Today is the second in what I hope will be a series of guest postings. I started submitting picks this year to The Prediction Tracker, and I became curious about Todd Beck, who runs the site, so I sent him a series of questions and I'm pleased to present his responses! -- Scott
I grew up in a small town south of Dallas Texas. I attended Texas A&M University where I got a undergrad degree in Math and a graduate degree in Statistics. For some reason I had a interest in trying to predict football games even way back when I was a kid. I used to have a notebook where I guessed at every NFL game for years. So that life-long interest and the fact that our statistics department had a college football pool got me started in coming up with an automated way to make predictions for any game.
For the 19 years since leaving college I have been a statistician at a hospital/medical school in Chicago, primarily in the research area of Alzheimer's disease and other aging related issues.
Most of the people who get involved in sports predictions are sports fans. Do you have a rooting interest in any particular teams? What's your favorite sport?
I was definitely a sports fan growing up. Being close to Dallas I grew up a fan of the Cowboys (Boo! -- Scott), Rangers, and Mavericks. Even though I have lived in Chicago for a long time now I have never been able to get into any of there teams and haven't been able to follow baseball or basketball too much, but I remain a diehard Cowboys fan.
My favorites sports to play was always baseball for some reason. These days I really only follow pro and college football.
ThePredictionTracker.com provides a tremendous service to prediction community, but it seems like it must be a ton of work. What's your workflow like and how much time do you spend each week on it? I know you run ads on your site -- does that recoup your costs? How about your donations button -- does that bring in much support?
Because a majority of the things on thepredictiontracker.com are automated it doesn't require a lot of actual hands on time. The most time is always in the first couple of weeks of a season as I look to see which systems are going to continue on and potentially adding any new ones. I have basketball completely automated so it really takes no time at all once it is running. Football is a little different because it is a weekly event. Collecting the data is automated but I still manually run the program each day. So Monday usually needs a little more time to set up the new week of college games and Tuesday more time to set up for the NFL.
Initially I was never in favor of having ads. But I started on a free web server but I was soon hitting the bandwidth limits. So I switched to a different service and before long I started hitting limits there too. So if I wanted to continue the site I had to get a permanent domain name and start paying for the server. My traffic has continued to grow ever year and so the ad revenue has grown with it.
Originally with ads plus donations I was making just about the same amount as it was costing me, only now with a better ad service and with a larger audience am I bringing in more. But it is definitely not enough to quit the day job.
You don't seem to have a predictor or ranking system of your own. Do you use the information you gather on ThePredictionTracker.com yourself? How successful has that been?
I do not have any basketball systems of my own but I do several for football. My completely original system is the PerformanZ Ratings. But I also maintain Elo and Pythagorean ratings that I have modified for football. Plus there are all of the regression based rankings: Least squares regression, least absolute value regression and logistic regression. After so many years of doing this I still maintain that a simple least squares model is as good as anything out there. There is a scoring efficiency algorithm that I learned a long time from a book in my public library written by a professional gambler. And also a simple points per game comparison which doesn't do that bad.
But in some ways I guess you could say that the 'system' that I am most known for is the composite mean. Calculating the mean or median of all, or most, of the ratings, especially when you have 70+ of them, is always going to end up performing better than most individual systems.
I don't do any real gambling, any simulated gambling I have experimented with has almost always been using the system average. On my blog page I generally post probabilities (for football) that are based off the system average. Tracking the top picks of the week over the season almost always gives results in the 55-60% range for at least one of pro or college football. The NFL picks have been 62% each of the last two years. A terrible season pops up every few years for some reason I haven't figured out yet. Like this past year the NCAA picks were only 35%. I experimented with the same idea in college basketball last season and had huge success in the first half of the season before giving it all back once conference play started.