Why Most Backtests Are Misleading
A backtest that shows 90% win rate and 500% annual return sounds incredible, until you trade it live and lose money. The gap between backtest fantasy and live reality is almost always caused by one or more of these five common mistakes.
What makes these mistakes dangerous is that none of them throws an error. Your backtest runs cleanly, the equity curve climbs, and the metrics look professional. The flaw is invisible until real capital is on the line, which is exactly why so many traders fund an account on the strength of results that were never achievable. Understanding these pitfalls is essential before you trust any backtest result with real money, and before you confuse a strong backtest with a strong strategy. The difference between the two is the subject of backtesting vs forward testing.
How we identified these mistakes
Reviewed by Matthieu DAVID, proprietary trader since 2020, FTMO-funded, founder of Backtrex. Last updated June 2, 2026.
These five mistakes are the ones we see most often when reviewing community strategies and our own early backtests. We track them in every audit we run on Backtrex and confirm each one with side-by-side reruns: same strategy, mistake removed, watching how the equity curve changes.
Mistake 1: Overfitting (Curve Fitting)
Warning signs:
- Your strategy has 10+ parameters
- Small parameter changes dramatically alter results
- The strategy works on one specific date range but fails on others
Here is what overfitting looks like in practice. A trader builds a moving-average crossover, then adds an RSI filter to cut the losing trades, then a volatility band to cut more, then a time-of-day rule, then a specific stop distance tuned to the cent. Each addition makes the historical curve smoother. By the tenth rule the backtest is almost a straight line up, because the rules have been reverse-engineered from the exact trades that happened to lose. None of them describe the market. They describe the past. Run the same configuration on the following year and the edge evaporates, because next year's noise is different from last year's noise.
The Overfitting Trap
A strategy with 20 parameters can be tuned to show profits on virtually ANY historical dataset. This does not mean it has an edge. In-sample optimization without out-of-sample validation is the number one cause of blown accounts.
How to detect it: The cleanest test is the most uncomfortable one. Take the exact strategy you are proud of and run it on a period you never looked at during development. If a strategy that returned, say, 180% in-sample drops to 12% (or goes negative) out-of-sample, you did not find an edge, you found a curve fit. A genuine edge degrades gracefully on unseen data; an overfit one collapses.
How to avoid it: Use walk-forward analysis. Split your data into training (70%) and testing (30%) periods. Only trust strategies that perform consistently on unseen data, and prefer fewer parameters to more. A robust strategy with three rules beats a fragile one with fifteen every time it actually matters, which is live.
→ Deep dive: how to detect and prevent overfitting
Mistake 2: Look-Ahead Bias
Common sources:
- Using the current bar's close price to trigger an entry on that same bar
- Referencing indicators that use data not yet available at the decision point
- Using daily high/low to set intrabar targets
The most frequent version is also the most innocent-looking: entering a trade at the close of the very candle whose close triggered the signal. In a backtest the engine knows that close, so the fill is perfect. Live, you cannot act on a close until the candle has finished, by which point the price has moved on. That single timing error can turn a losing strategy into a winning backtest, because you are systematically buying at prices that were only knowable after the fact.
How to detect it: Shift every signal one bar into the future and rerun. If the strategy still works on close[1] logic, it has a real edge. If the performance falls apart the moment you stop using the current bar, the backtest was reading the future.
How to avoid it: Always use close[1] (the previous bar's data) for decision logic. Reputable backtesting platforms enforce this automatically through anti-repainting safeguards, so the bias becomes impossible rather than something you have to remember.
Mistake 3: Repainting Indicators
Notorious repainters:
- Zigzag indicator
- Some implementations of pivot points
- Certain smoothed oscillators with future lookback
Repainting is look-ahead bias hiding inside an indicator. The Zigzag is the classic example: it only confirms a swing point once price has moved far enough in the other direction, then it redraws the line back to where the real high or low was. On a chart it looks like it caught every top and bottom perfectly. In reality the signal did not exist at the moment you would have needed it. A backtest built on Zigzag entries can show a flawless equity curve that is physically impossible to trade.
How to Detect Repainting
Run your strategy on historical data, note the signals. Then wait for new bars to form and check if those signals moved. If they did, your indicator repaints and your backtest is unreliable.
How to avoid it: Use only confirmed, non-repainting indicators. Tools with built-in anti-repainting protection flag or block repainting logic before you run the test, which means you never ship a strategy that depends on signals that were never really there.
Mistake 4: Ignoring Transaction Costs
A strategy that makes 2 pips per trade sounds profitable, until you account for 1.5 pips of spread plus commission. Many strategies that look good in a zero-cost backtest become losers when realistic costs are applied.
Costs to include:
- Spread (bid-ask difference)
- Commission per trade
- Slippage (especially on larger positions or illiquid assets)
- Swap/overnight fees for positions held past rollover
The arithmetic is brutal for high-frequency strategies. Imagine a scalping system averaging 2 pips of gross profit per trade across 1,000 trades. On paper that is 2,000 pips. Now subtract a typical 1.5-pip spread and roughly 0.3 pips of commission and slippage per trade, and your net edge per trade is 0.2 pips, a tenth of what the backtest promised, and one bad week of widening spreads wipes it out entirely. The lower your per-trade edge, the more costs decide whether you are profitable, and the more a cost-free backtest lies to you.
How to avoid it: Always configure realistic transaction costs in your backtest settings. Test with slightly higher costs than expected, if the strategy still profits, it has a buffer for the days when liquidity dries up and spreads widen.
Mistake 5: Survivorship Bias
Testing only on assets that still exist and trade today ignores all the assets that were delisted, went bankrupt, or became illiquid. This artificially inflates your strategy's performance, because the very act of "still being tradeable in 2026" is itself a filter for success.
The effect is largest in equities and crypto. Backtest a momentum strategy on the current top 100 coins and you are implicitly betting on assets that already survived; the hundreds of tokens that went to zero never enter the test, so your drawdowns look impossibly mild. Forex is less exposed, the major pairs do not get delisted, but it still bites anyone testing exotic pairs or assuming a broker will always quote them.
How to avoid it: Include delisted assets in your data when possible. Be skeptical of strategies that only work on a handful of winning assets. Test across multiple assets and timeframes to validate stability, because an edge that only appears on the survivors is not an edge, it is hindsight.
How These Mistakes Compound
These five mistakes rarely show up alone, and that is what makes them so destructive. An overfit strategy that also enters on the current bar's close and uses a repainting indicator does not just exaggerate its returns a little, it multiplies three illusions together. Each layer adds plausible-looking performance, and together they produce a backtest so far from reality that the live result feels like a different strategy entirely. This is why traders are routinely shocked when a "90% win rate" system loses money in week one. They are not unlucky, they are seeing the sum of several invisible errors collected at once.
The practical takeaway is that fixing one mistake is not enough. You have to assume all five may be present and rule them out one by one, which is precisely what a disciplined process, or a platform that enforces it, is for.
The Right Way to Backtest
Keep It Simple
Use Out-of-Sample Data
Include Realistic Costs
Test Across Multiple Assets
Use Anti-Repainting Tools
→ New to the process? Start with our step-by-step guide to backtesting a strategy.
Backtesting is powerful, but only when done correctly. Avoid these five mistakes and you will be ahead of 90% of retail traders who trust flawed results. If you want an objective benchmark, the Backtrex Score grades every strategy from 0 to 100 across all assets, so an overfit result that shines on one market gets exposed when it ranks low everywhere else. Once your in-sample numbers survive that scrutiny, the next step is to confirm them forward, which is where the difference between backtesting and forward testing becomes decisive.
Frequently Asked Questions
Run your strategy on data it has never seen (out-of-sample). If performance drops significantly compared to the training period, you are overfit. Also check: does the strategy have more than 5-7 parameters? Do small changes in parameters cause big swings in results? Both are overfitting red flags.
Aim for 200+ trades minimum. Under 100 trades, random variance dominates your results. A strategy with 95% win rate on 20 trades proves nothing. The same strategy might have 45% win rate on 500 trades.
Yes. Tools like Backtrex enforce anti-repainting rules at the engine level: all indicators use close[1] (confirmed bar data only). This means repainting is impossible by design, unlike TradingView community scripts where repainting depends on the script author.
Shift every signal forward by one bar and rerun the backtest. A strategy with a genuine edge still performs on previous-bar (close[1]) logic. If the results collapse the moment you stop using the current bar's close, your backtest was using information that was not available at the decision point.
At minimum: the spread, commission per trade, and an allowance for slippage. For positions held overnight, add swap or financing fees. A good rule is to test with slightly higher costs than your broker quotes, so the strategy has a buffer for the days when spreads widen and liquidity thins out.
Less than equities or crypto, because the major currency pairs are not delisted. It still matters if you test exotic pairs or assume a broker will always quote a given instrument. The bias is most severe when backtesting stocks or tokens, where the assets that failed have quietly disappeared from your dataset.
Not sure what backtesting is or how to get started? Read our complete guide. If you trade SMC/ICT strategies, check out our dedicated guide on backtesting Smart Money setups.
Ready to backtest the right way? Start for free with built-in safeguards against all five mistakes listed above.