Quick note — in case you missed it on Saturday, most of the world watched Spanish soccer kings Barcelona and Real Madrid play “El Clásico” one more time. This one was particularly interesting because Madrid was ahead in the standings (a rarity lately) and riding a 15 game winning streak. However, Barcelona continued their dominance of José Mourinho (Real’s enigmatic, controversial coach), and won 3-1. Real hasn’t beaten Barcelona in La Liga since 2008 — it’s getting embarrassing. If you have further questions about the game (or about American progressivism), I direct you to my brother Conor, who studied in Barcelona for a year and is probably their biggest fan in the United States.
However, this post is really about football gambling, and by that I mean AMERICAN football gambling. I hear more about pick ’em contests (against the spread) then I do about fantasy football these days. I understand the draw —fantasy will take you down strange roads irrelevant to the outcome of the game. Pick ’em leagues make you cheer for meaningless garbage covers, but at least your focused on points, which is what really matters.
Football gambling is particularly interesting because there are so few games. Limited data means limited room for prediction based on past experience (whether informal or statistical). Even if you are a joker who never watches football, you will do ok in your pick ’em league. No one will get a huge advantage from watching games or digging up stats, unless they do it as seriously as a professional handicapper. Most people are just guessing, which makes it fun, since everyone has a shot (for further evidence supporting the “randomness is fun” theory, consider March Madness).
I’m interested in picking against the spread precisely because it’s difficult. I am awful at it. Despite what I just wrote, I would probably never win a friendly pick ’em league. Instead, I probe for statistical approaches. A couple years ago, I decided to look at whether different lines have different cover rates (fair warning that this has nothing to do with estimating team quality — this is a pure math exercise). I thought that the nuances of NFL scoring (sevens and threes) might generate weird behavior. In the graph below, you can see that NFL margins of victory have an odd distribution.
There’s one bar for each margin of victory from -42 to 59 (poor Titans). From the perspective of the home team, every point difference between -38 and 42 has occurred as least once. However, the distribution doesn’t look smooth at all. There are spikes everywhere. Over 15 percent of the time, the margin of victory is exactly three points (one way or the other).
But a lumpy distribution is not enough to generate betting opportunities. Vegas lines are extra sticky around “key points” because those values are more likely to occur. And when a line is sitting on -1, maybe all the three point losses balance out the three point wins that would cover for the home team. Still, it’s important to remember that Vegas doesn’t care whether favorites cover 50% of the time. They just want even money on both sides of the line. I don’t think most people visualize lumpy distributions, which might push lines to places where the NFL’s weird scoring can bite.
Here’s an example for clarification. A line of -2 signals that the favorite is a better team. Since teams are much more likely to win by 3 points rather than 1 or 2, and the better team usually wins, I would expect the favorite to cover -2 more often than not. The table below spells out the relevant probabilities since 2002 (through week 13 of 2011):
The line data is taken from Warren Repole’s database (final lines), and I give the favorite half credit for a push. Since there are very few lines over 14, I excluded them. For reference, 55% is a reasonable rule of thumb for a money-making gamble.
I was pretty excited when I first looked at this. As I suspected, lines from -1 to -2.5 tend to be covered by the favorite. The small distance from -7.5 to -9.5 looks similar. If you squint, you can convince yourself that the same is true (on average) between 3.5 and 6.5. That range is a little bigger, so I’m not surprised that the outcomes are less consistent. On the key points (3, 7, and 10), the favorite struggles, probably because they don’t quite get over the hump and end up pushing.
Even though my theory holds (some) water, I nearly missed the forest for the trees. Above 10, I found the truly poisonous lines; the cover percentages in this range are abysmal. There are over 200 lines between 10 and 14 since 2002 (granted, 64 of them are at 10), so sample size is not a problem. The percentages are pretty stable over time, too. Basically, all high lines are too high. It’s well known that the public bets too frequently on favorites. What surprised me is that sharps aren’t pushing these lines back down by heavily betting underdogs. The advantages from my lumpy points theory are smaller so perhaps not worth attacking. Taking dogs to go against the public is a pretty old school gambling technique, but amazingly it still seems to work.