TL;DR. Probability of Profit is the fraction of bootstrap-resampled alternate histories that end with positive total return. It's a path-level summary, not a per-trade win rate. Above 70% is the practical deployability threshold for most strategies.
What you'll be able to do
- Distinguish per-path probability of profit from per-trade win rate and know when each matters.
- Read the Path Luck panel's "P(profit)" KPI and interpret it correctly.
- Decide what P(profit) level your specific risk tolerance can accept.
Why a path-level probability
A backtest produces one equity curve and one final number. The strategy made money or it didn't. But the question that should drive a deployment decision is "what fraction of plausible alternate histories of this strategy ended profitably?"
That's a different question. A strategy with a positive expected return can still produce many losing equity curves if the variance is high enough — the long-run mean is positive but individual histories can land anywhere within a wide distribution. Probability of Profit measures the fraction of those histories that ended above zero.
How it's computed
The Robustness Lambda runs your trade list through stationary block bootstrap ten thousand times. Each resample produces a complete alternate equity curve with its own total return. The fraction of those total returns above zero is the Probability of Profit.
P(profit) = (count of resamples with total_return > 0) / total_resamples
The number isn't conditioned on anything else — not on size, not on prop firm rules, not on regime. It's the unconditional probability of a positive total return across alternate histories of the strategy's edge.
The Robustness Lambda's specific bootstrap
By default, the Lambda applies the stationary block bootstrap with the Politis-White auto-detected block length. Both the resampling method and the block length are recorded in the manifest, so you can see exactly what assumption produced your P(profit).
When the user explicitly opts into iid bootstrap (with the data-independence confirmation), P(profit) is computed against the iid resamples instead. The numbers are different because iid produces narrower distributions — see Bootstrap for the why.
Probability of Profit vs. win rate
These are different numbers measuring different things.
- Win rate is the per-trade probability of a winning trade. Useful for understanding the strategy's local behavior.
- Probability of Profit is the per-path probability of a profitable equity curve over the full backtest window. Useful for understanding the strategy's deployable reliability.
A strategy can have a 60% win rate and a 50% P(profit) — frequent small wins paired with rare large losses that frequently bring the path under zero. A strategy can have a 35% win rate and a 90% P(profit) — infrequent but large wins that dominate the path-level outcome.
When the two diverge sharply, you've learned something about the strategy's payoff structure. Equal divergence in either direction is a real feature of the strategy worth understanding before deploying.
In AlgoLift
P(profit) shows up as a KPI in the Path Luck sub-tab and contributes to the overall verdict in the Robustness Score waterfall (indirectly, through the Path Stability component).
- The Path Luck KPI strip shows P(profit) alongside median return, p5 return, p95 max DD, and median Sharpe.
- The hero verdict tier uses P(profit) > 70% as one of the signals for the "Ship" tier when other components also support it.
P(profit) in the Path Luck KPI strip
Reading the score
| P(profit) | Reading |
|---|---|
| Above 85% | Strong — most alternate histories ended positive. |
| 70–85% | Healthy — typical for a reliably-profitable strategy. |
| 50–70% | Borderline — closer to coin-flip than to skill. |
| Below 50% | More likely to lose than profit on any given path. Don't deploy. |
The 70% threshold is empirical, not mathematical. It's roughly where the Robustness Lambda's experience shows strategies transitioning from "fragile, depends on luck" to "honest edge that survives noise." Above 85%, the strategy is reliably profitable across the kinds of histories the resampler can generate; the remaining 15% are typically tail scenarios that you'd accept losses on anyway.
Worked example
Three strategies, three relationships between win rate and P(profit)
What this shows. Strategy A wins less than half its trades but produces profitable equity curves 88% of the time — the wins are large enough and the variance is bounded enough that the path-level outcome stays in the green even when most trades lose. Strategy B has a healthy per-trade win rate but a narrower P(profit) because the wins are small and the losing streaks compound. Strategy C is the rarest case — a true asymmetric-payoff strategy where most trades fail but the rare wins are so large that the alternate histories stay positive at 79% rate.
Common misunderstandings
- "P(profit) is 60% but my win rate is 75% — something's wrong." Nothing's wrong. The two measure different things. A 60% P(profit) with a 75% win rate suggests big losses or correlated losing streaks; investigate the Tail Discipline and Edge Concentration components.
- "I should optimize for P(profit) directly." Risky. P(profit) responds to selection bias the same way Sharpe does. The right primary optimization metric is something tied to expected utility, not P(profit) directly.
When P(profit) is the right summary
- Deployment screen. The first number to look at before a deploy. Below 70% means stop and investigate.
- Sizing decisions. P(profit) drops as you size up because variance scales. The Sizing panel's frontier chart shows the cross-over point where P(profit) drops below an acceptable threshold.
When it's less informative
- Comparing strategies with very different time horizons. A multi-year backtest and a one-month backtest will produce different P(profit) just because of how the variance scales. Compare like-with-like only.
- Strategies with no closed terminal date. Forward-deployed strategies don't have a fixed window for which P(profit) is meaningful. The Forward Testing tab uses different metrics.
Probability of Profit is the first deployability screen on the Robustness tab. It says, in one number, how reliable the strategy's profitability is across plausible alternate histories. Above 70% is the practical threshold for "deployable under tested assumptions."