Modelling a Multivariate Transaction Process

In this paper a model for the specification of the multivariate density of the transaction process is presented. The transaction process is characterized by four marks: price changes, transaction volumes, bid-ask spreads and trade durations. The resulting four dimensional density is decomposed into its conditional and unconditional densities. The density of price changes is modelled with an integer count hurdle model and the densities of transaction volume, bid-ask spread and trade duration are specified with fractionally integrated autoregressive conditional models, which allow for long memory effects. The proposed models help to gain profound insights into market microstructure behavior. In the empirical section these models are compared against several alternatives. The database used consists of one month of transaction data of twelve stocks traded at the New York Stock Exchange in May, 2001. Evidence of long memory effects is presented, and based on standard statistical inference tools, it is shown that the fractionally integrated models perform much better than their standard alternatives. JEL classification: C22, C25, C41, G10