Asymmetric Responses of Conditional Variances to Earnings Surprises: A Bayesian Approach

Abstract

This paper studies the relation between stock returns and innovations to the conditional variance on earnings announcement dates using a simple Bayesian information processing model. The distribution of the earnings announcement returns greatly influence the way investors update their beliefs about the firms' equity value. On the earnings announcement date, most of the large returns are negative and these large negative returns are more informative than the positive returns. This evidence is supported by the mixture of distribution hypothesis and the discretionary disclosure theory. If investors use this information while updating their prior beliefs about the firms' equity value, then they are more likely to ignore the large positive returns as outliers and not revise their prior beliefs. Thus perceived uncertainty will not be affected by the large positive news. On the other hand if they observe a large negative return they will be forced to damp their prior beliefs and reformulate their posterior thus increasing their uncertainty about the firms' equity value. This explains why returns and conditional variances are negatively correlated, and why conditional variances respond more strongly to negative returns than to positive returns on announcement dates. We formally present this argument using a simple Bayesian updating problem with non-normally distributed prior beliefs. The behavior of option-implied measures of conditional variances around earnings announcements is consistent with our proposed explanation but is at odds with the leverage hypothesis.