FanPost

ERA TEST

**Glad to share the Excel Doc if anyone wants.

I decided to do a little experiment following one of those recurring debates about ERA vs. xFIP and other advanced stats. I got to wondering if the advanced stats could be more accurately predictive if you added ERA to their equations. I decided to do this in a very simple way--by weighting the advanced stat (.75) and ERA (.25) to see if this prediction is more reliable than the advanced stat on its own.

This is a small sample size. I just decided to work with the three current Braves starters who started the last three seasons and the legendary three: Smoltz, Glavine, Maddux. And I only looked at three years to predict the next three. In the case of the legends, I looked at the first three years xFIP existed (2002-2004). So I don't know how these trends hold up across baseball history, but I think 4 years of these six pitchers is a decent sample (elections are predicted on like 1% of the data). Either way, it's a relevant relative frame for Braves fans.

I didn't weight the previous three years in any complicated way--just 2 for the previous year, 1.5 the year before that and 1 the third year back. I then subtracted the prediction from the actual ERA and took the absolute value(s).

After doing this for the six individual pitchers, I created a top sheet that averaged all the differences of all the pitchers.

What I found is that the previous three ERAs were a better predictor of the next year's ERA than the previous three xFIP's. That having been said, xFIP was a better predictor for Folty and Smoltz--the power pitchers. (Also, all three were very close and very accurate for Keuchel.)

I know the answer I found on ERA accuracy over xFIP accuracy was not supposed to happen, and I expect I'll be told that it wouldn't be the case if we added more pitchers.

However, I'm curious what would happen if you subtracted power pitchers from the sample -- how much more predictive would ERA be of that pool than it is when you include power pitchers? I'd predict that trend would hold.

What I'm relatively sure of is that these models should be relative to pitcher-profile--like the Parallax view of Baseball Statistics.

It also seems very likely that we are overvaluing power pitchers, because we are overvaluing the models that better predict power pitcher success. Those models won't predict non-power pitchers to have success, but that doesn't mean they won't have more success than those models predict, because those models are better predictors of power-pitcher success.

Unfortunately I don't see an easy way to do this with 100 power pitchers and 100 non-power pitchers to see if it holds up.

TEHERAN ERA FIP xFIP FIP-ERA xFIP-ERA
2016 3.21 3.69 4.13 3.57 3.44
2017 4.49 4.95 4.96 4.84 4.61
2018 3.94 4.83 4.72 4.61 4.14
2019p 3.96 4.62 4.67 4.45 4.14
2019a 3.55 3.55 3.55 3.55 3.55
diff 0.41 1.07 1.12 0.90 0.59

KEUCHEL

ERA FIP xFIP FIP-ERA xFIP-ERA
2016 4.55 3.87 3.53 4.04 4.30
2017 2.9 3.79 3.32 3.57 3.01
2018 3.74 3.69 3.84 3.70 3.77
2019p 3.64 3.76 3.60 3.73 3.63
2019a 3.59 3.59 3.59 3.59 3.59
diff 0.05 0.17 0.01 0.14 0.04

FOLTY ERA FIP xFIP FIP-ERA xFIP-ERA
2016 4.31 4.24 4.18 4.2575 4.2125
2017 4.79 4.33 4.6 4.445 4.6475
2018 2.85 3.37 3.77 3.24 3.54
2019p 3.82 3.88 4.14 3.87 4.06
2019a 4.46 4.46 4.46 4.46 4.46
diff 0.64 0.58 0.32 0.59 0.40

Smoltz ERA FIP xFIP FIP-ERA xFIP-ERA
2002 3.25 2.39 2.83 2.61 2.94
2003 1.12 1.12 2.5 1.12 2.16
2004 2.72 2.76 2.66 2.75 2.68
2005p 2.30 2.13 2.64 2.17 2.56
2005a 3.06 3.06 3.06 3.06 3.06
diff 0.76 0.93 0.42 0.89 0.50

Glavine ERA FIP xFIP FIP-ERA xFIP-ERA
2002 2.96 4.2 4.59 3.89 4.18
2003 4.52 4.74 4.84 4.69 4.76
2004 3.6 4.24 4.36 4.08 4.17
2005p 3.76 4.40 4.57 4.24 4.37
2005a 3.53 3.53 3.53 3.53 3.53
diff 0.23 0.87 1.04 0.71 0.84

Maddux ERA FIP xFIP FIP-ERA xFIP-ERA
2002 2.62 3.43 3.61 3.23 3.36
2003 3.96 3.89 3.69 3.91 3.76
2004 4.02 4.36 3.5 4.28 3.63
2005p 3.69 4.00 3.59 3.92 3.61
2005a 4.24 4.24 4.24 4.24 4.24
diff 0.55 0.24 0.65 0.32 0.63

Multipliers Inputs
three years ago multiplier 1
Two years ago multiplier 1.5
Last Year Multiplier 2
ERA Multiplier 0.25
xFIP Multiplier 0.75
FIP Multiplier 0.75
Pitcher ERA diff FIP diff xFIP diff FIP+ERA diff xFIP+ERA diff
Glavine 2002-2004 for 2005 0.23 0.87 1.04 0.71 0.84
Maddux 2002-2004 for 2005 0.55 0.24 0.65 0.32 0.63
Smoltz 2002-2004 for 2005 0.76 0.93 0.42 0.89 0.50
Teheran 2016-2018 for 2019 0.41 1.07 1.12 0.90 0.59
Keuchel 2016-2018 for 2019 0.05 0.17 0.01 0.14 0.04 All of them are very close
Folty 2016-2018 for 2019 0.64 0.58 0.32 0.59 0.40
Average 0.44 0.64 0.59 0.59 0.50
Power Pitchers ERA diff FIP diff xFIP diff FIP+ERA diff xFIP+ERA diff
Smoltz 2002-2004 for 2005 0.76 0.93 0.42 0.89 0.50
Folty 2016-2018 for 2019 0.64 0.58 0.32 0.59 0.40
Average 0.70 0.75 0.37 0.74 0.45
xFIP Superior to ERA 0.33
Non Power Pitchers ERA diff FIP diff xFIP diff FIP+ERA diff xFIP+ERA diff
Glavine 2002-2004 for 2005 0.23 0.87 1.04 0.71 0.84
Maddux 2002-2004 for 2005 0.55 0.24 0.65 0.32 0.63
Teheran 2016-2018 for 2019 0.41 1.07 1.12 0.90 0.59
Keuchel 2016-2018 for 2019 0.05 0.17 0.01 0.14 0.04
Average 0.31 0.59 0.71 0.52 0.52
Superior to xFIP 0.39
Conclusions: Relative to the 6 current and legendary Braves pitchers
xFIP is a better predictor of a power pitcher's future ERA than ERA
ERA is a better predictor of non-power pitcher's future ERA than xFIP
ERA was a better predictor for Glavine and Maddux than xFIP
This is consistent with the hypothesis that different predictors better predict different pitchers.
AND that the current tendency is to overvalue the predictors that prefer power pitchers.
This is consistent with a Braves tradition we used to hear more about, "pitching to contact,"

This FanPost does not express the views or opinions of Battery Power.