How Well Does Our Grade Predict Future Success?
In our last article we predicted how well a player would perform based on his stuff. This was done by predicting the SwStr% for each pitch type he threw and then taking his fastball, best breaking ball and changeup and predicting his xFIP. Other variations of this model considered a player’s walk rate and how often they threw the pitch to project their xFIP. The purpose of wanting to project xFIP based on stuff would be to allow us to project players when they had limited amounts of stats, but enough data on their arsenal that we could try to get an idea of how they’d perform.
In this article we will look at how well the player’s predicted SwStr% on each pitch type predicted the following years xFIP. When taking people who had three pitches and at least forty innings pitched in back to back years, I wound up with 691 pitchers. The best model again turned out to be the one considering the pitch types with how often they were used and BB%. That one had roughly a .12 R^2 and a 0.85 RMSE and 0.68 MAE. The model considering just pitch types had about a .05 R^2, while the one considering usage and pitch types without BB% had about a .09 R^2.
While this could still be useful, especially when faced with limited data, it isn’t as high as I was expecting. More could be done to improve these numbers, such as factoring in age, but the goal here was just to see how well their arsenal predicted their xFIP.
The model’s biggest misses for both projecting current year and following year xFIP do seem to be on the extremes of players who performed poorly or very well though. This makes sense and in the next article we’ll look at the model’s biggest year to year misses and what may lead to people beating their arsenal projection.
A link to following year projections and the amount they missed by can be found here:
You can also view 2020 Predictions at the same link: