Comparing Regime-Switching Models In Time Series: Logistic Mixtures vs. Markov Switching

Abstract

The purpose of this thesis is to review several related regime-switching time series models. Specifically, we use simulated data to compare models where the unobserved state vector follows a Markov process against an independent logistic mixture process. We apply these techniques to crude oil and heating oil futures prices using several explanatory variables to estimate the unobserved regimes. We find that crude oil is characterized by regime switching, where prices alternate between a high volatility state with low returns and significant mean reversion and a low volatility state with positive returns and some trending. The spread between one-month and three-month futures prices is an important determinant in the dynamics of crude oil prices.