I'll be using this blog to record my notes and thoughts on learning to model stock prices. The basic problem I am interested in is whether, given a time series of stock prices, we can predict the next price in the series. We can look at this goal from two points of view: *maybe*, MAYBE, there are some basic technical patterns that can illuminate future stock prices, in whic case we can learn to exploit them, or more likely in my opinion, stocks move locally at random and the best we can hope us more along the lines of a buy and hold strategy. There is already plenty of research in this area, most of which to my knowledge suggests that timing the market is futile and at best any gains are erased by transaction fees. If and when I find some time I'll share any reading I can find on this topic.
A little bit about me: I've conducted machine learning research for over 15 years, first in robotics and vision and currently in internet service security. I have tons of experience building regression models and binary classifiers. This little project is purely a labor of love and mostly aimed at proving to myself that I can't do much better than buy and hold. ;-). I have literally no prior experience in this problem space (except a few dot-bomb writeoffs) so we shall learn as we go.