Search

The last 60 years of research striving to explain the post-earnings announcement drift (PEAD) have resulted in numerous potential explanations. This ”zoo” of explanations, limited academic consensus, and a literature relying on thousands of earnings announcement make researchers able to detect subtle and complex effects with little practical significance. This paper exploits new capabilities of inference via machine learning to systematically examine leading variables explaining the PEAD. First, we identify a small set of variables associated with momentum, liquidity, and limited arbitrage that directly and consistently affect the PEAD. Secondly, we demonstrate the danger of hand-picking a small set of control variables, which can lead to unreliable results, overestimation of coefficients, and underestimation of their standard errors. Finally, we explore multiple variables related to general equity risk premia and find a more prominent role for price trends than otherwise suggested. We demonstrate the benefits of merging traditional hypothesis-driven research with new methods from machine learning that enable high-dimensional inferences. As the post-earnings announcement drift (PEAD) literature is governed by a "zoo" of explanations, limited academic consensus on model design, and the reliance on massive data, it will serve as a leading example to demonstrate the challenges of high-dimensional analysis. We identify a small set of variables associated with momentum, liquidity, and limited arbitrage that directly and consistently explain PEAD. The framework can be applied broadly in finance.

We show that machine learning (ML) algorithms improve one-day-ahead forecasts of realized variance from 29 Dow Jones Industrial Average index stocks over the sample period 2001 - 2017. We inspect several ML approaches: Regularization, tree-based algorithms, and neural networks. Off-the-shelf ML implementations beat the Heterogeneous AutoRegressive (HAR) model, even when the only predictors employed are the daily, weekly, and monthly lag of realized variance. Moreover, ML algorithms are capable of extracting substantial more information from additional predictors of volatility, including firm-specific characteristics and macroeconomic indicators, relative to an extended HAR model (HAR-X). ML automatically deciphers the often nonlinear relationship among the variables, allowing to identify key associations driving volatility. With accumulated local effect (ALE) plots we show there is a general agreement about the set of the most dominant predictors, but disagreement on their ranking. We investigate the robustness of ML when a large number of irrelevant variables, exhibiting serial correlation and conditional heteroscedasticity, are added to the information set. We document sustained forecasting improvements also in this setting.