TL;DR. Edge Concentration measures how much of your total profit comes from a small number of trades. A high score means edge is spread across many decisions; a low score means a handful of trades carried the backtest. Concentrated edge is the most common pattern of in-sample overfitting on the platform.
What you'll be able to do
- Read an Edge Concentration score and know if your strategy depends on a few lucky trades.
- Decide whether to investigate specific high-impact trades or accept the strategy's concentration profile.
- Avoid the most common false-positive pattern in retail backtesting.
Why concentration matters
A strategy with an edge produces small advantages on many decisions that compound into a meaningful equity curve. A strategy without an edge — but with enough free parameters to fit historical noise — produces an equity curve dominated by a few large lucky wins. The two cases can produce identical headline metrics. The first will probably repeat. The second won't.
Edge Concentration looks at the distribution of per-trade PnL and asks: how concentrated is it? Distributed-edge strategies score high. Lucky-cluster strategies score low.
The pattern is so common it deserves a name: most fragile backtests on the platform have above-target Sharpe driven by 5–15 trades out of several hundred. Edge Concentration is the single number that catches this without the user having to manually inspect every trade.
The math
Gini coefficient of per-trade PnL
The Gini coefficient is an inequality measure originally developed for income distribution. It runs from 0 (perfect equality — every trade contributes identically) to 1 (perfect inequality — one trade carries everything). The platform computes Gini over the absolute values of per-trade PnL so wins and losses both count toward concentration.
EdgeConcentration = (1 − Gini(|per_trade_pnl|)) × 100
A perfectly even strategy scores 100. A strategy where 5% of trades produce 80% of profit scores below 25. The mapping is intentionally aggressive — small Gini differences translate to meaningful score differences because the practical reliability difference between Gini 0.6 and Gini 0.7 is large.
Why absolute values? A strategy that has 3 huge losses and 50 small wins is just as concentrated — and just as fragile — as one with 3 huge wins and 50 small losses. Sign doesn't matter for the diagnosis; magnitude does.
In AlgoLift
Edge Concentration is computed once per Robustness run from the original trade list — it's not bootstrapped because the concentration of the original trades is what's being judged.
The component drives:
- A 15% weight in the overall Robustness Score.
- The mc-path-luck insight chip when concentration is in the bottom third, with copy that calls out the top contributing trades.
- The Raw Scenarios sub-view lets you sort the metrics arrow by total return and inspect what the lucky-tail sims looked like.
The score waterfall when Edge Concentration is the bottleneck
Reading the score
| Score | Reading | Interpretation |
|---|---|---|
| 75–100 | Profit is widely distributed across trades. | Real edge, repeatable structure. |
| 50–74 | Moderate concentration. | Normal for strategies with selective entries or asymmetric exits. |
| 25–49 | Concentrated edge. | A small number of trades dominate. Investigate them before deploying. |
| Below 25 | Severe concentration. | One or two trades likely drive the equity curve. High overfitting risk. |
Some strategies are intrinsically concentrated. Long-only breakout strategies that hold trends often have a few huge wins among many small losses — that's a legitimate edge structure, not noise. The score still flags it; the user judges whether the concentration is signal or fitting.
Worked example
Three strategies, three concentration profiles
What this shows. Strategy A is concentrated but legitimately so — long-trend-following is structurally about capturing rare large moves. Strategy B is a textbook distributed edge; nothing to investigate. Strategy C is the warning. When you look at the top three trades, they're all clustered in March 2020 and depend on a specific intraday move that probably won't recur. The strategy's headline Sharpe drops 60% when you exclude those three trades — which is what a real out-of-sample window would have done anyway.
Common misunderstandings
- "My EdgeConcentration is low but my Path Stability is high. Which do I trust?" Both. They measure different failure modes. Path Stability says your equity curve was a typical outcome of the trade list you have. Edge Concentration says the trade list is fragile in its own right. Both can fail simultaneously, and both can be fine simultaneously.
- "Adding a 'don't take small trades' filter raises Edge Concentration." It usually does, but for the wrong reason — you've narrowed the distribution by truncating the tails of the distribution itself. The component goes up; the underlying robustness doesn't.
When Edge Concentration matters most
- Strategies built with many filters. Each filter removes trades. The remaining trades are by construction the strongest in-sample — exactly the wrong direction for Edge Concentration.
- Strategies validated against short time windows. A backtest covering one full market cycle has more chances to spread its edge than a backtest covering eighteen months — the longer one tends to score higher on this component for structural reasons.
When it matters less
- Strategy archetypes that are structurally concentrated. Long-only momentum, long-vol, long-tail. Concentration is the feature, not the bug; the right cross-check here is whether the concentration trades have a logical reason to recur.
A low Edge Concentration score is the most common signal of in-sample overfitting on the platform. Investigate the top-contributing trades before deploying — if their setup conditions are unlikely to recur, neither will the backtest's returns.