respect to a capital asset pricing model. The capital asset pricing model beta estimates used to fit the model are calculated from monthly regressions of the previous 36 month returns on the market excess return. The share of idiosyncratic volatility is one minus the yearly average of the R-squares of the monthly regressions. The price-cost margin (PCM) is operating profits (before depreciation, interest, special items, and taxes) over sales. The excess price-cost margin (EPCM) is defined as the firm’s PCM minus the industry value-weighted average PCM. EPCM, equally weighted, is the difference between the firm’s PCM and the industry equally weighted average PCM. Herfindahl-Hirschman index is calculated as the sum of the squared market shares (sales over total industry sales) of firms in the industry, where industries are defined using two-digit level SIC codes. The C4 concentration index and the C10 concentration index are the sum of market (sales) shares of the top four and top 10 firms in each industry, respectively.

the Herfindahl-Hirschman index of its industry; this measure is therefore meant to capture interindustry differences in pricing power. We define industries using two-digit level SIC codes. Table 1 presents summary statistics for our main variables. Median annual idiosyncratic volatility is 0.23, or 48% standard deviation per year.8 The median share of systematic volatility, calculated as the R-square from the capital asset pricing model, is around 16%. Both these numbers are similar to those reported by Campbell et al. (2001). The average PCM in our sample is about 9%, slightly lower than estimates reported in 1980s studies (e.g., Domowitz, Hubbard, and Petersen 1986a, 1986b) but close to values reported in more recent papers (e.g., Aghion et al. 2002). The average EPCM is negative, a mechanical consequence of the positive correlation between size and profitability and of the fact that we use value weights to calculate industry averages. For comparison, we report the average EPCM calculated using equal weights and find it, as expected, statistically equal to zero.9 Finally, to put our figures for the Herfindahl-Hirschman index into context, we also report values for the sample C4 and C10 market concentration indexes, the sum of market shares of the largest four and the largest 10 firms in the industry, respectively. Overall, the top four firms in each industry have around 49% of the market, while the top 10 firms hold 70% of the total market share. 8. The corresponding number using value-weighted percentiles is 0.10, or 31% standard deviation per year. 9. In the remainder of this article, we focus on results using EPCM calculated with value weights. Results using EPCM calculated with equal weights are exactly similar and are available upon request.

3132

IV. A.

Journal of Business

Results Preliminary Evidence: Deregulated Industries

A good starting place to investigate the impact of market power on volatility can be found among companies that have suffered exogenous shocks in their market power. In this regard, the best examples should come from industries that were subject to one-shot events of deregulation that increased substantially their degree of competitiveness. An example of an industry that went through a major deregulation process is the telecommunications industry. The U.S. Justice Department sued AT&T in 1974 for monopolizing both telephone manufacturing and the telephone service supply. In 1982, the parties entered into a court-approved settlement that became known as the modified final judgment (MFJ), which required AT&T to divest its holdings of local telephone assets.10 As a result, “the telephone equipment and services sector changed from a tranquil, regulated monopoly into a set of increasingly competitive markets” (Crandall 1991, 123). Companies such as MCI and Sprint created their own networks, becoming strong competitors for AT&T in the market for long distance service. Local companies also experienced new forms of competition from “bypass” access providers.11 Figure 1 presents the evolution of the PCM (Lerner index), the Herfindahl-Hirschman index, and idiosyncratic volatility in the telecommunications industry. The years after the MFJ indeed coincide with a significant and persistent drop in the profitability, along with a very substantial increase in idiosyncratic volatility. The average PCM of the industry dropped from an average 27%, for the period of 12 years before deregulation, to 19% afterward. The Herfindahl-Hirschman index of concentration decreases by half, from 55% to 23%. At the same time, idiosyncratic volatility almost tripled, from 10% to 28% per year, while the share of systematic volatility dropped from 27% to 22% (all differences statistically significant at the 1% level). This example is a good illustration of the relation that our hypotheses predict. Do these results extend to other industries? Table 2 shows the prederegulation and postderegulation averages of the four variables mentioned above for other industries with major deregulation experiences (Winston 1998): airlines, electricity, natural gas, telecommunications, and transportation (the latter being composed of railroad and trucking). The numbers show that, for all 10. The judgment, implemented in 1984, created seven separate holding companies, known as “Baby Bells,” that owned 22 local Bell Operating Companies (BOCs). The MFJ prohibited the BOCs from providing long distance telephone services between newly created local access transport areas (LATAs), from supplying information services, and from manufacturing telecommunications equipment. “Local exchange carriers” would provide services within a LATA, while “interexchange carriers” (that had networks between but not within LATAs) purchased access service from local carriers. 11. Access providers connect users directly to an interexchange carrier, bypassing the local carrier and its access fee.

Volatility and Competition

3133

Fig. 1.—Market power and idiosyncratic volatility in the telecommunications industry. This figure presents (a) averages of the price-cost margin (Lerner index), Herfindahl-Hirschman index, and (b) idiosyncratic volatility and the share of systematic volatility for the telecommunications industry (see the appendix for a precise definition of the SIC codes used). The price-cost margin (Lerner index) is operating profits (before depreciation, interest, special items, and taxes) over sales. The HerfindahlHirschman index is the sum of the squared market shares of firms in the industry. Idiosyncratic volatility is the yearly sum of squares of daily residuals with respect to a capital asset pricing model. The capital asset pricing model beta estimates used to fit the model are calculated from monthly regressions of the previous 36-month returns on the market excess return. The share of systematic volatility is the yearly average of the R-squares of the monthly regressions.

3134

TABLE 2

Market Power and Idiosyncratic Volatility in Deregulated Industries Industry and Deregulation Year Airlines (1978)

Electricity (1978)

Natural Gas (1978)

Before

After

Test Result

Before

After

Test Result

.196

.259

***

.052

.064

***

.059

.077

.291 .108

.229 .059

*** ***

.335 .232

.146 .202

*** ***

.287 .147

.114

.104

*

.025

.023

***

.060

Before

Transportationa (1980)

Telecoms (1982) Before

After

Test Result

Before

After

***

.087

.223

**

.165

.181

.246 .106

*** ***

.216 .229

.230 .185

.230 .089

.200 .085

*

***

.054

**

.551

.322

**

.071

.128

*** **

After

Test Result

Test Result

Panel A. 5-Year Averages Idiosyncratic volatility Share of systematic volatility PCM Herfindahl-Hirschman index

Panel B. 12-Year Averages Idiosyncratic volatility Share of systematic volatility PCM Herfindahl-Hirschman index

.205

.277

***

.061

.075

***

.067

.082

***

.103

.281

***

.191

.243

.322 .122

.196 .063

*** ***

.289 .266

.164 .213

*** ***

.252 .166

.213 .107

*** ***

.266 .265

.219 .185

*** ***

.241 .094

.242 .092

.129

.117

*

.026

.025

**

.076

.058

***

.546

.228

***

.106

.113

Journal of Business

Note.—This table presents averages of the price-cost margin (PCM), idiosyncratic volatility, and the share of systematic volatility for industries that were subject to severe one-period deregulation shocks meant to introduce or enhance competition. See the appendix for a precise definition of the SIC codes used to define each industry. The deregulation events considered for each industry are in the same order as in the table, that is, the Airline Deregulation Act, the Public Utilities Regulatory Act, the Natural Gas Policy Act, the AT&T Settlement, the Motor Carrier Reform Act, and the Staggers Rail Act. Idiosyncratic volatility is the yearly sum of squares of daily residuals with respect to a capital asset pricing model (CAPM). The CAPM beta estimates used to fit the model are calculated from monthly regressions of the previous 36-month returns on the market excess return. The share of idiosyncratic volatility is one minus the yearly average of the R-squares of the monthly regressions. PCM is operating profits (before depreciation, interest, special items, and taxes) over sales. Herfindahl-Hirschman index is the sum of the squared market shares of firms in the sector. Panel A calculates averages of these variables for a time period of 5 years before and 5 years after the deregulation year. Panel B does the same for a time period of 12 years before and 12 years after the deregulation year. a Railroad and trucking. * Significant at the 10% level for the two-tailed Wilcoxon hypothesis test that the difference between averages in the two periods equals zero. ** Significant at the 5% level for the two-tailed Wilcoxon hypothesis test that the difference between averages in the two periods equals zero. *** Significant at the 1% level for the two-tailed Wilcoxon hypothesis test that the difference between averages in the two periods equals zero.

Volatility and Competition

3135

industries, the average PCM and average concentration have decreased and that this decrease coincided with (i) an increase in idiosyncratic volatility and (ii) a decrease in the explanatory power of the capital asset pricing model. The only exception is the transportation sector, which does not exhibit a statistically significant reduction in the Lerner index after deregulation.12 These results show preliminary evidence of the validity of our hypothesis that less market power implies more sensitivity to idiosyncratic shocks. These findings, however, also stress the need to control carefully for possible confounding effects in our analysis. An increase in idiosyncratic volatility could be a mechanical result of the increase in the number of firms and the corresponding decrease in the average firm’s size. Taking the telecommunications industry as an example, it is well known that the late 1980s coincided with a series of major technological changes (e.g., mobile communications, fiber optic cable, and broadband connections) that have substantially increased the underlying riskiness of the business. Furthermore, many firms (particularly after the mid-1990s) opted for substantial increases in leverage to finance aggressive expansion plans. It is therefore important that we control for these, as well as other, firm characteristics in our regression analysis. B.

Regression Results

Table 3 presents results from testing our hypotheses in a cross-sectional regression framework. We run a series of least-squares cross-sectional regressions LIVit p a 0,t ⫹ b  X it ⫹ ␧it i p 1, … , Ntt p 1962, … , 2001,

(1)

where LIVit represents the log of idiosyncratic volatility for firm i at year t and X it is a matrix containing our measures of market power as well as control variables. Our set of control variables is similar to the one employed by Pa´stor and Veronesi (2003), to which we add some other important variables suggested by the literature. The rationale for their use, along with corresponding references, can be found in the appendix. Industry dummies are included in all specifications. Fama-McBeth (1973) parameter estimates are obtained by averaging the coefficients across estimation years. The t-statistics we use are adjusted for the possibility of first-order autocorrelation in the parameter estimates, using the correction of Litzenberger and Ramaswamy (1982).13 Our results concerning the control variables match roughly the findings of previous literature. Older firms and dividend-paying firms have lower idiosyncratic volatility (Pa´stor and Veronesi 2003); firms with high trading volume 12. This result might be due to the fact that both railroad and trucking having been through significant consolidation processes, particularly the railroad industry, where many uneconomic routes were closed. 13. We also replicate our findings using three other methods for calculating t-statistics: the usual Fama-McBeth t-values, the Litzenberger and Ramaswamy (1979) correction for heteroskedasticity, and the simple average of the heteroskedasticity-consistent standard errors from each regression (Welch 2004). Our results are the same using all these methods.

3136

TABLE 3

Market Power and Idiosyncratic Volatility Dependent Variable: Log of Idiosyncratic Volatility (0)

(2)

(3)

(4)

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

⫺1.206 ⫺.048 2.907 1.155 ⫺.044 ⫺.001 .004 ⫺.183 .154 ⫺.018 ⫺.064 ⫺.081 .255 ⫺.494 ⫺.541

⫺8.02*** ⫺1.20 2.61** 2.45** ⫺.89 ⫺5.70*** .53 ⫺1.54 2.33** ⫺4.15*** ⫺1.41 ⫺1.89* 3.99*** ⫺9.94*** ⫺3.45***

⫺1.318 ⫺.051 2.864 1.161 ⫺.048 ⫺.001 .002 ⫺.188 .155 ⫺.018 ⫺.066 ⫺.082 .255 ⫺.494

⫺9.07*** ⫺1.18 2.50** 2.63** ⫺.95 ⫺5.66*** .28 ⫺1.49 2.36** ⫺4.26*** ⫺1.40 ⫺1.92* 4.04*** ⫺10.08***

⫺1.275 .023 2.489 .521 .053 ⫺.000 .006 ⫺.269 .156 ⫺.013 ⫺.008 .004 .220 ⫺.320

⫺7.62*** 1.30 2.52** 1.72* .78 ⫺3.25*** .50 ⫺1.53 3.19*** ⫺2.83*** ⫺.34 .29 3.09*** ⫺8.19***

⫺1.208 ⫺.051 2.858 1.214 ⫺.071 ⫺.001 ⫺.001 ⫺.341 .154 ⫺.018 ⫺.065 ⫺.082 .265 ⫺.503

⫺7.87*** ⫺1.17 2.52** 2.62** ⫺1.26 ⫺5.56*** ⫺.05 ⫺2.18** 2.40** ⫺4.27*** ⫺1.32 ⫺1.92* 4.24*** ⫺10.04***

⫺1.178 .022 2.528 .546 .032 ⫺.000 .006 ⫺.437 .157 ⫺.013 ⫺.003 .006 .226 ⫺.323

⫺6.39*** 1.25 2.57** 1.68 .51 ⫺3.18*** .55 ⫺2.22** 3.17*** ⫺2.86*** ⫺.14 .49 3.16*** ⫺7.79***

⫺.500

⫺3.11***

⫺.400

⫺3.11***

Journal of Business

Constant Size Trading volume Volatility of profits Leverage Age Market-to-book Return on assets Beta Price Spin-off dummy S&P500 dummy Nasdaq dummy Dividend dummy PCM EPCM

(1)

Coefficient

⫺.865 .101 ⫺.010 Yes .63 2,113

Yes .63 2,113

⫺5.35*** .44 ⫺2.70** Yes .69 1,919

⫺.235

⫺2.41**

Yes .62 2,113

⫺.286 ⫺.888 .089 ⫺.010

⫺2.11** ⫺5.43*** .38 ⫺2.79** Yes .68 1,919

Note.—This table presents results of yearly cross-sectional regressions of idiosyncratic volatility on measures of market power and other firm characteristics. Idiosyncratic volatility is the yearly sum of squares of daily residuals with respect to a CAPM model. The CAPM beta estimates used to fit the model are calculated from monthly regressions of the previous 36-month returns on the market excess return. The price-cost margin (PCM) is operating profits (before depreciation, interest, special items, and taxes) over sales. Excess price-cost margin (EPCM) is defined as the firm’s PCM minus the industry value-weighted average PCM. Herfindahl-Hirschman index is calculated as the sum of the squared market shares (sales over total industry sales) of firms in the industry, where industries are defined using two-digit level SIC codes. Size is measured by market capitalization. Trading volume is the average monthly number of shares traded over total shares outstanding. Volatility of profits is calculated as the root mean squared error of a regression of the firm’s return on equity (ROE) on its lag. Leverage is total long-term debt over total assets. Age is the number of months since the date the stock first appears in the CRSP database. Market-to-book is the ratio between the market value and the book value of equity. Return on assets is operating earnings over total assets. Beta is the yearly average beta estimate from the estimation of idiosyncratic volatility defined above. Price is the firm’s stock price recorded in CRSP. The spin-off dummy takes a value of one if a company is either a “parent” or a “child” in that year or if it was a “parent” or a “child” the year before. The Nasdaq and the S&P500 dummies indicate that the company traded on the Nasdaq or was a constituent of the S&P 500 index in that year, respectively. The dividend dummy takes a value of one if Compustat reports a positive dividend being paid in that year. Institutional ownership is the ratio of a firm’s shares held by institutional investors relative to total shares outstanding. Investor turnover is the weighted average of the portfolio turnover of the firm’s investors over the four quarters of the year, calculated following Gaspar, Massa, and Matos (2004). Analysts is the number of analysts following the stock. The parameter estimates shown are time-series averages of the individual cross-sectional parameter estimates. Fama-McBeth t-statistics are adjusted for first-order autocorrelation using the Litzenberger and Ramaswamy (1982) correction. * Significant at the 10% level for the two-tailed hypothesis test that the coefficient equals zero. ** Significant at the 5% level for the two-tailed hypothesis test that the coefficient equals zero. *** Significant at the 1% level for the two-tailed hypothesis test that the coefficient equals zero.

Volatility and Competition

Herfindahl-Hirschman index Institutional ownership Investor turnover Analysts Industry dummies Average adjusted R2 Average N

3137

3138

Journal of Business

and firms listed on Nasdaq have significantly higher idiosyncratic volatility. We keep volatility of profits as a right-hand-side variable to control for technologies with different risks. Note that the coefficient of volatility of profits is positive but that, contrary to Pa´stor and Veronesi (2003), it is not statistically significant in all specifications. In addition, institutional ownership and analyst following seem to be negatively correlated with idiosyncratic volatility. The most important conclusion from table 3 is that the impact of market power on idiosyncratic volatility is negative and statistically significant. Columns 1 and 2 of table 3 use the EPCM as a measure of market power, while columns 3 and 4 use the Herfindahl-Hirschman index of concentration (for reference purposes, column 0 presents results with the unadjusted PCM). The point estimate for EPCM is around ⫺0.5 (t-statistic p ⫺3.11), implying that a one standard deviation increase in EPCM implies a ⫺2% change in volatility standard deviation per year.14 The t-statistic of the Herfindahl-Hirschman index coefficients ranges from ⫺2.11 to ⫺2.48, which seems remarkably strong given the number of effects we control for and the fact that concentration measures only reflect interindustry variation in market power.15 C.

Alternative Measures of Market Power

Table 4 reports several important robustness checks of our results that use different variants of the definition of market power. The first set of robustness checks addresses intraindustry differences in market power. Besides the ability to price above marginal cost, two firm characteristics are usually associated with the presence of monopoly power: age and size. Firms with substantial market power tend to be older because it takes time to develop a dominant position in the product market. In addition, size provides the firm with bargaining leverage vis-a`-vis its suppliers or business customers. Although size and age can affect idiosyncratic volatility for reasons unrelated to competition, we expect, as a robustness requirement, that (conditional on a firm’s age and size) the same amount of pricing power has a stronger impact on idiosyncratic volatility the older or larger the firm is.16 The first two columns of table 4 present the result of interacting the EPCM with industry-normalized measures of age (col. 1) and size (col. 2). Both interactions have a negative statistically significant sign, showing that the negative impact of market power on idiosyncratic volatility is enhanced for larger and older firms. The second set of robustness checks in table 4 addresses different measures 14. The standard deviation of EPCM is 0.17, and the average log of idiosyncratic volatility (the dependent variable) is ⫺1.46, corresponding to 0.23 in variance terms and 48% in standard deviation terms. A one standard deviation change in EPCM changes the log of idiosyncratic volatility to ⫺1.46 ⫹ 0.17 ∗(⫺0.5) p ⫺1.55, or to 46% in standard deviation terms. The difference between the two values is ⫺2%. 15. As a final significance test, we checked the percentage of cross sections where our variables are significant at the 5% level using the heteroskedasticity-adjusted standard errors from each cross section. The percentage was 87% for EPCM, and it was 65% for the Herfindhal index. 16. We thank an anonymous referee for pointing this out to us.

Volatility and Competition

3139

of interindustry market power. Column 3 shows that results are similar if we use a Herfindahl-Hirschman index of concentration constructed using total firm assets instead of firm sales. Columns 4 and 5 report results using C4 and C10 indices of concentration instead. Again, results are statistically and economically similar. All the coefficients maintain the same sign as well as roughly the same levels of significance, and the results concerning control variables are also unchanged. Our conclusion from table 4 is that our results do not seem to depend on the particular measure of market power we use. These findings provide further confirmation that the degree of product market competition plays an important role in determining idiosyncratic volatility. D.

Industry-Level Analysis

In our final robustness test, we replicate our results aggregating firms into industries. It is important to confirm our results at the industry level because most of the industrial organization literature looks at market power from an industry viewpoint, assuming, for example, that pricing power is constant within industries (e.g., Domowitz et al. 1986a). Our results for the HerfindahlHirschman index show that interindustry differences in market power matter in firm-level regressions. It seems natural to ask, as a robustness check, whether such differences also matter in industry-level regressions. Table 5 reports the results of performing cross-sectional regressions at the industry level. For each year, the values of all variables are averaged across firms within each industry, using either value-based weights (panel A) or equal weights (panel B). Since the notion of EPCM is not applicable in this context, we report results using the industry PCM, as well as the HerfindahlHirschman index of concentration, as measures of market power. The main message of table 5 is that our results hold true at the industry level. The PCM coefficient (with t-statistics ranging from ⫺2.11 to ⫺2.32) and the Herfindahl-Hirschman index coefficient (with t-statistics from ⫺1.94 to ⫺2.33) are also negative and significant. Comparing our results with the previous tables, we see that among our control variables the dividend dummy coefficient still has the statistically strongest impact on volatility. We conclude that our findings are very robust to a wide variety of stringent tests. V.

Market Power and Uncertainty

A.

The Two Channels of Influence of Market Power

Our hypotheses posit that market power affects idiosyncratic volatility through two distinct channels. The first channel relies on the ability of companies with dominant market positions to hedge away idiosyncratic shocks by passing them on to their customers. The second channel is based on the fact that competition increases uncertainty about the firm’s average profitability and therefore the idiosyncratic volatility of returns. Our analysis so far has treated

3140

TABLE 4

Alternative Measures of Market Power Dependent Variable: Log of Idiosyncratic Volatility (1)

(3)

(4)

(5)

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

⫺1.277 .022 2.489 .523 .054 .000 .006 ⫺.265 .156 ⫺.013 ⫺.006 .003 .220 ⫺.320 ⫺.388 ⫺.001

⫺7.60*** 1.25 2.51** 1.77* .81 ⫺3.24*** .51 ⫺1.48 3.18*** ⫺2.82*** ⫺.27 .27 3.07*** ⫺8.27*** ⫺2.99*** ⫺2.01**

⫺1.276 .023 2.485 .514 .053 .000 .006 ⫺.268 .156 ⫺.013 ⫺.010 .004 .221 ⫺.320 ⫺.387

⫺7.71*** 1.35 2.51** 1.59 .78 ⫺3.24*** .50 ⫺1.53 3.17*** ⫺2.83*** ⫺.42 .31 3.10*** ⫺8.17*** ⫺3.07***

⫺1.181 .022 2.528 .544 .026 .000 .006 ⫺.438 .157 ⫺.013 .004 .007 .226 ⫺.323

⫺6.34*** 1.25 2.57** 1.66 .41 ⫺3.21*** .56 ⫺2.22** 3.18*** ⫺2.86*** .14 .52 3.17*** ⫺7.74***

⫺1.149 .022 2.528 .545 .036 .000 .006 ⫺.438 .157 ⫺.013 ⫺.009 .006 .225 ⫺.323

⫺6.15*** 1.24 2.57** 1.71 .61 ⫺3.20*** .53 ⫺2.22** 3.11*** ⫺2.86*** ⫺.39 .52 3.13*** ⫺7.88***

⫺1.111 .022 2.525 .543 .040 .000 .006 ⫺.438 .157 ⫺.013 ⫺.009 .006 .224 ⫺.323

⫺5.90*** 1.23 2.57** 1.68 .68 ⫺3.20*** .53 ⫺2.22** 3.10*** ⫺2.86*** ⫺.35 .48 3.13*** ⫺7.84***

Journal of Business

Constant Size Trading volume Volatility of profits Leverage Age Market-to-book Return on assets Beta Price Spin-off dummy S&P500 dummy Nasdaq dummy Dividend dummy EPCM (EPCM) # (Excess age)

(2)

.000

⫺.864 .098 ⫺.010

⫺5.34*** .43 ⫺2.68** Yes .69 1,919

⫺.865 .103 ⫺.010

⫺2.06**

⫺5.38*** .44 ⫺2.61 Yes .69 1,919

⫺.277

⫺2.02**

⫺.887 .090 ⫺.010

⫺5.45*** .39 ⫺2.79** Yes .67 1,919

⫺.145

⫺2.68**

⫺.887 .095 ⫺.010

⫺5.39*** .40 ⫺2.78** Yes .68 1,919

⫺.166 ⫺.887 .096 ⫺.010

⫺2.12** ⫺5.39*** .40 ⫺2.76** Yes .68 1,919

Volatility and Competition

(EPCM) # (Excess size) Asset-based HerfindahlHirschman index C4 concentration index C10 concentration index Institutional ownership Investor turnover Analysts Industry dummies Average adjusted R2 Average N

Note.—This table presents results of yearly cross-sectional regressions of idiosyncratic volatility on alternative measures of market power and other firm characteristics. (EPCM) # (Excess age) is the product of EPCM (defined as the firm’s price-cost margin minus the industry value-weighted average price-cost margin) times the difference between the firm’s age and the average age of firms in the industry. (EPCM) # (Excess size) is the product of EPCM times the difference between the firm’s size and the average size of firms in the industry. Asset-based Herfindahl-Hirschman is the sum of the squared asset shares (assets over total industry assets) of firms in the industry. Industries are defined using two-digit level SIC codes. The C4 concentration index and the C10 concentration index are the sum of market (sales) shares of the top four and the top 10 firms in each industry, respectively. All other variables are calculated according to the definitions given in the caption of table 3. The parameter estimates shown are time-series averages of the individual cross-sectional parameter estimates. Fama-McBeth t-statistics are adjusted for first-order autocorrelation using the Litzenberger and Ramaswamy (1982) correction. * Significant at the 10% level for the two-tailed hypothesis test that the coefficient equals zero. ** Significant at the 5% level for the two-tailed hypothesis test that the coefficient equals zero. *** Significant at the 1% level for the two-tailed hypothesis test that the coefficient equals zero.

3141

3142

TABLE 5

Industry-Level Analysis Dependent Variable: Log of Idiosyncratic Volatility A. Value Weighted (1)

Constant Size Trading volume Volatility of profits Leverage Age Market-to-book Return on assets Beta Price Spin-off dummy S&P500 dummy Nasdaq dummy Dividend dummy PCM Herfindahl-Hirschman index Institutional ownership Investor turnover Analysts Average adjusted R2 Average N

B. Equally Weighted

(2)

(3)

(4)

(5)

(6)

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

⫺1.747 .000 4.227 1.121 ⫺.384 .000 .002 ⫺.177 .236 ⫺.011 .906 ⫺.169 .185 ⫺.524 ⫺.245

⫺5.72*** ⫺1.72* 1.90* 1.45 ⫺1.77* ⫺1.82* .10 ⫺1.44 1.98* ⫺3.10*** 1.78* ⫺2.87*** .97 ⫺4.05*** ⫺2.32**

⫺1.927 ⫺.002 4.919 1.044 ⫺.250 .000 .010 ⫺.145 .273 ⫺.010 ⫺.751 ⫺.226 .023 ⫺.402

⫺6.15*** ⫺1.95* 1.96* 1.57 ⫺2.03** ⫺2.24** .92 ⫺1.38 2.15** ⫺2.99*** ⫺1.23 ⫺2.26** .21 ⫺3.22***

⫺1.665 ⫺.001 4.169 .390 ⫺.312 .000 .021 ⫺.209 .259 ⫺.009 ⫺1.222 ⫺.130 .054 ⫺.346

⫺3.59*** ⫺1.34 2.32** 1.02 ⫺1.17 ⫺1.29 1.05 ⫺2.19** 2.11** ⫺2.61** ⫺2.78** ⫺2.35** .76 ⫺2.06*

⫺4.03*** ⫺1.64 1.67 1.29 ⫺1.73* ⫺1.83* 1.87* ⫺1.50 1.41 ⫺3.75*** .79 ⫺1.37 .42 ⫺4.05***

⫺.829 ⫺.001 1.099 .210 ⫺.237 .000 .033 ⫺.781 .166 ⫺.010 .300 ⫺.061 .122 ⫺.820

⫺2.27** ⫺1.68 1.63 .64 ⫺.72 ⫺.77 1.11 ⫺1.83* 1.99* ⫺2.02* 1.04 ⫺.27 .77 ⫺3.19***

⫺2.26**

⫺.119 ⫺.669 ⫺.283 ⫺.001

⫺1.94* ⫺2.09** ⫺.57 ⫺.25

⫺4.63*** ⫺.74 1.83* 1.17 ⫺1.27 ⫺1.75* 1.43 ⫺1.86* 1.54 ⫺3.19*** 1.58 ⫺.65 .17 ⫺4.40*** ⫺2.11**

⫺1.080 ⫺.002 2.050 .324 ⫺.334 ⫺.001 .050 ⫺.531 .168 ⫺.014 .257 ⫺.177 .055 ⫺.840

⫺.112

⫺1.080 .000 2.428 .545 ⫺.256 ⫺.001 .031 ⫺.395 .163 ⫺.014 .453 ⫺.068 .025 ⫺.852 ⫺.408

⫺.193

⫺2.19**

⫺.240 ⫺1.045 ⫺.106 .003

⫺2.33** ⫺1.75* ⫺.15 .90

.78 56

.76 56

.78 57

.79 56

.80 56

.82 57

Journal of Business

Note.—This table presents results for a series of yearly cross-sectional regressions of idiosyncratic volatility on measures of market power and other firm characteristics, performed at the industry level. Industries are defined using two-digit level SIC codes. All the variables are defined as before (see note to table 3), except that they are aggregated at the industry level using value weighting (panel A) and equal weighting (panel B). The parameter estimates shown are time-series averages of the individual cross-sectional parameter estimates. Fama-McBeth t-statistics are adjusted for first-order autocorrelation using the Litzenberger and Ramaswamy (1982) correction. * Significant at the 10% level for the two-tailed hypothesis test that the coefficient equals zero. ** Significant at the 5% level for the two-tailed hypothesis test that the coefficient equals zero. *** Significant at the 1% level for the two-tailed hypothesis test that the coefficient equals zero.

Volatility and Competition

3143

these two effects jointly. This section asks whether we can find separate evidence for both of them. We endogenize volatility of profits to check for a direct impact of market power on volatility of profits and, separately, on idiosyncratic volatility.17 We build a simultaneous equation system that has idiosyncratic volatility and volatility of profits as endogenous variables and estimate it using two-stage least squares: LIVj,t p g0 ⫹ g1VOLPj,t ⫹ g2 Xj,t ⫹ g3 XEq1 VOLPj,t p d 0 ⫹ d1 LIVj,t ⫹ d 2 Xj,t ⫹ d 3 XEq2

j,t

j,t

⫹ ␧1j,t ,

⫹ ␧ 2j,t .

(2)

Identification of the system is provided by specific variables for each equation. We use the firm’s beta and price (two stock market–related variables) to identify the idiosyncratic volatility equation and the proportion of fixed costs and spending on R&D (two accounting-related variables) to identify the volatility of profits equation. The rationale for choosing the latter is that firms with a high proportion of fixed costs tend to exhibit greater fluctuations in profits due to operational leverage, while technologically intensive firms have more volatile profits relative to firms in established industries with mature technologies. The set of remaining control variables is similar to the one used in previous specifications, with the exception that we add yearly time dummies to accommodate for existing within-firm autocorrelation. Table 6 reports estimation results using the EPCM (panel A) and the Herfindahl-Hirschman index (panel B) as measures of market power. Panel A shows that the EPCM is significant in the idiosyncratic volatility equation, with t-statistics ranging from ⫺3.60 to ⫺5.93. Moreover, economic significance is unchanged, with EPCM displaying a point estimate of ⫺0.44. More important, the EPCM is strongly negatively significant in the volatility of profits equation (t-statistics ranging from ⫺6.70 to ⫺14.2). This means that indeed pricing power affects volatility of profits, as implied by the “market power as hedge” argument, but that at the same time it also affects return volatility directly, as implied by the “uncertainty” argument. Results in Panel B are broadly similar. The point estimate of the Herfindahl-Hirschman index in the idiosyncratic volatility, of around ⫺0.23, is identical to our previous estimate, implying equal levels of economic significance. The only difference is the significance level of the Herfindahl-Hirschman index in the volatility of profits equation, which seems lower albeit still statistically positive. One noticeable observation about table 6 is that the t-statistics seem rather high compared to those of previous tables. This raises the issue of whether the 17. The single-equation analysis uses volatility of profits as a control variable for exogenous business risk. Since volatility of profits also depends on market power, the single-equation specification can be interpreted as a “horse race” between, on the one hand, the autonomous impact of market power and, on the other hand, the joint impact of business risk, market power, and any other variable that affects volatility of profits. The coefficient of market power is consistently significant, while the one of volatility of profits is not.

3144

TABLE 6

Market Power, Idiosyncratic Volatility, and Volatility of Profits Panel A: Specifications Using EPCM as Measure of Market Power Dependent Variable Log Idiosyncratic Volatility (1) Coefficient

t-Statistic

2.062

6.65***

⫺.441 ⫺1.444 .000 .569 ⫺.119 ⫺.001 ⫺.022 ⫺.011 .090 ⫺.017 ⫺.090 ⫺.121 .319 ⫺.500

⫺5.93*** ⫺9.95*** 2.95*** 6.55*** ⫺2.90*** ⫺12.13*** ⫺3.86*** ⫺.89*** 14.07*** ⫺9.34*** ⫺2.19** ⫺4.57*** 20.09*** ⫺24.25***

Yes Yes .49 77,691

Coefficient

t-Statistic

.030 ⫺.195 .033 .000 .058 .075 .000 .013 ⫺.013

7.10*** ⫺14.20*** 1.96* ⫺4.03*** 2.80*** 7.98*** ⫺2.34** 15.79*** ⫺4.61***

.013 ⫺.002 ⫺.005 ⫺.016 .075 ⫺.130

1.06 ⫺.45 ⫺1.43 ⫺4.24*** 6.69*** ⫺3.59***

Yes Yes .12 77,691

Log Idiosyncratic Volatility (3) Coefficient

t-Statistic

1.960

5.23***

⫺.262 ⫺1.225 .000 1.552 ⫺.071 .000 ⫺.012 ⫺.139 .035 ⫺.012 ⫺.026 ⫺.009 .294 ⫺.329

⫺3.60*** ⫺35.24*** 5.55*** 20.74*** ⫺1.53 ⫺9.35*** ⫺2.02** ⫺2.88** 3.58*** ⫺6.22*** ⫺.62 .39 19.91*** ⫺18.26***

⫺.748 ⫺.130 ⫺.010

⫺14.14*** ⫺1.36 ⫺8.65*** Yes Yes .58 38,368

Volatility of Profits (4) Coefficient

t-Statistic

.030 ⫺.141 .057 .000 .071 .064 .000 .011 ⫺.030

3.02*** ⫺6.70*** 3.40*** ⫺3.11*** 2.26** 4.71*** ⫺.91 9.28*** ⫺2.22**

.014 ⫺.001 ⫺.015 ⫺.014 .061 ⫺.098 ⫺.050 .134 .001

1.16 ⫺.35 ⫺3.39*** ⫺2.61** 3.95*** ⫺2.42** ⫺3.62*** 5.59*** 2.01** Yes Yes .11 38,368

Journal of Business

Volatility of profits Log idiosyncratic volatility EPCM Constant Size Trading volume Leverage Age Market-to-book Return on assets Beta Price Spin-off dummy S&P500 dummy Nasdaq dummy Dividend dummy Fixed costs R&D spending Institutional ownership Investor turnover Analysts Industry dummies Time dummies R2 N

Dependent Variable Volatility of Profits (2)

Log Idiosyncratic Volatility (1) Coefficient Volatility of profits Idiosyncratic volatility Herfindahl-Hirschman index Constant Size Trading volume Leverage Age Market-to-book Return on assets Beta Price Spin-off dummy S&P500 dummy Nasdaq dummy Dividend dummy Fixed costs R&D Inst. ownership Investor turnover Analysts Industry dummies Time dummies R2 N

t-Statistic

1.788

5.26***

⫺.200 ⫺1.278 .000 .604 ⫺.114 ⫺.001 ⫺.017 ⫺.083 .089 ⫺.018 ⫺.093 ⫺.122 .336 ⫺.517

⫺2.85*** ⫺8.44*** 2.26** 6.94*** ⫺2.96*** ⫺11.71*** ⫺2.78*** ⫺4.47*** 14.46*** ⫺9.43*** ⫺2.34** ⫺4.46*** 20.15*** ⫺22.10***

Yes Yes .53 77,691

Dependent Variable Volatility of Profits (2)

Coefficient

t-Statistic

.047 ⫺.058 .095 .000 .024 .062 .000 .013 ⫺.029

11.82*** ⫺3.52** 5.52*** ⫺6.07*** 1.16 6.67*** 1.17 15.95*** ⫺8.26***

.013 ⫺.001 ⫺.008 ⫺.011 .069 .001

1.06 ⫺.23 ⫺2.15** ⫺3.03*** 6.05*** .02

Yes Yes .11 77,691

Log Idiosyncratic Volatility (3) Coefficient

t-Statistic

.872

2.95***

⫺.234 ⫺1.114 .000 1.686 ⫺.001 .000 ⫺.009 ⫺.355 .043 ⫺.013 ⫺.011 ⫺.009 .296 ⫺.350

⫺3.37*** ⫺35.70*** 4.59*** 27.74*** ⫺.03 ⫺9.98*** ⫺.46 ⫺6.19*** 5.57*** ⫺6.17*** ⫺.30 ⫺.44 21.28*** ⫺20.72***

⫺.870 .019 ⫺.010

⫺16.45*** .82 ⫺9.54*** Yes Yes .69 42,032

Volatility of Profits (4) Coefficient

t-Statistic

.061 ⫺.048 .121 .000 .044 .048 .000 .012 ⫺.070

11.08*** ⫺2.02** 2.18*** ⫺3.70*** 1.90* 3.71*** .50 10.27*** ⫺4.99***

.017 .001 ⫺.027 ⫺.003 .044 ⫺.051 ⫺.047 .076 .001

1.37 .34 ⫺6.47*** ⫺.74 3.00*** ⫺1.39 ⫺2.93*** 3.25*** 4.87***

Volatility and Competition

Panel B: Specifications Using Herfindahl-Hirschman Index as Measure of Market Power Dependent Variable

Yes Yes .09 42,032

3145

Note.—This table presents two-stage least squares estimates of the simultaneous relation between idiosyncratic volatility and volatility of profits in order to investigate the joint impact of market power on both measures. The measures of market power used are EPCM (panel A) and the Herfindahl-Hirschman index (panel B). Fixed costs is the ratio between selling, general, and administrative expenses and total operating costs. R&D spending is R&D expenditures over sales. All other variables are calculated according to the definitions given in the note of table 3. The t-statistics are calculated using robust clustered (by firm) standard errors. * Significant at the 10% level for the two-tailed hypothesis test that the coefficient equals zero. ** Significant at the 5% level for the two-tailed hypothesis test that the coefficient equals zero. *** Significant at the 1% level for the two-tailed hypothesis test that the coefficient equals zero.

3146

Journal of Business

statistical significance of our results is overstated in spite of the use of robust standard errors. To refute this possibility, we reestimate the system for every individual cross section. The (unreported) regressions display uniformly lower t-statistics. Furthermore, the results show that, in the idiosyncratic volatility equation, EPCM loads negatively at the 5% significance level (or lower) in 81% of the cases and at the 1% level (or lower) in 75% of the cases. In addition, in the volatility of profits equation, EPCM loads negatively at the 5% significance level (or lower) in 74% of the cases. Similar, slightly lower figures are obtained for the Herfindahl-Hirschman index. We conclude that the high t-statistics are probably an artifact of estimating the system by stacking all the observations for different years in a single pooled equation. We conclude that there exists sufficient evidence to support the claim that competition affects idiosyncratic volatility through two different channels: one related to profit volatility, and an independent channel, which we claim is linked to uncertainty. B.

Competition and Uncertainty

We have argued that the presence of competition is associated with greater firm-level uncertainty and that this uncertainty is one of the channels through which the competitive environment affects volatility. If this hypothesis is true, we should find evidence of this phenomenon by looking at measures of uncertainty surrounding the firm. Table 7 reports results from cross-sectional regressions having as the main dependent variable a measure of information uncertainty, the dispersion in analysts’ earnings-per-share forecasts (Barry and Brown 1985; Diether, Malloy, and Scherbina 2002; Qu, Starks, and Yan 2004). The specifications used are similar to those of table 3. If our hypothesis is correct, we should find that industries in concentrated sectors exhibit less disagreement in analyst forecasts. This conjecture is strongly supported by the data. Analyst forecast dispersion is negatively associated with profitability (i.e., ROA) and measures of market power, both intraindustry and interindustry, and positively associated with profit volatility and leverage, although the results are not consistently strong. The negative correlation between market power and analyst dispersion helps to shed light on the findings of Hou and Robinson (2003), who show that industries in concentrated sectors earn persistently lower returns. Lower information risk—proxied by lower analyst dispersion—could be one of the possible factors explaining their findings.18

18. Hou and Robinson (2003) hypothesize that their findings are due to the ability of firms in concentrated industries to insulate themselves from aggregate shocks, an argument similar to the one made here with respect to firm-specific shocks.

Market Power and Firm-Level Uncertainty Dependent Variable: Analyst Forecast Dispersion (1)

Constant Size Trading volume Volatility of profits Leverage Age Market-to-book Return on assets Beta Price Spin-off dummy S&P500 dummy Nasdaq dummy Dividend dummy EPCM Herfindahl-Hirschman index Institutional ownership Investor turnover Analysts Industry dummies Average adjusted R2 Average N

(2)

(3)

(4)

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

Coefficient

t-Statistic

.361 ⫺.004 .664 .640 .289 .000 .001 ⫺.652 ⫺.013 ⫺.003 .031 ⫺.032 ⫺.014 ⫺.124 ⫺.319

1.88 ⫺.35 1.35 4.48*** 1.71* 2.02* .09 ⫺2.12** ⫺.88 ⫺1.77* .38 ⫺2.07** ⫺.50 ⫺2.15** ⫺2.88***

.507 ⫺.034 .765 .101 .351 .000 .008 ⫺.728 ⫺.006 ⫺.003 .059 ⫺.032 ⫺.037 ⫺.128 ⫺.359

1.85 ⫺1.04 1.46 .74 1.72* 1.82* 1.04 ⫺2.10** ⫺.42 ⫺1.62 .72 ⫺1.59 ⫺1.24 ⫺1.85* ⫺2.09**

.432 ⫺.006 .680 .664 .275 .000 .000 ⫺.750 ⫺.015 ⫺.003 .032 ⫺.029 ⫺.010 ⫺.128

1.91 ⫺.49 1.35 4.65*** 1.73* 2.06** .01 ⫺2.45** ⫺1.06 ⫺1.66 .41 ⫺1.89* ⫺.38 ⫺2.12**

.594 ⫺.034 .817 .114 .334 .000 .007 ⫺.839 ⫺.008 ⫺.003 .062 ⫺.028 ⫺.033 ⫺.131

2.05* ⫺1.04 1.50 .80 1.75* 1.87* .93 ⫺2.33 ⫺.62 ⫺1.60 .78 ⫺1.42 ⫺1.16 ⫺1.86*

⫺.154 ⫺.372 .004

⫺1.87* ⫺1.63 .89

⫺.309

⫺2.55**

⫺.327 ⫺.173 ⫺.387 .004

⫺2.59** ⫺2.00** ⫺1.68 .82

Yes .05 1,423

Yes .04 1,543

Yes .04 1,423

Volatility and Competition

TABLE 7

Yes .04 1,543

3147

Note.—This table presents results of yearly cross-sectional regressions of analyst forecast dispersion (a measure of firm-level uncertainty) on measures of market power and other firm characteristics. Analyst forecast dispersion is the coefficient of variation (standard deviation divided by the mean) of one-quarter horizon earnings-per-share forecasts, calculated from I/B/E/ S. All other variables are calculated according to the definitions given in the note for table 3. The parameter estimates shown are time-series averages of the individual cross-sectional parameter estimates. Fama-McBeth t-statistics are adjusted for first-order autocorrelation using the Litzenberger and Ramaswamy (1982) correction. * Significant at the 10% level for the two-tailed hypothesis test that the coefficient equals zero. ** Significant at the 5% level for the two-tailed hypothesis test that the coefficient equals zero. *** Significant at the 1% level for the two-tailed hypothesis test that the coefficient equals zero.

3148

VI.

Journal of Business

Conclusion

This article investigates the link between a firm’s competitive environment and its idiosyncratic volatility. We argue that the competitive positioning of a firm can influence the impact of firm-specific shocks. We posit two effects: a natural hedge effect and an uncertainty effect. According to the first effect, market power works as a hedging instrument that smoothes out idiosyncratic fluctuations. According to the second effect, higher market power implies lower information uncertainty for the investors and therefore lower stock return volatility. We find strong support for both effects. We find that indeed the level of idiosyncratic noise in a firm’s returns is lower if a firm enjoys more market power relative to that of its peers. This result is valid both within and among industries, and it is robust to different ways of measuring pricing power. We separately identify the channels through which competition affects idiosyncratic volatility and show that competition is positively correlated with profit volatility as well as with a proxy for information uncertainty like dispersion in analysts’ forecasts. Anecdotal evidence suggests that competition has increased in the past decades due to deregulation and globalization. Our results therefore suggest that product market competition has probably contributed to the recent upward trend in idiosyncratic volatility. Our results also show that different competitive environments, by affecting the extent to which the firm can smooth out idiosyncratic shocks and the uncertainty surrounding the firm, might potentially affect the value of the firm if idiosyncratic risk is priced.

Appendix This appendix describes in detail the construction of the variables and the rationale for their use, as well as the original references whenever necessary. Idiosyncratic volatility is the sum of squared residuals with respect to the capital asset pricing model. For each month in the sample, we run a time-series regression of the previous 36 monthly returns of a stock on the market. We then fit the resulting beta estimates to daily returns in the current month. Idiosyncratic volatility for the current month is the sum of the squares of the daily residuals. Following Duffee’s (1995) suggestion, we do not take into account possible autocorrelation of the residuals within the month. Many firms could have within-month estimates of autocorrelation lower than ⫺0.5, which would imply negative variance estimates. The yearly idiosyncratic volatility is the sum of the monthly idiosyncratic volatilities within the year. The price-cost margin, also referred to as the Lerner index, is defined as operating profits (before depreciation, interest, special items, and taxes) over sales. The numerator is calculated as sales (Compustat annual data item 12) minus cost of goods sold (41) minus selling, general, and administrative expenses (189). Whenever this calculation is not possible, we use operating income (178). Excess PCM (EPCM) is defined as the firm’s PCM minus the industry valueweighted average PCM. The Herfindahl-Hirschman index is calculated as the sum of

Volatility and Competition

3149

the squared market shares (sales over total industry sales) of firms in the industry, where industries are defined using two-digit level SIC codes. The asset-based Herfindahl-Hirschman index is calculated in the same way but it uses firm total assets (item 6) instead. The C4 concentration index and the C10 concentration index are the sum of market shares of the top four and the top 10 firms in each industry, respectively. The sectors subject of deregulation used in table 2 consist of the companies having the following SIC codes: 4512 (airlines); 4911 and 5063 (electricity); 4923–4925 (natural gas); 4813 (telecoms); and 4212–4215 and 4231 for trucking and 4011 and 4013 for railroads (transportation). Trading volume, the CRSP (Center for Research in Security Prices) average monthly number of shares traded over total shares outstanding, controls for the observed positive relationship between volatility and volume. Volatility of profits controls for the riskiness of the firm’s underlying technology (some firms might be subject to higher levels of idiosyncratic shocks because their business is more volatile, independently of their ability to exert market power). Volatility of profits is calculated, following Pa´stor and Veronesi (2003), as the root mean squared error of a regression of the firm’s return on equity (ROE) on its lag. We use quarterly items to have sufficient degrees of freedom to run the regression. ROE is calculated as quarterly earnings over previous year book value. Quarterly earnings are calculated as income before extraordinary items available to stockholders (quarterly data item 25) plus deferred taxes from the income statement (quarterly data item 35). Book equity is constructed as stockholder’s equity plus balance sheet deferred taxes and investment tax credit (annual data item 35) minus the book value of preferred stock. Stockholder’s equity is annual data item 216, or 60 ⫹ 130, or 6 ⫺ 181, in that order. Preferred stock is calculated as item 56, or item 10, or item 130, in that order. Leverage, total long-term debt (item 9) over total assets, controls for the increase in shareholder risk brought about by increased indebtedness (Black 1976). Age, the number of months since the date the stock first appears in CRSP, controls for investor uncertainty (Pa´stor and Veronesi 2003). Market-to-book, the ratio between market value and book value of equity, controls for the existence of growth opportunities. Return on assets, the ratio between operating profits and total assets, controls for the level of profitability of the business. Beta, the yearly average CAPM beta estimate, controls for the fact that idiosyncratic volatility is a measure of substitutability between stocks (Wurgler and Zhuravskaya 2002). We use beta given that comovement proxies for the degree to which certain large investors consider stocks to be substitutable (Barberis, Shleifer, and Wurgler 2005). Price, the firm’s stock price recorded in CRSP, controls for the impact of possible microstructure noise on volatility estimates for lowpriced stocks. Institutional ownership, the ratio of a firm’s shares held by institutional investors relative to total shares outstanding, controls for the positive relationship between institutional investment and idiosyncratic volatility (Sias 1996; Malkiel and Xu 2003). Investor turnover controls for the volatility-inducing “transience” of institutional investors (Bushee and Noe 2000). Following Gaspar, Massa, and Matos (2006), define the “churn rate” of investor i in firm k’s shares at time t as CR k,i,t p

FNk,i,t Pk,t ⫺ Nk,i,t⫺1 Pk,t⫺1 ⫺ Nk,i,t⫺1 DPk,t⫺1F

(Nk,i,t Pk,t ⫹ Nk,i,t⫺1 Pk,t⫺1 )/2

.

3150

Journal of Business

Denote as S the set of shareholders in company k and as wk,i,t the weight of investor i in the total percentage held by institutional investors in quarter t. The investor turnover variable is the weighted average of the (time-)average churn rates of its investors over the four quarters of the year: Turnover of firm k p

冘 (冘 wk,i,t

i苸S

1 4

4 rp1

)

CR k,i,t⫺r⫺1 .

Analysts, the number of analysts following the stock (obtained from I/B/E/S), controls for the fact that idiosyncratic volatility reflects firm-specific information in developed capital markets with strong investor protection (Morck, Yeung, and Yu 2000). The spin-off dummy takes a value of one if a company spun off a unit or was itself spun off during the current or the past year. The spin-off dummy controls for the Campbell et al. (2001) suggestion that conglomerate breakup might increase idiosyncratic volatility because risks that were bundled together are now traded separately. Spin-off data is taken from the SDC New Issues database. The S&P500 and the Nasdaq dummies are obtained from CRSP and Compustat, respectively. These dummies are employed to see to what extent our results are particular to the subset of a subset of bigger, mature firms or to small, high-tech firms. The dividend dummy, which takes a value of one if Compustat reports a positive dividend being paid in that year, controls for the lower impact of learning uncertainty on volatility for dividend-paying firms (Pa´stor and Veronesi 2003). Fixed costs, the ratio between selling, general, and administrative expenses and total operating costs (items 189/(189⫹41)), controls for operating leverage. R&D spending, the ratio of R&D expenditures (item 46) over sales, controls for the maturity of the firm’s technology. Analyst forecast dispersion, the coefficient of variation (standard deviation divided by the mean) of one-quarter horizon earnings-per-share forecasts from I/B/E/S, is a measure of information uncertainty (Barry and Brown 1985; Diether et al. 2002; Qu et al. 2004).

References Aghion, P., N. Bloom, R. Blundell, R. Griffith, and P. Howitt. 2002. Competition and innovation: An inverted U relationship. NBER Working Paper no. 9269. National Bureau of Economic Research, Cambridge, MA. Bali, T. G., N. Cakici, X. Yan, and Z. Zhang. 2005. Does idiosyncratic risk really matter? Journal of Finance 60, no. 2:905–29. Barberis, N., A. Shleifer, and J. Wurgler. 2005. Comovement. Journal of Financial Economics 75, no. 2:283–317. Barry, C. B., and S. J . Brown. 1985. Differential information and security market equilibrium. Journal of Financial and Quantitative Analysis 20:407–22. Bernard, A. B., J. B. Jensen, and P. K. Schott. 2006. Survival of the best fit: Exposure to lowwage countries and the (uneven) growth of U.S. manufacturing plants. Journal of International Economics 68, no. 1:219–37. Black, F. 1976. Studies of stock price volatility changes. In Proceedings of the 1976 Meetings of the Business and Economic Statistics Section, American Statistical Association, 177–81. Brown, D. P., and M. A. Ferreira. 2003. Information in the idiosyncratic volatility of small firms. Working paper, School of Business, University of Wisconsin–Madison and ISCTE Business School, Lisbon. Bushee, B., and C. Noe. 2000. Corporate disclosure practices, institutional investors, and stock return volatility. Journal of Accounting Research 38, suppl.:171–202. Campbell, J. Y., M. Lettau, B. G. Malkiel, and Y. Xu. 2001. Have individual stocks become more volatile? An empirical exploration of idiosyncratic volatility. Journal of Finance 56:1–43.

Volatility and Competition

3151

Campbell, J. Y., and R. J. Shiller. 1988. The dividend-price ratio and expectations of future dividends and discount factors. Review of Financial Studies 1:195–228. Carlton, D. W., and J. M. Perloff. 1989. Modern industrial organization. Glenview, IL: Scott, Foresman. Crandall, R. W. 1991. After the breakup: U.S. telecommunications in a more competitive era. Washington, DC: Brookings Institution. Diether K. B., C. J. Malloy, and A. Scherbina. 2002. Differences of opinion and the cross section of stock returns. Journal of Finance 57, no. 5:2113–41. Domowitz, I., R. G. Hubbard, and B. C. Petersen. 1986a. Business cycles and the relationship between concentration and price-cost margins. RAND Journal of Economics 17, no. 1:1–17. ———. 1986b. The intertemporal stability of the concentration-margins relationship. Journal of Industrial Economics 35, no. 1:13–34. Duffee, G. R. 1995. Stock returns and volatility: A firm level analysis. Journal of Financial Economics 37:399–420. Fama, E. F., and K. R. French. 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, no. 1:3–56. Fama, E. F., and J. D. McBeth. 1973. Risk, return, and equilibrium: Empirical tests. Journal of Political Economy 81:607–36. Federal Trade Commission. 1992. Horizontal merger guidelines. Washington, DC. Gaspar, J., M. Massa, and P. Matos. 2006. Favoritism in mutual fund families: Evidence on strategic cross-fund subsidization. Journal of Finance 61, no. 1:73–104. Goyal, A., and P. Santa-Clara. 2003. Idiosyncratic volatility matters! Journal of Finance 58: 975–1008. Herfindahl, O. C. 1950. Concentration in the U.S. steel industry. PhD diss. Columbia University. Hicks, J. R. 1935. The theory of monopoly. Econometrica 3, no. 1:1–20. Hirschman, A. O. 1945. National power and the structure of foreign trade. Berkeley: University of California Press. Hou, K., and D. T. Robinson. 2003. Industry concentration and average stock returns. Dice Center working paper no. 2003–30, Department of Finance, Ohio State University, and Fuqua School of Business, Duke University. Lerner, A. P. 1934. The concept of monopoly and the measurement of monopoly power. Review of Economic Studies 1, no. 3:157–75. LeRoy, S. F., and R. D. Porter. 1981. The present-value relation: Tests based on implied variance bounds. Econometrica 49, no. 3:555–74. Litzenberger, R. H., and K. Ramaswamy. 1979. The effect of personal taxes and dividends on capital asset prices. Journal of Financial Economics 7, no. 2:163–95. ———. 1982. The effects of dividends on common stock prices: Tax effects or information effects? Journal of Finance 37, no. 2:429–43. Malkiel, B. G., and Y. Xu. 2001. Idiosyncratic volatility and security returns. Unpublished manuscript, Department of Economics, Princeton University, and School of Management, University of Texas at Dallas. ———. 2003. Investigating the behavior of idiosyncratic volatility. Journal of Business, 76, no. 4:613–45. Merton, R. C. 1987. A simple model of capital market equilibrium with incomplete information. Journal of Finance 42, no. 3:483–510. Morck, R., B. Yeung, and W. Yu. 2000. The information content of stock markets: Why do emerging markets have synchronous stock price movements? Journal of Financial Economics 58, no. 2:215–60. Pa´stor, L., and P. Veronesi. 2003. Stock valuation and learning about profitability. Journal of Finance 58, no. 5:1749–90. Qu, S., L. Starks, and H. Yan. 2004. Risk, dispersion of analyst forecasts and stock returns. Unpublished manuscript, Department of Finance, University of Texas at Austin. Ryan, D. J. 1997. Trends in market power and productivity growth rates in U.S. and Japanese manufacturing. Economics Letters 57:183–87. Shepherd, W. G. 1982. Causes of the increased competition in the U.S. economy, 1939–1980. Review of Economics and Statistics 64, no. 4:613–26. Sias, R. W. 1996. Volatility and the institutional investor. Financial Analysts Journal 54, no. 2: 13–20. Suppanz, H., M. Wise, and M. Kiley. 2004. Product market competition and economic perfor-

3152

Journal of Business

mance in the United States. OECD Economics Department working paper no. 398. Organisation for Economic Cooperation and Development, Paris. Vuolteenaho, T. 2002. What drives firm-level stock returns? Journal of Finance 57:233–64. Wei, S. X., and C. Zhang. 2006. Why did individual stocks become more volatile? Journal of Business 79:259–92. Welch, I. 2004. Capital structure and stock returns. Journal of Political Economy 112:106–31. Winston, C. 1998. U.S. industry adjustment to economic deregulation. Journal of Economic Perspectives 12, no. 3:89–110. Wurgler, J., and E. Zhuravskaya. 2002. Does arbitrage flatten demand curves for stocks? Journal of Business 75:583–608.