Detecting News in Aggregate Accounting Earnings: Implications for Stock Market Valuation

How much new information is there in accounting earnings? A long line of research dating back to Ball and Brown (1968) investigates the informativeness of earnings at the firm level using simple regressions of stock returns on earnings changes. The estimated slope coefficient is commonly referred to as the earnings response coefficient (ERC) and indicates how much response there is in stock prices for a change in earnings. Studies at the firm level typically employ the earnings change as a measure of cash flow news and, consistent with the fundamental equality that prices are discounted expected future cash flows, find evidence of a significantly positive ERC. In contrast to firm-level findings, recent research extending the investigation at the aggregate level provides puzzling evidence of a weak, or even negative, association between aggregate earnings changes and stock market returns. This result is puzzling because one would expect that positive aggregate earnings changes translate into favorable cash flow news and, in turn, higher stock market prices.

The decomposition of Campbell (1991) provides the basis for elucidating the relation between earnings changes and stock returns. Campbell (1991) decomposes stock returns at time t into expected returns at time t-1 and revisions in expectations at time t about future cash flows (cash flow news) and future stock returns (discount rate news). With this decomposition in mind, it becomes clear that the relation between earnings changes and stock returns depends not only on the covariation of earnings changes with cash flow news but also on the covariation of earnings changes with the remaining components of realized stock returns.

In terms of this standard decomposition, literature in accounting and finance proposes two divergent views of the weak earnings-returns relation at the aggregate level. The first view, initially proposed by Kothari et al. (2006), suggests that aggregate earnings changes are informational. Specifically, aggregate earnings changes are mostly unanticipated and correlated with contemporaneous value-relevant news that causes investors to revise their expectations about not only future cash flows but also discount rates. A link between aggregate earnings changes and discount rate news implies that stock market returns can be weakly, or even negatively, related to aggregate earnings changes unless expected future cash flows increase by enough to offset an increase in expected future stock market returns.

In contrast, Sadka and Sadka (2009) propose that aggregate earnings changes are mostly anticipated and therefore provide little or no value-relevant news. This view suggests that aggregate earnings changes are non-informational and merely confirm investors' expectations. To explain a negative earnings-returns relation at the stock market level, they further argue that aggregate earnings changes are negatively related to lagged expected stock market returns. The argument goes as follows. Investors can anticipate aggregate earnings changes more than one year in advance, and as they predict higher future earnings, they demand a lower discount rate.

Which of the two views prevails is an open empirical question with important implications for the informational role of accounting earnings at the aggregate stock market level. My study directly addresses this question. I compile a comprehensive sample of U.S. publicly traded firms over the period from 1981 to 2009 and investigate the association of aggregate earnings changes with proxies for value-relevant revisions in expectations. I proxy for the unobservable ex ante expected future stock market return using the implied cost of capital method of Easton et al. (2002). The changes in the implied cost of capital estimates proxy for revisions in expectations about future stock market returns. I employ aggregate revisions in security analysts' projections of future rates of return on equity to proxy for revisions in expectations about future cash flows.

My study provides direct evidence that aggregate accounting earnings are correlated with new information that is relevant for valuation at the stock market level. First, I find a significantly positive association between aggregate earnings changes and revisions in expectations about future stock market returns. The link between aggregate accounting earnings and discount rate news extends across all components of the expected future stock market return. Specifically, revisions in expectations about the real riskless rate, future inflation, and the implied equity risk premium (i.e., the excess of the implied cost of capital over the nominal riskless rate) jointly explain as much as 28 percent of the time-series variation in aggregate earnings changes. I also find a significantly positive association between aggregate earnings changes and aggregate revisions in security analysts' projections of subsequent profitability rates. Aggregate analysts' revisions explain 48 percent of the time-series variation in aggregate earnings changes. In contrast, most of the time-series variation in aggregate earnings changes is not explained by lagged realized and expected stock market returns.

The link between aggregate earnings changes and new information about all components of the expected future stock market return implies that, univariately, the earnings-returns association can be weak, or even negative, at the aggregate level. This is because stock prices react to all value-relevant news (i.e., both cash flow news and discount rate news) embedded in accounting earnings. Indeed, a two-stage analysis reveals that over the sample period studied, cash flow news and discount rate news in aggregate earnings changes covary positively and have almost offsetting impacts on stock market prices. As a result, at the aggregate level, the simple ERC estimate is indistinguishable from zero. After I expand the right-hand side of the traditional ERC regression model to include contemporaneous revisions in expectations about all components of the expected future stock market return (i.e., the real riskless rate, expected inflation, and the expected equity risk premium), the multiple ERC estimate soars from virtually zero to +3.08 with a t-statistic of 3.59 (significant at the 1 percent level). The multiple ERC regression model also improves explanatory power with the adjusted R^{2} rising from below zero percent to 43 percent.

To address potential concerns about the construct validity of Easton's et al. (2002) implied cost of capital as a proxy for the ex ante expected future stock market return, I implement an instrumental-variables two-stage procedure. In the first stage, I estimate a regression of implied cost of capital on the real riskless rate, expected inflation, and a composite index computed as the first principal component of a comprehensive array of instrumental variables for the unobservable ex ante expected equity risk premium. Using the fitted value from this first-stage regression, I construct “cleaned-up” proxies for the expected future stock market return and discount rate news. In the second stage, I find that the multiple ERC estimate strengthens at +3.28 with a t-statistic of 3.89 (significant at the 1 percent level). In addition, the coefficient estimates from the multiple ERC regression model converge to theoretically predicted values when I directly take into account the time-series properties of aggregate profitability rates and discount rate components.

I conclude that, although the simple earnings-returns relation is weak at the stock market level, it is not the case that aggregate earnings changes are non-informational. Instead, if aggregate earnings changes are introduced “properly” along with new information about discount rates, then the earnings-returns relation improves considerably. Stated otherwise, the traditional ERC regression model fails to detect the stock market's reaction to value-relevant news embedded in aggregate accounting earnings because it suffers from a correlated omitted variables problem. Discount rate news is (1) positively associated with contemporaneous aggregate earnings changes and (2) negatively associated with contemporaneous stock market returns. My results highlight that researchers interested in detecting whether and how new information in aggregate accounting earnings maps into stock market prices need to carefully consider the overlap between aggregate earnings changes and revisions in expectations about all components of the future stock market return. My study contributes to research in accounting and finance by providing direct evidence on the association between accounting earnings and new information that is relevant for valuation at the aggregate stock market level. An interesting direction for future research would be to decompose aggregate earnings changes to identify which components are more correlated with value-relevant new information (e.g., revenues or expenses, accruals or cash flows). The findings are for all firms, on average. Further research may find cross-sectional variation in the link between earnings changes and news that is relevant for valuation at the stock market level. Such research would further broaden our knowledge of the link between accounting earnings and the aggregate economy.