medium · FRM Part 2 Market Risk
A portfolio's daily loss distribution is modeled as a Student-t with low degrees of freedom (heavy tails). An analyst observes that, for this portfolio, the ratio of 99% Expected Shortfall to 99% VaR is substantially larger than the value that would obtain under a normal distribution. A colleague argues that switching the risk measure from VaR to ES eliminates the need to worry about tail-heaviness because ES 'captures the average of the tail.'
Which statement most accurately characterizes the situation?
- ES is more sensitive to tail-heaviness than VaR, so a fatter tail widens the ES/VaR ratio; ES does not eliminate model risk because the ES estimate itself depends on the assumed shape of the extreme tail beyond the VaR threshold.
- Because ES averages losses beyond the VaR threshold, it is treated as invariant to the degrees-of-freedom parameter of the Student-t, so the higher observed ES/VaR ratio must instead signal an error in the VaR estimate itself.
- The elevated ES/VaR ratio is taken as proof that the loss distribution violates subadditivity at the 99% confidence level, meaning ES is therefore not a coherent risk measure for heavy-tailed data and VaR should be kept instead.
- ES and VaR are assumed to converge as the return tails grow heavier because both risk measures become dominated by the very same single worst-case quantile outcome, so their ratio should approach one as the degrees of freedom fall further.
Sign up free to see the explanation and track your rank →
More FRM Part 2 Market Risk practice
- A leptokurtic distribution, often modeled by EVT, is characterized by which of the followi
- If a bank records 11 exceptions in a 250-day backtesting window for 99% VaR, what is the r
- In the GPD framework, if the threshold u is chosen too low, what is the most likely error
- In the Kupiec Likelihood Ratio test, what does the null hypothesis (H_0) state?
- The Hill estimator is primarily used to provide a direct estimate of which parameter?
- What happens to the mean of a GPD-distributed variable if the tail index ξ ≥ 1?
- What happens to the VaR estimate if we move from a thin-tailed (Gumbel, ξ = 0) model to a
- What is the base capital multiplier (m) applied to a bank's internal model market risk cap