The dropout test: kill a strategy, see what survives
Backtests assume every strategy keeps working. Markets do not sign that contract. Dropout testing removes strategies at random during simulation and asks whether the portfolio still deserves its numbers.
Data window: 12-month empirical sample (May 2025 – Apr 2026) · Monte Carlo: 1,500 paths × 3-year horizon · Last verified: June 2026 · Figures refresh quarterly.
Every portfolio in the roster is validated with a dropout layer in the Monte Carlo: on each simulated path, strategies are randomly disabled, roughly a 25 percent dropout on the futures configurations and 20 percent on the swing side. The published blow rates and payout figures already include those crippled paths. The clean version of the portfolio would score better; the clean version is also a fantasy.
Why this matters
Edges decay. A regime shifts, a session changes character, and a strategy that earned for a year goes quiet, the retirement criteria covered in when to retire a strategy. A portfolio whose numbers depend on every member performing is fragile in exactly the way a single backtest cannot reveal: it has no answer to the question "what if one of these stops working", because the historical path never asked it.
What dropout exposes
Concentration wearing a costume. A six-strategy portfolio where one strategy produces most of the net looks diversified and tests as a single point of failure: drop that member and the path quality collapses. The configurations that survive dropout are the ones whose production is genuinely distributed, the same property that drives the low blow rates on uncorrelated pairs. Dropout and decorrelation are two views of one robustness.
Running your own version
The crude version needs no framework: re-run the portfolio backtest with each strategy removed in turn and look at the worst stretch of every reduced set against the buffer. If any single removal turns the portfolio unsurvivable, the allocation is wrong regardless of the headline numbers. The fuller method, resampling with random dropout across thousands of paths, slots into the standard Monte Carlo pipeline, and the wider validation stack lives in the systematic trading guide.
FAQ
What is dropout testing in a trading portfolio?
Randomly disabling strategies during Monte Carlo simulation and measuring the portfolio on those crippled paths. It answers the question a clean backtest never asks: what happens when a member stops working.
Why publish numbers that include dropout paths?
Because edges decay in live trading, and figures that assume every strategy performs forever overstate reality. Numbers that survive dropout are closer to what a deployed portfolio will actually experience.
How can I dropout-test my own portfolio?
At minimum, re-run the backtest with each strategy removed in turn and check the worst stretch of every reduced set against your buffer. If one removal makes the portfolio unsurvivable, the weights are concentrated regardless of appearances.
Not financial advice. Performance figures are hypothetical, modeled outputs (12-month sample; ~1,500-path Monte Carlo where noted). Past performance does not guarantee future results. Verify every prop-firm rule with the firm directly.