How to backtest a trading strategy in Python

How do you backtest a trading strategy in Python?

To backtest a trading strategy in Python, you need three parts: historical price data, the strategy rules as code, and a loop that feeds the data to the rules in time order while a simulated broker fills the orders. The output is a trade list, an equity curve and statistics such as drawdown and profit factor.

You can write that loop yourself, use one of the common open source Python libraries, or use a backtesting engine that brings the data and the broker model with it. The rules are the easy part. The data, the fills and the costs decide whether the result is honest.

Should you write your own Python backtester?

Writing your own Python backtester is a good way to learn how backtests work, but a trustworthy one takes more than a for loop. Every part below has to be built, tested and kept correct, and each one can quietly flatter the result when it is wrong.

What are the steps of a Python backtest?

A Python backtest follows six steps, whether it runs in your own script or in an engine:

1. Write the rules as code: entry, exit, stop, target and position size, each based only on data up to the current bar.
2. Load the data for the market and period, in the resolution the strategy needs. Tight stops need minute bars or ticks.
3. Set the costs: spread, commission per lot and side, and overnight swap at your broker's values.
4. Run the backtest and read the trade list, not only the final profit.
5. Optimize carefully and keep part of the data out of the optimization.
6. Test robustness with walk forward windows and a Monte Carlo reshuffle of the trades.

What goes wrong in Python backtests?

Python backtests most often go wrong through lookahead bias, missing costs and overfitting. Python makes it easy to compute a whole column of signals at once, and the same convenience lets a future value slip into today's decision without any error message.

• Unshifted signals: a signal computed on the full price series and not shifted by one bar trades on a close it could not have known. See overfitting and lookahead bias.
• Fills at the close: when the stop and the target both lie inside one bar, bar data cannot say which came first. Tick data settles it.
• No costs: a strategy with many small trades can turn from profit to loss on spread and commission alone. See spread, commission and swap.
• Too many parameters: values tuned until they fit one period. See strategy optimization and walk forward analysis.

Which trading platforms run Python strategies?

cTrader runs Python cBots, indicators and plugins since version 5.4. MetaTrader 5 offers Python for data and trading, while its strategy tester runs MQL5. TradingView uses Pine Script only. The Orion Quant Technologies Engine takes a strategy written as a Python class and validates it before the run.

A Python strategy that compiles to the engine's strategy format can be exported as a cTrader cBot, an MT5 Expert Advisor (.mq5) or a Pine script, with a list of every construct that was translated. The export is exact or it refuses, never a guess. See the MT5 strategy tester alternative and cTrader backtesting.

How do you backtest Python strategies in the Orion Quant Technologies Engine?

In the Orion Quant Technologies Engine a strategy is a Python class. Validate parses it before the run and lists every import the sandbox will refuse, so nothing fails halfway through a test. You can write it in the app, or open it in VS Code or PyCharm and keep writing there: the app picks up the file when you save.

The run uses the built in archive or your own CSV bars, with the costs and the fill tier as settings and no lookahead by construction. Afterward you can optimize with grid or genetic search, check the result with walk forward and Monte Carlo, and run the prop firm check. Plugins for data sources, indicators and exporters are small Python modules as well, see the plugin library. The engine is in early access through the waitlist.

Frequently asked questions