≡ Menu

The Smarter (Sigh) Football Betting Market

Economists (I am one) have historically been trained to believe in the efficiency of markets. The simplest way to think of this is that market prices capture all relevant information. Of course, this is sometimes not quite right, or even close to right. All the mortgage-backed securities that helped bring down our economy were horrendously mispriced, for example, despite lots of people seeing the warning signs. Even then, people betting against those securities provided information about their true value. They were just drowned out for too long by people clamoring to buy that worthless stuff.

The sports betting market, though, is a case that we might actually expect to work better. Unlike mortgage-backed securities, everyone making a wager in Las Vegas is incentivized to get the price right. There’s nobody who’s pushing a bad wager on their clients, for example.1 Therefore, we might expect efficient markets to mostly work in Vegas and that the odds would converge to the correct number.

Mostly, it seems like that’s what’s going on. Whatever information is not contained in the initial odds may be quickly corrected as people swoop in to take advantage. I’ve experienced this first-hand. Last year, I went to Vegas about a week after the first season win-totals for 2013 came out. I found the numbers online and came up with this list of wagers I was interested in.

Team
Side
Number of wins
Odds
Share of total
SEAOver10-1201/3
KCOver6.5-1101/3
INDUnder8.5-1101/6
ARIOver5.5Even1/6

So I left for Vegas planning to bet 1/3 of my money on the Seahawks to go over 10 wins (betting $120 to win $100), 1/3 on the Chiefs over 6.5, 1/6 on the Colts under 8.5, and 1/6 on the Cardinals over 5.5. As you might have guessed, I never got the chance. By the time I arrived, the Seahawks were at 10.5 and the odds were up to -150 (needed to bet $150 to win $100). The Chiefs had moved even more. I didn’t record exactly how far it moved, so I don’t know the exact number. I think they were at 7.5 in at least one sportsbook. By the time the season started, they were at 7.5 and you had to lay $160 to win $100 on the over, and a different book listed them at 8.

Alas, this story doesn’t have the happy ending it would have if I’d gotten in on the IPO. I didn’t bet on either the Seahawks or the Chiefs because I decided the price had moved too much. The Cardinals and Colts prices stayed put, so I made those bets. I was too late. A potential nice win became a wash.2

The movements in those Seahawks and Chiefs lines likely reflected efficient markets at work. Those initial numbers that looked so appealing to me caused money to crash in to take advantage. What did the bettors see? Probably the same things I saw. On the Seahawks, it was basically that they’d been #1 in DVOA in 2012. More simply, they had an estimated win total based on that DVOA that was substantially better than their actual wins in 2012.3 My research suggested that teams like the Seahawks tended to beat their win total most of the time of the following year. For the Chiefs, I was mainly attracted by their extremely easy schedule and likelihood of bouncing back from a two-win season.

The Seahawks’ situation was one that seemed more likely to perhaps be mispriced, in general. It was at least feasible that teams with misleading win/loss records would mean opportunities. To see if that was the case, I collected data on season wins lines from 2003-2013, with two years (’05 and’07) not included. I combined that data with Football Outsiders’ estimated win totals from the previous year. Even though those numbers are a great way to measure previous performance, we still might not expect them to have much predictive power given the Vegas lines, which incorporate all the free agent signings, injuries, and the draft. Still, FO’s numbers do have some predictive power that would have been profitable to utilize, at least in the past.

Betting on Underachieving Teams by DVOA

Football Outsiders reports a team’s estimated wins, which is close to using DVOA to predict what would have happened against an average schedule in a given year. Every year, there are teams that substantially outperform or underperform their estimated wins. The 2012 Colts were the biggest deviation in the sample, winning 4.8 games more than their 6.2 estimated wins.

Looking first at teams that underachieved, winning at least one game less in the previous season than their estimated win total according to DVOA, have there been opportunities to win by betting on these teams? The data seem to suggest the answer has been yes. The table below summarizes the outcome of betting $100 on the over for every team that had estimated wins at least one higher than their actual wins in the previous year.

Year
Team
Estimated Wins Prev Yr
Actual Wins Prev Yr
Win Total
Line
Result
Profit/Loss
2003BAL8.877.5-135OVER74.07
2003CHI6.347.5140UNDER-100
2003CIN3.625.5-130OVER76.92
2003DEN10.299.5-120OVER83.33
2003JAC7.566.5-135UNDER-100
2003KC10.789-140OVER71.43
2003MIA10.199.5-160OVER62.5
2003NYJ10.29OFF------
2004CLE6.357110UNDER-100
2004JAC7.957.5-145OVER68.97
2004NYG5.546.5110UNDER-100
2004NYJ8.668.5-110OVER90.91
2004OAK5.947.5-110UNDER-100
2004PIT7.468150OVER150
2004SD6.144.5-115OVER86.96
2004TB9.579-110UNDER-100
2006BAL7.668-150OVER66.67
2006GB5.146-160OVER62.5
2006HOU3.425.5-150OVER66.67
2006NO4.837140OVER140
2006OAK6.746.5140UNDER-100
2006PHI7.868.5-130OVER76.92
2006SD11.599110OVER110
2008BAL6.756115OVER115
2008CIN8.477-125UNDER-100
2008KC5.345.5-175UNDER-100
2008MIA4.415.5-110OVER90.91
2008PHI9.888.5-170OVER58.82
2008TB10.398110OVER110
2009CLE5.746.5-120UNDER-100
2009DET2.105140UNDER-100
2009GB8.669130OVER130
2009JAC8.558-135UNDER-100
2009KC3.526-120UNDER-100
2009NO9.188.5-165OVER60.61
2009PHI11.69.59.5-150OVER66.67
2009SD10.389.5-170OVER58.82
2009WAS9.188-110UNDER-100
2010BAL12910-110OVER90.91
2010DEN9.287-110UNDER-100
2010MIA8.478.5-120UNDER-100
2010NE11.2109.5-110OVER90.91
2010TB4.835.5-135OVER74.07
2010WAS6.647.5EvenUNDER-100
2011BUF5.545.5-135OVER74.07
2011CIN6.645.5130OVER130
2011CLE6.757-110UNDER-100
2011DEN5.346-120OVER83.33
2011DET7.568-145OVER68.97
2011HOU7.969-160OVER62.5
2011MIA8.577.5EvenUNDER-100
2011PHI11.91010.5-150UNDER-100
2011TEN8.666.5-115OVER86.96
2012BUF7.167-110UNDER-100
2012CLE5.545.5EvenUNDER-100
2012MIA7.767.5-110UNDER-100
2012MIN4.636EvenOVER100
2012SEA8.177-110OVER90.91
2012TB5.546-110PUSH0
2012WAS6.356.5EvenOVER100
2013CAR8.877-160OVER62.5
2013CLE6.256-135UNDER-100
2013DEN14.71311.5EvenOVER100
2013DET7.648105UNDER-100
2013NE13.41211110OVER110
2013SEA131110.5-130OVER76.92

If you had made these bets over the nine years in the data, you would have won a total of $881, or an average of about $13.50 on the 65 bets you would have made. You would have won 39 bets, lost 25, and pushed on 1.4

It’s a little hard to tell if things have changed over time, but you would have particularly done well in the early years of the data. The graph below summarizes the profit or loss from betting on these estimated win underachievers for each year in the data.

Figure 1

From this, it at least seems possible that there were more opportunities to bet on teams like the 2013 Seahawks in the past. On the 28 bets from 2003-2008, you would have won about $27 per $100 bet, compared to only $3 since 2008. Of course, this result depends on how you divide the data and we’re dealing with small samples. Still, the previously successful strategy has not worked in three of the last five years.

Betting on Overachieving Teams by DVOA

If betting on teams that underperformed compared to their expected wins has made money in the past, how about betting against teams that won more than would have been expected based on their play? Things actually look pretty different here. Betting against teams like the 2012 Colts actually hasn’t worked, on average, going back to 2003. These kinds of teams definitely do worse the previous year, but that regression is baked into the price so that there’s been no opportunity there.

The table below summarizes all the outcomes from betting $100 on the under for all teams that won at least one game more than their estimated wins in the previous year.

Year
Team
Estimated Wins Prev Yr
Actual Wins Prev Yr
Win Total
Line
Result
Profit/Loss
2003ARI3.555.5-180UNDER55.56
2003GB9129.5-110OVER-100
2003IND8.1109.5105OVER-100
2003MIN4.368-110OVER-100
2003NYG7.9109-140UNDER71.43
2003PHI10.61210.5-170OVER-100
2003PIT8.210.59.5-110UNDER90.91
2003TEN9.6119.5-105OVER-100
2004CAR7.7119-165UNDER60.61
2004DAL8.4109-200UNDER50
2004KC10.6139-135OVER-100
2004NE11.71411-155OVER-100
2004PHI10.91210165OVER-100
2004STL8.21210-230UNDER43.48
2004TEN10.81210-190UNDER52.63
2006CHI8.2119160OVER-100
2006JAC10.7129-120UNDER83.33
2006MIN6.698-130UNDER76.92
2006SF2.345-110OVER-100
2006STL3.967-110OVER-100
2006TB8.5118-110UNDER90.91
2008ARI5.487.5115OVER-100
2008CAR5.877.5120OVER-100
2008DAL11.11310.5115UNDER115
2008DET4.976.5110UNDER110
2008GB11.1138.5-160UNDER62.5
2008NYG8.2108.5120OVER-100
2008SD9.41110.5115UNDER115
2008STL1.836.5120UNDER120
2008WAS7.997.5EvenOVER-100
2009ARI7.298.5-120OVER-100
2009ATL8.4118.5-120OVER-100
2009CAR10.3128.5-125UNDER80
2009DEN6.387-160OVER-100
2009HOU6.488.5-135OVER-100
2009IND10.41210-120OVER-100
2009MIA8.5117-110PUSH0
2009MIN7.1109105OVER-100
2009TEN11.8139110UNDER110
2010ATL7.999.5EvenOVER-100
2010CIN8108.5-120UNDER83.33
2010DET025.5-110OVER-100
2010IND11.11411-150UNDER66.67
2010MIN10.4129.5-140UNDER71.43
2010NO11.21310.5-110OVER-100
2010OAK3.956105OVER-100
2010SD10.41310.5-110UNDER90.91
2010SEA3.157-145PUSH0
2011ARI3.157.5-125OVER-100
2011ATL11.31310-120PUSH0
2011CHI8.3118-135PUSH0
2011IND8.210OFF----------
2011JAC6.586.5-110UNDER90.91
2011KC8.3107.5EvenUNDER100
2011NO9.21110135OVER-100
2011STL5.477.5-115UNDER86.96
2011TB8.4108-140UNDER71.43
2011WAS4.766-115UNDER86.96
2012ARI4.987-120UNDER83.33
2012BAL10.61210-125OVER-100
2012GB13.31512-105UNDER95.24
2012SF10.81310-135OVER-100
2013ATL9.11310-140UNDER71.43
2013CIN8.7108.5105OVER-100
2013HOU8.31210.5-140UNDER71.43
2013IND6.2118.5-130OVER-100
2013MIN8.8107.5-130UNDER76.92
2013TEN3.366.5-130OVER-100

If you’d followed this strategy, you would have lost about $3 per $100 bet since 2003. This loss is almost entirely because of Vegas’s commission on the bets. You would have won 31 bets, lost 32, and pushed on 4, by betting this way.

And there’s little reason to think that much has changed over time, either. The graph below shows the trend in profitability of betting against these likely-regression teams. As before, it’s a little hard to say, but I think the most reasonable reading of the graph is that betting against the overachievers has consistently lost a small amount of money over time.

Figure 2

2014 Win Totals: Learning from the Past

The first thing I remember thinking when I saw the 2014 lines was that there was almost nothing that looked appealing.5 I thought the Packers and Patriots looked appealing at 10 and that the Jags looked appealing at 4.5. I didn’t see any unders that excited me. The most amazing thing was how much regression the oddsmakers built in. Houston opened at 8.5 at the Cantor sportsbooks in March.6 Really? For a 2-14 team? That’s a huge adjustment, but perhaps not surprising now that investment bankers are making the lines at some of the sportsbooks. And then there is the team with the biggest gap between 2013 estimated and actual wins. The Falcons had 6.5 estimated wins compared to their actual four. Again, the price already seems to build in the expected rebound in 2014 wins. The Falcons opened at eight wins, similar to where the Lions ended up last year after a four-win season.

I think the market is likely pricing all of these things close to correctly now. I also think it’s pretty likely that there was a window of opportunity to profit within the last ten years. The strategy of betting on teams that either underperformed in the win column or against those that overperformed would have, on average, yielded about $5 in profit for every $100 bet over the last ten years, with all of that coming from betting on the underperforming teams. Betting just on the underperformers would have yielded $13 in profit, although I think the greater success there may be about random chance rather than something special about underperforming teams relative to overperforming ones.

By the way, give me a time machine, and here’s my favorite bet in the data: the 2004 San Diego Chargers. They had four wins in 2003, but 6.1 estimated wins. Somehow, they went off at 4.5 wins, with -115 the line on the over. I find it pretty hard to believe that line would be anything below 5.5 today. As the data suggest if they cannot quite prove, the market probably just does not provide these kinds of enticing opportunities anymore.

  1. These perverse incentives have been going on a long time, too. Check out Michael Lewis’s Liar’s Poker for fascinating stories of investment bankers pushing junk on their clients. []
  2. Obviously, the Seahawks and Chiefs bets would actually still have worked out well had I jumped in at the less attractive prices. []
  3. For details on estimated wins and DVOA, see here. []
  4. Note that the 2003 Jets do not count in these calculations since they were off the board in my data. []
  5. See here for the complete set of initial lines and his comparisons of those lines to projections using 2013 DVOA []
  6. The Superbook had them at 7.5 in May. []
{ 20 comments }
  • Kevin June 19, 2014, 12:42 am

    We looked at game-by-game betting in a paper a while back (http://harvardsportsanalysis.org/?p=4391). As you wrote, a lot of the profitability depends on what the vig would be on your bets.

    Reply
  • C Bolton June 19, 2014, 4:44 am

    Some people have a very good knowledge of football based on watching the games, or perhaps having played or coached in them, and not because of any statistical analysis. Those people might be able to outperform sport-analytics people.

    Reply
    • Anders June 19, 2014, 8:54 am

      Not really because watching a team tells you nothing about things that might regress (closer win/loss, fumbles recovery and over/under performing point differential or dvoa)

      Reply
      • Andrew Healy June 19, 2014, 11:43 am

        I guess I lean towards these kinds of stats-based metrics, too, even if they’re probably harder to find now. But they can be paired with other qualitative info, too. The chapter in Nate Silver’s book has some good examples of that with Bob Voulgaris, the basketball bettor, who I think followed players’ Twitter feeds for info that might inform his bets.

        Reply
    • Andrew Healy June 19, 2014, 11:36 am

      Thanks for the comment. I agree. Any kind of general strategy based on stats stuff doesn’t rule out case-by-case strategies based on knowing more specific info.

      Reply
      • Bryan Frye June 19, 2014, 12:49 pm

        The argument of former players/coaches outperforming stat geeks makes me think of the former players/coaches on the pregame shows who tend to fair quite poorly when picking winners. That doesn’t even take spreads into account; they can’t pick the winners straight up.

        Reply
        • Chase Stuart June 19, 2014, 2:05 pm

          Yeah, I had the same thought.

          Reply
        • Andrew Healy June 19, 2014, 5:02 pm

          Yup, let’s hope nobody is betting on the picks made by Terry Bradshaw, Jimmy Johnson, etc.

          I do think it’s a good idea to add to the numbers what your eyeballs tell you, too. I think the process on my Colts under bet was not great. My models suggested that was a good bet in part b/c the numbers were a little less excited about Luck than RG3 or Wilson in 2012. I knew some of the reasons for the numbers being a little lower (Arians liking to throw downfield, for example), but the bigger thing was just that Luck so clearly passed the eye test. He was better than the numbers suggested. That was the bet I made that didn’t feel great as I made it and I probably could have avoided it.

          Reply
  • Chase Stuart June 19, 2014, 12:02 pm

    Another killer post by Andrew. Thanks again.

    Reply
  • Ty June 19, 2014, 2:09 pm

    While I would agree that the market for opening win totals are close to efficient, you can still find value during the season. Carolina, for example, started 1-3, 2 of those losses being very close, and 2 were against playoff caliber teams (one was clearly the best team in the league). I’m not sure what the win total came down to after the 1-3 start, but it wouldn’t surprise me to see it come down to where there was an inefficiency. You could use the numbers (or even your eyes) that Carolina wasn’t nearly as bad as their record.

    IMO, for major sports (and possibly minor sports, but less so), you need to be able to use numbers/stats/metrics to win long term, if you aren’t using numbers, then either you are one of the smartest people in the world, you know someone that uses numbers, or you are losing. How can you tell a number has value when you don’t use numbers, especially if the number is of by 0.5? You probably can’t.

    Reply
    • Ty June 19, 2014, 2:10 pm

      I meant to say *off by 0.5

      Reply
    • Andrew Healy June 19, 2014, 2:44 pm

      That’s a great example, I think. Online, the Panthers’ playoff odds when they were 2-3 were something like 2-1 or 3-1 and FO’s models had their chances at much higher than that. That was a clear opportunity (one I might have taken advantage of :-) ).

      Reply
  • Richie June 19, 2014, 2:57 pm

    I’m not sure what the magnitude of the effect is, but are you using FO’s originally published “estimated wins” for the early years, or the adjusted numbers? FO has tweaked their formula over the years. (Also the site has gotten much more popular.) Is it possible that the early success you show is because the “estimated wins” known at the time were influencing the Vegas win totals for those years?

    Reply
    • Andrew Healy June 19, 2014, 3:55 pm

      I’m using the currently posted numbers. The early success would come, I think, because estimated wins/Pythagorean kind of stuff were not influencing the numbers enough in the past. If the conclusion is right, that stuff is influencing Vegas’s totals more now.

      Reply
  • Kriss Berg June 20, 2014, 12:19 am

    I’ve gone at 3-2 or better every year for the past 5 by using FO’s prospectus for the coming year against Vegas season over unders. Too much season to season variation to use previous year’s numbers. Also, I’m doubling my money over a season using Advanced Nfl Stats probabilities each of the past 3 seasons ATS. The inefficiencies are still there, you just have dig.

    Reply

Leave a Comment