Monday, August 26, 2013

Home underdogs are no longer a good bet (in case you were wondering)

Underdog-Macys-1979In an oft-cited 2004 paper, Steven Levitt (of Freakonomics fame) investigated the behavior of gambling markets. Using a unique dataset comprised of bets made in an NFL handicapping contest, Levitt concluded that, contrary to conventional wisdom, bookmakers do not set point spreads to equalize the amount of money placed on either side of the spread. Instead, Levitt postulated that bookmakers willingly accept these imbalances in an effort to maximize their own profits.

A bookmaker can increase profits beyond their nominal "vig" by exploiting bettor biases. Levitt cited home underdogs as one example of this bettor bias that bookmakers could exploit. Using data from the 1980-2001 seasons, there were 1,483 games in which the home team was the underdog. Of those games, the home team covered 53.3% of the time (significant at the 0.5% level).

Results from the 1989-2012 Seasons

I don't have data going back to 1980, but my dataset (from does go back to 1989, and the results from 1989-2003 (Levitt published in 2004) are consistent with the 1980-2001 data. From 1989-2003, there were 1,121 games in which the home team was the underdog. Of those games, the home team covered 53.5% of the time (significant at the 0.8% level).

So how have home underdogs fared since Levitt published his 2004 paper? Not so well. From 2004-2012, there were 766 games in which the home team was the underdog. They covered just 47.8% of the time, nearly significant at the 10% level (but in the wrong direction). The graph below summarizes the results by season and compares them to the "breakeven" percentage of 52.38% (how often a bettor would have to win in order to break even against the standard vig).

The above chart illustrates why gambling on sports can be so challenging to do successfully. In 2004, you had a betting strategy with a plausible rationale and statistical significance out the wazoo. And you would have gone broke following it.


  1. It's a minor digression, but I frequently wonder why Vegas hasn't adopted auction or parimutuel-like wagering for sports betting. It's not hard to come up with structures where the casino is - mostly - acting as a clearinghouse and can collect a vigorish (or commission) with minimal capital risk.

    I few months ago, I briefly entertained formally modelling a system where bookies are trying to maximize expected returns (with variance aversion) against a pool of bettors (with some diverse perceptions) and a gaussian distribution of game outcome probabilities, but then I remembered that I'm lazy.

    1. Levitt covers a lot of that ground in his 2004 paper.

      The problem with a pari-mutuel system is that you don't know what odds you're buying until betting closes, which puts a sportsbook at a competitive disadvantage to one that offers fixed odds. (based in the UK) uses the clearinghouse model, where they're just matching bettors on either side of the wager. The upside is that their vig is lower than a standard sportsbook. The downside is that your bet volume is limited by the amount of bettors out there willing to take the other side.

    2. I wonder how hard it is to catch the bookies with their finger on the scale...

    3. Armed with your work on the Cantor Gaming NFL spreads, I looked for games where the expected spread (based on home GPF and away GPF) had the largest mismatch with the Cantor spread.

      As might be expected, the two London games show up until I treat both teams as away for those games There are a lot of week 1 match-ups with unusually high differences between GPF spread and expected spread. Based on that, the week 1 picks should be:

      The Jets (+3 Hosting Tampa)
      San Diego (+3 Hosting Houston)
      Baltimore (+9.5 at Denver)
      Buffalo (+10.5 Hosting New England)

      The Baltimore game, in particular is interesting - they're the reigning champions, and I'm loathe to pick them on the road because they have such a huge home field advantage.

      Conversely, Houston is a playoff team against San Diego a non-playoff team, and the home field advantage calculus favors the Chargers.

  2. That paper is pretty awesome reading -- thanks for pointing me into that direction.

  3. The bookies probably have good absolute game prediction tech, they certainly have information about how the bets are stacking up, and they know how scoring is distributed (

    With those pieces of information they can probably make iterative corrections to the line for maximum profitability, and quickly recognize dangerous shifts in bettor behavior, and bookies have the 'safe' option of offering the 'true' line if gambler's behavior becomes unpredictable.

  4. Posting these to start the season...
    HFA Model picks:
    Broncos -7.5
    Cincinnati +3
    Atlanta +3
    Detroit -5.5
    Carolina +3.5
    Kansas City -4
    St Louis -4.5
    San Francisco -4.5
    Giants +3.5
    Philadelphia +3.5

    Over/Under Model Picks:
    Dolphins @ Browns - Under 41
    Giants @ Cowboys - Over 49

    ATS History Model Picks:
    Ravens +7.5
    Patriots -9.5 (The line for this looks all over the place.)
    Dolphins PK
    Falcons +3
    Steelers -7
    Giants +3.5
    Eagles +3.5

  5. This is quiet awesome blog. Now enjoying the real betting on NFL Sports Games and grab the chance to win exciting prizes.

  6. Wow, this is actually really relevant. I have always assumed that home underdogs cover at a slightly higher rate than the rate needed to profit. Thanks for the info! With teasers you can add points to the original line to increase your chance of winning - I wonder if there are any nfl handicapping statistics on how often a home underdog covers with an additional +6.5 to +8.5 points.

  7. I would like to thank you for sharing a amazing blog.
    Great to know about this article.

  8. You have shared the best blog. This is what I really need immediately. online terpercaya dan

  9. I have always assumed that home underdogs cover at a slightly higher rate than the rate needed to profit.
    gclub จีคลับ
    gclub casino