Statistical methods for conditional volatility modeling with applications to finance

Type: dissertation-thesis

Abstract

This doctoral thesis focuses on developing innovative statistical methods for conditional volatilityforecasting, integrating machine learning techniques, and improving numerical optimizationstrategies. With an emphasis on the highly volatile Bitcoin market, the research introduces theVariable Selection Log-TGARCHX (VS-LTGARCHX) algorithm, which helps identify keyexogenous variables to improve volatility predictions and mitigate overfitting. The thesis alsopresents the Cross Entropy with Symmetry Breaking and KMeans (CESBKM) algorithm, refiningclustering techniques for capturing regime-switching behaviors in financial time series. Additionally,it introduces the Cross Entropy with Decision Trees (CEDT) algorithm for Open Loop ThresholdAutoregressive (OLTAR) models, combining global optimization with decision tree learning to bettercapture regime dynamics. Empirical applications demonstrate that the proposed modelsoutperform traditional benchmarks, offering useful methodological tools with practical applicationsin financial modeling.

Recommended citation: Citation information from ORCID (Work Type: dissertation-thesis)
Download Paper