TL;DR. Regime Breadth is the minimum probability of positive Sharpe across detected market regimes. It carries 10% of the Robustness Score and exists specifically to prevent a strategy from looking good on average while being unprofitable in a specific kind of market.
What you'll be able to do
- Read a Regime Breadth score and know your worst-regime performance.
- Distinguish strategies that are intrinsically regime-specific from ones that are accidentally so.
- Decide when low Regime Breadth is acceptable.
Why the minimum, not the average
Market regimes change. Strategies that work brilliantly in trending environments often lose money in chop. Strategies that excel in low volatility often blow up when volatility spikes. The average performance across a multi-regime sample is misleading because you don't trade against the average — you trade against whichever regime is in front of you right now.
A strategy with a Sharpe of 1.8 across the full backtest could have a Sharpe of 2.5 in trend regimes and -0.7 in chop. The trend regimes happen to be the majority of the sample, so the average looks fine. The reality is that when chop arrives — and it does, repeatedly — you bleed money.
Regime Breadth catches this by scoring the worst regime, not the blended one. It's deliberately anti-cherry-pick. You can't earn a high Regime Breadth score by ensuring your backtest had more good regimes than bad ones.
The math
RegimeBreadth = min over regimes of P(Sharpe > 0 | regime) × 100
The platform's regime classifier produces between 2 and 5 regime labels per backtest based on volatility, trend strength, and time-of-day clustering. For each regime, the resampled paths are filtered to trades that occurred during that regime, and the probability of positive Sharpe is computed.
The minimum is taken across regimes. If the strategy has a 92% P(Sharpe > 0) in trends, 85% in mixed, and 31% in chop, Regime Breadth is 31. The other regimes don't matter — they can't compensate for the bad one.
The fallback when regimes aren't detected
Not every backtest has enough trades in enough regimes to compute Regime Breadth directly. When the classifier fails, the platform falls back to a 60/40 volatility split — top 40% of the sample by realized vol vs. bottom 60% — and computes the same minimum. The score is annotated with a "vol-split fallback" tag in this case so users know the regime detection wasn't fine-grained.
In AlgoLift
Regime Breadth's role in the score is small (10%), but its diagnostic value is large. A low Regime Breadth score is often the only signal pointing at why a strategy fails out-of-sample.
- The Robustness Score waterfall shows Regime Breadth as one of the smaller segments.
- The Tail Risk sub-tab's percentile bands are filterable by regime in the Advanced view (when present), letting you see the regime-specific equity curve cloud.
- The mc-regime-fragility insight chip fires when the minimum-regime P(Sharpe > 0) is below 50%.
Reading the score
| Score | Reading | Interpretation |
|---|---|---|
| 75–100 | Every regime has high probability of positive Sharpe. | The strategy generalizes across conditions. |
| 50–74 | Modest cross-regime variation. | Normal for most strategies. |
| 25–49 | At least one regime is a meaningful drag. | Investigate which regime and whether you can avoid it. |
| Below 25 | At least one regime is consistently unprofitable. | Strategy is regime-specific. Either gate live trading or accept the limitation. |
Worked example
Two strategies across detected regimes
What this shows. Both strategies have similar overall Sharpe ratios. Strategy A's breadth is 78 — it works across regimes with predictable degradation in chop. Strategy B's breadth is 22 — it's a great strategy in trending markets and broken in range-bound ones. The overall Sharpe hides this because the backtest sample happened to be majority trending. Live trading is going to expose Strategy B to extended range periods, and the equity curve will look nothing like the backtest during those windows.
Common misunderstandings
- "My backtest covers a 6-year window — that's plenty of regime variety." Often, but not always. The 2010s had unusually long low-volatility regimes. A backtest dominated by one type of market doesn't have enough information for the regime classifier to detect alternatives.
- "I should add a regime filter to raise the score." Sometimes. Other times the strategy's intrinsic structure simply doesn't work in some regime, and removing those trades reduces drawdown but doesn't actually rebuild the missing edge. Run the strategy with and without the filter through the Robustness tab and compare components, not just the headline.
When Regime Breadth matters most
- Strategies deployed without a regime gate. If you're trading the same parameters around the clock and across market conditions, Regime Breadth is a binding constraint.
- Strategies tested over short windows. A 12-month backtest with a single dominant regime can't tell you about your strategy's other-regime behavior. The platform falls back to vol-split in this case but the diagnostic value is reduced.
When it matters less
- Regime-gated live deployments. If you only trade when a detector confirms the favorable regime, the unfavorable-regime score is informational only.
- Single-regime strategies by design. A volatility-event strategy that only trades on FOMC days legitimately doesn't have meaningful range-regime exposure; low Regime Breadth here describes the strategy's design rather than its weakness.
Regime Breadth is the anti-cherry-pick component. A high overall Sharpe with a low Regime Breadth tells you the strategy was rewarded for the regimes that happened to dominate the backtest — and that the live experience may be very different when those regimes pass.