A0960
Title: Forecasting of intraday trading volume using Bayesian nonlinear ACV models for a VWAP trading strategy
Authors: Roman Huptas - Krakow University of Economics (Uniwersytet Ekonomiczny w Krakowie) (Poland) [presenting]
Abstract: The aim is to evaluate the accuracy of intraday volume point forecasts obtained from alternative Bayesian nonlinear autoregressive conditional volume (ACV) models in terms of a daily volume weighted average price (VWAP) trading strategy. The VWAP trading strategy is the most popular algorithmic trading strategy due to its operational simplicity. Its goal is to split large orders into smaller-sized orders and execute them during the trading day to achieve an average price that is close to the VWAP. The VWAP strategy needs to be based on accurate intraday volume point forecasts, which are crucial to accomplish its goal. More accurate predictions result in a better-executed VWAP and lower execution risk. Different specifications of ACV models are analyzed and compared: a linear, logarithmic and Box-Cox ACV model. The exponential and the generalized gamma distributions of error terms are examined. Additionally, a log-normal distribution for innovations is also explored since it has received no attention in the literature on this type of model. The ACV models are compared to benchmarks such as the naive method and the rolling means technique. Two types of VWAP replication strategies, static and dynamic, are considered, and VWAP tracking mean squared error is applied to measure the VWAP order execution risk. The empirical part includes forecasting 10-minute volume data of selected widely traded stocks from selected well-developed stock exchanges.
