Robustness Checks

A. The Accrual Effect within Industries

Working capital requirements vary across lines of business, so in a line of activity in which inventories and receivables represent a small fraction of total assets, accruals (excluding depreciation) are likely to be relatively low. The information contained in accruals about future returns is likely to be meager in such cases. To bring out more clearly the predictive power of accruals, we examine return spreads associated with accruals across firms within the same industry, so they are relatively homogeneous. In particular, we apply the same sort procedure as in table 3 to form four portfolios within each of the industry groupings defined by Fama and French (1997). The spread in abnormal returns between the bottom and top quartile portfolios for each of the three years

Earnings Quality and Stock Returns 1073

TABLE 11 Differences in Excess Returns across Portfolios Sorted by Accruals within Industries

Code Industry Name AR1 AR2 AR3

1 Food .016 ⫺.031 .038

7 Medical equipment .017 ⫺.001 ⫺.069

8 Drugs ⫺.037 ⫺.005 ⫺.023

9 Chemicals .024 .010 ⫺.021

10 Rubber .086 .074 .081

11 Textiles .059 .090 .060

12 Building materials .058 .070 .036

13 Construction .162 .084 ⫺.007

14 Steel .036 ⫺.053 ⫺.016

15 Machinery .061 .049 .059

16 Electrical equipment .086 .063 ⫺.016

17 Autos .081 ⫺.016 .041

18 Mines ⫺.045 .004 .002

19 Energy ⫺.023 .005 .023

20 Utilities .035 .047 .011

21 Telecommunications .066 .014 .002

22 Business services .041 .035 .000

23 Computers .094 ⫺.046 ⫺.009

24 Chips .050 .084 .109

25 Lab equipment .053 .039 .008

26 Paper .062 .061 ⫺.013

Note.—The sample comprises all domestic common stocks (except financial firms) on NYSE, AMEX, and NASDAQ with coverage on CRSP and Compustat, and with available data. At the end of April each year from 1971 to 1995, all stocks in a given industry are ranked by accruals relative to average total assets and assigned to one of four portfolios (assuming a reporting delay of four months from the end of the fiscal year).

Annual buy-and-hold returns are calculated over the subsequent year and measured in excess of the return on a control portfolio matched by size and book-to-market. The difference between the average excess returns across the bottom and top quartiles within each industry in each of the first to third years following portfolio formation (AR1, AR2, and AR3) are reported. Also reported is the average return difference over industries and the t-statistic for the mean difference. Definitions of industries are taken from Fama and French (1997).

after portfolio formation is reported in table 11, along with the simple mean and t-statistic over all industries.¹⁵

The abnormal return spread in the first postformation year averages 5%

across all industries (with a t-statistic of 6.86) and is positive in 29 out of 32

15. We require the average number of firms in an industry over the sample period to be at least 20. Also, the number of years in which an industry contains fewer than 20 firms cannot be over five. Firm-year observations that do not meet these criteria are assigned to the “other”

category in the table.

1074 Journal of Business

cases, so the accrual effect is robust. Notably, the spread tends to be larger in industries in which working capital is high such as toys (recreational prod-ucts) and construction and is smaller in lines of business with low working capital levels such as meals (restaurants, hotel and motel), utilities, and transportation.

The results in table 11 also provide another means of discriminating between the competing explanations for the accrual effect. In the case of cash-based businesses in which earnings roughly match cash flows, there is less scope for managers to distort reported income. Under the manipulation hypothesis, therefore, the predictive power of accruals should vary across industries de-pending on the levels of working capital. On the other hand, if unanticipated changes in business conditions are the force driving both accruals and sub-sequent returns under the underreaction hypothesis, the effect of accruals should be relatively uniform across industries. The dominant components of working capital are inventory and accounts receivable (see table 1). In table 11, the rank correlation between inventory plus accounts receivable (relative to average total assets) and the abnormal return spread AR1 is 0.27. The rough correspondence between the size of the return spread and the level of working capital thus provides support for the manipulation hypothesis.

B. Evidence from U.K. Stocks

All the existing studies that document the association between accruals and future stock returns are based on evidence from the United States. This raises the question whether the accrual effect is specific to U.S. stocks. If the evidence does not generalize to other markets with similar accounting conventions, the suspicion arises that the association is spurious and may have no economic rationale. Accordingly, as another robustness check we explore whether ac-cruals predict returns in a foreign market. In particular, accounting standards in the United Kingdom are similar to those in the United States, and (as of the time of writing) the U.K. equity market is the second-largest in the world.

From the standpoint of the potential motivation to manipulate earnings, it is also noteworthy that, generally speaking, in the United Kingdom, management compensation schemes (as well as the behavior of research analysts and in-vestors) more closely resemble those in the United States than those of other countries.

In table 12 we replicate our analysis using data on U.K. stocks. The sample comprises all domestic, primary, nonfinancial U.K. stocks that are available in the Xpressfeed database, with data on accruals and returns. Portfolios are formed at the end of April each year from 1991 to 2000, assuming a four-month reporting delay.

The results for the U.K. data confirm an association between accruals and future returns. Further, the sort by accruals (panel A) generates larger spreads in returns than the U.S. results. The difference in raw returns between the

Earnings Quality and Stock Returns 1075

TABLE 12 Returns for Portfolios Sorted by Accruals and Selected Accrual Components, U.K. Data

AR1 .060 .045 .049 .012 ⫺.024 .012 .004 ⫺.046 ⫺.060 ⫺.048 .108 AR2 .030 .015 ⫺.024 .022 .017 ⫺.020 ⫺.030 .005 ⫺.047 ⫺.038 .067 AR3 .008 ⫺.037 .002 ⫺.047 ⫺.014 ⫺.026 ⫺.003 ⫺.035 ⫺.052 ⫺.031 .039

B. Ranked by Change in Accounts Receivable

R1 .122 20.095 .120 .098 .108 .104 .077 .107 .159 .137 ⫺.015 R2 .131 .120 .104 .100 .093 .084 .100 .061 .109 .135 ⫺.005 R3 .090 .075 .112 .074 .060 .068 .089 .101 .058 .119 ⫺.029 AR1 ⫺.004 ⫺.019 .013 .006 .005 ⫺.003 ⫺.020 ⫺.006 .040 ⫺.012 .008 AR2 ⫺.003 .007 .012 .004 .011 ⫺.028 ⫺.005 ⫺.059 ⫺.010 ⫺.002 ⫺.001 AR3 ⫺.024 ⫺.046 .008 ⫺.019 ⫺.031 ⫺.015 ⫺.022 ⫺.009 ⫺.059 ⫺.024 .000

C. Ranked by Change in Inventory

R1 .130 .132 .160 .140 .172 .090 .104 .094 .056 .049 .081

R2 .130 .120 .132 .092 .174 .146 .074 .042 .067 .064 .066

R3 .063 .090 .124 .093 .116 .089 .075 .078 .052 .070 ⫺.007 AR1 .009 .016 .055 .026 .061 ⫺.021 .001 ⫺.016 ⫺.056 ⫺.072 .080 AR2 ⫺.010 .023 .017 ⫺.004 .059 .044 ⫺.033 ⫺.055 ⫺.057 ⫺.051 .042 AR3 ⫺.049 ⫺.025 .015 .003 .005 ⫺.015 ⫺.028 ⫺.025 ⫺.052 ⫺.065 .016

D. Ranked by Change in Accounts Payable

R1 .117 .088 .111 .097 .132 .090 .109 .126 .133 .122 ⫺.005

R2 .127 .105 .098 .111 .099 .080 .089 .095 .122 .112 .016

R3 .087 .089 .085 .069 .084 .085 .059 .063 .109 .122 ⫺.034 AR1 ⫺.010 ⫺.021 .003 ⫺.004 .027 ⫺.020 .016 .007 .020 ⫺.017 .007 AR2 ⫺.003 ⫺.011 ⫺.001 ⫺.002 ⫺.006 ⫺.015 ⫺.005 ⫺.021 .007 ⫺.015 .012 AR3 ⫺.023 ⫺.030 ⫺.016 ⫺.027 ⫺.013 ⫺.015 ⫺.045 ⫺.052 ⫺.008 ⫺.012 ⫺.011 Note.—The sample comprises all domestic common U.K. listed stocks (except financial firms) with available data on the Xpressfeed database. At the end of April each year from 1991 to 2000, all stocks are ranked by total accruals (panel A) or by a component of accruals (panels B–D) relative to average total assets and assigned to one of 10 portfolios (assuming a reporting delay of four months from the end of the fiscal year). Annual buy-and-hold returns are calculated over the subsequent year, as well as returns in excess of the return on a control portfolio matched by size and book-to-market. Average returns in each of the first to third years following portfolio formation (R1, R2, and R3 respectively) and excess returns in each of the first to third postformation years (AR1, AR2, and AR3) on the equally weighted decile portfolios are reported.

lowest- and highest-ranked decile groups is 13.5% in the year following port-folio formation, and the corresponding difference in excess returns is 10.8%.

When we look at the components of accruals, the evidence generally echoes that in the United States. The major contribution to the predictive power of accruals comes from the inventory component. The sort by inventory changes (panel C) produces spreads in raw and excess returns (about 8%) that are close to the results for total accruals. In short, our evidence from the second-largest equity market, which is relatively free from data-snooping biases, confirms the predictive power of accruals for returns and also highlights the importance of the inventory component.

1076 Journal of Business