Hacker News Reader

Looking at your first chart there, is there a reason (other than market conditions) you were making significantly more at the end of '09 than mid '10?

View on HN

To be honest I don't know exactly what happened. My theory is that over time more and more market participants started integrating the types of analysis I was doing which rendered my program ineffectual. It's a pretty normal pattern that there is some inefficiency in the market and over time it disappears.

Being a machine learning program, how much of it did you tell it to forget?

Were you compounding the data always, or telling it to forget what was going on several months ago? Or somewhere in between?

(I'm pretty unfamiliar with machine learning, apologies if this is obvious or something)

This. If the 2009 model worked well, did you try letting your algorithm to 'forget' the 2010 data and see if the model worked better?

It's possible that your algorithm is sensitive to market volatility.

Reply on news.ycombinator.com