Abstract:
This thesis is composed of two main research lines. The first line, developed in
chapters 2 to 4, deals with frequentist and Bayesian estimation of regime-switching
GARCH models and its application to risk management on energy markets, while the
second part, which corresponds to chapter 5, focuses on forecast rationality testing
within a Bayesian framework.
Chapter 2 presents a unified mathematical framework for characterizing the class
of MS-GARCH models based on collapsing the regimes in order to eliminate the usual
path dependence problem. Within this framework, two new models (identified as
Basic model and Simplified Klaassen model) are proposed as alternative specifications
of the MS-GARCH model. Using Maximum Likelihood Estimation, we estimate the
parameters of the different models within this family and compare their performance
on both simulation and empirical exercises. Chapter 3 proposes new efficient Monte
Carlo simulation techniques based on multiple proposal Metropolis. The application
to approximated inference for regime-switching GARCH models is there discussed.
In Chapter 4, we provide an extension of our efficient Monte Carlo simulation
algorithm to a multi-chain Markov switching multivariate GARCH model and apply
it to risk management in commodity market. More specifically we focus on futures
commodity market and suggest a dynamic and robust minimum variance hedging
strategy which accounts for model parameter uncertainty. In chapter 5, we propose
a new Bayesian inference procedure for testing the monotonicity properties of second
moment bounds across several horizons presented in Patton and Timmermann [2012].
