A comparative evaluation of VaR models using Monte Carlo simulations

• Autor(a) principal: Gomes, Diogo Filipe Meireles
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/21164
• topic: Quantile regression; Value-at-Risk; Garch; Monte Carlo simulations; Regressão de quantis; Simulação de Monte Carlo

Quantile regression models emerges as an alternative Value-at-Risk (VaR) methodology that does not require any specific distribution assumption. This dissertation describes and tests a recent quantile regression VaR model that introduces a nontrivial transformation enabling the use of Generalized Autoregressive Conditional Heteroskedasticity (GARCH) volatility models, proposed by Zheng et al. (2018). His study has shown that this approach to the conditional quantile estimation for a GARCH(1,1) model provides promising results. We test this new model against a group of benchmarks composed by traditional VaR methods and other quantile regression VaR models. In order to evaluate the performance of this new VaR model, we generate returns by Monte Carlo simulations following a GARCH(1,1) process similar to what was carried out by Zheng et al. (2018). After, we change the returns process parameters to, in our opinion, more realistic assumptions on the daily persistent volatility. We confirm the superior performance of the new proposed model when the return generating process is simulated with Zheng et al. (2018) parameters, however, the same does not happen when a more reasonable parametrization is simulated. New results show that the benchmark group is heavily penalized by Zheng et al. (2018) parametrization.