Internet Appendix for “Accrual Quality, Skill, and the Cross-Section of Mutual Fund Returns”
by
Suresh Nallareddy
Maria Ogneva
This draft: November 2016
Introduction
This appendix reports theresults of ancillary tests including the tests of stock-level association between accrual quality (AQ) and equity returns and robustness tests. For the relevant equations and methods, refer to the main paper.
1.1AQ and equity returns: stock-level analysis
We document an economically and statistically significant AQ premium in mutual fund returns. These findings stand in stark contrast to prior evidence based on stock returns (e.g., Core et al. 2008; Ogneva 2012). Our sample starts in 1991, which is significantly later than in prior studies. To ensure that this difference in sample periods does not cause the discrepancy in results, we replicate the stock-based analyses within our sample period. If the positive AQ premium documented in mutual fund returns arises due to using a later period, we should also observe a significantly positive AQ premium for this sample of stocks.
Table A1 contains the stock-level results. All analyses are based on a sample of stocks that are held by at least one mutual fund in a given month. Panel A reports descriptive statistics and future returns for 10 AQ portfolios formed by sorting stocks each month into decile portfolios based on the AQ measure. The reported values are the time-series means of monthly portfolio averages. The associations between stock characteristics and AQ resemble those reported earlier for mutual funds. Specifically, worse AQ stocks tend to be smaller and have higher betas, lower book-to-market ratios, and higher lagged returns.
Table A1, Panel A, also reports time-series means of excess, FF3, and FF5 returns for the AQ portfolios. The FF3 and FF5 return calculations are performed analogously to those described above for the calculation of mutual fund returns. We report both the differences in returns on extreme AQ decile portfolios and the slopes from regressions of portfolio returns on AQ ranks. In contrast to results based on the mutual fund returns, we find that neither the differences in returns between extreme AQ portfolios nor the AQ slope estimates are statistically significant for the stock sample.
Panel B of Table A1 replicates regressions of future realized stock returns on ranked stock characteristics. The results differ sharply from those of our prior analysis based on mutual fund returns. First, the estimated AQ premiums (the slopes on AQ decile ranks) are not only statistically insignificant but also negative. After we control for market beta, market capitalization, book-to-market, momentum, and liquidity ranks, the AQ premium is equal to -0.018%,when we use excess returns, and -0.021% (-0.008%),when we use FF3 (FF5) returns. The insignificant AQ premiums for stocks held by the mutual funds suggests that it is not the selection of stocks per se, but rather the greater weights that skilled managers put on stocks with better prospects, that drives our fund-level results.
Second, in contrast to our mutual-fund-based results, other characteristics (such as market beta, market capitalization, book-to-market, and liquidity) have statistically significant premiums. As previously noted, a greater dispersion in characteristics for individual stocks as compared to well-diversified mutual fund portfolios likely increases the power to detect characteristic-related premiums.
Overall, the fact that we find no evidence of a positive AQ premium in returns on stocks held by mutual funds suggests that the positive AQ premium documented using mutual fund returns is not driven by the later sample period and is unlikely to be entirely driven by institutional factors restricting mutual fund ownership.[1]
1.2Additional analyses and robustness checks
Stock prices and competition for shares
Two prior studies find that AQ is associated with realized returns for certain stock subsamples. Kim and Qi (2010) document a positive AQ premium after restricting the sample to stocks with prices over $5 at the beginning and the end of each month, which they interpret as controlling for market microstructure effects. Armstrong et al. (2011) document significantly higher returns for poor AQ stocks for a subsample of firms with a smaller number of shareholders, consistent with their conjecture that information asymmetry caused by poor AQ is priced only in imperfectly competitive capital markets. If mutual funds invest in stocks with higher prices or lower capital market competition, then our results may be related to the findings of Kim and Qi (2010) or Armstrong et al. (2011).
We first differentiate our results from the market microstructure effect in Kim and Qi (2010). Untabulated results suggest that mutual funds indeed invest in stocks with higher prices as compared to the CRSP-Compustat universe. In particular, the average price score (a weighted-average decile rank of stock prices) is 7.60, higher than the CRSP-Compustat average of 4.5. However, when we restrict our sample to stocks with prices above $5 at the beginning of each month, we find no evidence of a significant positive association between AQ and stock returns in our sample. Specifically, the untabulated AQ premium (the coefficient on rAQ in regression (3) based on individual stock returns) for the restricted sample is consistently negative and ranges from -0.022 to -0.008 depending on the return measure used (i.e., excess, FF3, or FF5 return). Kim and Qi (2010) may obtain a significant AQ premium because they restrict stocks to those with end-of-month prices above $5. Such a restriction effectively removes observations where prices drop below $5, which can be equivalent to omitting negative returns for low priced stocks that are more likely to have poor AQ.
We next show that our results are unrelated to Armstrong et al.’s findings. Mutual funds, on average, invest in stocks with stronger capital market competition. Specifically, the average number of shareholders (weighted-average decile rank of the number of shareholders) is 7.11, which is higher than the average rank of 4.5 for the CRSP-Compustat universe. Thus, by focusing on mutual fund returns, we find evidence of an AQ premium in a competitive setting, which is different from Armstrong et al.’s conjecture that AQ is priced as a measure of information asymmetry only in imperfectly competitive capital markets.
In a study related to Armstrong et al., Akins et al. (2012) similarly predict that information quality is priced only in the presence of imperfect competition among investors. They use either the proportion of shares held by institutional investors or the concentration of institutional ownership to measure competition and to document that AQ premiums are significantly lower for firms with higher institutional ownership.[2] These findings are consistent with the lack of statistically significant AQ premiums for a sample of stocks held by mutual funds that we document in Section 1.2. Overall, our findings are unlikely to be related to the competition effects documented in Armstrong et al. (2011) and Akins et al. (2012).
Raw AQ measure
Our main results use decile ranks of AQ, which facilitates interpretation of the regression coefficients. However, we also re-estimate our main tests using the raw weighted-average AQ of mutual fund holdings. All main findings are robust to using this alternative measure. In particular, the AQ premium—the coefficient on the raw AQ measure in regression (3)—is significantly positive across all regression specifications. The skill-based results are also similar to the results described in Section 4.5 of the draft. First, AQ premiums increase monotonically from the low- to the high-skill partition. Second, differences in AQ premiums between the two extreme skill terciles are statistically significant in all test specifications described in Section 4.5.
Alternative skill measure
Our main tests rely on the return gap and the degree of active fund management (FF3 R2 and FF5 R2) as measures of skill. In this section, we describe results that use an alternative skill measure: recent 12-month fund returns. Past returns (or alphas) are used as a measure of skill in several prior studies (e.g., Amihud and Goyenko 2013). However, given that the validity of past returns as a skill measure is highly contested—there is some debate as to whether high past returns represent skill or simply capture luck (Carhart 1997; Barras et al. 2010; Fama and French 2010)—we do not include tests based on this measure in Section 4.5 but instead describe them here for completeness.
The results based on prior returns as a measure of skill are qualitatively similar to those described in Section 4.5. After including other holdings characteristics as regression controls, we find that the AQ premiums in the high-skill partition (0.043%, 0.034%, and 0.040% when we use excess, FF3, and FF5 returns, respectively) are higher than premiums in the low-skill partition (0.013%, 0.007%, and 0.013%, respectively). The differences in premiums between high-skill and low-skill partitions are statistically significant.
Other robustness tests
We perform additional robustness checks to differentiate our results from the accrual anomaly present in mutual fund returns (Ali et al., 2008) and an alternative explanation for the association between AQ and returns: operating risk (Liu and Wysocki, 2007). Specifically, we include signed accrual and earnings volatility decile ranks as additional controls in the regressions reported in Table 4. The coefficients on AQ (untabulated) remain statistically significant after we control for signed accruals. After we control for operating risk, the AQ premiums remain statistically significant when the regressions are based on excess returns or FF5 returns; the premiums based on FF3 returns become smaller and lose statistical significance for the full sample of mutual funds. However, we continue to find statistically significant premiums on AQ using all return measures in the highest skill partitions. Our overall conclusions therefore remain unchanged. While these results suggest that some of the documented AQ effect may be related to operating risk, an alternative explanation is that, by controlling for earnings volatility, which is a proxy for environmental uncertainty and thus a determinant of AQ, we are controlling in part for information risk. As a result, we do not include these analyses in our main results and report them only as robustness checks.[3]
REFERENCES
Akins, B., Ng, J., & Verdi, R. (2012).Investor competition over information and the pricing of information asymmetry.Accounting Review, 8, 35–58.
Ali, A., Chen, X., Yao, T., & Yu, T. (2008). Do mutual funds profit from the accruals anomaly? Journal of Accounting Research, 46, 1–26.
Amihud, Y., & Goyenko, R. (2013). Mutual fund’sR2as predictor of performance. Review of Financial Studies, 26, 667–694.
Armstrong, C., Core, J., Taylor, D., & Verrecchia, R. (2011). When does information asymmetry affect the cost of capital? Journal of Accounting Research, 49, 1–40.
Barras, L., Scaillet, O., & Wermers, R. (2010). False discoveries in mutual fund performance: Measuring luck in estimated alphas. Journal of Finance,65, 179–216.
Carhart, M. (1997).On persistence in mutual fund performance.Journal of Finance, 52, 57–82.
Core, J., Guay, W., & Verdi, R. (2008). Is accruals quality a priced risk factor? Journal of Accounting and Economics, 46, 2–22.
Dechow, P., & Dichev, I. (2002). The quality of accruals and earnings: The role of accrual estimation errors. Accounting Review, 77, 35–59.
Fama, E., & French, K. (2010).Luck versus skill in the cross-section of mutual fund returns. Journal of Finance,65, 1915–1947.
Kim, D., & Qi, Y. (2010). Accruals quality, stock returns, and macroeconomic conditions. Accounting Review, 85, 937–978.
Liu, M., & Wysocki, P. (2016).Cross-sectional determinants of information quality proxies and cost of capital measures.Quarterly Journal of Finance, forthcoming.
Mashruwala, C., Mashruwala, S. (2011). The pricing of accruals quality: January versus the rest of the year. Accounting Review,86, 1349–1382.
Ogneva, M. (2012). Accrual quality, realized returns, and expected returns: Controlling for cash flow shocks. Accounting Review, 87, 1415–1444.
Ogneva, M., Piotroski, J., & Zakolyukina, A. (2017).Accounting fundamentals and systematic risk: Corporate failure over the business cycle Working paper, University of Southern California.
Table A1. Accrual quality and individual stock returns
The table documents an association between stock returns and AQ for a sample of CRSP-Compustat stocks held by mutual funds in a given month. Panel A provides descriptive statistics for AQ decile portfolios. Every month, stocks are sorted into 10 portfolios based on their AQ measure. The panel reports time-series averages of mean variable values calculated each month for each decile portfolio. Excess returns are stock returns less the risk-free rate. FF3 (FF5) returns equal an intercept plus a residual from the by-stock regressions of excess returns on Fama-French three factors (plus momentum and liquidity factors). All other variables are defined in Table 1. 10-1 indicate differences between the tenth and the first deciles of AQ. Slope indicates an average slope estimate from Fama-MacBeth regressions of returns on AQ rank. Panel B reports Fama-MacBeth regressions of excess, FF3, and FF5 returns on AQ ranks and control variables. rAQ, rBeta, rMV, rBM,rLagret, andrLiqrefer to decile ranks based on accrual quality, CAPM beta, market value, book-to-market ratio, prior 12-month returns, and liquidity, respectively.t-statistics with Newey-West correction for autocorrelation are reported in parentheses. Returns are in percentages. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
Panel A. Descriptive statistics and univariate results
AQDecile / Nobs / AQ / MV / CAPM
Beta / BM / Lagret / Liq / Excess Ret / FF3 Ret / FF5 Ret
(Best AQ) / 51531 / 0.011 / 5.91 / 0.752 / 0.789 / 0.133 / -0.119 / 0.825 / 0.16 / 0.161
2 / 51645 / 0.019 / 4.79 / 0.909 / 0.742 / 0.148 / -0.086 / 0.948 / 0.202 / 0.299
3 / 51661 / 0.025 / 4.39 / 0.969 / 0.713 / 0.155 / -0.046 / 1.001 / 0.239 / 0.352
4 / 51647 / 0.031 / 3.76 / 1.014 / 0.716 / 0.173 / -0.056 / 1.009 / 0.224 / 0.327
5 / 51624 / 0.037 / 2.98 / 1.080 / 0.733 / 0.168 / -0.115 / 0.96 / 0.165 / 0.305
6 / 51695 / 0.045 / 2.36 / 1.113 / 0.717 / 0.178 / -0.049 / 1.041 / 0.243 / 0.382
7 / 51672 / 0.055 / 1.50 / 1.154 / 0.737 / 0.176 / -0.098 / 1.045 / 0.237 / 0.378
8 / 51636 / 0.068 / 1.09 / 1.203 / 0.730 / 0.191 / -0.089 / 0.97 / 0.163 / 0.352
9 / 51670 / 0.089 / 0.97 / 1.250 / 0.708 / 0.197 / -0.143 / 1.087 / 0.331 / 0.511
(Worst AQ) / 51555 / 0.159 / 0.62 / 1.365 / 0.608 / 0.236 / -0.186 / 0.986 / 0.241 / 0.499
10-1 / 0.148*** / -5.29*** / 0.613*** / -0.181*** / 0.103 / -0.067 / 0.161 / 0.081 / 0.339
t-stat / (17.03) / (-5.71) / (5.55) / (-3.19) / (1.60) / (-0.43) / (0.34) / (0.28) / (1.07)
Slope / 0.015 / 0.008 / 0.029
t-stat / (0.56) / (0.39) / (1.11)
Table A1. (continued)
Panel B. Accrual quality premium
Excess Returns / FF3 Returns / FF5 ReturnsrAQ / -0.021 / -0.018 / -0.023* / -0.021 / -0.010 / -0.008
(-1.52) / (-1.23) / (-1.74) / (-1.51) / (-0.63) / (-0.49)
rBeta / 0.060* / 0.036 / 0.014 / -0.009 / 0.034 / 0.015
(1.70) / (1.19) / (0.48) / (-0.37) / (1.21) / (0.57)
rMV / -0.082** / -0.084*** / -0.076*** / -0.075*** / -0.082*** / -0.082***
(-2.54) / (-2.67) / (-3.25) / (-2.95) / (-3.29) / (-3.05)
rBM / 0.064*** / 0.062*** / 0.031 / 0.033** / 0.024 / 0.028*
(3.18) / (3.19) / (1.52) / (2.24) / (1.18) / (1.71)
rLagret / 0.035 / 0.029 / 0.038
(0.63) / (0.55) / (0.87)
rLiq / 0.021** / 0.020** / 0.017**
(2.53) / (2.59) / (2.51)
Intercept / 0.889*** / 0.723 / 0.532*** / 0.363 / 0.563*** / 0.335
(3.19) / (1.42) / (3.16) / (0.93) / (3.02) / (0.99)
Obs. / 516336 / 516280 / 516336 / 516280 / 516336 / 516280
Adj. R2 / 0.032 / 0.040 / 0.013 / 0.020 / 0.013 / 0.019
1
[1] Stocks held by mutual funds may be systematically different from the CRSP-Compustat population due to institutional restrictions imposed on mutual fund holdings. For example, mutual funds may avoid investing in small, illiquid stocks that tend to have worse AQ (Dechow and Dichev 2002).
[2]Specifically, they document a significantly negative interaction between AQ and institutional ownership in regressions of realized stock returns on stock characteristics.
[3] We perform additional tests (untabulated) to investigate whether the January effect in AQ premiums documented by Mashruwala and Mashruwala(2011) is present in our mutual fund sample. Consistent with their findings, the AQ premiums become weaker after we exclude January observations from our sample. However, we continue to find statistically significant AQ premiums when using excess returns or FF5 returns (all return measures) for the full sample of mutual funds (within the highest skill partitions).