High stakes, experts, and recency bias: evidence from a sports gambling contest

ABSTRACT

Sports betting markets provide a unique opportunity to test market efficiency and, in this case, study the presence of recency bias and experts in forecasting sports outcomes. This paper uses a panel data set of individual NFL sports gamblers in a season-long contest known as the SuperContest. The contest is used to track the picks of individual gamblers against the spread on games throughout the 2013–18 NFL regular seasons. We find that overall, the SuperContest entrants are no better at picking winners than a coin flip and exhibit a recency bias. However, more skilled contestants are less susceptible to the recency bias and their skill advantage forecasting outcomes may be their avoidance of a recency bias trap.

KEYWORDS:

• Sports wagering
• forecasting
• recency bias
• NFL

JEL CLASSIFICATION:

• D81
• D91
• L83
• Z20

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ https://www.espn.com/chalk/story/_/id/19740480/the-united-states-sports-betting-where-all-50-states-stand-legalization

² For more on SC, see https://www.westgateresorts.com/hotels/nevada/las-vegas/westgate-las-vegas-resort-casino/casino/supercontest/

³ In gambling parlance, winning ATS implies either the underdog team wins or loses by less than expected point difference set by the bookmaker. For more, see https://ats.io/sports-betting/what-is-against-the-spread/

⁴ A complete list of the most recent season payouts can be found at: https://www.westgateresorts.com/hotels/nevada/las-vegas/westgate-las-vegas-resort-casino/supercontest-payouts/

⁵ The data contains contestants who submit selections for all 17 weeks. 20% of contestants starting a season do not finish. Dropouts are due to missing the deadline or quitting in the final weeks without a chance at a payout. Thus, the sample will be biased towards having more skilled gamblers than a sample including all starting contestants.

⁶ We do not include teams who experience a bye in the following week as they cannot be selected, recency bias would likely be different if there is 1 or 2 weeks between games.

⁷ As previously noted, some contestants drop out indicating that gambler behaviour may change near the end of the season even for those contestants in our sample. As a check of the results in Table 2 we omit the last four weeks from the sample and note coefficients are of similar magnitude and statistical significance.