A comparison of financial duration models via density forecasts

Using density forecasts, we compare the predictive performance of duration models that have been developed for modelling intra-day data on stock markets. The compared models are the autoregressive conditional duration (ACD) models, their logarithmic versions, in each case with three distributions (Burr, Weibull, and exponential), and the stochastic volatility duration (SVD) model. A pilot Monte Carlo study is conducted to illustrate the relevance of the approach. The evaluation is done on transaction, price, and volume durations of 4 stocks listed at the NYSE. The results lead us to conclude that ACD and Log-ACD models often capture the dependence in the data in a satisfactory way, that they fit correctly the distribution of volume durations, that they fail to do so for trade durations, while the evidence is mixed for price durations.