Backtesting, honestly

A backtest tells you what a rule would have done. That is useful and it is not a promise.

Validating a strategy on historical data before risking money is a step almost nobody skips deliberately — they skip it because every free tool either demands an account or requires learning a scripting language first. That barrier is why most traders run on intuition and discover the cost live.

The four numbers that matter

Win rate alone is the most misleading figure in trading. A rule winning 70% of the time with tiny wins and enormous losses is a losing system. Read win rate and profit factor together or not at all.

The comparison everyone forgets

Why results always flatter

1. Fees. Include both sides of every trade. A rule that trades often can be profitable before fees and dead after them.
2. Slippage. Backtests fill at the closing price. Real orders fill worse, especially in the fast moves that trend rules depend on.
3. Funding. On perpetuals, every day held costs money that no simple backtest counts.
4. Overfitting. Tuning parameters until the curve looks beautiful produces a rule that describes the past perfectly and predicts nothing.
5. Survivorship. Testing on coins that still exist ignores every one that went to zero.

How to test properly

• Test the same rule on several pairs. An edge that works on one coin and nowhere else is probably noise.
• Test across timeframes. A rule that only works on four-hour candles is fragile.
• Change one thing at a time. If adding a stop improves results, that is information. Changing six settings at once is not.
• Judge drawdown before return. The strategy you cannot hold through is worthless regardless of its final number.

What comes next

A good backtest is permission to test small with real money, nothing more. Log those trades in the journal and compare live results with the backtest after twenty entries. The gap between them is where slippage, hesitation and your own psychology live — and that gap is the real education.

Overfitting, and how to recognise it in yourself

Overfitting is tuning a rule until it fits past data beautifully and describes nothing about the future. It is not a technical failure — it is a psychological one, because the process feels like research.

Four warning signs:

• The rule needs an unusual parameter. If it works at period 37 but not 35 or 40, you found noise.
• It works on one pair only. A real edge usually appears, weakly, across similar markets.
• Adding conditions keeps improving results. Each new filter removes losing trades from the sample. That is curve-fitting, not discovery.
• The equity curve is suspiciously smooth. Real edges are lumpy. Perfection is a red flag.

Testing the same rule properly

1. Run it across several pairs. Our backtester lets you switch markets without changing anything else — the fastest test of whether an edge is real.
2. Run it across timeframes. A rule that only works on four-hour candles is fragile.
3. Change one variable at a time. If adding a stop improves things, that is information. Changing six settings at once is guessing.
4. Test different market regimes. Most trend rules look brilliant in a trending sample and lose steadily in a ranging one.
5. Compare against buy and hold, always. Beating the asset is the only result that justifies the risk and effort of trading it.

What a backtest cannot include

Every backtest is optimistic, and knowing precisely how keeps you honest.

• Slippage. Fills happen at the closing price in a simulation and worse in reality — especially on the fast moves trend rules depend on.
• Funding. On perpetuals, every day held has a cost no simple backtest counts.
• Liquidity. The test assumes your size fills instantly. Run it through the exit test to see whether that is true.
• You. The simulation follows the rule perfectly through a 40% drawdown. The person will hesitate, skip trades and intervene — and that gap is usually larger than every other cost combined.

From backtest to real money

1. Trade it small first. A quarter of your normal size for twenty trades.
2. Journal every one, including whether you followed the rule exactly.
3. Compare live results with the backtest. The difference is your execution cost — slippage, hesitation and the trades you talked yourself out of.
4. Only then size up, and only if the live sample resembles the test.

Most strategies fail at step three, and that is the point of doing it. Finding out with a quarter position costs a fraction of finding out with a full one.

Reading the four numbers together

Profit factor below 1 means the rule lost money before slippage. Stop there.

Profit factor above 1.3 with a shallow drawdown is the profile worth pursuing — modest returns you can actually hold through beat spectacular ones you cannot.

High win rate with negative expectancy means many small wins and a few enormous losses. This profile feels wonderful and ends accounts.

Low win rate with strong expectancy is the trend-following signature: frequent small losses and occasional large wins. It is mathematically sound and psychologically brutal, which is why most people abandon it.

Common questions

How much history do I need?

Enough to include different market conditions — at minimum a trending period and a ranging one. A thousand candles on the daily timeframe covers several years and multiple regimes; a thousand five-minute candles covers three days and proves nothing.

Is a 60% win rate good?

Unanswerable alone. At a 1:1 reward ratio it is decent; at 0.5:1 it loses money. Win rate is only meaningful next to the reward ratio, which is why the backtester reports profit factor and expectancy alongside it.

Why does my live trading underperform the backtest?

Three usual causes, in order: slippage on entries and exits, funding on held positions, and deviation from the rule. The journal isolates the third, which is almost always the largest.