Simple yet powerful backtesting framework in python/pandas.
21.05.2015 - Author's note: This package has long been abandoned, but now I am planning to move it to version 0.2. Main goals are:
- Smarter handling of possible backtest initialization arguments, such as different ways to specify signals, price data and such.
- Better multi-asset backtest capabilites.
- Order type specification. Currently there are two ways to specify backtest: simply pass signals and get next-bar-market-on-open backtest, or add trade prices and get stop order-like execution, but without proper possibility of trade checking. You will still be able to do these things, with added option to futher customize execution and simulate more production-like environment.
It allows user to specify trading strategies using full power of pandas, at the same time hiding all boring things like manually calculating trades, equity, performance statistics and creating visualizations. Resulting strategy code is usable both in research and production setting.
Strategies could be defined as simple this:
ms = pandas.rolling_mean(ohlc.C, 50)
ml = pandas.rolling_mean(ohlc.C, 100)
buy = cover = (ms > ml) & (ms.shift() < ml.shift())
sell = short = (ms < ml) & (ms.shift() > ml.shift())And then tested like this:
pybacktest.Backtest(locals())
We use it in our research and production operations.
git clone https://github.com/ematvey/pybacktest.git
cd pybacktest
python setup.py install
If you don't install it in virtualenv, you might need to prepend last line with sudo.
Tutorials are provided as ipython notebooks in folder examples. You run it from cloned repo or watch via nbviewer.
Single-security backtester is ready. Multi-security testing could be implemented by running single-sec backtests and then combining equity. Later we will add easier way.