A0771
Title: Trading volume alpha
Authors: Chao Zhang - HKUST(GZ) (China) [presenting]
Abstract: Rather than focusing on predicting asset return moments, the economic benefits are modeled for predicting individual stock trading volume. Volume forecasts are translated into a component of expected trading costs, and their value is analyzed through a portfolio framework. By recasting the volume prediction problem into a portfolio optimization problem that trades off tracking error versus net-of-cost performance, volume predictions are quantified into economic outcomes. Incorporating the economic loss function directly into a machine learning algorithm yields better out-of-sample performance than commonly used statistical loss functions. While volume is only one component of what drives trading costs, it is highly predictable, readily available, and its economic benefits are as large as those from stock return predictability.
