Backtesting is an essential tool for evaluating trading strategies, but historical performance can be dangerously misleading. A strategy that looks highly successful on past data may simply be benefiting from biases, unrealistic assumptions, or overfitting. This webinar explores how to use backtesting as a tool for estimation rather than proof, and how to distinguish a potentially durable signal from historical noise.
Speaker
Grigory Chikishev is a Team Lead and Quantitative Trader at Quantum Brains with over 9 years of experience.
What Backtesting Can and Cannot Tell You
Backtesting simulates how a trading strategy would have performed under historical conditions. It can help estimate potential returns, risks, drawdowns, and behavior across different market environments, but it cannot prove that a strategy will work in the future. A strong historical result should therefore be treated as a hypothesis rather than evidence of future performance.
Where Backtests Go Wrong
There are many ways for a backtest to produce misleading results. Look-ahead bias, incorrect timestamps, survivorship bias, unrealistic execution latency, and ignoring fees, funding, or slippage can all make a strategy appear more profitable than it would be in real trading.
These issues are particularly dangerous because the resulting performance can look convincing while relying on assumptions that could never hold in live markets.
The Overfitting Trap
One of the biggest risks is overfitting: creating a strategy that has learned historical noise rather than a durable market signal. Overfit strategies can demonstrate excellent in-sample performance while performing poorly on unseen data. The more parameters, variations, and hypotheses are tested, the easier it becomes to find a strategy that appears successful purely by chance.
To reduce this risk, strategies need to be validated on data that was not used during development. Time-aware approaches such as walk-forward testing, cross-validation, and combinatorial purged cross-validation provide different ways to test whether performance remains robust outside the original sample.
From the Best Result to a Reliable Estimate
The purpose of backtesting should not be to find the strategy with the highest historical return. Instead, results should be used for calibration and estimation: understanding expected return and risk, drawdown and recovery behavior, turnover and transaction costs, parameter sensitivity, and performance across different market regimes.
A robust strategy is not necessarily the one with the most impressive backtest. It is the one whose behavior remains reasonably stable under different assumptions and unseen conditions. The goal is therefore not to optimize for the best historical result, but to build a more reliable estimate of future risk and performance.
Conclusion
Backtesting is most useful when treated as a framework for testing hypotheses rather than proving them. Reliable evaluation requires controlling for data and execution biases, limiting overfitting, validating against unseen periods, and focusing on robustness and uncertainty.
The key lesson is simple: a backtest should not tell you that a strategy will work. It should help you understand how plausible its performance is, where the risks lie, and how much confidence you can reasonably place in the result.