Semiparametrické metody Value at Risk

The goal of this thesis is to examine the performance of semiparametric Value at Risk (VaR) methods to see if they deliver better results compared to parametric and nonparametric alternatives on market data. Four models are estimated and compared: the nonparametric Historical Simulation (HS) VaR model, the parametric Delta Normal VaR model, the semiparametric Volatility-Weighted Historical Simulation (VWHS) VaR model using Exponentially Weighted Moving Average (EWMA) volatility estimation, and the semiparametric Age-Weighted Historical Simulation (AWHS) VaR model. The models are applied to the Nikkei 225 index returns series over the period 20 April 2016 to 20 April 2026, a time window containing the COVID-19 price shock of 2020 and the price crash of August 2024. Backtesting framework includes the failure rate and Kupiec's proportion test. The results of the tests provide strong evidence that the semiparametric Volatility-Weighted Historical Simulation VaR model is better than other selected models when applied to data that contain price shocks and volatility regime changes.