1. Overlooking Transaction Costs and Slippage
A backtest that ignores commission fees, exchange charges, and bid-ask spread is not a simulation—it is a fantasy. Transaction costs compound with trade frequency, and strategies that appear profitable on paper often collapse once realistic frictions are applied. Slippage is equally destructive: in fast markets, your assumed entry price may never fill. A scalping system showing a 60% win rate can turn net-negative when each round trip costs 0.1% and the average edge per trade is only 0.08%. Always model maker/taker fees, financing rates for leveraged positions, and a conservative slippage assumption based on historical order book depth, not optimistic midpoint fills.

2. Survivorship Bias in Asset Universes
Testing a strategy on today’s S&P 500 constituents is a classic flaw. Companies that went bankrupt, were delisted, or merged out of existence are absent from the dataset, creating an artificially strong universe of survivors. This bias inflates returns by 1–4% annually in equity strategies, according to multiple academic studies. The correction is straightforward: use point-in-time databases that include delisted securities, or at minimum reconstruct historical index membership. Without this, you are backtesting a world where every company succeeded—a world that never existed.

3. Look-Ahead Bias and Data Leakage
Look-ahead bias occurs when your model uses information that was not available at the time of the trading decision. A common example: calculating a moving average using the closing price of the same bar you intend to trade on. Another: using financial statement data on the day the quarter ended rather than the day it was actually reported. Data leakage is subtler—normalizing features across the entire dataset before splitting into training and test sets, or using future volatility to size past positions. Rigorous backtests enforce a strict time barrier: at timestamp T, only data with timestamp ≤ T is accessible, and fundamental data must be lagged by its real-world reporting delay.

4. Overfitting Through Excessive Parameter Tuning
The more parameters you optimize, the more you curve-fit noise. A strategy with 12 adjustable thresholds tested across 5,000 parameter combinations will almost certainly find a configuration that looks brilliant historically and fails live. This is the “backtest overfitting” problem quantified by Bailey, Borwein, and López de Prado: the expected maximum Sharpe ratio from random data grows with the number of trials. Defenses include limiting degrees of freedom, using walk-forward optimization instead of a single in-sample fit, and applying the deflated Sharpe ratio to adjust for multiple testing. If your strategy needs 47 rules to work, it does not work.

5. Ignoring Market Regime Changes
A backtest spanning 2010–2021 may capture only low-volatility, rising-rate environments. A strategy calibrated on that period can disintegrate during 2008-style liquidity crises or 2022-style regime shifts. Markets alternate between trending, mean-reverting, high-volatility, and low-volatility states. Failing to segment performance by regime—or worse, assuming stationarity—produces fragile systems. Test across multiple interest rate cycles, bull and bear markets, and volatility spikes (VIX > 40). If the strategy only profits in one regime, it is a bet on that regime persisting, not a robust edge.

6. Incorrect Position Sizing and Compounding Assumptions
Many backtests assume fixed dollar position sizes, then report percentage returns as if capital compounded. This mismatch distorts drawdowns and risk-adjusted metrics. Others use full Kelly sizing without accounting for estimation error, leading to simulated returns that are mathematically impossible to achieve live. Position sizing must reflect actual account equity at each trade, margin requirements, and the sequence of wins and losses. A strategy with a 55% win rate and 1.5 profit factor can still go bankrupt if bet sizing is too aggressive and losses cluster early. Always simulate equity-curve-dependent sizing, not static allocations.

7. Neglecting Liquidity Constraints
A backtest that assumes you can buy 50,000 shares of a small-cap stock at the closing price ignores market impact. Real orders move prices, especially in illiquid names. If your strategy trades more than 1% of average daily volume, your fills will be worse than modeled. High-turnover strategies are particularly vulnerable: a mean-reversion system trading 20 times per day in mid-cap equities may show 30% annual returns before liquidity adjustment and 5% after. Incorporate volume-based slippage models—such as the square-root law of market impact—and cap position sizes relative to ADV.

8. Using the Wrong Benchmark and Risk-Free Rate
Comparing a long-only equity strategy to a 0% risk-free rate during a period when T-bills yielded 5% is misleading. Similarly, benchmarking a market-neutral strategy to the S&P 500 conflates alpha with beta exposure. Every backtest must state its opportunity cost: the return available from a comparable-risk passive alternative. Use the actual historical risk-free rate for Sharpe calculations, and choose a benchmark that matches the strategy’s universe and leverage. Otherwise, you are measuring skill against an irrelevant yardstick.

9. Cherry-Picking Start and End Dates
Beginning a backtest in 2009 and ending in 2021 flatters almost any long-biased strategy. Trimming the first six months of losses or ending before a major drawdown is a form of data dredging. The fix is to use the longest available clean data series, or to pre-register the test period before running the simulation. If you would not have chosen those exact dates in advance, the result is not evidence—it is a narrative. Walk-forward analysis with rolling windows reduces this bias by forcing the strategy to perform across all sub-periods.

10. Confusing Correlation with Causality in Signals
A backtest may show that when copper rises and the VIX falls, your strategy wins 70% of the time. But if both variables are driven by a third factor—say, Fed liquidity—the signal is spurious. Overlapping data, autocorrelation, and non-stationary relationships create false predictive power. Techniques like Granger causality tests, cointegration analysis, and out-of-sample cross-validation help separate genuine edges from coincidental patterns. A signal that only works when tested on one asset or one time frame is likely noise dressed as insight.

11. Failing to Account for Short-Selling Constraints
Short backtests often assume you can borrow any stock at any time at zero cost. In reality, hard-to-borrow names carry fees of 5–50% annually, and some equities are simply unavailable. During short squeezes, borrow costs spike or recalls force premature covering. A long-short strategy showing 15% annual returns may lose half its edge to borrow fees. Model realistic short availability using historical borrow data, and stress-test for periods when shorting becomes expensive or prohibited.

12. Ignoring Dividends, Splits, and Corporate Actions
Price-only backtests miss dividend income (or payments on shorts) and mishandle stock splits, mergers, and spinoffs. A strategy holding high-dividend equities for months can understate returns by 2–4% annually if dividends are excluded. Splits unadjusted in the price series create artificial gaps that trigger false signals. Always use total-return adjusted data, verify corporate action handling in your data source, and confirm that dividend withholding taxes (for international stocks) are modeled correctly.

13. Using Too Few Trades for Statistical Significance
A backtest with 30 trades cannot distinguish skill from luck. The standard error of a win rate estimate with 30 observations is roughly 9 percentage points. You need hundreds of trades—ideally 500+—to achieve reasonable confidence. For lower-frequency strategies, extend the backtest across decades or multiple instruments. Reporting a Sharpe ratio on 20 monthly returns is statistically meaningless. Always report the number of independent bets, the t-statistic of returns, and the confidence interval around key metrics.

14. Neglecting the Impact of Taxes
For taxable accounts, short-term capital gains are taxed at ordinary income rates—up to 37% in the U.S.—while long-term gains max at 20%. A high-frequency strategy that looks stellar pre-tax may be mediocre after-tax. Wash sale rules disallow loss deductions when repurchasing the same security within 30 days. Backtests that ignore taxes overstate net returns by 5–15% for active strategies. Either model after-tax returns explicitly or clearly state that results are pre-tax and suitable only for tax-advantaged accounts.

15. Assuming Perfect Execution and Immediate Fills
A limit order backtest that assumes every order fills at the specified price is optimistic. In reality, limit orders fill only when the market trades through them, and queue position matters. Market orders fill at the next available price, which may be far from the last print. Stop orders become market orders when triggered, often filling at worse prices during gaps. Simulate order-level execution using tick data or at minimum bar-level assumptions with conservative fill logic. A strategy that relies on capturing the exact high or low of a bar is untradeable.

16. Ignoring Psychological and Operational Frictions
A backtest has no fear, no margin calls, and no internet outages. Live trading introduces hesitation, forced liquidation during drawdowns, and execution delays. A strategy with a 40% maximum drawdown may be mathematically sound but psychologically impossible to follow. Operational frictions—API rate limits, broker downtime, data feed errors—cause real losses that backtests never capture. Stress-test the equity curve against human tolerance: if you would abandon the system after six consecutive losses, the backtest’s return is irrelevant.

17. Using Inconsistent or Low-Quality Data
Bad data produces bad backtests. Common issues include missing bars, incorrect timestamps, unadjusted prices, and stale quotes. A single erroneous price spike can trigger a cascade of false signals. Before backtesting, audit your data: check for gaps, outliers, and timezone alignment. Cross-validate against a second source. For intraday strategies, verify that your data includes pre-market and after-hours sessions if your strategy trades them. Garbage in, garbage out—and in trading, garbage out means real money lost.

18. Ignoring Portfolio-Level Interactions
Testing each strategy in isolation ignores correlation between positions. Two strategies may each show a 1.5 Sharpe ratio, but if they both go long tech during the same regime, the combined portfolio is far riskier than either alone. Backtest at the portfolio level: aggregate signals, net exposures, and margin usage across all strategies. Calculate portfolio drawdown, value-at-risk, and correlation matrices. A collection of individually profitable strategies can produce a combined equity curve that is worse than any single component due to unintended factor overlaps.

19. Failing to Perform Out-of-Sample and Walk-Forward Testing
In-sample optimization is necessary but insufficient. The only meaningful test is out-of-sample performance on data the strategy has never seen. Walk-forward analysis—rolling optimization windows followed by test windows—approximates live trading more closely. A strategy that passes in-sample but fails out-of-sample is overfit. A strategy that passes both but fails on a third holdout period is fragile. Use at least three disjoint time periods: training, validation, and final test. Never touch the final test set until the strategy is frozen.

20. Misunderstanding Drawdown and Recovery Mathematics
A 50% drawdown requires a 100% gain to break even. A 75% drawdown requires a 300% gain. Backtests that report average returns without maximum drawdown and recovery time hide the true risk. Calculate the longest drawdown duration, the ulcer index, and the pain index. A strategy with a 20% annual return and a 60% drawdown is worse for most traders than one with 12% return and a 15% drawdown. Risk-adjusted metrics like the Calmar ratio (return / max drawdown) and the Sterling ratio reveal what raw returns obscure.

21. Overreliance on a Single Metric
Optimizing solely for Sharpe ratio, profit factor, or win rate creates blind spots. A high win rate with tiny wins and huge losses is a losing strategy. A high Sharpe ratio with negative skewness hides tail risk. A high profit factor with few trades is statistically unreliable. Evaluate multiple dimensions: expectancy, payoff ratio, maximum adverse excursion, time in market, and consistency across sub-periods. No single number captures strategy quality. Use a dashboard of metrics and reject any strategy that looks good on only one axis.

22. Ignoring the Cost of Capital and Opportunity Cost
Capital tied up in a backtested strategy cannot be used elsewhere. If your strategy returns 8% annually while a risk-free bond yields 5%, the excess return is only 3%—and that is before taxes and effort. For leveraged strategies, the cost of borrowing (margin interest) must be subtracted. For long-only strategies, the opportunity cost of not holding a passive index matters. Always compare net returns to a passive benchmark with similar risk. A strategy that beats cash but trails the S&P 500 after costs is not adding value.

23. Using Static Parameters in a Dynamic Market
A 20-period moving average worked in 2015 but may fail in 2025. Markets evolve: tick sizes change, algorithms adapt, and volatility regimes shift. Backtests that fix parameters for a decade assume the market’s statistical properties are constant. Adaptive strategies—those that adjust lookback windows, thresholds, or position sizes based on recent volatility—tend to be more robust. Alternatively, re-optimize periodically using walk-forward analysis. Static parameters are a hidden bet that nothing changes, which is almost always wrong.

24. Neglecting the Sequence of Returns
Two strategies with identical average returns can have wildly different outcomes depending on the order of wins and losses. A strategy that loses 10% in year one and gains 30% in year two ends with a different compounded return than one that gains 30% then loses 10%. For leveraged or path-dependent strategies, sequence risk is amplified. Monte Carlo simulations that shuffle trade order reveal the range of possible equity curves. A backtest that reports only the final compounded return hides the sequence risk that can force liquidation mid-path.

25. Failing to Document and Reproduce Results
A backtest that cannot be reproduced is not evidence. Without version-controlled code, documented data sources, and explicit parameter logs, you cannot verify whether a result is real or a bug. Reproducibility also enables debugging: when live results diverge from backtest, you need to trace every assumption. Use containerized environments, seed random number generators, and store raw data snapshots. If you cannot hand your backtest to a colleague and have them reproduce the same equity curve, the result is meaningless.