A0237
Title: Modelling the conditional distribution of durations via mixture distributions
Authors: Jennifer Chan - The University of Sydney (Australia) [presenting]
Rasika Yatigammana - University of Sydney (Australia)
Abstract: Conventionally, positive valued financial time series such as trade duration are generally unimodal, positive skewed with excessive near-zero values occurring during period of active trading from optimistic traders. Drawing from financial theory on market microstructure that divides the traders into distinct categories, primarily based on access to information as informed and uninformed traders, presumably the distribution of durations is derived from a mixture of distributions. We adopt this mixture approach using mixtures of exponential and generalized Beta type II distribution with Weibull as a special case to capture the distinct empirical traits including the high proportion of near-zero durations. In addition, the flexible and often heavier tails capture extreme outliers. The model performance is evaluated based on predictive log likelihood of the forecasts. The tail-risk measures such at time-at-risk (TaR) and conditional TaR are assessed via several criteria such as violation rates, quantile loss function and mean and maximum absolute deviations.
