Bayesian estimation of generalized hyperbolic skewed student GARCH models

Abstract

Abstract. Efficient posterior simulators for two GARCH models with generalized hyperbolic disturbances are presented. The first model, GHt-GARCH, is a threshold GARCH with a skewed and heavy-tailed error distribution; in this model, the latent variables that account for skewness and heavy tails are identically and independently distributed. The second model, ODLV-GARCH, is formulated in terms of observation-driven latent variables; it automatically incorporates a risk premium effect. Both models nest the ordinary threshold t-GARCH as a limiting case. The GHt-GARCH and ODLV-GARCH models are compared with each other and with the threshold t-GARCH using five publicly available asset return data sets, by means of Bayes factors, information criteria, and classical forecast evaluation tools. The GHt-GARCH and ODLV-GARCH models both strongly dominate the threshold t-GARCH, and the Bayes factors generally favor GHt-GARCH over ODLV-GARCH. A Markov switching extension of GHt-GARCH is also presented. This extension is found to be an empirical improvement over the single-regime model for one of the five data sets. Séminaire d'économétrie,

The impact of parameter and model uncertainty on market risk predictions from GARCH‐type models

“…To analyze PITs below 5%, we begin by rescaling and normalizing them (see Christoffersen & Pelletier, 2004). Then, we carry out the ARCH, JB and LR tests suggested in Deschamps (2012). The ARCH test is an F-test of the nullity of autoregression coefficients (intercept excluded) in a six-order autoregressive process of the squared PITs.…”

“…Weights and parameters of the different pools are computed from past data. Methods for weight computation include Bayesian model averaging with predictive likelihoods (Eklund & Karlsson, 2007), as well as optimal pooling or OP (Geweke & Amisano, 2011, 2012. Broadly speaking, the former method averages measures of past predictive performance to form the weights, while the latter looks for the weights that maximize past predictive performance.…”

Econometric Modeling of Exchange Rate Volatility and Jumps

“…There are other definitions of skewed-Student distribution (see for example Hansen, 1994;Mittnik and Paolella, 2000;Aas and Haff, 2006;Dark, 2010;Deschamps, 2011). For instance, Aas and Haff (2006) extend the skewed-Student distribution to the Generalized Hyperbolic skewed-Student distribution (GHSST), while Deschamps (2011) proposes a Bayesian estimation of GARCH models with GHSST errors. Forsberg and Bollerslev (2002) use a GARCH model with Normal Inverse Gaussion (NIG) error distributions on exchange rate data.…”

Econometric Modeling of Exchange Rate Volatility and Jumps

“…There are other definitions of skewed-Student distribution (see for example Hansen, 1994;Mittnik and Paolella, 2000;Aas and Haff, 2006;Dark, 2010;Deschamps, 2011 …”