Dynamic volatility models for market risk and portfolio analysis

• Ano de defesa: 2020
• Autor(a) principal: Arthur Tarso Rego
• Orientador(a): Thiago Rezende Dos Santos
• Banca de defesa: Fábio Nogueira Demarqui, Cristiano de Carvalho Santos, Fernando Ferraz do Nascimento, João Batista de Morais Pereira
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Minas Gerais
• Programa de Pós-Graduação: Programa de Pós-Graduação em Estatística
• País: Brasil
• Palavras-chave em Português: Financial Time Series; Dynamic Models; Multivariate Stochastic Models
• Link de acesso: http://hdl.handle.net/1843/34384

Dealing with financial time series brings many challenges to data modeling, given the existence of heavy tails and extreme return values caused by external events, such as politics, natural disasters, economical events, or even speculation. Providing reliable and interpretable insights to market agents, based on statistical models, to base strategic decisions on building assets portfolios, taking arbitrage decisions, and managing investment risks is crucial for avoiding losses and correctly pricing assets for developing successful investment strategies. Rego and Santos (2020) proposed the Non-Gaussian Stochastic Volatility Model with Jumps (NGSVJ) for market volatility evaluation, which includes automatic inference procedure that allows the model to be fast enough to bring tangible results for the user, using an ordinary home computer, to perform trading operations. The Dynamic Models (DM) class, on which the NGSVJ is based, has a flexible structure that enables the inclusion of new features on the models and has implementation simplicity from the computational perspective. The DM class of models is still unexplored for financial applications when compared to the other classes of models commonly used on literature, mainly based on Stochastic Volatility (SV) and Generalized Autoregressive Conditional Heteroskedasticity (GARCH) classes of models. In this thesis, several developments are made using as basis the DM class and the NGSVJ model. For dealing with a single asset, or univariate, financial time series, developments are made to the NGSVJ to be able to estimate the degree of freedom of the gamma mixture parameter and include a Hiden Markov (HMM) to give flexibility and interpretability to the model for applications on arbitrage intraday market operations. For dealing with multiple assets portfolio, or multivariate, financial time series the Multivariate Stochastic Volatility Model with Jumps (MSVJ) was developed, based on DM structure, to enable financial agents to estimate the volatility and correlation between portfolio assets and effectively develop a risk management strategy. This thesis provides a wide set of statistical models, based on DM class, that can be used in finance for taking arbitrage and investment decisions, whether it is used for analyzing a single asset or a portfolio. Simulation studies are presented as well as applications on the S\&P 500 market index, commodity derivatives, and exchange rates, to illustrate model performance. The proposed models have highly interpretable results, bringing major developments to the DM class of models and their applications on finance. The proposed models are robust in the sense to incorporate several stylized characteristics of return data, bringing major developments to the NGSVJ and their applications.