Variants for ERA - for Pyth - Baseball

[Romanowski]

In this thread, UVA laid out a method for creating a line for baseball

it creates an expectation, to be compared with the market

and to find inefficiencies...i.e edges

The basic formula is:


where Win% is the winning percentage generated by the formula.

The expected number of wins would be the expected winning percentage multiplied by the number of games played.

there are variations to this formula, with different quotient values, in the one above everything is to the 2 power..

however others use 1.83

and some use a floating quotient, thats team dependent

where (runs per game[scored or allowed] to the 0.287 power) = the quotient for the various teams

to calculate runs scored by each team, in the early stages of the season

UVA took the ERAs of both starting pitchers (projected CLONE #s) and simulated them over an entire season

like this:

Masterson 4.36 x 162 = 706.32
Floyd 4.33 x 162 = 701.46


here are a couple changes or tweaks you can make to determine runs scored rather than true ERA


if you have any other ideas, such as using OPS, please add them here...


xERA=expected ERA

= (0.575*Hits allowed per nine) + (0.94*HRs allowed per nine) + (0.28*Walks allowed per nine)

- (0.1 * Strikeouts per nine) - (2.68)


if a Pitchers xERA is higher than is true ERA, than he was fortunate or lucky ( and will regress)

or conversely if his true ERA is higher than his xERA, he pitched better than his numbers indicate

Ken Pomeroy uses a similar "luck" factor in his college basketball model


another is FIP

(FIP) ERA = Fielding Independent ERA

designed to measure only the variables the pitcher has complete control over

Namely, home runs, strikeouts, and walks

everything else, basehits, errors, runs, could be by chance

Author of this formula now works for the Seattle Mariners..


= (HRs*13)

+ (Walks + Hit batters) *3

- ( Strikeouts *2) / (Innings Pitched)

+ 3.2

FIP ERA = [(HRs*13) + (BB +HBP)*3)-(K's*2)/Innings Pitched)] +3.2


One negative w/FIP is a pitcher gets hurt a lot by giving up more than the league average of HRs and could otherwise be good...


FIP can be tweaked however by replacing the pitchers' HR rate to the league average HR per fly ball rate.

This is called xFIP, and tends to be a better indicator of future performance.

[uva3021]

i don't know, not the biggest proponent of arbitrary formulations based on a vast database

while vital to the information i typically use, and i understand kenpom's whole method relies on 50 some years of data, the elements that make up the calculations will change as the years progress

its basically a temporal aggregation that is hypersensitive to a time series, and the coefficients and placeholders for the sake of a nice round number seems inherently capricious

i guess using the same nonperformance variables to measure FIP for all pitcher has validity in that its consistent, but the number itself when compared to general ERA, i don't find it applicable

maybe a relationship by way graph between the two will show otherwise

think there is a better way of doing things, like a more practical formulation of an FIP era

[Romanowski]

worth testing at least, all valid sabermetrics

not sure however how often used to plug in for pyth, we shall see

chone's projections right now are finding inefficiencies

[uva3021]

did a correlation measure of the relationship between last year's FIP / ERA of all starting pitchers, and find the correlation to be slightly above semi-linear, what is interesting though is there is a stronger correlation of FIP to WHIP than FIP to ERA, the number approaches 70% compared to 60%

[Romanowski]

Originally Posted by smitch124

I'm actually working on creating a conversion from OPS against to runs allowed/inning, I think it would be more accurate than ERA for estimating totals runs and has the added benefit of using numbers that are less "public."

smith did you ever uncover something using ops to use for a pyth expectation