问一道题：NO.PZ2019052001000139

问题如下：

A profitable derivatives trading desk at a bank decides that its existing VaR model, which has been used broadly across the firm for several years, is too conservative. The existing VaR model uses a historical simulation over a three-year look-back period, weighting each day equally. A quantitative analyst in the group quickly develops a new VaR model, which uses the delta normal approach. The new model uses volatilities and correlations estimated over the past four years using the RiskMetrics EWMA method.

For testing purposes, the new model is used in parallel with the existing model for four weeks to estimate the 1-day 95% VaR. After four weeks, the new VaR model has no exceedances despite consistently estimating VaR to be considerably lower than the existing model’s estimates. The analyst argues that the lack of exceedances shows that the new model is unbiased and pressures the bank’s model evaluation team to agree. Following an overnight examination of the new model by one junior analyst instead of the customary evaluation that takes several weeks and involves a senior member of the team, the model evaluation team agrees to accept the new model for use by the desk.

Which of the following statements about the risk management implications of this replacement is correct?

选项：

Delta-normal VaR is more appropriate than historical simulation VaR for assets with non-linear payoffs.

Changing the look-back period and weighting scheme from three years, equally weighted, to four years, exponentially weighted, will understate the risk in the portfolio.

The desk increased its exposure to model risk due to the potential for incorrect calibration and programming errors related to the new model.

A 95% VaR model that generates no exceedances in four weeks is necessarily conservative.

解释：

Given the quick implementation of the new VaR model and the insufficient amount of testing that was done, the desk has increased its exposure to model risk due to the increased potential for incorrect calibration and programming errors. This situation is similar to the JP Morgan London Whale case in 2012, where a new VaR model was very quickly introduced for its Synthetic Credit Portfolio response to increasing losses and multiple exceedances of the earlier VaR model limit in the portfolio.