To beat or not to beat? The importance of analysts’ cash flow forecasts

See Ramnath et al. (2008) for a recent literature review and Brown (2007) for abstracts of articles in this literature.

One reason for the importance of cash flows is that cash flow forecasts (or their mathematical equivalents) underlie firm valuation models (Ohlson 1995).

For simplicity, we use the term “beat” in lieu of the more precise but cumbersome “meet or beat.”.

See FASB Statement of Financial Accounting Concepts No. 1 (1978).

Ghosh et al. (2005) find that firms with revenue-supported earnings increases have higher earnings quality.

We use split-unadjusted I/B/E/S data.

We require all firm-quarters to have at least one cash flow forecast and one earnings forecast in I/B/E/S. The former requirement makes our sample relatively small vis-à-vis samples that simply require at least one earnings forecast.

Unlike good news earnings and good news cash flows which are represented by nonnegative ES and CS respectively, good news accruals is represented by negative AS. In contrast to good news earnings and cash flows, the accruals surprise is good news if actual accruals are below predicted accruals. Conceptually, for a given earnings surprise, the larger the accruals component is, the less the earnings surprise is supported by cash flow performance, which implies lower earnings quality. We therefore classify greater than expected accrual surprises (AS >0) as bad news accruals and all other cases as good news accruals. Our resulting GNA variable theoretically captures not only absence of earnings management but also cases of downward earnings management. On the one hand, one can view any type of earnings management, including downward earnings management, as a negative signal, i.e., as bad news. On the other hand, downward earnings management may not be bad news in our meet-or-beat setting if the firm would have met the earnings target even in the absence of downward earnings management. In addition, while downward earnings management may be undesirable from an accounting standpoint, the market may favorably view a firm’s ability to manage earnings downward to build “cookie jar reserves.” Indeed, extant evidence indicates the market rewards earnings smoothing (Kirschenheiter and Melumad 2002) and upward earnings management is regarded as more egregious than downward earnings management (Eccles et al. 2001; Levitt 1998). Hence, the concern that our definition of good news accruals reflects downward earnings management cases is mitigated. Nevertheless, to the extent our GNA variable is noisy by encompassing downward earnings management cases, and the market is concerned about downward earnings management, our ability to obtain results for hypotheses tests is weakened. That is, ceteris paribus, one would not expect a greater market response to good news earnings when AS <0 (GNE_GNA) than when AS >0 (GNE_BNA) if downward earnings management renders the AS <0 accruals surprise to be perceived as negatively as the AS >0 accruals surprise.

Unlike Melendrez et al. (2008) who measure cumulative size-adjusted returns from 1 day after the last I/B/E/S forecast summarization date before fiscal year-end to 90 days after fiscal year-end, we use a short 3-day window. In addition, unlike Melendrez et al. (2008) who do not require the earnings announcement date to be close to the cash flow announcement date, we require cash flows to be announced within 1 day of earnings to ensure we properly capture the market response to both the cash flow and earnings announcement.

Melendrez et al. (2008) only examine future earnings performance. We examine both future return on assets and future cash flow performance.

Our findings based on quarterly data differ from those of Melendrez et al.’s results (2008) which are based on a pre-SOX sample of firm-years. Melendrez et al. (2008) document that prior to the passage of SOX, firms on average reported bad news cash flows and bad news accruals on an annual basis.

The mean difference is significant at the 0.10 level and the median difference is significant at the 0.001 level.

Bartov et al. (2002) find ES to be statistically significant. The fact that ES is not statistically significant in our Model 2 may be due to differences between our sample and the one used in Bartov et al. (2002). For example, we investigate a more recent time period than Bartov et al. (2002) do. In addition, unlike Bartov et al. (2002), our sample is limited to firm-quarters with analysts’ cash flow forecasts and actual cash flow data on I/B/E/S, and we impose the requirement that actual earnings are released within 1 day of actual cash flows.