A carregar...
Publicação
Models for Forecasting Value at Risk: A comparison of the predictive ability of different VaR models to capture market losses incurred during the 2020 pandemic recession
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.67 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
The purpose of this study is two-fold. First, it aims at providing a theoretical overview of the most widely adopted methods for forecasting Value-at-Risk (VaR). Second, through a practical implementation, it proposes a methodology to compare and evaluate the predictive ability of different parametric, non-parametric and semi-parametric models to capture the market losses incurred during the COVID-19 pandemic recession of 2020. To evaluate these models, it is applied a two-staged backtesting procedure based on accuracy statistical tests and loss functions. VaR forecasts are evaluated during a volatile and a stable forecasting periods. The results of the study suggest that, for the volatile period, the Extreme Value Theory with a peaks over threshold (EVT-POT) approach produces the most accurate VaR forecasts across all different methodologies. The Filtered Historical Simulation (FHS), Volatility Weighted Historical Simulation (VWHS) and the Glosten, Jagannathan and Runkle (GJR) GARCH with skewed generalized error distribution (GJR GARCH–SGED) models also produce satisfactory forecasts. Moreover, other parametric approaches, namely the GARCH and EWMA, despite less accurate, also produce reliable results. Furthermore, the overall performance of all models improves significantly during the stable forecasting period. For instance, the Historical Simulation with exponentially decreasing weights (BRW HS), one of the worst performers during the volatile forecasting period, produces the most accurate VaR forecasts, with the lowest penalty scores, during the stable forecasting period. Lastly, it was also found that as the level of conservativeness of the model increases, the overestimation of the actual incurred risk seems to a be recurrent event.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Statistics and Information Management, specialization in Risk Analysis and Management
Palavras-chave
COVID-19 Financial Loss Pandemic Recession Stock Indexes Value at Risk