Causal Hangover Effects

TL;DR

This paper investigates whether off-court nightlife in party cities yields a causal hangover effect on next-day performance in NBA and MLB. It introduces latent measures of partying (discrete indicator and continuous nightlife index) and exploits back-to-back games within 24 hours, using bookmaker spreads and money lines to identify effects while controlling for fatigue, jet lag, and home advantage. The results show a statistically significant decline in meeting the spread for NBA and in probability of winning for MLB after visiting party cities, with evidence from both team- and player-level analyses and robustness checks including placebo tests. The study demonstrates potential profitable betting opportunities and discusses limitations, such as unobserved travel schedules, and outlines avenues for future research across leagues and longer horizons.

Abstract

It's not unreasonable to think that in-game sporting performance can be affected partly by what takes place off the court. We can't observe what happens between games directly. Instead, we proxy for the possibility of athletes partying by looking at play following games in party cities. We are interested to see if teams exhibit a decline in performance the day following a game in a city with active nightlife; we call this a "hangover effect". Part of the question is determining a reasonable way to measure levels of nightlife, and correspondingly which cities are notorious for it; we colloquially refer to such cities as "party cities". To carry out this study, we exploit data on bookmaker spreads: the expected score differential between two teams after conditioning on observable performance in past games and expectations about the upcoming game. We expect a team to meet the spread half the time, since this is one of the easiest ways for bookmakers to guarantee a profit. We construct a model which attempts to estimate the causal effect of visiting a "party city" on subsequent day performance as measured by the odds of beating the spread. In particular, we only consider the hangover effect on games played back-to-back within 24 hours of each other. To the extent that odds of beating the spread against next day opponent is uncorrelated with playing in a party city the day before, which should be the case under an efficient betting market, we have identification in our variable of interest. We find that visiting a city with active nightlife the day prior to a game does have a statistically significant negative effect on a team's likelihood of meeting bookmakers' expectations for both NBA and MLB.

Figures (6)

• Figure 1: We examine the relationship between last game location and next day opponent in the NBA, sorted by distance from New York City. Top left: notice that the most likely scenario across all seasons is that a game will be played at home, and the last game was also at home. This makes sense; 1/2 of all games are played at home, and 1/2 are played away, so clearly each team is most likely to be found playing a home game for any given game. When exactly one of the last game or current game was played away, we observe that some combinations cease to exist in our data; e.g. we never observe Charlotte Bobcats play Utah Jazz on the road and then fly directly back home for a game. Lastly, notice the patterns that emerge when both the last game and current game are played on the road. E.g. we see that when last game location was in Los Angeles, next day opponent likely to be a west coast team. There appears to be less of a correlation between last game location in New York or Brooklyn and next day opponent, perhaps because flying from east-to-west is less burdensome as it relates to timezone changes.
• Figure 2: Empirically, in NBA any given team is apt to meet the spread half the time across a particular season; in MLB, after conditioning on bookmakers' odds of success, successful prediction of game outcome is like a coin-flip. In general, systematic deviations represent unexplained mechanisms.
• Figure 3: Within the NBA, we examine travel distance to current game location as a function of whether the last game was in a (non) party city. We see that in general, when traveling from a non party-oriented city to next game location, the distribution of travel-distance has fatter tails. This makes sense: as we saw in our previous heat-map, next day opponent geographically correlated with current game location. Since LA and NY are on the coast, their next day opponent is likely to be nearby; compare this to inland cities, for example, who may have to travel farther to reach their nearest neighbor. We mention that the same plot is not interesting to look at within the MLB, since games are played in series, and therefore most travel-distances between games are zero miles.
• Figure 4: We plot our continuous measure of nightlife for each city within the NBA and MLB. We add a small amount of random noise to our x-coordinate such that cities with both an NBA and MLB team are visible.
• Figure 5: Betting performance by season. In each season, the circled point represents the worst out of pocket expense incurred. We first train a model on historical data then use said model to predict on "current year" data, plotting profits as a function of time over the "current" season. For each relevant game, we place a bet of $100 if the expected value of the bet is positive after accounting for the bookmaker's cut. E.g. in the early 2016 season, at its worst we are out of pocket $89 dollars. By season end, we have recouped all of our initial investment plus an additional $11.5k. We are profitable in each season, and mention that 2015 is the "worst", but we still come out ahead.