Is accruals quality a priced risk factor?

https://doi.org/10.1016/j.jacceco.2007.08.001Get rights and content

Abstract

In a recent and influential empirical paper, Francis, LaFond, Olsson, and Schipper (FLOS) [2005. The market pricing of accruals quality. Journal of Accounting and Economics 39, 295–327] conclude that accruals quality (AQ) is a priced risk factor. We explain that FLOS’ regressions examining a contemporaneous relation between excess returns and factor returns do not test the hypothesis that AQ is a priced risk factor. We conduct appropriate asset-pricing tests for determining whether a potential risk factor explains expected returns, and find no evidence that AQ is a priced risk factor.

Introduction

In a recent and influential paper, Francis, LaFond, Olsson, and Schipper (FLOS, 2005) examine whether accruals quality (AQ) is a determinant of the cost of capital. FLOS conclude that “information risk (as proxied by accruals quality) is a priced risk factor” (p. 296), and that “accruals quality plays a statistically and economically meaningful role in determining the cost of equity capital” (p. 315). FLOS base their inference, in part, on coefficients from time-series regressions of contemporaneous stock returns on returns to portfolios that mimic exposure to AQ, the market, size, and book-to-market. We explain that FLOS’ time-series regressions of contemporaneous stock returns on contemporaneous factor returns do not test the hypothesis that AQ is a priced risk factor.1 We then conduct appropriate tests for determining whether a risk factor is priced, and find no evidence that AQ is a priced risk factor.

Whether information risk is diversifiable is an open question in the literature. Traditional asset-pricing theory (e.g., Fama, 1991) takes the position that information risk is diversifiable and should not affect expected returns. More recently, Easley and O’Hara (2004) develop a model in which firms with less public and more private information have greater information risk and higher expected returns. They argue that uninformed investors are not able to adjust their portfolio weights in the same way as informed investors and, therefore, information risk cannot be diversified away. Lambert et al. (2007, pp. 396–397), however, argue that when the number of traders becomes large in the Easley and O’Hara (2004) model, the information effect is diversified away. If the Lambert et al. claim is correct, the Easley and O’Hara (2004) model provides no support for the hypothesis that information risk or accounting quality is priced.

In addition, even if information risk is not diversifiable, it is still debatable whether it should be included as an additional risk factor in asset-pricing models. Lambert et al. (2007) study how accounting information could affect the cost of capital in an economy with multiple assets. They develop a model consistent with the CAPM in which accounting information affects investors’ assessments of the covariance of firm cash flows with those of the market, and therefore affects firm beta. Consequently, this model suggests that information risk affects firm beta, but that a well-specified, forward-looking beta would fully capture cross-sectional differences in expected returns. However, if beta is measured with error, a proxy for information risk could appear to be priced if it proxies for measurement error in beta. In a similar vein, Hughes et al. (2007) study information risk in the context of a multi-factor asset-pricing model, and develop a model that suggests that information signals are either diversifiable or are captured by existing factor risk premiums.

FLOS document a positive and significant coefficient on an AQ mimicking portfolio in firm-specific time-series regressions that correlate returns contemporaneously with the AQ_factor and the Fama and French (1993) factors (market, size, and book-to-market). Specifically, they find that returns are positively correlated with an AQ_factor, where the AQ measure is greater for firms with poorer accounting quality. As we illustrate in more detail below, the fact that the average coefficient on the AQ_factor is positive and statistically significant in FLOS’ regressions does not imply that the AQ_factor is a priced risk factor. Rather, the average positive coefficient indicates that, on average, the firms in the contemporaneous time-series regressions have positive exposure to the AQ_factor. To put the FLOS result in a familiar context, by itself a positive coefficient in a contemporaneous regression of stock returns on the market portfolio does not imply that the market factor is priced, but simply confirms that the average beta in a random sample of firms is positive and mechanically close to one.

There are a number of methods to test whether a proposed risk factor is priced. The most common method in the literature is a two-stage cross-sectional regression technique (2SCSR) that estimates factor betas in the first stage, and the factor risk premiums in the second stage. This method provides a well-specified test of the hypothesis that a proposed risk factor explains cross-sectional variation in expected returns, and as such, significant factor risk premiums are taken as evidence that a given risk factor is priced. This method has been used over time to test the CAPM (Fama and MacBeth, 1973), the conditional CAPM (Jagannathan and Wang, 1996), the intertemporal CAPM (Brennan et al., 2004; Petkova, 2006), the two-beta model (Campbell and Vuolteenaho, 2004), and to test whether default risk or takeover risk are compensated and priced factors (e.g., Vassalou and Xing, 2004; Cremers et al., 2007). In addition, Barth et al. (2006) use this method to test the hypothesis that greater financial statement transparency, as proxied by the value-relevance of earnings, is associated with a lower cost of capital. We apply the 2SCSR technique to test whether AQ is a priced risk factor. Our results suggest that AQ is not a priced factor since it does not carry a positive risk premium with respect to returns.

In addition to the 2SCSR tests, we also examine the pricing of AQ using several other approaches that are found in the literature. One such test is to examine whether there is a significant unconditional time-series mean annual risk premium on the AQ factor. We find that FLOS’ AQ_factor generates a mean annual risk premium of about 3%, which is not statistically different from zero. The lack of a significant risk premium suggests that AQ is unlikely to be priced (Shanken and Weinstein, 2006).2 Another pricing test is to examine whether firm characteristics predict future excess returns (e.g., Fama and French, 1992; Easley et al., 2002). We find no evidence that AQ as a characteristic predicts future excess returns. Finally, we conduct a time-series pricing test that is similar to the approach used by Aboody et al. (2005), where we examine whether a mimicking portfolio strategy that buys (sells) firms with high (low) AQ beta earns positive abnormal returns. This test is commonly used in finance and accounting as an alternative method of documenting a relation between firm characteristics and expected returns. We find that the AQ hedge portfolio strategy does not earn positive returns during the full 1971–2003 period (although we find significantly positive hedge returns in the January 1985–November 2003 sub-period used by Aboody et al. (2005)). To the extent that longer time-series are advisable (Lundblad, 2005), our results suggest that there is no evidence that AQ is positively associated with future returns.

Like us, Aboody et al. (2005, p. 659) recognize that “positive loadings [on factors in contemporaneous time-series regressions] do not in themselves imply a non-zero risk premium.” Although their primary focus is on whether insiders trade profitably on private information due to information asymmetry resulting from poor accounting quality, they also conduct a hedge portfolio analysis to assess the pricing of AQ. They find no statistically significant evidence from asset-pricing tests that AQ is priced, but they offer the reader a different conclusion (pp. 665–666):

Our results show that the evidence is in fact weaker than one might surmise from factor loading estimates alone. Of course, this difference in conclusion is at a quantitative level only. In the next section, we show [that the spread between low-quality and high-quality] firms is positively correlated with insider trading profits, a finding that further supports the notion that the systematic component of the asymmetric information factor is priced.

Thus, while a reader who is familiar with asset-pricing tests might glean from Aboody et al. that the FLOS result is weak, the point of our paper is to illustrate that the returns-based tests in FLOS do not test whether AQ is a priced risk factor, and that our tests provide no evidence that AQ is priced.

In addition, the Aboody et al. results that we replicate are based on equal-weighted returns to portfolios that are rebalanced monthly. Similarly, Ecker et al. (2006) and Nichols (2006) examine equal-weighted daily returns to AQ hedge portfolios that require daily rebalancing, and find significant abnormal returns. An equal-weighted returns strategy requires frequent rebalancing, which can lead to biases in computed returns due to bid–ask spread bounces (Blume and Stambaugh, 1983). These biases can be systematic when the portfolio formation procedure is correlated with size. Because AQ is highly correlated with size, it is possible that these return biases are present in the equal-weighted portfolios of Aboody et al. (2006), Ecker et al. (2006), and Nichols (2006). We examine this possibility by first replicating the 20% annualized return that Ecker et al. document based on daily returns to an AQ hedge portfolio that is rebalanced daily to equal weights. We then show that the AQ hedge portfolio earns no significant returns when we instead examine daily buy-and-hold returns to an AQ portfolio that is rebalanced to equal weights once per year. We also find that the AQ hedge portfolio return (using daily returns and daily rebalancing to equal weights) becomes insignificant if we exclude stocks with price below $5; for these low-price stocks, the bias from daily rebalancing to equal weights is expected to be larger.

FLOS’ conclusions regarding AQ as a determinant of the equity cost of capital are not based solely on the contemporaneous time-series regressions described above, but also on tests that correlate AQ with industry-adjusted earnings/price ratios and with the interest rate on debt (FLOS, 2005), and with implied cost of capital estimates (FLOS, 2004). With respect to the FLOS (2005) results, Liu and Wysocki (2006) show that AQ loses significance in the earnings-to-price and interest rate regressions when idiosyncratic risk is introduced. Khan (2007) shows that the relation between AQ and earnings-to-price is sensitive to whether the earnings-to-price ratio is log transformed and the method used to control for industry effects. On the other hand, as we show in supplemental analysis, the relation between AQ and the implied cost of capital measure used by FLOS (2004) appears to be very robust. Although the implied cost of capital measure is significantly positively correlated with future realized returns, we do not find evidence that this positive correlation is driven by accounting quality.

In conclusion, Francis et al., 2004, Francis et al., 2005 claim four sets of results as evidence that AQ is priced. We show that returns-based tests provide no evidence that AQ is a priced risk factor, and contemporaneous research argues that the earnings-to-price tests and interest-rate tests are not robust. Only the implied cost of capital results appear robust, and they are based on a much smaller sample than the return results. Finally, although the implied cost of capital measure used by FLOS is significantly positively correlated with future realized returns, we do not find evidence that this positive correlation is driven by accounting quality.

The remainder of the paper proceeds as follows. In the next section, we replicate FLOS (2005) and illustrate that time-series regressions do not test the hypothesis that a factor is priced. In the third section, we test whether AQ is a priced risk factor using the 2SCSR approach, and also test whether AQ as a characteristic predicts future returns. In the fourth section we conclude.

Section snippets

Replication of FLOS and critique of FLOS time-series approach

FLOS calculate an equal-weighted portfolio that ranks firms into quintiles based on AQ and buys (sells) firms in the two highest (lowest) AQ quintiles, where lower AQ is considered to be higher accounting quality. To estimate AQ, we follow FLOS (2005, p. 302), and estimate a regression of total current accruals (TCA) on lagged, current, and future cash flows plus the change in revenue and PPE.

All variables are deflated by average total assets.TCAj,t=φ0,j+φ1,jCFOj,t-1+φ2,jCFOj,t++φ3,jCFOj,t+1+φ4,

Two-stage cross-sectional regressions (2SCSR)

In this section we test whether the AQ_factor is a priced risk factor using a two-stage cross-sectional regression approach (2SCSR), where excess returns are regressed on risk factor betas. This method presents one approach to test whether a candidate variable is a priced risk factor (Cochrane, 2005). Following FLOS and Aboody et al. (2005), and a large asset-pricing literature (e.g., Pastor and Stambaugh, 2003; Petkova, 2006), our tests examine whether AQ is a priced risk factor after

Conclusion

Whether accounting quality in general, and accruals quality (AQ) in particular, affects the cost of capital is an important subject for researchers and practitioners. While the intuition of many people is that information matters for the capital markets, there is no well-accepted theory that proves that information risk is not diversifiable. In light of recent research, the theoretical case for information effects on the cost of capital has grown weaker. Lambert et al. (2007) show that the

References (49)

  • A. Brav et al.

    Using expectations to test assets pricing models

    Financial Management Autumn

    (2005)
  • M.J. Brennan et al.

    Estimation and test of a simple model of intertemporal capital asset pricing

    Journal of Finance

    (2004)
  • J.Y. Campbell et al.

    Bad beta, good beta

    American Economic Review

    (2004)
  • S. Chen et al.

    Does the pricing of financial reporting quality change around dividend changes?

    Journal of Accounting Research

    (2007)
  • J.H. Cochrane

    Asset Pricing

    (2005)
  • Cohen, D., 2003. Does information risk really matter? An analysis of the determinants and economic consequences of...
  • M. Cremers et al.

    Takeovers and the cross-section of returns

    Review of Financial Studies

    (2007)
  • K. Daniel et al.

    Overconfidence, arbitrage, and equilibrium asset pricing

    Journal of Finance

    (2001)
  • D. Easley et al.

    Information and the cost of capital

    Journal of Finance

    (2004)
  • D. Easley et al.

    Is information risk a determinant of asset returns?

    Journal of Finance

    (2002)
  • P. Easton et al.

    An evaluation of the reliability of accounting based measures of expected returns: a measurement error perspective

    The Accounting Review

    (2005)
  • F. Ecker et al.

    A returns-based representation of earnings quality

    The Accounting Review

    (2006)
  • E. Elton

    Expected return, realized return, and asset pricing tests

    Journal of Finance

    (1999)
  • E. Fama

    Efficient markets: II

    Journal of Finance

    (1991)
  • Cited by (247)

    • Textual disclosure complexity and analysts’ weighting of information

      2024, Journal of Contemporary Accounting and Economics
    • Is the cash-returns relationship risk induced?

      2024, North American Journal of Economics and Finance
    • Explaining accruals quality over time

      2023, Journal of Accounting and Economics
    • Why corporate political geography matters for stock returns

      2023, International Review of Economics and Finance
    View all citing articles on Scopus

    We appreciate the comments of Jennifer Francis, Bob Holthausen, Mo Khan, S.P. Kothari (the editor), Ryan LaFond, Christian Leuz, Vinay Nair, Craig Nichols, Per Olsson, Scott Richardson, Katherine Schipper, Lakshmanan Shivakumar (a referee), Susan Shu, a second anonymous referee, and seminar participants at Stanford University and the Wharton School, and gratefully acknowledge the financial support of the Wharton School. Rodrigo Verdi is also grateful for financial support from the Deloitte & Touche Foundation and from the MIT Sloan School of Management.

    View full text