The standard intuition about drawdown recovery is straightforward: faster is better. A strategy that recovers quickly from a drawdown is preferable to one that takes a long time to recover. Recovery speed is a positive indicator. This is the way most retail traders think about drawdown analysis, and it is the way most retail backtest reports present it.
This intuition is wrong in a specific and identifiable case. There is a category of recovery that is too fast — recovery that happens in fewer trades than the strategy’s expectancy mathematically permits. When recovery is faster than expectancy allows, the explanation is not that the strategy has unusually strong edge. The explanation is that the strategy’s position sizes increased during the losing streak. The recovery is fast because the recovery trades are larger than the trades that produced the losses. The strategy is exhibiting a martingale signature, whether the developer intended it or not.
This article explains how the Streak Resilience test in Edge Matrix uses recovery speed as a martingale detection tool, why the mathematical relationship between expectancy and recovery time produces a clean diagnostic signal, and what the specific scoring thresholds mean for interpreting your own backtest results.
The Mathematical Relationship Between Expectancy and Recovery Time
Every trading strategy has an expectancy: the expected dollar value of a single trade, computed as the win rate multiplied by the average win plus the loss rate multiplied by the average loss. A strategy with a 58% win rate, an average win of $120, and an average loss of $95 has an expectancy of $0.58 × $120 + $0.42 × $-95 = $69.6 – $39.9 = $29.7 per trade. Each trade, in expectation, adds about $29.70 to the account.
When the strategy experiences a drawdown — a worst losing streak that drops the account from a previous high by some amount — recovering that drawdown requires generating enough profit to climb back to the previous peak. The expected number of trades to recover is the drawdown amount divided by the per-trade expectancy. If the worst streak produced $1,500 in losses and the strategy’s expectancy is $29.70 per trade, the expected number of trades to recover is $1,500 / $29.70 = approximately 51 trades.
This is not the only possible recovery time. The actual recovery time depends on the specific sequence of trades that follows the streak. A favorable sequence with several winning trades in a row recovers faster than the expectancy predicts. An unfavorable sequence with continued losses recovers slower or not at all. The variation around the expected recovery time is real and substantial.
But there is a floor below which the recovery time cannot reasonably fall, given the strategy’s position sizes. If the strategy is sizing its trades consistently, the realized recovery time will fluctuate around the expectancy-based estimate but should not drop dramatically below it. Recovering a $1,500 drawdown in 10 trades when the expectancy says it should take 51 trades is not statistical noise within normal variation. It is evidence that the recovery trades are not the same size as the trades that produced the loss.
Why Larger Recovery Trades Are a Martingale Signature
The only way to recover a fixed dollar drawdown in dramatically fewer trades than expectancy predicts is to deploy more capital per trade during the recovery period than was deployed during the streak. There are several mechanisms by which this can happen, and they are all variants of position sizing that increases after losses.
Explicit martingale doubles position size after each loss. A strategy that loses on trade 1, then loses on trade 2 with double size, then loses on trade 3 with quadruple size, then wins on trade 4 with eight times the original size produces the classic martingale equity curve: extended periods of normal trading punctuated by sudden recovery trades that recover the entire prior losing sequence in a single move. The recovery trade is dramatically larger than the trades that produced the loss, so the recovery happens in a single trade that cancels out many trades of losses.
Anti-martingale or progressive position sizing can produce similar patterns, although the mechanism is different. A strategy that uses a fixed fractional position sizing model where position size scales with account equity will systematically increase position size during winning periods and decrease it during losing ones — which is the opposite of martingale and produces the opposite signature. But many retail EAs implement fixed lot sizing with periodic “boost” trades or grid recovery mechanisms that fire after specific drawdown thresholds. These hybrid systems behave normally during ordinary trading and only deploy larger position sizes when the system identifies that it is in a recovery phase. The backtest looks normal for most trades and shows occasional large recovery trades that close drawdowns disproportionately fast.
Grid systems exhibit the same pattern from a different mechanism. A grid system opens multiple positions as price moves against the original direction, accumulating exposure. When price reverses, the entire grid closes at a small profit relative to the total accumulated risk. The “winning trade” that closes the grid is, in terms of position size, much larger than any individual trade in the grid because it is closing the combined position. The backtest’s trade-by-trade record shows several small losses followed by one large win that recovers all of them — which produces a recovery efficiency that is mathematically impossible to achieve with consistent position sizing.
What unites all of these mechanisms is the position size relationship: the trades that produce the recovery are larger than the trades that produced the drawdown. The aggregate backtest statistics can still look attractive — the system was profitable during the historical period because the recoveries occurred frequently enough and were large enough to overcome the accumulated losses. The risk profile is fundamentally different from a consistent-position-sizing strategy with the same aggregate result.
The Recovery Efficiency Calculation
The Streak Resilience test in Edge Matrix formalizes this analysis through a single metric called recovery efficiency. The calculation is straightforward:
Recovery Efficiency = Expected Recovery Trades / Actual Recovery Trades
Where the expected recovery trades is the worst streak’s damage amount divided by the strategy’s per-trade expectancy, and the actual recovery trades is the number of trades observed between the end of the worst losing streak and the equity returning to its pre-streak peak.
A recovery efficiency of 1.0 means the actual recovery took exactly as many trades as expectancy predicted — a perfectly average recovery. Below 1.0 means recovery was slower than expectancy predicted, which is a moderate negative signal but consistent with normal variation. Above 1.0 means recovery was faster than expectancy predicted, and the interpretation of this depends on how far above 1.0 the value sits.
A recovery efficiency between 0.7 and 1.5 is the healthy range. This includes some recoveries that were faster than the central estimate and some that were slower, all within the range of variation that consistent position sizing can produce. The scoring assigns 90 points or above for this range. A recovery efficiency between 1.5 and 2.5 means recovery was somewhat slow but not alarmingly so — the scoring drops from 90 to 65 in this range, reflecting concern about edge consistency without flagging a structural problem.
The diagnostic threshold is at recovery efficiency below 0.3. When actual recovery happens in less than 30% of the expected trade count, the score is explicitly capped at 50 — well below the 90-plus that would otherwise be assigned for “fast recovery is good.” The 50 score communicates: this recovery was suspiciously fast given the strategy’s stated expectancy, and the most likely explanation is that position sizes during the recovery were larger than position sizes during the streak. Verify the strategy’s position sizing logic before trusting this recovery as evidence of edge.
Why the Score Caps at 50 Rather Than Penalizing Further
The choice to cap the score at 50 for fast recovery — rather than driving it lower — reflects a specific epistemic position. Recovery efficiency below 0.3 is strong evidence of inconsistent position sizing during the streak, but it is not by itself proof of intentional martingale. The same signature can be produced by legitimate grid systems, by strategies with intentional pyramiding logic, by news event positioning, or by occasional manual interventions that show up in the backtest as larger-than-typical trades.
A score of 50 communicates concern without making a judgment. The trader is informed that the recovery pattern is mathematically inconsistent with consistent position sizing, but the test does not claim to know the reason for the inconsistency. The next investigation step is to examine the actual position sizes during the worst streak versus the recovery — information that is in the trade log but typically not summarized in aggregate backtest reports.
If the recovery trades genuinely have larger position sizes than the streak trades, the strategy is exhibiting a martingale-like risk profile regardless of what the developer intended or how the logic is described. The aggregate backtest figures understate the actual capital at risk during drawdown periods. The maximum drawdown reported in the backtest is the historical maximum — and historical events sometimes happen to favor the strategy by reversing before the position-sizing escalation reaches its breaking point. Live trading exposes the strategy to drawdown sequences that may not reverse, and the position-sizing escalation produces drawdowns that are dramatically larger than the historical figures suggest.
The Four Components of Streak Resilience
Recovery efficiency is one of four components in the Streak Resilience test. The full test combines: expected versus observed streak length at 30% weight, loss clustering at 25% weight, worst streak damage at 25% weight, and recovery speed at 20% weight.
The expected versus observed streak component compares the actual maximum losing streak in the backtest against the streak length that random probability would predict given the strategy’s loss rate. The formula is log(N) / log(1 / loss_rate), where N is the total trade count. For a strategy with a 42% loss rate and 400 trades, the expected maximum streak is log(400) / log(1 / 0.42) = approximately 6.9 trades. If the actual maximum streak was 7, the strategy is exhibiting normal random behavior. If the maximum streak was 14, the strategy is showing a streak that is twice as long as random chance predicts, which is a regime-dependency signal.
The loss clustering component measures whether consecutive losses occur more frequently than the loss rate would predict if losses were randomly distributed. A strategy with a 40% loss rate should have, on average, 40% of trades following a loss also being losses. If the actual frequency of consecutive losses is 65% — much higher than the 40% baseline — the strategy is exhibiting loss clustering that random distribution does not explain. This is the signature of regime dependency: market conditions that produce one loss tend to produce more losses, indicating the strategy’s edge is concentrated in some regimes and fails in others.
The worst streak damage component measures how many average wins are required to recover the dollar amount lost in the worst streak. A damage ratio of 3 means the worst streak’s losses equal three average winning trades — easily recovered. A damage ratio of 12 or higher means the worst streak’s losses equal more than 12 average wins, which is a long recovery period that exposes the strategy to additional drawdown risk during the recovery itself.
The four components together provide a multi-dimensional view of how the strategy behaves during adverse trade sequences. A strategy with normal streak length, distributed losses, manageable streak damage, and normal recovery efficiency receives a high composite score and is exhibiting healthy adversity handling. A strategy with abnormally long streaks, clustered losses, large streak damage, and either very slow or suspiciously fast recovery is exhibiting multiple warning signs simultaneously and warrants close examination before live deployment.
A Concrete Example: The Three Recovery Profiles
Consider three EAs with identical aggregate statistics: 58% win rate, $120 average win, $95 average loss, expectancy $29.70 per trade, and a worst losing streak of $1,500. The expectancy-based estimate of recovery time is 51 trades.
EA A recovered the $1,500 drawdown in 48 trades. Recovery efficiency: 51 / 48 = 1.06. The recovery happened slightly faster than expectancy predicted, which is normal variation. Score: 92. This is the profile of consistent-position-sizing strategy with genuine edge.
EA B recovered the $1,500 drawdown in 73 trades. Recovery efficiency: 51 / 73 = 0.70. The recovery was slower than expectancy predicted, but within normal variation for an edge that fluctuates over time. Score: 78. This is the profile of a strategy whose edge weakened during the recovery period — possibly a regime change, possibly just unfavorable sequencing. Worth monitoring but not structurally alarming.
EA C recovered the $1,500 drawdown in 11 trades. Recovery efficiency: 51 / 11 = 4.6. The recovery happened in less than a quarter of the trades that expectancy predicted. Score: capped at 50 with martingale warning. The only way this recovery is mathematically possible at consistent position sizing is if the strategy generated 11 consecutive winning trades averaging $136 each — possible but extraordinarily unlikely given a 58% win rate. The far more probable explanation is that the recovery trades had larger position sizes than the streak trades, and the strategy is exhibiting hidden martingale behavior that the aggregate statistics conceal.
All three EAs show the same headline figures. The Streak Resilience component scores tell three different stories about their actual risk profiles.
What to Do When Recovery Efficiency Flags a Strategy
A low recovery efficiency score is a diagnostic finding, not a verdict. The appropriate response is investigation rather than rejection. Several specific checks are warranted.
The first check is the position size column in the backtest trade log. Most MT5 and cTrader backtest exports include the lot size or volume for each trade. Filter the trades to those that occurred during the worst losing streak, compute the average position size, then filter to the trades during the recovery period and compute the average position size. If the recovery period’s average is meaningfully larger than the streak period’s average, the position sizing is not consistent, and the strategy is exhibiting the position-sizing-escalation pattern that recovery efficiency detected.
The second check is the strategy’s logic, if it is available. Search the source code or documentation for terms like grid, martingale, pyramid, recovery, boost, scale-in, or layered. Many EAs implement position-sizing-escalation logic explicitly under names that do not obviously announce the behavior. Documentation may describe the behavior as adaptive risk management, dynamic position sizing, or intelligent recovery — euphemisms that all describe the same fundamental pattern.
The third check is the worst drawdown trade-by-trade. The largest single losing trade in the backtest, the largest dollar loss in any 24-hour period, the deepest peak-to-trough drawdown — these specific events often reveal the position-sizing pattern even without explicit access to the strategy logic. A strategy whose worst single losing trade is 3 times its average losing trade is consistent with normal volatility. A strategy whose worst single trade is 15 times the average is exhibiting the position-sizing escalation pattern in its most extreme form.
The fourth check is what Edge Matrix’s Monte Carlo simulation reveals about drawdown distribution. Strategies with hidden position-sizing escalation produce Monte Carlo P95 drawdowns that are dramatically larger than the historical maximum drawdown — often 3 to 5 times larger. A consistent-position-sizing strategy typically has a Monte Carlo P95 drawdown 1.5 to 2 times the historical maximum. A ratio above 3 is consistent with the position-sizing pattern that recovery efficiency flagged.
Why This Test Exists in the Edge Matrix Suite
Martingale and grid systems are among the most common categories of EA available on MQL5 Market and other retail trading marketplaces. They are popular precisely because they produce attractive-looking backtests with smooth equity curves and few apparent drawdowns. The aggregate statistics — win rate, profit factor, average trade — typically look excellent. The drawdown figures look manageable. The strategies appear robust.
The structural risk only becomes visible when the position sizing pattern is examined directly. Most retail validation tools do not perform this examination. The standard backtest report does not summarize position sizes by streak versus recovery period. The Sharpe ratio does not detect position-sizing escalation. The profit factor does not detect it. Even Monte Carlo simulation does not always detect it cleanly, because the simulation typically resamples the trade results without regard to position size variation across the sequence.
Recovery efficiency provides a single, mathematically grounded diagnostic that flags the pattern from a direction the developer did not anticipate. A martingale strategy can fool many tests by simply producing favorable aggregate statistics. It cannot easily fool the recovery efficiency calculation because the calculation is grounded in the mathematical relationship between expectancy and recovery time — a relationship that consistent position sizing must respect and that escalating position sizing necessarily violates.
The Streak Resilience test in Edge Matrix is one of 18 tests in the validation suite. Its 7% weight in the composite Edge Score reflects its importance as a structural detector for one of the most common retail EA failure modes. The recovery efficiency component within that test is what catches the specific signature of position-sizing escalation — the pattern that aggregate statistics are designed not to reveal.
Edge Matrix is available at ergodiclabs.co with a 7-day free trial. The Streak Resilience test, including recovery efficiency analysis and the martingale signature detection, is included in the Edge Matrix Score tab for all uploaded backtests. The free Monte Carlo analyzer provides drawdown distribution analysis with no account required.