Another Everyman's Algorithm
Editor’s Note: This is the first (of hopefully many) guest contributor posts. Hope you enjoy!
I am not a professional sports bettor. I think it sounds like a romantic career concept, but even if you have a process that can beat the books, they don’t have to let you – they can cut you off if you’re too good at taking their money. And sports betting is a billion-dollar industry. It’s profitable for casinos and online sports betting sites; when the books take a bath, it’s newsworthy because it’s unusual. But when you take into account a vigorish of -110, the breakeven point for the bettor is 52.38%. Can you win 52.38% of your bets against the spread? Of course! Right? Well maybe it’s not as easy as it sounds, so for now I spend too much time on the NFL mini super contest with Chris.
I am not a gut picker. I hear about bettors who make picks based the eye test or what their gut is telling them. Three previous winners of the prestigious Westgate SuperContest were briefly profiled, and only one of them was using a statistically-driven process! I shamelessly thought I could easily use statistics to correctly predict who would cover the spread. But I quickly learned that sports betting is analogous to stock picking – the market is pretty efficient. If there were such statistics, there would be no sure profit for the house. In his book A Random Walk Down Wall Street, Burton Malkiel said that if you think you see a $100 bill on the ground don’t even bother to reach down and pick it up. If were there, someone else would have already seen it and scooped it up before you. This analogy explains that there is no easy money picking stocks, and I think the same is true for sports betting. Sure some people get in early on the next hot stock, and some people hit a big parlay. But can they do it consistently? Where is the inefficiency? Where is the edge? And critically, where can it be found repeatably and reliably?
Some people scuba dive for sunken treasure, others hunt for lost gold. Some people apparently win the Las Vegas SuperContest making picks with their gut. I search for an algorithm to make NFL picks. My goal is to create a process that can consistently make picks above 60% against the contest spread, which will allow me to take those picks without letting the narratives in my head and in the media challenge them. For example, in week 2, my process predicted the Jacksonville Jaguars, as road underdogs, would cover the spread against the Tennessee Titans. My gut said no way - Jacksonville, despite the Week 1 win, was still regarded as potentially the worst team in the league, while Tennessee was still riding high from last year’s performance and regarded as a top 10 team. And the public was betting against the road underdog at a 2:1 clip. But Jacksonville did cover, just like the algorithm predicted. Maybe those narratives aren’t the right signals, and my process had a better idea? That’s the goal – to create an algorithm that allows me to get out of my own way.
My overall algorithm consists of combining multiple smaller algorithms, which I refer to as signals, into a final selection of five picks. The signals come from three groups: wisdom-of-the-crowds, statistical performance, and market movement. The wisdom-of-the-crowds approach looks at power rankings across the internet and converts those rankings into spreads. One of the statistical signals uses a particle-swarm-optimization machine learning algorithm to predict the spread based on assigning weights to a set of key performance statistics. One of the market movement signals looks at lines across multiple online betting sites, some considered to be sharp and others considered to be square and compares those spreads against the contest spread. In the end, I combine the top picks from each signal to arrive at five final picks each week.
I review the market movement information all the way through Sunday kick-offs, so my process doesn’t lend itself to posting my final picks ahead of time. Here is a snapshot of my Week 4 dashboard on Wednesday, 9/30. One of the five picks it suggested was Miami against Seattle – I really hoped the market-movement signals would move off Miami, because I didn’t want to go against #LetRussCook!
One day later, after updating some performance and market information before Thursday’s kickoff, the dashboard looks a little different. I tend to avoid Thursday games unless a lot of signals are aligned. In this case, there are a lot of signals in conflict (so I will not make a play), but they lean in favor of Denver. I didn’t want to play this game anyway, with Denver’s performance completely unpredictable due to starting a 3rd string quarterback and the dumpster fire that is the New York Jets.
Finally, we get to Sunday morning. After updating the schedule to remove the PIT-TEN game and accounting for the fact that New England would be without Cam Newton against Kansas City, and updating the market movement signals, this is what the dashboard looked like before kickoffs:
I listened to the algorithm and made the 5 picks it suggested – JAX, MIN, SF, MIA, KC. The most painful was Miami; I definitely didn’t want to go against Seattle. Unfortunately, I was right about that one, and the Top 5 picks went only 2-3 for a 40% week. However, there is a silver lining - out of the 13 games that the algorithm made a pick, it got 8 correct and pushed one for 8.5 points. That’s 65% out of 13; it sounds like I need a better way to figure out which picks should be in the Top 5!
Through 4 weeks, this is the overall algorithm performance:
The algorithm has 12 contest points, right on the 60% goal. However, my actual entry in the contest only has 10 points because I couldn’t resist overriding the algorithm a couple of times already this season…and we see how that turned out. The contest leader has 15 points for 75% through 4 weeks! My entry with 10 points is tied for 80th, while a 12-point entry would be tied for 33rd. The evidence seems clear – I need to stick with my algorithm’s picks and hope its Top 5 picks start outperforming the overall picks as expected so I can get some 4s and 5s to catch up in the contest!