Analysis of Transaction Data in Stock Markets: the PCD Model

Abstract

Advances in data collection make available transaction-by-transaction data in financial markets. These data contain information about market microstructure such as price recovery and information processing. In this talk, we propose a statistical model for studying jointly price change and duration between transactions of stocks traded in the New York Stock Exchange (NYSE). The joint model consists of several conditional models that are designed to capture the characteristics of high-frequency financial data. Consider, for example, the price change. It is well-known that price is not a continuous variable in high-frequency. It only assumes a multiple of tick size in the NYSE. In the proposed PCD model, a price change is partitioned into direction and size of change and the two components are modeled separately. The duration between transactions is measured in seconds and may indicate market volatility. For illustration, we apply the proposed model to the IBM transaction data. Differences between the proposed model and other models available in the literature are given. Finally, we discuss possible applications of the PCD model to other scientific areas. (This is joint work with Robert E. McCulloch).