Sports Info Solutions Analyzes Team Over/Unders

SIS has a lot of good proprietary data which they've used to run simulations of the season. their methodology looks pretty similar to vegas' to me, though they obviously have no need to balance a book. this model's record last year looks pretty good, though if these were real vegas lines I dont think I could bet the vegas (10.4), NO (10.7), and chicago (10.7) unders fast enough.

Quote:

We’ve reached prediction season for the NFL. A big part of dissecting the upcoming season is analyzing the schedule and predicting which teams will overperform or underperform their expectations. Expectations these days come from the sportsbooks, especially through win totals. These markets have been posted and bettors have begun forming their stance on each individual team. The lines will move all offseason long, as more information and opinions will move the lines in one direction or another. At SIS, we have created a pre-game prediction model to predict the spreads, totals, team totals, and moneylines of each game using a multitude of data points that we collect. We also constructed the model to be able to run for the entire season, allowing us to run simulations on the season and create win total prices of our own. The model went 18-13-1 in over-under predictions on team’s preseason win totals last season.

Giants perform slightly better than vegas in their model at 7.1 but more interesting is their most underperforming team relative to vegas is Philly at 6.8 wins and almost 4 wins lower than the vegas o/u.

Quote:

Team DraftKings Win Total Model Projected Win Totals Difference (sorted from most overperforming teams to least) Raiders 6.5 10.4 Over 3.9 Saints 7.5 10.7 3.2 Bears 8.5 10.7 2.2 Lions 10.5 12.3 1.8 Buccaneers 8.5 10.3 1.8 Browns 8.5 9.8 1.3 Packers 9.5 10.7 1.2 Seahawks 7.5 8.7 1.2 Jets 9.5 10.6 1.1 Patriots 4.5 5.5 1 Ravens 11.5 12.1 0.6 Bills 10.5 11.1 0.6 Giants 6.5 7.1 0.6 Cowboys 10.5 11.1 0.6 Rams 8.5 9 0.5 Cardinals 6.5 6.9 0.4 Bengals 10.5 10.9 0.4 Texans 9.5 9.4 Under -0.1 Panthers 5.5 5.2 -0.3 Dolphins 9.5 9 -0.5 Chiefs 11.5 10.9 -0.6 Broncos 5.5 4.7 -0.8 Steelers 8.5 7.4 -1.1 Jaguars 8.5 7.4 -1.1 Falcons 9.5 8.1 -1.4 Chargers 8.5 6.9 -1.6 49ers 11.5 9.8 -1.7 Colts 8.5 6.1 -2.4 Vikings 6.5 4 -2.5 Titans 6.5 4 -2.5 Commanders 6.5 3.2 -3.3 Eagles 10.5 6.8 -3.7

Quote:

Model Overview (How did we get these numbers?) Our model utilizes 18 features that are a combination of game, team, and player level metrics. Game level metrics include whether or not the game is in the postseason, weather predictions like wind speed and precipitation probability, and a home team indicator. Some of the team level metrics include weighted points for and points against averages, both offensive and defensive weighted penalty yards, and some possession metrics in the form of snaps per game or drive. Each of these metrics is computed for the team’s past 7 games, with the most recent weighted more heavily. We use a weighted 16-game average with our Total Points for the player metrics. We calculate the average Total Points for each skill (Passer, Rusher, Pass Rush, etc.) for every player per snap, and then aggregate the projected values for each game to the team level using each player’s projected snap counts for the game. Then, we utilize Monte Carlo simulation to illustrate variance in player performance by simulating 1,000 games for each player. After each simulation, all of the metrics are incorporated into a Lasso regression model to predict the team’s point total. The distributions of the predictions are aggregated to an average and spread so that point estimates and alternate point estimates can be drawn from the distribution. Now that there are lines for each game, we can simulate 1,000 regular seasons based on the moneyline output and take the average win total for each team.

Analyzing NFL Win Totals with the SIS Prediction Model - ( New Window )