A very barebones backtesting library for python
Should be considered beta at the moment. As I use it more and I feel more confident about its stability that will change. But for now I may make
changes to how it works. ====
This is a module used to replay historical data and place limit/stop orders.
You can use contingent orders that wont become active until a parent order
is filled. For example if you have a buy order and want stop loss/take profit orders associated with them that only become active when the initial
buy order is filled. It's pretty basic, and I may add to it as time passes. It's biggest assumption is that your data is OK. It does no verification about the
sanity of your data, nor does it try to sync up time series if you are
testing multiple input sources. This is quite difficult in practise and
requires a bunch of assumptions about what is correct behaviour. So I side step the issue by ignoring it, with the assumption you will fix your
data as you see fit.
To use, subclass the BackTest object, and write your strategy logic in
bar_close which is called for each bar of your data.
You need to load up your data and tell backtest about it with bt.add_input,
which takes the symbol (e.g. 'SPY'), an object that will return one period
of data with each readline(), and the object used to convert the text data
to an OHLC bar. Once you have told bt about your data, and set up your bar_close, call run() to kick things off.
There is a built in data to bar convertor for Yahoo Finance data. You will
need to reverse the data (see revfile.py).
For an example see maeg.py which implements a 200 period moving average crossover system. run: python maeg.py SPY.csv Limit and Stop orders are filled which the bars range contains the limit or
stop level. Market orders are filled immediately, at whatever level you specified when the order was placed. If you want to use market orders, in
bar_close create the order with the close of the current bar (b.cl), and it will be filled and turn into an open position on the next bar.
You can use Order.verify to make sure your limit/stop orders make sense. Sell orders are always for a negative quantity, buy orders are always for
a positive quantity. Occasionally, you might have 2 orders that cancel out one another. If they relate to some parent order, that parent order will be unwound as if
it was never executed. This is rare but usually happens when a bar hits
both a stop loss and take profit, if they are a very close together. If it happens often, you need to change your strategy or use lower time frame data. Check bt.poslist.rewinded for any rewound positions. Prices should be of type decimal.Decimal
Once it's done you can bt.eqvals will have the equity curve. You can also
look through all your positions and orders as well.
====
If you actually find use in this please let me know and I will be more
dilligent about improving it, adding docs etc. For now take a look at the
example strategy and the unit tests to get a feel for how the parts all
work. I wrote this as something that is "good enough" for my own research, and
am making it available if you want to verify something I write about or
have your ideas to test. Let me know how you go
Regards,
Pete
peter dot werner at g mail dot com