The following is an excerpt from the Ed Miller and Matthew Davidow book Interception: The Secrets of Modern Sports Betting, from a section on practical angles bettors can use when they take on the extensive betting menus offered at most modern books.
Misalignment Between a Sport’s Rules and the Model’s Rules
This category is somewhat similar to the last one, but it’s more pure modeling error than miscommunication. Leagues often tweak the rules of a game. Recently, MLB added a pitch clock and changed the size of the bases. NFL changes kickoff and overtime rules semi-regularly. College football changes the rules about when clocks run and stop. College basketball moves the three-point line. And so on.
Furthermore, sometimes leagues switch rules up within a season. In MLB, every extra inning starts with a runner on second base. Except not in the playoffs. The NFL’s overtime rules are different in preseason, during the regular season, and in the playoffs. In the past, MLB has ended the second game of a double header after the seventh inning rather than playing a full nine.
Content companies must build and maintain models for these sports. It’s their job to stay on top of these rule changes and adjust their models as needed. But they often miss rule changes. Especially these little ones that switch back and forth within a season.
For example, just prior to this book’s publication, we were watching a Browns-Eagles game during the NFL preseason. The Browns led 18-10 in the fourth quarter, and we noticed a large modern sportsbook offering a three-way line. You could bet Browns, Eagles, or tie. The book had applied an extremely heavy 30 percent overround to the market. (Overround is a mathematical shorthand useful for quickly estimating the sportsbook’s hold in multiway markets. You convert the odds to break-even percentages and add them up. The amount that number exceeds 100% is the overround. So in this case, the break-even percentages of the three options summed to 130%. Not good for the bettor.)
Despite the extreme hold the sportsbook put on the market, however, they were not safe. There was still a good bet among the three! They were offering the tie at +1400.
This would be a terrible bet using the NFL’s regular season overtime rules, which no doubt the content company responsible for the market was using in their model. But in the NFL preseason, they play no overtime. We priced a tie at about 9 percent, making +1400 quite a good bet. As it turned out, the Eagles scored a touchdown and two-point conversion, and the game ended 18-18.
The runner on second base in extra innings rule change remains today a bugaboo for models pricing baseball totals. Not only does the rule get repealed for the playoffs, but we’ve noticed that many models in use in the market today simply don’t yet fully account for it. This is because many models used to price baseball derivatives have two flaws. First, they are simply out of date—again the business incentives in this industry are always to build more, more, more. New, new, new. It’s not to go back and make sure what you’ve already built is correctly priced. (The incentive to make sure your models are correctly pricing things in the first place often doesn’t exist, let alone to update it with minor rule changes.)
Second, even when updating models, often model builders will rely too heavily on past data. Instead of simulating a game accurately, applying all the most current rules, the model will rely instead on aggregating data from the results of past games. To the extent that models incorporate data from games before this rule was instituted, they will be inherently inaccurate.
Concretely, say you’re a content provider in charge of making baseball totals. From the market making sportsbooks, you can source a moneyline, run line, and market total. Your job is to build a model that uses these inputs to produce prices for every possible total from 3.5 to 23.5.
Broadly speaking, there are two approaches. One, you can just look at past games with similar moneylines and totals and then count how many games finished with each final score. You then just make your prices by assuming the percentage of game that lands on each number represents the correct break-even percentage.
Or you can try to build a game simulation that uses the current rules and players.
It’s much easier to just count up the results of past games. (To be clear, we’re oversimplifying the modeling process here to make the point.) When you count game results, you’re likely including data from both before and after the rule change. Thus, your answers will correspond to some hybrid rule that doesn’t exist—not quite right for either actual ruleset.
The upshot of this is that we currently see in-play MLB models get the totals wrong in games likely to go to extra innings. It’s much more likely in games that use the runner on second rule that a given inning will end exactly 1-1, thereby extending the game while adding 2 runs to the game total. So the higher alternate totals are often good bets as they are priced as unlikelier than they actually are.
In 2006, there was a seemingly small college football rule change that had a major modeling impact. In the previous season, the clock would not start after a change of possession (a punt or turnover) until the first snap of the new drive. For the new season, they would start the clock as soon as the ball was spotted.
This likely seemed like a small change to most people, and many people likely didn’t adjust their models. But this rule change has an outsized impact in the most critical part of the game—protecting a lead late game. Under the new rules, a team leading by one score with, say, two minutes remaining could punt the ball, and the receiving team would lose precious seconds between the spot and their first snap. This made punting in these situations more strategically attractive, and therefore made playing conservatively on the previous drive more attractive as well.
This led to lower scoring second halves as leading teams were more inclined to nurse their lead and try to bleed out the clock. We bet under in the second half in at least 75 percent of the games that year and won almost 60 percent of them (no doubt we ran good along the way). While we noticed the rule change and thought it might be important, it was our friend Jan Suchanek (who went by perpetualczech on gambling social media and internet forums) who realized exactly how important the change would be, and who pushed us to lean into the angle to the maximum from the beginning of the season.
Jan was particularly good over a very long time at picking up on angles like this, and he was generous in sharing his insights with us throughout the years. Unfortunately, Jan passed away recently, and we wanted to share this story here to memorialize him and his many contributions to sports modeling and betting. Our readers who believe in eternal energy will understand.
Little rule changes like this one happen all the time, and there will always be opportunities to be one step ahead of the modelers to identify the most critical impacts the change will have on the game and therefore exactly which bets are most likely to be affected.
Ready to talk Interception? Join us on Discord and look for the channel #book-club.