Expected and realized jumps on earnings announcement days: forecasting and the variance risk premium

<p>In this paper we analyze the realized and expected jumps in daily variance of stock returns using a sample of 36 S&amp;P 500 constituents. Firstly, in the section on realized jumps we find strong evidence of jumps in variance on earnings announcement days by fitting a Heterogeneous Autoregressive model (HAR) using the Parkinson (1980) and Alizadeh et al. (2002) proxies and including an additional dummy variable to reflect the earnings announcement days.</p><p>Secondly, in panel data analysis of expected jumps in variance, we suggest a model for daily variance as a sum of a baseline forecastable component and the expected jump in variance that we estimate using options-implied data. We find considerable support for our model when we do not deal with the 3 largest expected jumps. Controlling for 2 additional variables, we outline that there is only partial evidence of the relevance of a firm&rsquo;s size, as measured through the market cap., in determining the magnitude of jumps, and no evidence at all for the degree of variation in brokers&rsquo; forecasts of EPS prior to the reports.</p><p>We also check for the existence of a variance risk premium. Our investigation results show that in the absolute majority of cases when the expected earnings announcement-specific jump in variance is not extremely large, there is no significant evidence of the existence of a variance risk premium. However, once we include the 3 above mentioned largest expected jumps in our estimation, the evidence for the existence of a variance risk premium becomes significant. Hence we conclude that for the largest expected values of the jump in daily variance, investors&rsquo; estimates are biased upwards as the true jumps happen to be significantly smaller, which is exactly the case when a variance risk premium exists (see Carr and Wu (2005)).</p>