Welcome to our NFL game predictions page. Every week you will find our estimated win probability and expected spread (Home EPR Point Spread) for the upcoming games. We will also keep a running tally on how our model has performed compared to the betting markets, specifically evaluating it against the closing spread. We will only count the predictions where the model meaningfully differs from the market, which we’ve defined as when the EPR advantage differs by 2 or more from the spread (i.e. a team has an EPR advantage of 5.5 but is only favored to win by 3 according to the spread). These matchups will be indicated in the table by highlighting in yellow the team that we are expecting to “cover” the spread. The Vegas spread is sourced from Pinnacle via Lee Sharpe’s nflgamedata.com. Lines are pulled on Wednesday morning and may move prior to the matchups.

The win probabilities are calculated using each team’s EPR, home-field advantage, and rest advantages. Each team’s EPR is a prediction of how many points we’d expect that team to beat an average team by on a neutral field, so to come up with the expected spread we take the difference in the two team’s EPR and adjust for home-field advantage (which is no longer a factor for the 2020 season as of Week 10) and rest advantages. You can read more about EPR and the model here. You can also find our playoff projections piece here, which shows each team’s probability of making the playoffs  and advancing to each subsequent round (including winning the Super Bowl). 

Week 15 performance against the spread: 3-3

Season long performance against the spread: 57-51-2

Some very interesting research was published last week by FiveThirtyEight. The piece is largely about college football, and how unpredictable it has been this year, but has some analysis on the NFL betting markets as well. It turns out that at least so far, NFL favorites have only covered the spread on 45 percent of games this year, the lowest percentage since 2006. We often hear the phrase, “Any given Sunday” to describe how unpredictable the NFL can be. But that phrase has a cousin, with arguably larger implications: “Any given year.”

When building a model, we test its performance on previous years to ensure it’s worth its salt. And EPR is! You can learn more about why I’m confident in the process here. However, it’s very difficult to predict which assumptions we typically believe to be true (home-field advantage, favorites covering 50 percent of the time) are going to deviate from the norm in a given year, and by how much. Sometimes these just deviate for a year and bounce back to normal, sometimes they represent changing trends. We don’t know until we have more data. This is what makes sports modeling an always challenging, interesting problem. It is also what can make it frustrating — as we’ve seen this year, with EPR just above the break-even point despite performing better in my testing on previous years.

I trust the process. Play-by-play efficiency data is an incredibly powerful tool that allows us to accurately predict team strengths. Not every year will be a home run, and 2020 was always going to be a strange, unpredictable year, but I’m not going to rest until The Athletic is home to the best team strength model on the internet. I hope you all continue to follow the journey.