How well do Vegas futures predict playoff results in each sport?

Spurs_Tim_Duncan_Heat_LeBron_James_2013

Spurs and Heat… again.

On Friday, I wrote a post for FiveThirtyEight discussing the rampant inequality and predictability found in the NBA. I have a few more thoughts on that subject…

While the NBA has successfully suckered tens of millions of fans into believing the 2014 playoffs have been thrilling and unpredictable, with the LA Times even declaring, “March Madness suddenly seems like such a bore by comparison” (this is a pretty odd comparison considering the Final Four concluded with a No. 7 seed beating a No. 8 for the title), I’ve had a much different opinion, and this year’s predictable results simply fall in line with recent history.

While my original post mostly analyzed the odds distribution in each sport, I didn’t much consider the actual results, and it got me wondering… How well do those futures odds predict what actually happens on the court, as well as in other sports?

To analyze this, I ranked all the teams in each sport according to their preseason championship odds* (i.e. for 2014, Miami is No. 1, Bulls No. 2, etc.) I then plotted this rank against their eventual playoff results (0 = missed playoffs, 1 = lost in first round, 2 = made conference semifinals, 3 = made conference finals, 4 = made finals, 5 = won title) over the last four seasons. For comparison, I also included the NHL, a league that also has an 82-game regular season and four rounds of best-of-seven playoffs:

*Because any betting odds will add up to greater than 100 percent, I simply divided all the futures by the sum of all the teams’ odds to scale them down to 100 percent.

nba presason vs results

nhl presason vs results

The NBA has a clear strong correlation between preseason championship probability rank and playoff results, while the NHL is much weaker. Over the last four years, no NBA team ranked lower than fifth in preseason championship odds has made the finals. Compare that to the NHL, which has had six of eight non-top five finals teams in that time.

Here are the NFL and MLB distributions too:

NFL preseason vs results

(0 = missed playoffs, 1 = lost in wild card round, 2 = lost in divisional round, 3 = lost in conference championship game, 4 = lost Super Bowl, 5 = won Super Bowl)

MLB preseason vs results

(0 = missed playoffs, 1 = lost in Wild Card game, 2 = lost in LDS, 3 = lost in LCS, 4 = lost World Series, 5 = won World Series)

It’s easy to see how these other leagues’ plots differ from the NBA’s. Here are the correlations for each graph:

NBA = .647

NHL = .523

NFL = .440

MLB = .371

Clearly, the preseason favorites go further in the playoffs with a much higher frequency in the NBA than in any other sport.

Dominance of Miami

Much of the predictability in the NBA is due to the havoc the Miami Heat have waged on the wagering markets. To put Miami’s sports book dominance during the LeBron-Wade-Bosh era in perspective, here’s a scatter plot of the all the seasons in the MLB, NHL, NBA, and NFL over the last four seasons, plotting each team’s championship odds at the start of the season against their odds at the start of the playoffs.

AllOdds2

(Sorry, no pretty graphic with this one—just a screenshot of an R plot. Chartbuilder doesn’t take to kindly to doing scatter plots with multiple series.)

Only once in the last four years have the Heat not been favorites to win the title—the start of the 2011 playoffs, following the Big Three’s underwhelming debut regular season. Now, to really see how Miami’s recent run stands out, here are those odds when the four years for each team are averaged together.

AllCombo2

Over the last four seasons, the Heat have averaged 24.6 percent and 28.4 percent chances of winning the title at the start of the regular season and postseason respectively. Compare that to the Sacramento Kings, which have averaged 0.3 percent and 0.0 percent respectively over that time.

Another interesting takeaway from this chart is that the Patriots have often wildly outperformed preseason expectations during the regular season, accruing more than double the title odds by the time the playoffs start. The Lakers, on the other hand, wildly underperform (or are simply very overvalued at the start of the year), nearly slicing their preseason title odds in half over the course of the regular season.

Odds Distributions

Probably the most striking image from my FiveThirtyEight story was this line chart of the distribution of championship odds at the start of the season in each sport.

Pre line chart

When you zoom in on the bottom two thirds of the chart, you can really see the near-zero odds of the barren bottom 50 percent in the NBA.

Pre line zoom

This NBA inequality is still apparent come playoff time. (I separated the MLB data into the years in which there was only one wild card per league and years in which there were two.)

Play line chart

While the NBA may look like it’s made up some ground here equality-wise, it’s important to note that the NFL and MLB have byes and play-in games, a structural advantage that automatically adds greater disparity in championship odds. The average top NBA playoff team, which has to win four rounds for the championship, still has better title odds than the average top MLB and NFL teams, which only have to win three rounds.

Solutions

The day after my FiveThirtyEight story ran, the NY Times published a story that touched on very similar topics. Most interestingly came a quote from former Stanford professor Roger Noll, discussing one of my most despised regulations in sports:

“If you didn’t have an individual cap,” Noll said in a telephone interview, “if LeBron James was in a position to sell himself to the highest bidder, his salary would be much higher and you wouldn’t have a small number of top teams with more than half the superstars in the league.”

“It’s a big mistake, and the N.B.A. hasn’t adjusted. So if you have as many as 25 teams that know before the first game is played that they probably won’t even be in the conference finals, doesn’t that make the regular season seem almost meaningless, more of an exhibition than a pathway to a consequential championship?”

I couldn’t agree more (and plan to write more on this in the NBA off-season).

In previous posts, I’ve argued against the randomness and extreme parity of college basketball, much of which is due to the 68-team single elimination playoff format—a huge contrast to the NBA’s best-of-seven series. If that sounds like the complete opposite of my above criticism of the NBA, then you’re right. The NCAA lies too far toward the extreme randomness/parity end of the spectrum, while the NBA goes too far to the extreme predictability/inequality end.

You don’t want seasons to feel like they’re extremely predictable like the NBA, but at the same time, you want the true best team to win with a relatively high frequency so that it feels like teams are justly rewarded. It’s critical to have a good balance between the two, and finding that equilibrium is something sports leagues have wrestled with—and will continue to wrestle with—for decades.

 

Advertisements

NHL Playoffs Statistical Preview

The stats love the Kings.

The stats love the Kings.

(Note: From time to time, I’ll take a break from proposing new sports ideas. This is one of those times.)

While the hockey stats revolution is still about a decade or so behind baseball, there are still some pretty telling stats in our toolbox today. My favorites are Corsi, Fenwick, and PDO.

Here’s a description of each from the awesome Stats.HockeyAnalysis.com:

  • Fenwick = Shots + Missed Shots
  • FF% = Fenwick For% = Fenwick For / (Fenwick For + Fenwick Against)
  • Corsi = Shot Attempts = Shots + Missed Shots + Blocked Shots
  • CF% = Corsi For% = Corsi For / (Corsi For + Corsi Against)
  • PDO = Shooting Percentage + Save Percentage while on the ice

There’s obviously a lot of overlap, but we’ll just ignore that for now. The main takeaway is that possession and shots are extremely important.

I compiled the FF%, CF%, and PDO on all 30 NHL teams from the 2013-14 regular season and standardized them into a z-score. For Fenwick and Corsi, a higher z-score is good. For PDO, a lower z-score is good.

I averaged these figures up in the “Total” column ((FF + CF – PDO) /3) and sorted them by that metric. To compared them to the actual regular season results, I also added the regular season standings points and goal differential.

Team FF% FFZ CF% CFZ PDO PDOZ Total Points Goal Dif
Los Angeles 56.1 1.83 56.8 2.01 1000 0.00 1.28 100 32
New Jersey 53.6 1.07 54.4 1.29 985 -1.33 1.23 88 -11
Chicago 55.4 1.62 55.5 1.62 999 -0.09 1.11 107 47
San Jose 54.6 1.38 53.7 1.09 998 -0.18 0.88 111 49
Florida 50.8 0.23 51 0.28 980 -1.77 0.76 66 -72
NY Rangers 52.6 0.77 52.4 0.70 997 -0.27 0.58 96 25
St. Louis 53.7 1.10 53.1 0.91 1007 0.62 0.46 111 57
Vancouver 51.6 0.47 51.3 0.37 995 -0.44 0.43 83 -27
Ottawa 51 0.29 52.4 0.70 997 -0.27 0.42 88 -29
NY Islanders 49.1 -0.28 49.4 -0.19 984 -1.42 0.32 79 -42
Detroit 51.5 0.44 51.5 0.43 1000 0.00 0.29 93 -8
Winnipeg 50.5 0.14 50.1 0.02 994 -0.53 0.23 84 -10
Carolina 49.6 -0.13 50.3 0.08 994 -0.53 0.16 83 -23
Nashville 49.4 -0.19 48.5 -0.46 989 -0.97 0.11 88 -26
Dallas 50.9 0.26 50.5 0.14 1002 0.18 0.07 91 7
Tampa Bay 51.3 0.38 51 0.28 1007 0.62 0.02 101 25
Boston 53.4 1.01 53.9 1.15 1025 2.21 -0.02 117 84
Phoenix 50 -0.01 50.5 0.14 1003 0.27 -0.05 89 -15
Philadelphia 49.2 -0.25 50 -0.01 1003 0.27 -0.18 94 1
Pittsburgh 49.2 -0.25 48.7 -0.40 1001 0.09 -0.24 109 42
Calgary 47.7 -0.70 46.3 -1.11 988 -1.06 -0.25 77 -32
Columbus 49.8 -0.07 49.9 -0.04 1008 0.71 -0.27 93 15
Minnesota 48.8 -0.37 48.6 -0.43 1010 0.89 -0.56 98 1
Washington 47.1 -0.88 47.7 -0.69 1002 0.18 -0.58 90 -5
Montreal 47.9 -0.64 46.7 -0.99 1005 0.44 -0.69 100 11
Anaheim 50.1 0.02 49.8 -0.07 1024 2.13 -0.72 116 57
Edmonton 44.1 -1.78 44.3 -1.70 990 -0.89 -0.87 67 -67
Buffalo 42.8 -2.17 43 -2.09 982 -1.59 -0.89 52 -91
Colorado 46.7 -1.00 47 -0.90 1018 1.59 -1.17 112 30
Toronto 42.3 -2.32 42.9 -2.12 1013 1.15 -1.86 84 -25

 
It’s not hard to see why Toronto and their notoriously anti-analytics front office fell off the rails, but how did Colorado manage to achieve a No. 1 seed in the West? (Answer: Unsustainably over-achieving goaltending from Semyon Varlamov.)

For the sake of comparing these stats to regular season results, here’s how Total looks plotted against goal differential:

playoff data 2

(Sorry some of those are cut off. The far left is Toronto, and the far right are Los Angeles and New Jersey.)

The correlation is only .1845, which suggests that these metrics should probably only be used as a guideline in telling us what teams may have over- and under-achieved this season relative to their actual performance.

First round match-ups

Ok, so how do these teams’ stats compare for the first round match-ups?

Home Ice Total Road Ice Total Difference
Colorado -1.171 Minnesota -0.512 -0.66
St. Louis 0.106 Chicago 0.905 -0.80
Anaheim -0.575 Dallas 0.289 -0.86
San Jose 0.926 Los Angeles 1.233 -0.31
Boston 0.101 Detroit 0.356 -0.25
Tampa Bay 0.018 Montreal -0.377 0.40
Pittsburgh -0.200 Columbus -0.483 0.28
NY Rangers 0.753 Philadelphia -0.244 1.00

 
In the West, every matchup’s lower seed has a higher total, which suggests that upsets could be rampant. Most notably, Minnesota and Dallas have a considerable edge over the two No. 1 seeds. And whoever emerges from the San Jose vs. Los Angeles showdown (two of the three highest Totals in the playoffs) should be a serious contender to make the Stanley Cup Finals.

In the East, it’s a bit more chalk, but again, a No. 1 seed (Boston) has a lower total than its opponent (Detroit).

Does this mean all the No. 1 seeds are going down? No, probably not. But it does indicate they’re very susceptible to being upset—not just in the first round, but in every round of the playoffs.

Vegas odds

Finally, let’s take a look at how these Total figures for each team compare to Vegas. My (extremely unscientific and non-statistically sound) formula for a stat I call “BetFigure” is:

BetFigure = [(Total Differential + 2.5) * 0.2] / Vegas Win Percentage

From my (again, unscientific) experience, anything over 1.50 is generally a profitable bet.

Team Line Win Pct Total Opponent Total Total Diff BetFigure
Detroit 235 29.9% 0.356 Boston 0.101 0.25 1.85
Dallas 165 37.7% 0.289 Anaheim -0.575 0.86 1.78
Columbus 210 32.3% -0.483 Pittsburgh -0.200 -0.28 1.37
Minnesota 115 46.5% -0.512 Colorado -1.171 0.66 1.36
Chicago -105 51.2% 0.905 St. Louis 0.106 0.80 1.29
Los Angeles 125 44.4% 1.233 San Jose 0.926 0.31 1.26
NY Rangers -150 60.0% 0.753 Philadelphia -0.244 1.00 1.17
Tampa Bay -117 53.9% 0.018 Montreal -0.377 0.40 1.07
Montreal -103 50.7% -0.377 Tampa Bay 0.018 -0.40 0.83
Pittsburgh -250 71.4% -0.200 Columbus -0.483 0.28 0.78
San Jose -145 59.2% 0.926 Los Angeles 1.233 -0.31 0.74
Philadelphia 130 43.5% -0.244 NY Rangers 0.753 -1.00 0.69
Colorado -135 57.4% -1.171 Minnesota -0.512 -0.66 0.64
St. Louis -115 53.5% 0.106 Chicago 0.905 -0.80 0.64
Boston -280 73.7% 0.101 Detroit 0.356 -0.25 0.61
Anaheim -190 65.5% -0.575 Dallas 0.289 -0.86 0.50

 
So there you have it, Detroit and Dallas. Take it to the bank. Guaranteed*

*Maybe