The Third Generation ACD Model: A Semiparametric Approach

Abstract

Having concluded that thus far the question about the most appropriate type of nonlinear ACD model has not been satisfactorily answered, we intend to develop a novel ACD modelling methodology based on an iterative estimation algorithm and a semiparametric autoregressive process that not only allows the data to speak for itself, but also is robust across datasets without relying on some computational factors, such as the hypothesis about the probability density function of the standardised durations. We propose in this paper the Semiparametric ACD (SP-ACD) model, which can be considered a starting point of such a development. To address the problem about the unobservability of the conditional durations in practice, the current paper devises an iterative algorithm to estimate the unknown conditional duration process. In such a circumstance, it is essential to provide not only the mathematical justification of the estimation scheme, but also sound asymptotic results about the semiparametric and the adaptive data-driven estimators. This paper focuses mainly on the former and also on a number of simulation experiments, while the later is left for future study.

Keywords: Semiparametric ACD model, Conditional durations, Simulations

JEL Classification: G12, C14