Finding Player Upside by Predicting Fantasy WAR (Fantasy Football)
In recent years, the Wins Above Replacement, or WAR metric has become quite popular in the fantasy football setting. It tells us how many more fantasy wins a player would give you during the regular season than a replacement-level player, or someone you could likely pick up off of waivers. You can read more about this metric from its methodology on Pro Football Focus, the work of Jeff Henderson on FantasyPoints.com, or on our site, On the WARPath: Understanding Value Above Replacement.
In this article, I will explore a model I programmed to predict year-to-year fantasy WAR. All the WAR calculations you see are done for a 1 QB, 2 RB, 2 WR, 2 FLEX, 1 TE, and 1 D/ST squad. Calculations also are based on a deeper replacement level, meaning the model expects the top ~30 positional players to be rostered.
The model was trained on three years of fantasy data, and tested on one. I have plotted the testing results below. The graphic shows the ‘Expected WAR’ (xWAR), or my model’s prediction, vs. the ‘Actual WAR,’ or the WAR the given player earned in the predicted year. Points closer to the line show more accurate predictions.
When we break up these results by position, it is very apparent that the model is more accurate for TEs and WRs (and FBs, I guess). QB and RB predictions are still accurate, but a bit more random than the others which are pretty spot on.
On average, the model was 1.27 WAR off on each prediction, where most of the variation came from the extremes–not too shabby. If we say a prediction that lands closer than 1.27 WAR from the actual value is correct, the model correctly predicted 63.7% of its picks, which is a great number to be at.
Breaking it down further, we find 84.6% of predictions avoid big errors, where the xWAR is negative and the actual WAR is positive, or vice-versa. These types of errors can be dissected even more:
- Missing out on talent: xWAR is negative, but the player actually finishes with a positive WAR
- Costly Move: xWAR is positive, but the player actually finishes with a negative WAR
94.5% of the model’s predictions avoided costly moves and 90.1% of its predictions avoided missing out on talent. Excellent!
Now that we know the model is accurate, let’s take a look at its predictions for the 2023 season! Keep in mind the algorithm doesn’t account for player moves/acquisitions–you may even see some old legends in there since we can never really count out a Tom Brady comeback!
Below you will find the complete rankings in xWAR for the upcoming 2023 season. Here are some key takeaways from the rankings.
- Daniel Jones (2.66 xWAR) & Trevor Lawrence (2.13 xWAR) are projected to improve next season. The model typically projects regression from QBs, so keep an eye on these two.
- Najee Harris (1.52 xWAR) has a positive change in projection despite a rocky 2022 season. Look for improvements out of the third-year back.
- Noah Fant (1.17 xWAR) is projected to make big progress in 2023. After a negative-WAR season in 2022, the model predicts he will improve by over 1 WAR!
- Zamir White (0.14 xWAR) has taken the largest leap in projected WAR (+3.84). The second-year back could settle in nicely behind Jacobs in the Raiders’ backfield and may be a good risk to take in deeper leagues.
- Kyle Pitts (-0.63 xWAR) is projected to decline in production – not a great look for the second-year TE, who was a replacement-level player in 2022 (0 WAR).
That’s all for this article, feel free to reach out on Twitter @analytacist if you have any questions about this model!