BYU vs Utah: What we learn from the spread
 

The spread for tonight's game is BYU by 1.5 points. Currently the tradesports (http://tradesports.com) contract for BYU winning is at a bid price of 51 and an offer price of 56. So a reasonable estimate is that there is about 53.5% chance of BYU winning. Hovever, a couple of empirical facts work against BYU here that may suggest that the probability that BYU wins is lower than 53.5% (maybe more like 50/50). Levitt (2004, yes the same levitt who wrote Freakonomics) finds some important empirical patterns regard NFL betting that I will summarize with the following:

• In 50% of all games 2/3 of the bets (dollars) fall on oneside of the line.
• For the median game when the home team is favored, 58% of the bets (dollars) go to the home team.
• For the median game when the visting team is favored, 2/3 of bets (dollars) go to the visting team.
• Favorites beat the spread less than 50% of the time.
• Home teams cover the spread more than 50% of the time.
• Oddsmakers increase their profits by between 20-30% by taking advangtage of the "visiting team" and "favorite team" bias relative to a strategy of trying to get 50% of the money on each side of the bet.

These empirical results really only apply to NFL football betting, but it seems likely that oddsmakers do something similar for NCAA Basketball. It is possible that they don't because maybe oddsmakers don't have the same kind of informational advantage in college basketball as they do for NFL football. However, if they do then it suggests the following: The spread is probably too big because BYU is not only favored but a visiting team favorite which is where we see (at least for NFL football) the biggest betting bias. Thus my best guess is that this game is really close to a complete toss-up based on the spread.

My model predicts BYU by 8, but I don't feel that confident about the game.

A couple questions:

1. Is there enough NFL game data to really prove this stuff? How many games in a year? Under 300? If one year visiting team is 51% against the spread and next year they're 49% against the spread is that a statistically significant movement? That was one of my problems when I was trying to do this kind of analysis was that I thought I had a trend, but didn't feel confident about data sufficiency.

2. My research into college football lines seems to go against this data. One study I read compared 10 top computer models to Vegas lines. Vegas lines were more accurate in predicting final game score than any of the computer models. This suggests they're not monkeying around with any of these public betting trends, but just trying to predict the score. Maybe college football bettors are more sophisticated than NFL bettors?

It's empirical economics so nothing ever gets proved.:) Levitt has data from the 2001 (I think it is like 85 games). The nice thing about the data is that he observes wagers rather than just spread versus outcomes. The important contribution of his paper really is that oddsmakers are not trying to attract 50% of the money on each side of the spread. It is reasonable to worry about how generalizable Levitt's results are, and I think that is an open question. The paper is here:

http://pricetheory.uchicago.edu/levi...nction2004.pdf

If they're not trying to achieve even money on either side, then they're still shooting for some form of modified betting equilibrium, or else they can't easily manage their risk.

This is proabably true as well. Spreads are pretty efficient, but not perfect. Avery and Chevaliar have a nice paper on this I think in the 1999 Journal of Business that points out bias related to "hot teams." Remember, the differences have to remain small because the oddsmakers can't make the bias too big (52.4% is the threshold) or mechanical rules like always bet the home team will make money. So, in short I think you point in (2) is probably entirely consistent with Levitt's results.