Statistically based quarterly earnings expectation models for nonseasonal firms

Abstract

We present empirical evidence on the relative predictive power of statistically based quarterly earnings expectation models for firms that are characterized as nonseasonal in nature. We are particularly interested in nonseasonal firms for two reasons. First, it appears that a sizable and growing percentage of firms exhibit quarterly earnings patterns that are clearly nonseasonal in nature. We present new evidence that is consistent with this trend. Specifically, 36% of our sample firms (n = 296) are nonseasonal compared to 12% reported in Lorek and Bathke (J Acc Res 22:369–379, 1984) (n = 29); 17% in Brown and Han (J Acc Res 38:149–164, 2000) (n = 155); and 28.2% in Bathke et al. (J Business Inquiry 5:39–49, 2006) (n = 167). Second, we also find that 43.6% of the nonseasonal firms in our sample have no analyst coverage. Therefore, interest in the predictive ability of statistically based models for such firms is greatly enhanced. Our predictive findings indicate that the random walk model provides significantly more accurate pooled, one-step ahead quarterly earnings predictions across 40 quarters in the 1994–2003 holdout period than the first-order autoregressive model popularized in the literature. We attribute the superior performance of the random walk model to at least three contributing factors: (1) its parsimonious nature; (2) the reduced levels of autocorrelation observed in our quarterly earnings data relative to previous work; and (3) a significantly greater frequency of loss quarters evidenced by nonseasonal versus seasonal firms.