Season Win Total Market Inefficiency

Betting into the season win total market can be tricky, you are tying up bankroll for months on end, you are speculating that the team you back or fade will be on the right side of injury and turnover luck, and you are hamstrung about when you place your bet, before the number gets sharp or after you know some deeper information on the team you are handicapping. This point has been emphasized a lot in the traditional discussion about this market, a common saying is “all the value has been bet out of that number” as we see wins swing by 1, or 1.5 over the course of the offseason (Indy was a great example this offseason swinging from 6 to a juiced 7.5 on the Andrew Luck news alone). Does that mean it’s a sucker’s game to bet win totals in September? How much juice is too much juice to lay on a number that you have high confidence in? Should you have confidence in any side of a win total?

In an effort to answer these questions, we took the win totals market from the last three years and tried some simple strategies to see if there is any value in betting this market… the results were immediately apparent. On a gut level a couple of things never really added up about the wins total market:

  1.  If you think of the most evenly juiced market win total as the number of games that the market expects that team to win, there are so many teams expected to win around 8 games, very few expected to win fewer than 5 and very few expected to win more than 11; and yet every year the spread of actual games won is much wider than the implied spread looking at the win totals.
  2.  Patriots aside, the uncertainty from year-to-year in terms of actual wins is huge with teams commonly swinging up and down to the tune of 3 or 4 wins, and yet the market is lined implying some correlation between one season and the next (i.e. the teams previous win total informs the market/bettors more than it should).
  3.  Some totals take significant action and get juiced to the nines implying relatively high probability when in reality it only takes one fluke play, like Aaron Rodgers breaking his collar bone, to completely change the outcome for a given team and even beyond that, have a butterfly effect on the league.

Given these observations, it shouldn’t have been surprising that in analyzing the data, the win totals are not sharp numbers overall. Sure a handful of teams fall within a game or a half game of their expected total, but over the last three years on average the market win total was off by 2.3 games (or off by a median of 2 games which may be more meaningful than the average misfit).  This is a huge number and suggests that there may be some way to find value in the market so the alternate win totals posted offshore were used in a simple experiment as discussed below. The data sets for 2015, 2016 and 2017 are tabulated below with the winning sides shaded in the table for ease of reference.

2015 Season Win Total Odds

2016 Season Win Total Odds

2017 Season Win Total Odds

A couple things stand out about this data set, we’ll use the 2017 Philadelphia Eagles as an example in explaining this. The mean win expectancy for the Eagles last year was just over 8.5, the implied probability of the alternate win odds suggested that 69% of the time, they would be expected to finish under 9.5 and 31% of the time they would be expected to finish under 7.5. Looking at the area under a normal probability density function between 31% and 69%, this implies that half the time the Eagles will fall into this super tight window and half the time they will be outliers high or low (obviously they were outliers high last year punching 13 wins).  If you think of this now as an “inside” or “outside” bet given this 2-win window with even juice, based on all thirty-two teams over the last three years, teams only fall inside of that window 28% of the time and outside 72%. This gives an “outside” bet a staggering 44% expected value.

2015 Implied %tages and Results

2016 Implied %tages and Results

2017 Implied %tages and Results

If the market is significantly under predicting variance then what is the best way to attack this? In reality there is no “inside/outside” market and the alternate win totals aren’t lined perfectly symmetrically. As you move away from the mean win expectation, not all half wins are created equally in the eyes of the lines-makers. Last year going from PHI o9.5 to o10 implied a drop of 12% in terms of win probability but going from o10 to o10.5 only implied a difference of 4%. This was a common occurrence among the alternate totals with the payout on the extremes diminishing as you get to the limits of what lines were offered.

Rather than back calculate an optimal strategy for betting into this inefficiency, we conducted a simple test where you blindly created a Polish Middle on every team.  If a line better than +150 is offered on any total over or under, bet it for a unit and if a line better than +200 is offered bet it for half a unit. Going back to the Philly example from 2017, this meant playing full units on under 8 at +162, and under 7.5 at +193, and betting a full unit on over 9.5 at +174 with half units on over 10 at +308 and over 10.5 at +388. With the Eagles coming in at a spectacular 13 wins in 2017 this resulted in a net payout of over 5 units. Repeating this for all 32 teams, our experiment in 2017 resulted in 84.5 units staked with a return of 39.4 units profit (47% ROI). In total, 13 teams lost money (none more than 4 units) and 19 teams won money, all of which swept their alt totals because they were at least 2 wins off the mean expectation.  This simple experiment was repeated for the 2016 and 2015 alt win totals and the results were equally impressive with 102 units staked, 44.2 units profit (43% ROI) in 2016 and 100.5 units staked, 49.2 units profit (49% ROI) in 2015.  Summary by team is presented below:

Test_Summary.jpg

Now obviously 100u is somewhat ridiculous to salt away for the duration of the season and applying handicapping, judgement and math to the process would likely result in even more significant returns. Similarly, this is only based on a three-year sample size and should be extended to more seasons to see how many teams fall within the 2 win window around their expectation on a long term basis; for instance it would be worth knowing if there are “black swan” type seasons where everything goes as expected and you get taken to the cleaners. In terms of applying this for the current season, all I've placed this lotto ticket parlay for fun.

We’ll circle back and update this with the 2018 alt total results in January. But for now, the key takeaways based on this study are that the alternate win total market under-predicts variance and therefore once you identify your side, it is worthwhile to consider spreading your stake across the extremes with greater payouts. Also for the love of Tim Tebow don’t lay the juice on these season long win total bets.