Backtesting a strategy

Backtesting is the process of applying a precisely defined set of trading rules to historical market data to see how the strategy would have performed. Its core purpose is to reject bad ideas cheaply, before they cost real money, and to give a rough, honest estimate of a strategy's statistical properties — win rate, average reward-to-risk, drawdown profile — before it is traded live. This is valuable and important, and any strategy that has not been tested in some form before being used with real or funded capital should be treated as an unverified hypothesis, not a plan.

What backtesting cannot do is guarantee future performance, account for your own execution discipline, or fully capture real-world frictions like slippage, variable spread, and the psychological difficulty of holding a losing position. A backtest that shows an 80% win rate will not protect you from the very real difficulty of sitting through the 20% of trades that lose, particularly if several of them cluster together. Backtesting answers the question 'does this rule set have a statistical edge in this historical data', not the question 'will I successfully execute this strategy'.

The most productive mindset for backtesting is scientific: you are trying to disprove your own idea as efficiently as possible, using the most rigorous test you can construct, rather than trying to find a version of the test that makes the idea look good. This distinction — testing to find the truth versus testing to confirm a belief — is what separates useful backtesting from a waste of time.

Backtesting requires exactly the same precision discussed in the trading-plan lesson: every entry condition, exit condition, and sizing rule must be specific enough to apply mechanically to a chart without judgement calls. This is often the point where a trader discovers that their 'strategy' was never actually fully defined — clauses like 'if it looks strong' or 'use discretion near support' cannot be backtested at all, because they require a human judgement that a backtest cannot replicate.

The discipline of writing rules precise enough to backtest is valuable even independent of the backtest result, because it forces the same clarity that a good trading plan requires. If you cannot define your strategy precisely enough to test it, you also cannot define it precisely enough to execute consistently under pressure, and this alone is often the single biggest realisation a new trader has when they attempt their first serious backtest.

Look-ahead bias occurs when a backtest accidentally uses information that would not have been available at the time of the trade — for example, using a daily candle's closing price to decide an entry that would supposedly have happened during that same day, before the close was known. This inflates results artificially and is one of the most common errors in manually constructed backtests, particularly when working from a static chart rather than a bar-by-bar replay.

Overfitting occurs when rules are adjusted repeatedly to maximise performance on a specific historical dataset, adding more and more specific conditions until the strategy performs beautifully on that data but has effectively memorised its noise rather than captured a real, repeatable pattern. A telltale sign of overfitting is a strategy with many highly specific numerical parameters (for example, a moving average of exactly 37 periods combined with an RSI threshold of exactly 61.5) that were tuned to fit one particular data set; such parameters rarely hold up on new, unseen data. The correct discipline is to test on one period of data, fix the rules, and then validate on a separate, later period the rules were never adjusted against — commonly called out-of-sample testing.

Survivorship bias is less relevant to short-term intraday forex or index strategies but matters greatly for anyone backtesting on individual stocks or baskets of instruments over long periods, since a dataset that only includes currently-listed companies silently excludes the companies that failed or were delisted, inflating historical returns. Even in forex, a related issue arises if a strategy is only tested on the small set of pairs that happened to work well over the test period, rather than tested consistently across a pre-decided universe of instruments.

A backtest with 15 trades tells you almost nothing reliable, no matter how good the results look, because the sample is too small to distinguish a real edge from luck. As a rough guide, most practitioners want at least 100 trades, and ideally several hundred, spanning multiple different market conditions (trending, ranging, high volatility, low volatility) before placing meaningful confidence in the statistics. A strategy that performed brilliantly during a single strong trending month but has never been tested during a ranging period has not really been tested at all.

It is also worth explicitly calculating the strategy's expectancy — the average result per trade, in units of risk, calculated as (win rate × average win) minus (loss rate × average loss). A strategy can have a low win rate and still be strongly profitable if its average win is large relative to its average loss, and conversely a high win rate can still be unprofitable if occasional losses are large relative to typical wins. Looking at win rate alone, without expectancy, is one of the most common misreadings of a backtest.

A backtest's equity curve — the running total of account value over the test period — tells you more than the final return figure alone. A smooth, steadily rising curve suggests a robust edge; a curve with one enormous winning trade responsible for most of the total return is fragile, because removing or missing that single trade (which is entirely plausible in live trading, given slippage or a missed entry) would change the entire performance picture. Always check what the results look like with the single best trade excluded.

Maximum drawdown — the largest peak-to-trough decline in the equity curve — deserves as much attention as total return, because it tells you what psychological and account-survival pressure the strategy would have put you under during its worst historical period. A strategy with an attractive total return but a 35% historical maximum drawdown is not compatible with most funded account rules and would also be extremely difficult to sit through emotionally, even if eventually profitable. The platform's /backtesting tool reports both total return and maximum drawdown together for exactly this reason — never evaluate one without the other.