I recently took a stab at writing an algorithm to predict the stock market, and was surprised to find out how well it worked out on my first trade day. The initial version took only 16 hours to complete, additional modifications put my time a little over twenty hours. I aggregate web sites, twitter, weather statistics, as well as do some traditional stock estimation with figures. I’m basically using event-driven arbitrage.

I am constantly making small improvements to the methods that I have implemented, and measure each change carefully. Measuring success is great, however, I wanted to actually see it perform with an automated trading system. I wrote an automated buy/sell program to purchase on the predictions (with fake money of course) in about 6 hours including testing.

I will be adding short selling, as well as an API for those who would like to play with the system as well very soon.

As I learn new and interesting things, I will post them here.

algorithmic training · event driven arbitrage · finance · stock market