Monte Carlo asks how sensitive a result is to which specific trades happened to occur, holding everything else fixed. Stress testing asks a different, complementary question: what happens if the conditions themselves get worse — costs a bit higher, positions a bit larger, correlations a bit less favorable than assumed — rather than just resampling from the same fixed assumptions.
Assumptions, not just outcomes, deserve scrutiny
Every backtest runs on a specific set of assumptions about costs and conditions — a particular slippage value, a particular spread, a particular leverage level. Those assumptions are estimates, not guarantees, and a strategy's real-world performance depends on how wrong those estimates are allowed to be before the result stops holding up. Re-running a strategy with meaningfully worse cost assumptions than the base case — not because those worse numbers are expected, but to see how much room for error the strategy actually has — is a different exercise than resampling the trades that already happened under the original assumptions.
What this looks like in practice
Deliberately widening slippage and spread beyond the base-case estimate and re-running the backtest shows how much of the apparent edge depends on costs staying close to their assumed values. Testing at reduced leverage, or with per-instrument costs pushed toward their less favorable end, shows whether the strategy's margin for error is comfortable or razor-thin. None of this is about predicting the future correctly — it's about knowing, in advance, how much the strategy actually depends on today's assumptions being right.
Why this matters alongside Monte Carlo, not instead of it
Monte Carlo and stress testing catch different failure modes. A strategy can be robust to which specific trades happened to occur and still be fragile to costs turning out to be worse than assumed — or the reverse. Checking both is what distinguishes a strategy whose edge holds up under real-world variation from one that only survives inside the exact conditions it happened to be measured under.
Full reference: docs · The workflow this maps onto: reamerlabs.com