CMStatistics 2017: Start Registration
View Submission - CFE
A0657
Title: Modelling higher moments and density forecasting: A comprehensive look Authors:  Xiaochun Liu - University of Alabama (United States) [presenting]
Richard Luger - Laval University (Canada)
Abstract: Many GARCH-type models have been proposed in the literature for the higher moments of financial returns and their conditional distributions. We examine comprehensively whether these models yield better out-of-sample density forecasts. Among a wide range of specifications for autoregressive conditional volatility, skewness, and kurtosis, we find that the most promising approach rests on a decomposition of returns into their signs and absolute values. This approach specifies the joint distribution of the return components by combining a dynamic binary choice model for the signs, a multiplicative error model for the absolute values, and a dynamic copula function for their interaction. This flexible specification captures well the time-varying conditional skewness process and provides more accurate density forecasts than competing models, especially for the left tail of financial returns.