Backtesting

Backtesting is the application of specified trading rules to historical market data to estimate how a strategy would have performed in the past. It produces hypothetical results, not a record of actual trades. The purpose is to assess a strategy before committing capital and understand its behavior under the conditions represented in the data.

How Backtesting Works

A test starts with explicit rules for entering and exiting trades, sizing positions, and managing available capital. It also needs a historical period, starting portfolio value, and assumptions about when and at what price orders could execute. These choices define what is being tested; changing them can change the result.

The simulation then moves through historical observations in time order. At each decision point, it uses only information that would have been available then, generates the trades required by the rules, and updates holdings and portfolio value. Backtesting can be done manually or with software; coding is not part of the definition. Explicit rules support systematic comparison, but human choices about rules and data can still bias the test.

In algorithmic trading, this process tests coded entry, exit and position-sizing rules before live deployment. Portfolio tests can examine how strategies or assets interact rather than viewing each in isolation. Risk-management tests can compare stop rules and sizing assumptions through their simulated losses and drawdowns; they cannot establish a safe maximum future loss.

For implementation, MetaTrader 5 provides a Strategy Tester for automated trading programs, while TradingView tests Pine Script strategies using a broker emulator. QuantConnect supports running coded backtests, and AmiBroker provides a backtesting engine with configurable portfolio and execution assumptions. These are examples of ways to implement a test, not interchangeable results or recommendations. Check the available data, rule implementation, cost settings and fill model before comparing outputs.

Interpreting Returns and Drawdowns

Results describe simulated returns, losses, and risk during the selected period. Portfolio value should reflect both realized trade outcomes and changes in the value of open positions. A drawdown is a decline from a previous portfolio-value peak; the largest observed drawdown shows the deepest such decline within that test.

Backtesting can help reveal whether gains were spread across the period or concentrated in a short stretch of favorable conditions. A profitable endpoint alone does not explain the losses experienced along the way. Neither the worst historical loss nor the largest historical drawdown is a ceiling on future losses.

Additional measures answer different questions. A Sharpe ratio relates excess returns over a chosen reference rate to return variability; its interpretation depends on the calculation period and assumptions. Win rate measures the fraction of profitable trades, not the size of wins relative to losses. Average trade duration describes how long positions remain open and helps interpret exposure and execution costs.

Data and Execution Assumptions

The usefulness of backtesting depends on data suited to the strategy's market, time horizon, and decision rules. Missing observations or incorrect prices can distort signals and measured returns. A dataset that includes only assets that survived to the present can also omit failures that would have affected a historical strategy.

Execution assumptions need equal attention. Commissions and bid–ask spreads reduce simulated net returns, while slippage represents a difference between an expected trading price and the price obtained. A simulation must make plausible assumptions about whether orders could fill, how quickly, and for what quantity. Treating every order as an immediate fill at an observed price can overstate performance, particularly when liquidity is limited.

Bias and Limits of Historical Results

Overfitting occurs when rules or parameter choices are adapted to noise in the historical sample rather than a relationship likely to persist. Repeatedly adjusting a strategy until its past returns look attractive can therefore weaken the test's value. Look-ahead bias occurs when a simulated decision uses information that was not yet available, such as a later earnings release or a closing price before that close occurred.

Testing on data withheld from strategy development can help assess whether results extend beyond the development sample, but cannot establish that conclusively. If the withheld results are repeatedly used to revise the strategy, that data no longer provides an independent check. Backtesting remains historical evidence under specified assumptions, not proof of future profitability or robustness. Market conditions can change, and live execution can differ from the simulation even when the trading rules are unchanged.

Scenario analysis changes assumptions or applies specified shocks to examine possible outcomes, rather than simply replaying recorded history. Forward testing applies fixed rules to newly arriving data. Paper trading can implement that test with simulated orders, but does not establish that real orders would receive the same fills. These approaches complement historical testing without proving future profitability.