For the past several months I have become deeply involved in baseball quantitative analysis. It started last September when I read an article written by a "sabermatician" (what baseball-oriented statisticians call themselves) and it struck me that the analysis profiled in the article was not statistically well grounded. I did a little work, and it turns out I was correct. By then I was hooked on baseball statistics. Big time. It turns out that predicting baseball win/losses is the hardest categorical discriminant analysis I have ever done. And that is saying something because particle physics provides most of the meanest categorical discriminant problems known to man; literally billions of events are recorded over the lifetime of an experiment (nearly all of which are uninteresting background events), and often times important physical measurements are made on a mere handful of those events.
Baseball quantitative win/loss analysis has filled a deep void that was left in my life after my forced exit from particle physics. It is simply part of my psyche that I need to do research. And the harder the research problem, the better.
And, like I said, accurately predicting baseball wins/losses is very hard indeed. I wrote a bunch of perl scripts to spider web sites to collect the huge database of statistics that I use, and then I came up with a model that appears to work very well. It is based on only 5 well-motivated statistics and a simple multivariate categorical discriminant procedure. When I run the model on the 2003 through 2007 seasons, starting with an assumed betting pot of $1,000 at the start of each season and a conservative betting scheme, I make a total of almost $40,000 dollars. That's a return of almost 8 times the investment. The model predicts that I would expect to lose money only around one out of every 8 seasons, and that the average amount of money won per season is $6,000. And if I had rolled over my winnings from each season to the next, the model predicts I would make around $150,000 (all from the initial pot of $1000 at the beginning of 2003). Take me out to the ballll gaaammme....
I've been working for months on this, but while it is extremely fulfilling in many ways, it is also lonely work. I finally recently hooked up with a friend of mine from my undergraduate days. I remembered him and another friend were always sitting in a corner muttering about odds and matchups, and I knew that he used to bet frequently on sports and did simple (ie; no computers involved) statistical analyses back then to determine which bets to make (and he made more money than he lost too, so his simple models were effective). He got his PhD in medical physics, and now works with a company that uses neural networks to perform cancer diagnostics on mammogram images. It has been really, really great hooking back up with him (we had lost touch for over 12 years...we both moved too many times), and it is wonderful to chat everyday with someone who is as excited about sports quantitative analysis as I am. Plus he knows a lot about all sports (I know bugger all about football) which will be a bonus as I expand my quantitative analysis methods to other sports.
So that is what I've been up to lately. Does anyone want to buy baseball picks for the 2008 season?
;-)
