Conditional Scenario Sampling

A plain Monte Carlo stress test draws random scenarios from a model's overall distribution and looks at what happens to the portfolio across all of them. Conditional scenario sampling instead fixes one or more variables to a specified stressed value — say, oil up 20% — and samples the rest of the scenario (equities, rates, credit spreads) from their distribution given that condition, using the model's estimated dependence structure to fill in realistic knock-on moves rather than leaving them unconstrained.

Why "holding everything else constant" is the wrong default

A naive stress test that shocks oil while leaving every other variable at its unconditional average implicitly assumes oil moves are uncorrelated with everything else, which is rarely true — an oil spike historically comes with moves in energy equities, inflation breakevens, and often risk sentiment broadly. Conditional sampling uses the correlation or copula structure fitted from historical or simulated data to generate a distribution of complete scenarios consistent with the oil shock, then reports a distribution of portfolio outcomes rather than a single point estimate.

The output is a range, not one number: instead of "the portfolio loses $4 million if oil spikes 20%," a risk report can show that 90% of plausible co-scenarios consistent with that oil shock produce losses between $2 million and $9 million, which is far more useful for sizing a hedge or a capital buffer than a single deterministic shock.