Specifically, our tests assess if accruals capture relevant information after controlling for the information in cash flows. If an investor could obtain information about accruals and cash flows early in the company’s fiscal year, the investor who picks stocks based on cash flows and accruals will generate more current-period trading gains than another investor who only trades based on cash flow information. In that sense, an investor could benefit from including accruals in his/her trading decision. Alternatively, the accrual anomaly suggests that the investor who places a higher value on companies with larger accrual components of earnings in the current period experiences some future-period trading losses. Such losses would represent a cost to investors who rely on accruals.

Our hope is that this further clarification eases some of the concern about the meaning of “costs” and “benefits.”

The point Bob is making about cash flow v. accrual mispricing seems valid and is made in a number of other papers. For example:

- Desai Rajgopal Venkatachalam (TAR, 2004) find that after controlling for cash flows total accruals don’t predict returns
- Beaver McNichols Price (JAE, 2007) find that after disaggregating earnings into accruals and cash flows, cash flows but not accruals predict returns
- *selfish plug* Armstrong Foster Taylor (working paper, 2009) find that after controlling for cash flows, discretionary accruals at the time of the IPO do not predict future returns.

Collectively these studies (and Bob’s) suggest that cash flows are an important correlated omitted variable, that when included subsume the power of accruals. This finding is important in it’s own right, and flies in the face of much of the earnings management literature which interprets the Sloan (TAR, 1996) and Teoh, Welch, and Wong (JF 1998) evidence that firms successfully use earnings management to increase their stock prices.

That said, isn’t clear to me how one can assess “costs” or “benefits” based on a regression of returns on earnings (and its components). Suppose individual investors are driving the anomaly, and earnings information facilitates a wealth transfer from individuals to smart institutions, but limited arbitrage prevent immediate correction by smart institutions. There is a welfare gain on one-side and a welfare loss on their other. But it seems like that’s all we can say, unless we as the researcher or regulator assign a premium to the individual investors welfare. Which is fine, but these things should be made more explicit. This is why I enjoyed Bob’s article, because it highlights the issues with making statements about the societal benefits/costs of accounting standards.

However, it still seems to me that standard-setters need to go a bit further, because overestimation of the persistence of accruals implies one regulatory response, while underestimation of the persistence of cash flows implies another.

I would also note that RSST provided some evidence that the extent of mispricing varies with the reliability of accruals, while LLY don’t provide any such evidence. It isn’t immediately clear to me why the difference in method would generate this difference in results. If anyone has thoughts on this, I would love to hear them.

Another way of interpreting the LLY results is that firms with a strong cash component of earnings are initially underpriced by the market because the market knows that the average persistence of earnings is low (which is due mainly to accruals) and, not discriminating between accruals and cash flows, the market mistakenly undervalues firms with strong cash flows.

More generally, if the accrual component of earnings is less persistent than the cash component of earnings and investors fixate on earnings and don’t realize this, we expect two forms of mispricing:

1. firms with high accruals are overpriced

2. firms with high cash flows are underpriced

RSST jointly test for these two effects by regressing future stock returns on earnings and accruals. The coefficient on accruals then measures the difference between the coefficient on accruals and the coefficient on cash flows (because earnings=cash flows+accruals). I believe that this is the correct test to determine whether the market’s failure to discriminate between the differing persistence of accruals and cash flows results in mispricing (with a negative coefficient on accruals indicating that prices do not reflect the higher persistence of cash flows relative to accruals).

My overall view here. If we aggregate two components of earnings with differing levels of persistence and this results in mispricing because investors don’t discriminate between these differing levels of persistence, we have inefficient resource allocation. So there is a cost here. I agree with LLY that there are also benefits to including accruals because they also contain value-relevant information. But allowing more unreliable accruals is not necessarily a good thing. It is possible that the costs may exceed the benefits. I don’t think that the analysis in either RSST or LLY allows for firm conclusions about whether or not the costs exceed the benefits.