Estimates of Parameters of a Kyle-Type Model around Earnings Announcements

Abstract

Using NYSE TAQ data, we compute MLEs of the primitive parameters of a Kyle-type model, including the variance of fundamentals given only public information, the variance of errors in private signals, and the variance of uninformed liquidity trading (noise). An out of sample test shows that the Kyle-type model we use is a good candidate for describing the setting generating our data. Given rational behavior, our overall pattern of estimates suggests that earnings announcement windows are not periods with public information but periods with different public information. Liquidity noise is higher within an earnings announcement window. The variance of beliefs given only public information is also higher within an earnings announcement window, in line with the Fischer-Stocken (2004) argument that agents with more ability and opportunity to manipulate a disclosure cause the disclosure to be more variable. The variance of private information error is smaller in an event window, consistent with greater information acquisition to try and interpret a public announcement. We also document that Kyle's is higher in an event window, showing an overall increase in information asymmetry. We also find that abnormal trading volume in an event window is driven primarily by a greater diffuseness of beliefs given public information alone, and even more so, liquidity noise, and is not significantly related to acquisition of private information.