Modelling intraday trading activity using Box-Cox-ACD models

In this paper, I model the intraday trading activity based on volume durations, i.e. the waiting time until a predetermined volume is absorbed by the market. Since this concept measures the trading volume per time it is strongly related to market liquidity. I focus on volumes measured independently of the side of the market as well as on buy volumes, sell volumes and volumes measured on both market sides simultaneously. For econometric modelling of the different duration concepts, the performance of alternative types of Box-Cox-ACD models are analyzed. By evaluating out-of-sample forecasts, evidence is provided that Box-Cox-ACD models are a valuable tool for predicting volume durations. It is shown that volume durations measured independently of the side of the market have the best predictability. Furthermore, I illustrate that the inclusion of explanatory variables capturing past market activities concerning the price process and imbalances between the buy and sell side of the market. The empirical study uses IBM transaction data from the NYSE.