ABSTRACT
We propose a fully nonparametric approach to the analysis of the Autocorrelated Conditional Duration (ACD) process applied to durations between financial events. We use a recursive algorithm to estimate the nonparametric specification. In a Monte Carlo experiment, we analyse its forecasting performance and compare it with a correctly and a mis-specified parametric estimator. On a real dataset, the nonparametric estimator seems to mildly overperform in terms of predictive power. The nonparametric analysis can also provide guidance on the choice between alternative parametric specifications. In particular, once intraday seasonality is directly modelled in the conditional duration function, the nonparametric approach provides insights into the time-varying nature of the dynamics in the model that the standard procedures of deseasonalization may lead one to overlook.
JEL Classification: C14, C58, G17
