4 / 5 · 8 min
Forward testing: a check on a market you have never seen
A good result on history is necessary but not sufficient. Three things appear only going forward, and one of them costs more than the fee.
Why another stage is needed
A test on history answers whether this worked before. A forward test answers something else: whether it works on data that did not exist while you were tuning. The difference is fundamental. Everything you did during tuning — searching parameters, picking instruments, refining rules — could have adapted to the past. The future cannot adapt, because it is not there yet. That makes forward testing the only check you cannot fake even by accident.
First thing missing from history: the fill price
A test on history usually counts at the candle's closing price. In life you pay the spread and move the price with your own order. A live measurement of order books on 09.08.2026 shows how different this is across coins. On bitcoin a hundred-thousand-dollar order moves price by 0.001% — practically free. On a coin like GALA the spread is already 0.056% and the same size moves price by 0.282%. A round trip there costs about 0.68% instead of the assumed 0.1% — seven times more than the test showed.
How far your order moves price
| Coin | Spread | $1,000 | $10,000 | $100,000 |
|---|---|---|---|---|
| BTC | 0.000% | 0.000% | 0.000% | 0.001% |
| SOL | 0.013% | 0.000% | 0.000% | 0.006% |
| LINK | 0.012% | 0.000% | 0.009% | 0.031% |
| GALA | 0.056% | 0.038% | 0.054% | 0.282% |
Read the table together with your position size. At a thousand dollars slippage is invisible anywhere except small coins — and the test result is close to the truth. At a hundred thousand on an illiquid name you lose more than on fees, while the strategy stays exactly the same. Hence a useful consequence: a test result holds for the size at which it was computed and does not transfer to a larger one automatically.
Second: the market changed while you were tuning
The data on which you found the idea describes the market of that period: its volatility, its participants, its fees. If the idea was found in a quiet half-year and you started trading in the middle of a move, you are applying the rule to a different market. Forward testing catches this automatically: it runs on today's market, not yesterday's. That is exactly why ideas that were flawless on history often fall apart in the very first month — not because the test was wrong, but because it was about a different time.
Third: you appear
On history the trades are made not by a person but by a rule: it does not hesitate, does not skip an entry, does not close early and does not increase size after three wins. In a forward test an executor appears between the rule and the market, and that is where the edge is most often lost. The practical value of forward testing is not only checking the idea but measuring the gap: how many signals you skipped, how many times you exited early, how many times you entered off-rule. That difference is your personal execution tax, and knowing its size matters more than refining one more parameter.
Counting at the close and being pleased
The most common technical mistake in testing: the signal appears at the candle's close and the entry is counted at that same price. In life you only see the close once the candle has closed, and you enter at the next price — which on a fast move is noticeably worse. The cure is simple and makes the result honest immediately: count the entry at the open of the NEXT candle and add the spread. If the edge disappears after that correction, it was never there — it lived in an assumption rather than in the market.
Paper trading and its honest limits
Paper is the right way to run a forward test: trades are recorded at the moment of decision, the outcome is unknown, and picking after the fact is impossible. But it has limits and they should be known. It does not reproduce slippage at large size, does not let you feel a real loss, and does not test whether you will keep the rule when money is at stake. So paper answers whether the idea makes sense, while whether I will actually do this with my own money is settled only by a small real size.
How long to run a forward test
The answer is known from the sample lesson: as long as it takes to accumulate the trades your expected edge requires. A noticeable effect shows itself over two or three hundred trades; a small one never shows at all. The practical order is: paper until the required number of trades, then real money at minimum size for the same again, and only then normal size. Each step tests its own thing: the first the idea, the second the execution, the third only your discipline as the stakes grow.
Paper trading in the service runs on the server: the position lives even when the tab is closed, and the journal records entry and exit prices without your involvement. That is exactly the tool a forward test needs — you will not be able to remember a trade differently from how it was. Run your rule there for fifty trades before putting up money.
About paper tradingMeasure your execution tax
Take your last twenty trades and write down two prices for each: the price at which the rule produced the signal, and the price at which you actually entered. Compute the average difference in percent. Then add your instrument's spread and the slippage for your size — visible in the order book right now. Subtract the result from the outcome of your historical test. What remains is what you can count on; for most people it turns out to be half of what they expected.
A strategy showed 30% a year on history. What should I expect in life?
Less, and the question is how much less. Subtract three things. First, the difference in fill price: if you counted at the candle close, add the spread and the slippage for your size — fractions of a percent per round trip on a liquid coin, up to half a percent on a small one. Second, your execution tax: skipped entries and early exits. Third, the fitting effect, if there were many variants. After three corrections usually less than half the original figure remains, and that is a normal result rather than a sign of a bad strategy.
Can the forward test be skipped if the historical test was honest?
No, because an honest test answers a question about the past while you will be trading in the future. Forward testing catches two things history cannot show: whether the market has changed since then, and what happens once you appear between the rule and the market. The second matters most, because that is where the edge is usually lost — not in the idea but in its execution. Skipping the forward test means learning the size of those losses straight away with real money.