Assessing performances of density forecasts in duration models

Traditionally the forecast evaluation literature has focused on methods for assessing point forecasts and interval forecasts. Recently , however, density forecasting became very active and important area of research in the analysis of economic and financial time series with the introduction of simple and operational framework for density forecast evaluation by Diebold et al (1998). Density forecasts provide researchers with a full impression of the uncertainty associated with a forecast. In particular, measures of uncertainty surrounding a ”central tendency” (the point forecast) can enhance its utility. Given a few different choices of density forecasts, it is crucial to identify the correct density forecast for future variable of interest. In this paper we propose non-parametric density forecasts for the variable of interest, in particular, for duration models and asses their performances against their parametric competitors. Forecasts are obtained under the moment conditions imposed by the model and evaluated using the tools proposed by Diebold et al (1998). Our simulation results shows that the proposed non-parametric estimators perform better than parametric competitors when the densities are misspecified and equally well when the parametric densities are correctly specified. Also we suggest that these estimators can be used to identify the conditional mean duration process in duration models.