A backtest is supposed to tell you whether a strategy works. Too often it tells you what you want to hear. Here are the four traps we see most often, and a simple check for each.
All numbers on this page are fictional and used for illustration only.
1. No fees, no funding, no slippage
A backtest that ignores costs trades in a world that does not exist. In real markets you pay:
- Fees on every entry and exit.
- Funding on perpetual futures, every few hours, for as long as the position is open.
- Slippage and spread: the price you get is rarely the price on the chart, especially on stops and in fast markets.
Small edges are the most fragile. Imagine a strategy averaging +0.30R per trade before costs. If costs eat 0.25R per trade, what is left is +0.05R. That is a strategy with almost no margin for error.
Check: run the same backtest with realistic fees, funding and a pessimistic slippage. If the result flips, you never had an edge.
2. Too few trades
Twenty trades cannot tell luck from edge. A few big winners can make any rule look brilliant over a short sample.
There is no magic number, but orders of magnitude help: 20 trades could be luck, 100 trades is a first hint, several hundred trades across different market phases (trends, ranges, calm and volatile periods) is something you can start working with.
Check: split your results by period and by market regime. If all the profit comes from one month or one rally, you are looking at a story, not a strategy.
3. Overfitting
Overfitting happens when you tune settings until the past looks perfect. The strategy ends up memorising history instead of capturing something that repeats.
The classic symptom is fragility. Say your best setting is a 20-candle lookback. At 18 or 22, the result collapses into a loss. That edge was the tuning. A robust strategy degrades gently when you move a setting a little.
Check: move each setting by about 10% in both directions. Then test on a period you kept aside and never looked at while designing (out-of-sample), or roll the test forward through time (walk-forward).
4. Look-ahead bias
Look-ahead bias means the backtest uses information you could not have had at the time. The most common case: a rule based on a candle's close, but the backtest enters at that candle's open. If the signal is only known at 15:00, the earliest real entry is 15:00.
Other forms: using data that was revised later, or testing only on assets that still exist today (survivorship bias).
Check: for each rule, write down the exact moment its information becomes available. No order should happen before that moment.
The 4-point checklist
- Costs included: fees, funding, slippage.
- Enough trades, across several market phases.
- Stable when settings move, and confirmed on unseen data.
- No information from the future.
A strategy that passes all four is not guaranteed to make money. It is simply one you can start to trust.
How Sunia approaches it
Sunia's principle is "proven, or nothing": fees included, no look-ahead, and validation on data the strategy has never seen. Results show the drawdown, not just the curve, and every simulated trade can be explained by the rule that triggered it.