Know these first
- L5.1.07: Monte Carlo pricing
- L5.3.11: Bermudan swaptions
By the end you will be able to
- 1Price a Bermudan by simulation: go back through the exercise dates, estimating the value of waiting by a regression across paths, and choose the basis.
- 2Explain why the estimate on independent paths is biased low, and check it against a tree: the £100m 10nc1 payer at £5.122m against £5.115m.
- 3Build an upper bound from the same rule (Andersen and Broadie) and read the duality gap.
- 4Price the same trade under G2++ and the market models, and say when the curve's shape matters to the exercise decision.
Why it matters
L5.3.11: Bermudan swaptions priced the bank's 10-non-call-1 on a Hull-White tree: £5.115m for the payer. A tree has one dimension per factor. The market models of this module have forty forward rates; even cut to three factors (L5.5.06: Calibrating a market model), a lattice in three dimensions with nine exercise dates is slow and awkward to build. Simulation does not care how many factors there are, but it runs forwards in time, and an exercise decision needs the future. L5.1.07: Monte Carlo pricing promised the way round: regression on simulated paths. Here it is, checked against the tree, bracketed from above, and then taken where the tree cannot go.