Conditional autoregressive Value-at-Risk model estimates in financial markets
碩士 === 逢甲大學 === 統計與精算所 === 95 === Value-at-Risk (VaR) forecasting is required by all financial institutions (Basel II). For better VaR estimation, Engle and Manganelli (2004) proposed quantile regression to model VaR directly instead of modeling the underlying volatility generating process. They int...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en_US |
| Published: |
2007
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/47728487765504579564 |
| Summary: | 碩士 === 逢甲大學 === 統計與精算所 === 95 === Value-at-Risk (VaR) forecasting is required by all financial institutions (Basel II). For better VaR estimation, Engle and Manganelli (2004) proposed quantile regression to model VaR directly instead of modeling the underlying volatility generating process. They introduced the conditional autoregressive value at risk (CAViaR) model. Recent work shows that the nonparametric solution is a special case of the Skewed-Laplace distribution. This will allow development of likelihood/Bayesian to more rigorously estimate VaR. However, VaR may exhibit substatistical nonlinearity and a model that allows for structural change in VaR should be entertained. We extended the CAViaR idea to more flexible nonlinear CAViaR models. Some theoretical results were derived and Bayesian methods for model estimation were proposed. We further proposed a thorough comparison with GARCH-type models and RiskMetrics. Violation rates were used to assess the Value-at-Risk forecasting performance, and CAViaR models were found to exhibit superior performance.
|
|---|