On many Expert Advisor pages, the first number you see is the win rate: 85%, 90%, 95%. It is the easiest number to sell, because being right most of the time feels like safety.
It is also the number that hides the most risk. A win rate tells you how often a system gets paid, not whether it gets paid enough for the risk it carries.
Below, one worked example shows the problem. Then come the three checks I use to separate a real edge from a stop that was never really tested. Every market in AbacuQuant Portfolio had to pass them, and the free demo lets you check that yourself.
The problem: a win rate without its break-even line
A win rate means nothing until you compare it with the win rate the system needs just to break even.
Take a hypothetical EA with a 30-point take profit and a 300-point stop loss. Over 300 trades it wins 285 times, a 95% win rate. That is 8,550 points won (285 × 30), 4,500 points lost (15 × 300) and +4,050 points net. The equity curve looks smooth, almost a straight line.
Now ask what win rate this EA needs to break even:
break_even = SL / (SL + TP) = 300 / (300 + 30) = 90.9%
The EA is not "95% good". It sits 4.1 percentage points above its own break-even line, and that margin rests on just 15 losing trades. Here is what happens when a few of those 300 trades turn into losses (before costs):
| Win rate | Losing trades (of 300) | Net result |
|---|---|---|
| 95% | 15 | +4,050 points |
| 91% | 27 | +90 points |
| 90% | 30 | −900 points |
Twelve extra losses wipe out almost the entire profit. With fifteen, the EA wins 90% of its trades and still loses money. A drop of four or five points in win rate is the kind of shift a change in market regime can produce.
That is the hidden risk in high-win-rate systems. The losses are rare, so they are barely sampled, and each one is large, so every extra loss matters.
Three checks that separate an edge from an untested stop
Three questions separate a real edge from a lucky one, and a market has to pass all three. I apply them to any high-win-rate system, mine included.
1. Is the win rate above break-even by more than noise?
Subtract the break-even rate from the real win rate, then express that margin in standard errors, where n is the number of trades:
margin = win_rate - break_even t = margin / sqrt( win_rate * (1 - win_rate) / n )
A t of 2 or more means the margin is unlikely to be chance. The example above passes this check, with a t of about 3.3.
2. How many losses is the record built on?
The risk lives in the losses, so they have to be observed, not assumed. I require at least 40 losing trades. The example has 15, so it fails here, and the table above shows why that matters.
3. Do the losses actually close at the stop?
The break-even math assumes that a loss costs the full stop. If losing trades close early through time exits, opposite signals or manual action, the stop was never tested and the real cost of a loss is unknown. I require at least 95% of losing trades to close at their stop.
How AbacuQuant Portfolio applies the checks
Every market in AbacuQuant Portfolio passed all three checks in a stand-alone test on real ticks, in the exact version shipped in the EA. Of the 39 markets I researched, 22 did not make it.
Before those checks, each market also had to show that ranking configurations by one period predicted their ranking in the next. Ten individual US stocks failed that screening. Some of the later rejections show the checks at work:
- Crude oil: showed the strongest agreement between periods of any market (rank correlation +0.60), then failed on real ticks with a negative margin.
- S&P 500: passed the screening, then failed on real ticks. Its win rate sat barely above break-even (t = 0.53).
- DAX, CAC 40 and IBEX 35: failed the same check (t = 0.34, −0.85 and 0.16).
- AUDJPY: missed the screening threshold by 0.004 and was rejected anyway. The thresholds do not bend.
- GBPUSD: passed with a wide margin, but on only 18 losses. It ships switched off until a rebuilt preset reaches 40.
The eleven markets that passed all three checks:
| Market | Configurations | Margin over break-even (percentage points) | t | Losses observed |
|---|---|---|---|---|
| AUDUSD | 8 | +8.17 | 4.88 | 83 |
| USDJPY | 6 | +7.24 | 4.19 | 80 |
| Euro Stoxx 50 | 6 | +8.53 | 4.12 | 58 |
| EURUSD | 8 | +4.24 | 3.56 | 42 |
| Dow Jones | 9 | +6.03 | 3.50 | 127 |
| Australia 200 | 3 | +8.52 | 3.06 | 66 |
| FTSE 100 | 7 | +6.36 | 3.04 | 81 |
| Nikkei 225 (*) | 8 | +5.22 | 2.63 | 84 |
| Nasdaq 100 | 4 | +6.37 | 2.33 | 71 |
| XAUUSD | 7 | +4.41 | 2.27 | 123 |
| EURJPY | 5 | +6.62 | 2.21 | 72 |
In every accepted preset, 100% of losing trades closed at their stop. Gold, the Nasdaq 100 and EURJPY passed with the least room, and they are listed that way instead of being averaged into the others.
(*) For the Nikkei 225, the validation broker supplied real ticks for about half of the test history. The Strategy Tester modelled the rest.
Rejected markets that reached the real-tick stage stay in the code as labelled, switched-off presets. You can open them in the Strategy Tester and see exactly why they failed.
From eleven markets to one account
Eleven valid edges can still overload one account, so all presets run under one account-level risk engine:
- Each position is sized from its own stop distance, at 0.5% risk per trade by default.
- Three global ceilings, 30% each by default, cap total open risk, margin in use and the decline from the equity peak, across every market at once.
- No more than 10 positions can be open across the whole portfolio.
- Every order opens with its stop loss and take profit attached. No martingale, no grid, no averaging, no recovery logic.
The markets also follow the trading day: Asian indices in the Tokyo and Sydney sessions, European indices and currencies through London and Frankfurt, US indices into the New York afternoon. When one market is quiet or in a losing stretch, others are often active.
I ran all eleven presets together in the Strategy Tester with real ticks, from January 2020 to July 2026, at 0.5% risk per trade. The result: 11,333 trades, an 80.3% win rate and a deepest equity decline of 6.4%.
In the most recent period, May 2024 to July 2026, the profit factor was 1.38 and 24 of 27 months were positive. All eleven markets ended in profit, and none contributed more than 14% of the total.
The trade-off you should see before you rent
A high win rate with honest stops has a cost: across the combined test, the average loss was about 2.8 times the average win. You will see losing trades, and a single one can erase several winners.
That is not a malfunction. It is exactly what the break-even line measures, and the margin above that line is what pays for it.
A few more points, stated plainly:
- The second research period was also used to require consistency when configurations were chosen, so it is not a blind test. The stronger evidence is elsewhere: one process rejected 22 of 39 markets, and I have been running this engine on my own real-money account since August 2026.
- Losses grow with the risk setting. An earlier run of the portfolio at 2% per trade, four times the default, declined 22% from its equity peak.
- Every figure here comes from the Strategy Tester on historical data, at one broker with one clock setting. It describes what happened, not what will happen.
Run the checks yourself with the free demo
The free demo lets you run the same three checks on your own broker before you pay anything. It is the full EA, limited to the MetaTrader 5 Strategy Tester.
- Download the free demo from the AbacuQuant Portfolio product page.
- In the Strategy Tester, select the EA on an H1 chart and choose "Every tick based on real ticks".
- Set the deposit to the amount you actually plan to trade. At startup, the Experts log prints the balance each configuration needs on your broker.
- Open the trade list in the report. Count the losses, check that they closed at their stops, and compare each market's win rate with the break-even implied by its stop and target.
- Look closely at the losing trades. If that distribution does not fit the risk you accept, it is far better to learn it in the tester than with real money.
You need a hedging account; a netting account merges positions and the presets will not behave as validated. The validation used broker time GMT+2 in winter and GMT+3 in summer; if yours differs, enter its winter offset in the inputs.
As a reference, sizing works properly from about 3,000 for the three currency majors and about 20,000 for all eleven markets. Adding gold, EURJPY and the indices except the Nikkei needs about 8,000.
The same checks work on any EA. For a wider audit of closed trades, covering martingale, grid, payoff asymmetry and risk of ruin, you can use my free Hidden Risk of Ruin Auditor script. The article explains how it works.
Who it is for
AbacuQuant Portfolio is for traders who want to see the losses before they trade, not after. One H1 chart runs eleven validated markets and 71 configurations, with a hard stop on every order and risk measured across the whole account.
Rentals start at 49 USD per month and include ongoing updates. New markets are added only when they pass the same checks, and you get direct support from me in English and Spanish.
It is not for anyone looking for an equity curve without visible losses.
Download the free demo and read the full methodology
Trading currencies, indices and metals on margin carries a high risk of loss. Strategy Tester results describe the past and do not guarantee future results.
