Is your strategy skilled, or just lucky? Measure it against random trades
21 November 2026, 07:56
0
10
From experience
I develop trading strategies, and I used to believe that any manual trading method could be turned into an EA, as long as the rules were clear and complete. So I studied more and asked experienced traders for advice, then built EAs from many strategies. I adjusted functions and logic until the results passed the standards I had set, on the 3-5 years of data my broker provided.
But every time I tested them on a new period, the results turned completely upside down. At the time, I did not understand why.
Now I understand: tuning a strategy until it looks good on old data means fitting it to the luck of that period. Every strategy has to be confirmed on new data (a forward test). And while you wait for the forward test, there is another check you can run right away: compare your strategy with random entries.
1. The question a profitable backtest cannot answer
An EA runs through a backtest. The profit looks good, the statistics look good, so you feel sure the strategy works.
But ask one more question: "Was the market simply kind during the test period?"
The equity curve cannot answer this. It only tells you how much you made, not how much you would have made by entering at other times.
2. Let a "monkey" trade against you
A monkey test is how you get the answer.
The monkey changes only the entry time of each trade to a random time. Direction, lot size and holding time stay the same. Repeat this thousands of times and you get a pile of "random versions". Then see where your real result sits in that pile.
If it sits in the middle, your strategy enters about as well as random timing. If it sits above almost the whole pile, you have stronger evidence that the strategy has something real.
3. Keep the comparison fair
Careless randomizing gives misleading results. Watch three things:
- First, match the timing: pick random times on the same weekday and the same hour, in the following weeks.
- Second, use prices with the real spread and the costs you actually pay. Otherwise the random versions look better or worse than they should.
- Third, trades that are open at the same time, or opened as a set, must be moved together as one group, so the random versions keep the structure of the real strategy.
4. Traps most strategy developers do not know
1) Martingale, or adding to losing trades: it wins almost all the time, so it beats random timing almost every time, until one day it takes a huge loss. That day may simply not have happened yet in the test period. So this result cannot count as evidence.
2) Too few trades: 30-50 trades are still within the range of pure chance.
3) Trades closed too fast: trades held for only a few seconds cannot be compared with random versions built on 1-minute prices. The result is not reliable.
4) Picking the best of many strategies: there is a good chance you are only picking the luckiest one.
5. Example: a random EA and an EA that knows the future
To see whether this works, I tried two imaginary EAs on gold in early 2026.
- The random EA: its real result sat in the middle of the random versions, exactly as it should, because it trades at random just like the test does.
- The EA that knows the future (impossible in real life, otherwise we would all be rich): its real result was above every single random version.
6. Summary
Comparing with random trades does not tell you that a strategy will keep making money. It only tells you whether its entry and exit timing is better than random, and it tells you honestly when there is not enough data to decide.
If you have tick price data, you can build this test yourself in MQL5.
And always remember: a good result in this test is evidence, not proof. Every strategy still has to be confirmed with a forward test on new trades.


