Genetic Trading Algos – Genotick

genotick

Johnny Come Lately, as ever, I have become fascinated with machine learning.  Since my sphere is financial markets it was obviously the easiest dataset to use to further my enquiries.

I have started from a very low knowledge base and have mostly so far used Python and Python Libraries such as Scikit Learn, Theano and Keras.

I recently came across a ready built algo coded in Java called Genotick and have been playing around with it.

To quote the author:

It creates mechanical trading systems that can later be used for your day-to-day investment decisions. Systems are created automagically, without user’s intervention nor ideas. Genotick is capable of creating any kind of system: mean reverting, trend following, price action or even based on fundamentals. The equity you see here is built day-by-day, exactly as in real life. To stay realistic, Genotick learns trading with historical prices and THEN makes a prediction for the next day. Unlike other artificial intelligence trading software it doesn’t iterate over data more than once, so it avoids over-fitting and over-learning.

The above equity curve was created by using Genotick and SPX data out of the box.  The data was simply the daily OHLC.

As you can see it made most of its money in this particular back test by shorting the SPX during the 2008 / 2009 crash.

Early days since to understand the algo you need to go through the code line by line in a suitable IDE.  The code is open source and can be found at Github

There will be many naysayers. But hats off to the developer and thanks for open sourcing it.

 

Categories:

2 Comments

  1. Except that “automagically” created algos and maybe (?) machine learning in general is a dead end since it’s overfitting even with single runs in parallel and/or there is no logical methodology in the strategy to begin with. Keep up the good work and thanks for all the articles.

    Like

  2. I wonder. Despite the unfortunate use of language this algo does at least advance day by day (IE no K Fold cross validation, no training / validation set) but there is indeed no fundamental reasoning behind the “strategy”. Machine learning is very good at telling us what is (eg the Travelling Salesman Problem) but it may well be that neither machine learning nor any other method can tell us what will be – at least in terms of adaptive complex systems. Although…..I can’t help feeling that given relevant fundamental data a machine learning algo probably has as much chance as a human stock picker in choosing winners.

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s