The late-season bias: explaining the NFL's home-underdog effect

Abstract

This article examines price efficiency and out-of-sample predictability in the NFL point-spread betting market. Our main contribution to the existing literature is the identification of a persistent increase in bias magnitude during the final few weeks of each season. We demonstrate that this anomaly causes the much-documented home-underdog effect. We also offer evidence that the limits of arbitrage have enabled this phenomenon to persist for decades. Finally, we use several regression models to confirm our univariate analysis and show that these models can be used to implement profitable betting strategies. The predictive models presented differ from those in the prior literature by taking into account both short-term and aggregate biases.

Notes

¹ Coaches in the NFL may also behave sub-optimally. Research by Romer (2006) using NFL data suggests that simple models of optimizing behaviour do not seem to explain their observed decision making.

² Much literature has benchmarked 52.38% as the break-even threshold because of the 11 for 10 betting rule. However, this figure fails to account for the other important factors. We address this in detail later in this article.

³ Different betting and payoff systems exist in other sports markets. In baseball and hockey, an odds system is used in which a bet might be listed as (−195, +185). In this case, a bettor would receive $1.85 by betting $1.00 on a winning underdog. Alternatively, she could win $1.00 by betting $1.95 on a winning favourite. So, while the margin of victory is important to football and basketball bettors, baseball and hockey bettors care only about which team wins. Horse racing markets utilize a pari-mutuel betting system in which participants place bets on any number of horses to win, place, show, etc. As opposed to the odds and point-spread betting systems where the payoff is locked in at the time of the bet, the winning bettors in a pari-mutuel system divide the total amount of bet after deducting commissions. For instance, a bet originally made at a payoff of $20 to $1 may only pay $15 to $1 after all bets are placed. Transactions costs in pari-mutuel systems are generally a fixed percentage of the amount bet.

⁴ While the goal of the market specialist is strictly to match buyers and sellers, there is some evidence that the bookie may also intentionally take a position with respect to the outcome of the game if he is able to predict and capitalize on bettor biases (Levitt, 2004). However, even for the experts, predicting outcomes against the closing line is extremely difficult. Avery and Chevalier (1999) examine the historical success rate of sportswriters whose predictions of each game against the spread are published in newspapers. Out of the 12 experts who made predictions for more than one season, only four correctly predict better than 50% of the outcomes and the highest success rate is only 51.1%.

⁵ Although not shown, it is interesting to note that the most common spreads are 3 points (N = 580) and 7 points (N = 370) and the two most common winning margins are also 3 points (N = 703) and 7 points (N = 343).

⁶ Hirshleifer and Shumway (2003) demonstrate the psychological relationship between weather and performance.

⁷ Cold climate games are defined as those which are played after week 14 in Buffalo, Chicago, Cincinnati, Cleveland, Denver, Green Bay, New England, New York, Philadelphia or Pittsburgh.

⁸ Interestingly, it appears that this relationship is reversed when games are played during the hottest months of the season. That is, cold climate teams are at a disadvantage when playing at home against mild climate teams early in the season. Here, cold climate teams are predicted to win by 2.67 points, but win only by 1.32 points on an average. The difference is significant at the 10% level.

⁹ This type of behaviour can also be explained by a change in reference point that alters the preference order for prospects (Kahneman and Tversky, 1979).

¹⁰ The data in Table 6 were obtained from the Nevada Gaming Control Board. Records are kept dating back to September 1988. Data is recorded by sport and we are unable to separate college football from professional football.

¹¹ Moderate and extreme spreads are defined as 2 points or greater and 8 points or greater respectively. While ‘bet on all home teams’ and ‘bet on all home underdogs’ are two commonly used simple betting strategies, bets on moderate and extreme home underdogs are not. However, it was discovered during work on this article that as the spread increases, bets on home underdogs are more likely to cover.

¹² We can determine the marginal probability of a team covering the spread for each of the possible scenarios by multiplying the coefficient estimate by the sample mean standardization factor. In our regression, the mean standardization factor is 0.3922. So, for example, the marginal probability of a visiting favourite covering the spread is −0.3459 * 0.3922 = −0.1357. This implies that when away favourites from moderate climates play underdogs in a cold environment, they are on average 13.57% less likely to cover the spread than otherwise.

¹³ However, to exploit this inefficiency and develop a practical out-of-sample betting strategy, we must be able to accurately predict future statistics. Therefore, as we later demonstrate, conditioning out-of-sample bets on past statistics does not improve predictive accuracy.

¹⁴ When weather is added as an additional explanatory variable, out-of-sample accuracy increases to 54.28% late in the season.

¹⁵ These results are also robust with respect to various distributions.