Mendes, Diana E. AldeaMoura, Rita Vaz2019-01-162019-01-1620182018http://hdl.handle.net/10451/36449Tese de mestrado, Matemática Financeira, Universidade de Lisboa, Faculdade de Ciências, 2018As séries financeiras exibem algumas características particulares (como não-estacionarie-dade e volatilidade, por exemplo), possíveis de serem capturadas pelo modelo GARCH (Genera-lized Autoregressive Conditional Heteroskedastic) ou por algumas das suas vertentes, tais como o EGARCH (Exponential GARCH) ou o GJR (Glosten, Jagannathan & Runkle). No entanto, da-dos os resultados não totalmente satisfatórios obtidos pelos anteriores, em considerar mudanças súbitas em diferentes regimes de volatilidade, o modelo MRS-GARCH (Markov Regime Switching-GARCH) começou a ganhar bastante importância. O foco desta dissertação passa por comparar a capacidade preditiva dos dois tipos de modelos referidos sobre o Value-at-Risk e posteriormente, garantir que os modelos uni-regimes mais ade-quados para essa, vão surtir ainda melhores efeitos de Expected Shortfall. De forma a inferir sobre ligações entre modelos e medidas, pode encontrar-se uma análise e comparação dos parâmetros envolvidos e das respectivas previsões das mesmas, aplicadas sobre índices provenientes de paí-ses com desenvolvimento económico diferente, o PSI20 de Portugal e o DAX30 da Alemanha. Recorrer-se-á às distribuições Normal, t-Student e GED (Generalized Error Distribution) e a loss functions e técnicas de backtesting específicas para avaliar os resultados. Corrobora-se a ideia de que os MRS-GARCH e distribuições com caudas pesadas, proporci-onam, de certa forma, melhores resultados de Value-at-Risk diário: para o PSI20, o MRS-t é o preferível, e para o DAX30, garante-se que tanto o anterior modelo como o EGARCH-t são os mais adequados. Quanto aos steps temporais utilizados nas técnicas de backtesting, salientam-se os intermédios e o superior horizonte para os anteriores modelos, respectivamente, e a confiança de 95%. Relativamente ao Expected Shortfall, o EGARCH-t a duas semanas para o PSI20 e o EGARCH-N a um mês para o DAX30 são os favorecidos. O Expected Shortfall diário, expecta-velmente, é o que apresenta melhor performance na cobertura das verdadeiras perdas envolvidas.Financial series exhibit some particular characteristics (like non-stationarity and volatility), which can be captured by the GARCH (Generalized Autoregressive Conditional Heteroskedastic) model or some of its strands, such as EGARCH (Exponential GARCH) or the GJR (Glosten, Jag-annathan & Runkle). However, given the unsatisfactory results obtained by the previous ones, in considering sudden changes in different volatility regimes, the MRS-GARCH (Markov Regime Switching–GARCH) model began to gain importance. The focus of this thesis is to compare the predictive capacity of the two types of models refered below on Value-at-Risk and later on, ensure that the best single regimes models for this measure will provide even better Expected Shortfall effects. In order to infer connections between models and measures, an analysis and comparison of the parameters involved and their respective forecasts, applied on indices from countries with different economic development, the PSI20 of Portugal and the DAX30 of Germany can be found. Normal, t-Student and GED (Generalized Error Distribution) distributions and loss functions and specific backtesting techniques will be used to evaluate the results. It is corroborated that the MRS-GARCH and larger distributions provide, somewhat, better daily Value-at-Risk results: for PSI20, MRS-t is preferable and for DAX30, it is guaranteed that the previous one and also EGARCH-t are the most adequate. For the time steps used in backtesting techniques, the intermediate and the superior horizon for the previous ones, respectively, and the 95% confidence are emphasized. For the Expected Shortfall, the two-week EGARCH-t for the PSI20 and the one-month EGARCH-N for the DAX30 are prefered. The daily Expected Shortfall is, expectedly, the one that performs better in covering the real losses involved.porNão-LinearidadeMedida de RiscoMRS-GARCHAvaliação da VolatilidadeTeses de mestrado - 2018Markov-Switching: Value-at-Risk e Expected Shortfall Aplicado aos Índices PSI20 e DAX30master thesis202190706