Gaussian Process Mixture Model: A Covariate-Dependent Bayesian Nonparametric Approach

Gaussian Process Mixture Model: A Covariate-Dependent Bayesian Nonparametric Approach

• Autor
• GUILHERME JOSE LEMOS PIANTINO
• Co-autores
• Hedibert Lopes
• Resumo

• The Gaussian process mixture model (GPMM) is designed to capture regime changes and nonstationary dependence in time-ordered data without imposing a rigid Markovian structure. The model assumes a contiguous segmentation of the time index, assigning a distinct Gaussian process to each regime. To govern this segmentation structure, we employ a Pitman-Yor process adapted as a prior on compositions, which includes the Dirichlet Process as a special case. The estimation procedure utilizes a split-merge MCMC sampler to jointly learn the number and locations of regime boundaries alongside the GP hyperparameters, providing coherent uncertainty quantification. We evaluate the proposed approach through an empirical application to the CBOE Volatility Index (VIX), illustrating its ability to identify meaningful volatility regimes and capture complex nonstationary dynamics in financial time series.

• Palavras-chave
• Gaussian Processes; Regime Switching; Financial Econometrics; Financial Time-Series; Bayesian Nonparametrics; Uncertainty Quantification
• Modalidade
• Comunicação oral
• Área Temática
• Econometria Financeira

Voltar Download