When estimating tail risk, should I fit the whole return distribution or only the tail?

Question

AcadiFi · Accepted Answer

The tradeoff is bias versus variance. Fitting the whole return distribution uses more data and gives stable parameter estimates, but the model may fit the center well while missing the tail. Fitting only the tail focuses directly on the risk region, but it uses fewer observations and can become noisy. That is why tail modeling often requires judgment about threshold choice, sample length, and model purpose.

Suppose Meridian Equity Risk has 2,500 daily returns for a portfolio. A full-sample normal fit may underestimate 99 percent VaR if returns are skewed or fat-tailed. A tail-only extreme-value approach may better represent severe losses, but if only 125 observations are used, parameter uncertainty becomes important. Neither approach escapes assumptions.

For FRM purposes, the best answer is usually not "always use the whole sample" or "always use the tail." Ask what risk measure is being estimated, how much data exists, and whether the model will be validated through backtesting or stress comparison. Tail risk estimation is about building a defensible model of rare losses, not mechanically choosing the most elegant distribution. On the exam, also state the horizon, confidence level, data source, modeling assumption, and validation step, because those choices usually explain why two reasonable VaR answers differ materially. Practice more in our FRM question bank.

When estimating tail risk, should I fit the whole return distribution or only the tail?

Master Part I with our FRM Course

Related Questions

Practice Questions