Forex Backtesting Software

What backtesting software does

Forex backtesting software takes a set of rules and a file of past prices, steps through the prices, and records each trade the rules would have made. It then summarises them. Some tools are built into trading platforms, some are stand-alone, and some are code libraries you drive yourself. The method is the same in each, and is covered in how to backtest. This page is about judging the tool.

Any tool can run. The question is whether its answer can be believed. Four checks cover most of it.

The four checks

1. Data quality

• Source and time zone. Is it written down? Different feeds disagree by small amounts, and day boundaries move with the time zone.
• Detail. Ticks or bars? A bot that acts inside a bar needs detail the bar does not hold. See the strategy tester page.
• Gaps and bad ticks. A single wrong price can create a fake winning trade. Good tools show gaps or let you inspect them.
• Bid and ask. Does the data hold both sides, or one price with a fixed spread added?

2. A cost model

The tool should let you set the spread, commission, slippage and overnight charges, and let you change them for a stress test. A tool with no cost inputs reports a number that no account will see. Costs vary by broker and account type, so use your own.

3. No look-ahead

The tool must feed rules only what was known at the time. Ask: at which price does a signal on a bar fill? Is an indicator calculated with the unfinished bar? Can a rule reach a future bar by accident? See look-ahead bias. Many errors live in the user's code, not the tool, so the delay test below is worth running whatever the tool.

4. An export of every trade

The tool should give you the trade list with entry time, exit time, prices, size and costs. Without it you cannot recompute the headline, filter the result by year or by hour, or hand the trades to a Monte Carlo simulator.

A worked example: recompute the export

A tool prints a headline of +1.90 percent for a strategy. You export the 8 trades. Every number below is made up, in percent of the traded amount. The tool had no cost input, so you subtract a cost of 0.12 percent per trade yourself. The last column is the same rules run again with every signal delayed by one bar, also after costs. Everything was calculated by us on 29 Oct 2026 and is exact.

• Tool headline: +1.90 percent, the sum of the first column.
• Recomputed with costs: +0.94 percent. Eight trades at 0.12 take 0.96 points.
• With signals one bar later: -2.26 percent. Most of the result depends on acting on the very next price.

Two readings are possible. Either the strategy is quick and the live fill really will be hard to get, or the original run used a price it could not have had. The test cannot tell which. It tells you to look at how the tool fills signals before trusting it. Eight trades is far too few to conclude anything, and is used only to show the steps.

Other things worth checking

• Repeatability. The same inputs and data should give the same trades every run.
• Out-of-sample support. Can you hold part of the data back, or run walk-forward testing? Without it, tuning drifts into overfitting.
• Position sizing and margin. Does the tool model margin calls, or does it let a losing grid run forever? For grid and martingale bots, this decides whether the test means anything.
• Documentation. Does it say how it fills orders and handles the first bars of an indicator?

What no tool can tell you

Software cannot supply data it was never given, or events that did not happen in the sample. A clean report is a model result. It is followed by forward testing, not by trust.