Core Path · L6.2.08

Estimating expected shortfall

14 minExpected shortfallBootstrapStandard errorStressed ESLiquidity horizons

By the end you will be able to

  1. 1Estimate a 97.5% expected shortfall from history in several ways (empirical, fractional, kernel, weighted, filtered, parametric and GPD), and explain why they differ.
  2. 2Put a standard error on the empirical ES by formula and by bootstrap, and compare it with a simulated ES's error.
  3. 3Compare ES 97.5% with VaR 99% on real data and under normal and t distributions.
  4. 4Carry out an illustrative FRTB stressed calibration and a liquidity-horizon cascade on the sleeve.

Why it matters

The sleeve's 97.5% expected shortfall over the last 500 days is £2,248,211: the mean of its 13 worst days. Resample those days and the answer ranges from £1.93m to £2.54m. Weight recent days more and it is £1.61m; scale each day to today's GARCH volatility and it is £1.39m. FRTB's internal models approach puts this one number at the centre of capital. This lesson asks how well we can know it.