Know these first
- L5.1.07: Monte Carlo pricing
Math Atoms
By the end you will be able to
- 1Replace full revaluation by delta, delta-gamma(-theta) and pricing grids, and judge each on the scenarios the engine will see.
- 2Apply antithetic pairs, a control variate with a known tail and importance sampling to a Monte Carlo VaR, and explain when each helps.
- 3Measure an estimator's standard error from repeated seeded runs, and work out the paths it needs for a target error.
- 4Explain why a fast risk engine must still pass the profit and loss attribution test.
Why it matters
Run the sleeve's Monte Carlo VaR on 2,000 normal paths, change the seed, and run it again. Over 100 runs the answer wanders with a standard deviation of £64,381, 3.3% of the VaR. Getting that down to £10,000 would take 82,898 paths. Steering the paths towards the losses gets there with 2,482: 33 times fewer, for the same answer. On fifteen positions this is a curiosity. On a bank's whole trading book, revalued every night, it is the difference between a risk number in the morning and none.