Back in August, several statistics-based websites posted predicted NFL win totals. A predicted win total represents the number of cumulative wins predicted for each NFL team. These numbers can be easily compared to the Las Vegas line for each team (I used the ones set by sportsbetting.ag), and end-of-season results, to determine if these predictions are worth our time, or, in some cases, our money.
In this post, I attempt to figure out how accurate these predictions were, and to determine if any outperformed Las Vegas’ predictions.
Here are the sites I used: (if there are others that you’d like to compare, please send them along!)
Football Outsiders: I used the ones included in their preseason almanac, which are fairly close to the ones linked here.
Team Rankings: While, to the best of my knowledge, there is no available URL for these predictions, a loyal reader (@spidersvasports) sent these along.
Accuscore: Same as Team Rankings
Prediction Machine: Linked here.
Peter King (MMQB): Okay, fine, this one’s not a stats-based website, but I figured I’d throw in picks made by the league’s most notable reporter.
Next, here are my metrics
MAE: Averaged absolute error between the prediction and the win totals (lower is better). This represents the average number of wins that each site’s prediction was off by.
ROI %: This represents the percentage gain (or loss) from an investment using the Las Vegas line, and is directly tied to both the Vegas predicted win totals and the odds on each total. For example, if someone picked Carolina over 7 (-145), the return on a $100 bet would be $68.97, as the Panthers won 12 games.
Also, I used the difference between the Vegas line and each site’s predictions to generate a list of “Top-5” and “Top-10” bets for each site. The Top-5 picks for a site, for example, were the ones with the largest absolute differences between that site’s projections and the Vegas projection. (Note: Many bettors might use different strategies to figure out their favorite bets, but largest absolute differences seemed like one possibility)
Here’s a table of our metrics for each site.
Team Rankings (TR) boasts the best performance, both in terms of closest predictions (MAE = 2.08) and ROI %. For example, placing $100 on TR’s top 5 picks yielded four winners (unders on Atlanta, Chicago, Houston, and Washington) and a 52.5% profit. In terms of accuracy, Prediction Machine (PM) also out-dueled Vegas as judged by MAE, but one needed to bet all of PM’s sides in order to see a profit (the site allows for a more elaborate description of strategy here).
Peter King and AccuScore appeared to lag behind their peers, at least as judged using these metrics. On average, each boasted predictions that finished nearly two and a half wins away from each teams eventual results.
Here are each site’s picks, depicted in unique colors and symbols. I ordered teams by Las Vegas predictions (low to high, shown by red asterisks), in order to make it clear which teams outperformed or underperformed expectations. Each team’s final 2013 win total is shown in black (enlarge for detail).
Teams mostly nailed by the stats-sites included, among others, Houston (under) and Arizona (over). The sites had more difficulty predicting the win totals of Oakland, Washington, and San Diego. Peter King, while whiffing on the Jets and the Chargers, outperformed stat-based sites at predicting the strong seasons of Cincinnati, Indianapolis, and New Orleans.
What to conclude?
The stats-sites, for the most part, performed well. In October, I ran that several sites had a more difficult time with MLB win totals; results from the 2013 NFL season were, for the most part, stronger.
Before finishing, it’s important to note that I extracted Vegas lines towards the end of August, which is relatively late in the game, so-to-speak. For example, the Panthers bet (over 7 wins, at -145) may have opened up closer to -115 or -125, in which case a more aggressive bettor could have turned a stronger profit. While this could have effected each site’s ROI%, the MAE values would obviously not be impacted.