### Stupid Feats of Gambling Derring-Do

Loyal readers of my blog may have a memory, rattling around in their brain, of my talking about my "ranking algorithm", a mathematically simple way of ranking teams, calculating individual offensive and defensive power ratings and, when I'm feeling really ambitious, predicting the scores of upcoming games.

Without going into too much detail, the algorithm works something like this:
For each game you play, you either get + or - pts depending on the quality of the team you faced. A win over an otherwise undefeated team = 2 points. A win over a winless team = 1 point. A loss to an undefeated team = -1 point. A loss to an otherwise winless team = -2 points. You can see already that beating a terrible team is better than losing to a good team. That's because this algorithm is meant to do one thing only, and that is rank teams at year end. After 13 games, your record is the most critical factor in your ranking, and wins are better than losses. But among 12-1 teams, the ones with the hardest schedules rise.

The offensive ratings are a measure of how many points you score, relative to the quality of the defenses you've faced. If you face a team that gives up 30 on average, and you score 20, you get 0.667 (20/30). 1 = average, higher is better and lower is worse.
Defense is the same. If you face a team that scores 30 and you give up 20, you get 0.667 (20/30). 1 = average, higher is worse and lower is better.

And then, it simply uses your offensive and defensive ratings, your offensive and defense PPG averages and those of your opponents and calculates a predicted score. I'll leave the math elsewhere.

I did this for a couple of years in the late 90s just predicting scores, seeing how often I'd get the winner right (high 70s %age wise). Someone suggested I try spreads.

I do not gamble. I don't like gambling. I don't encourage people to gamble. I tested my algorithm against the spread solely for the fun of seeing if it worked. And what I found was, in a way, it did. If you just tally up on what %age of games, my score prediction was on the correct side of the spread (that is, if my algorithm said Texas would win by 8 and they were favored by 11, bet against Texas), it beat the magic 52% break even for the week with regularity. 80% of the time or so, if I remember correctly.

I ought to make picks by percentages. The algorithm, when applied properly, creates a set of simulated games. I ought to calculate what %age are on each side of the spread, but I have misplaced my template, and I don't have time to rewrite all the macros before kickoff tomorrow. For today, just a shortcut that creates 1 average game result.

But ... and this is a big caveat when you look at the chart below, the scores my algorithm predicted were often crazy taken on their own. A team would be a 13 point underdog. My algorithm would have them winning by 17. I say bet the underdog, and they lose by 6. I'm actually, on average, further from the correct score than the Vegas oddsmakers, but somehow, even after testing this for 5 years, I'm consistently able to beat the spread doing that. And really, I did - I even posted them online every week.

So the upshot is this:
For the first time in a couple years, I'll be doing this again. I'll probably make myself look foolish. I seem to be under the impression that Kentucky can beat Alabama (and I can tell you right now that I know why and I know my algorithm is being fooled by Kentucky's schedule). But I am not encouraging anyone to use my numbers below for anything other than amusement. If I hear even a rumor that *anyone* has used my numbers to place a wager, I will come to their house, steal their pennies, drink straight from their milk carton and walk all over their petunias in a fit of rage.

The spreads are the "Opening" column from USA Today.
Some numbers may not seem to add up, due to rounding.
The ones where my predicted score is furthest from the spread are at the top.
The "spread" column has a positive number next to the team getting points.
I've noted games where my prediction and the spread are within 1 pt. I don't tally those.