Extensions of the EKOP Model

Abstract

Easley, Kiefer, O'Hara and Paperman (1996) introduced a model that enables one to estimate the probability of informed trading in a stock using as input data the numbers of buyer and seller initiated trades over a period. Empirical testing reported in Venter and de Jongh (2004) indicated that this model does not fit data well and they formulated a model based on the Poisson Inverse Gaussian (PIG) distribution that fitted better. This model allowed for random daily news noise impacts on the trading intensities of the liquidity and informed traders. These news noise impacts were assumed to be independent over successive days. Further investigation of the model indicated that the independence assumption may not be realistic. In this paper we reformulate the model in terms of Poisson Log Normally (PLN) distributed news noise impacts with serial correlation. We show how the much more complicated likelihood function and other items of interest for inference purposes may be computed by means of efficient importance sampling.