Tagged: Pitchers

2014 MORPS Projections

MORPS projections are late coming this year.  I’ve delivered a set of baseline projections several weeks ago.  However, you’ll find that the actual projections have some drastic differences.  I’m always amazed by the amount of player movement during the off season.

2014 MORPS Batting Projections 20140227 (XLS)

2014 MORPS Batting Projections 20140227 (PDF)

2014 MORPS Pitching Projections 20140227 (XLS)

2014 MORPS Pitching Projections 20140227 (PDF)

Batting Projections

The Major-League Obie Role-Based Projection System (MORPS) uses four years of player performance data for all hitters. Since I started playing with Sabermetrics using Tango’s Marcel system, the first iteration of MORPS four years ago used the same formulas. After learning the basics, the batter formulas were adjusted to include the most recent four years of performance data. Adjustments were also made for player age, home ballpark data and expected playing time. The most complicated part of the system is the regression formulas. Tango provided formulas for his three year model. I had to crack open the math books to figure out how to transition the formulas to a four year model.

One of the most time consuming tasks in developing the system was determining the proper mean for player regression. If the goal was to ensure that the mean of all the projections competed favorably with end of year player means, the task would have been straight forward. However, my goal was to make the actual player projections as accurate as possible. “Role-Based” means that the player projections are regressed to position specific means. National League means are also separated from American League means.

While conducting research, I noticed that most projection systems used minor league stats as well as any available major league stats to project the future performance of young players. There are even formulas that anticipate player regression when entering the majors. The interesting thing is that Tango’s Marcel system does just as good at predicting rookie performance as other projection systems and he doesn’t use any minor league stats. Some players are great in the minors and simply can’t make the jump to the major leagues. Some players start out great, but find that major league pitchers start exploiting weaknesses they never knew they had. Others outperform all expectations. By calculating the reliability of a player’s projection using only major league data, MORPS adds a proportional dosage of a player’s positional mean to complete a rookie’s player projection. Since we are focused on individual player performance, I didn’t see the point of including all minor league stats when the results don’t seem to provide significant value. The last year of a rookies minor league or international season is included, with appropriate adjustments for competition, if no major league experience exists.  While efforts have been made to adjust projections to reflect anticipated playing time, players who have a roster flag of “N” are projected using baseline projections only.

Pitching Projections

The formulas used to create pitcher projections are very similar to those that we have already discussed with batters. MORPS uses four years of data to create a pitcher projection. Adjustments are made for age, home field and anticipated role. The reliability of a projection is calculated based upon the amount of data available for a particular player. Someone with low reliability will regress more to a position specific mean than someone that has faced a lot of major league batters over the last four years.

The big difference between projecting pitchers and batters is the usage disparity between relief pitchers and starting pitchers. A good relief pitcher may face 350 batters in a season. A top end starting pitcher may pitch to 900 batters in a season. The plate appearances for position players are typically not dependent on role. A first baseman and shortstop may both have 600 plate appearances over the course of a year. Their position means will be different. First basement will typically have higher power stats while shortstops have higher speed stats. But, they are similar enough that their projections can be calculated using the same basic formulas. The disparity between relief and starting pitchers forces them to be calculated very differently. For months I struggled with pitching projections. When I finally figured out that starting pitchers and relief pitchers had to be calculated separately, everything fell in place.

Projecting Pitchers – FIP versus ERA

I read some research late last year that altered the way that I drafted pitchers in 2011. Steamer Projections evaluated 2009 projection systems and determined that Marcel beat most other ERA projections by using Fielding Independent Pitching (FIP) in lieu of standard ERA projections (see article). Since my formulas were originally based on the Marcel system, I decided to use FIP in lieu of ERA on draft day. The results were mixed which motivated me to conduct my own research after the 2011 season was over.

The two tables below capture the cumulative number of individual MORPS projections in two categories. One category is projected ERA versus actual ERA. The other category is projected FIP versus actual ERA. One table is for relief pitchers. The other table is for starting pitchers. The top 175 used relief pitchers were evaluated (batters faced). The top 150 used starting pitchers were evaluated.

Relievers

ERA to ERA

FIP to ERA

Within 5%

19

20

Within 10%

38

43

Within 15%

59

63

Within 20%

79

84

Within 25%

96

101

Within 30%

111

112

Starters

ERA to ERA

FIP to ERA

Within 5%

23

26

Within 10%

64

56

Within 15%

79

71

Within 20%

96

90

Within 25%

107

103

Within 30%

122

117

As you can see, MORPS projected 81.3% of starter ERAs within 30% of actuals. Projected starter FIPs were lower at 78%. MORPS projected 63.4% of reliever ERAs within 30% of actuals. The FIP reliever projections were a little better at 64%. Although it seems that FIP may be just as good of a predictor of ERA for relievers as my standard projection formulas, it did not seem to be the case with the starters. Because the numbers were so close, I decided to perform the same analysis with ZiPS and Marcel.

Zips Relievers

ERA to ERA

FIP to ERA

Within 5%

20

17

Within 10%

40

41

Within 15%

62

60

Within 20%

74

73

Within 25%

87

85

Within 30%

96

97

Zips Starters

ERA to ERA

FIP to ERA

Within 5%

29

28

Within 10%

50

54

Within 15%

79

71

Within 20%

91

89

Within 25%

111

106

Within 30%

123

122

Marcel Relievers

ERA to ERA

FIP to ERA

Within 5%

21

21

Within 10%

40

45

Within 15%

58

65

Within 20%

81

85

Within 25%

92

100

Within 30%

107

112

Marcel Starters

ERA to ERA

FIP to ERA

Within 5%

26

27

Within 10%

53

48

Within 15%

78

77

Within 20%

98

92

Within 25%

113

112

Within 30%

117

123

ZiPS projections and conclusions were very similar to those outlined above for MORPS. Marcel starter and reliever projections were definitely better with calculated FIP in lieu of their projected ERAs. I have to conclude that FIP is a better projection than standard ERA formulas with less data since the major difference between MORPS and Marcel is the number of years that are used within the projection formulas. In order to test this hypothesis, I decided to alter my 2011 ERA projections to use FIP for those players that had less data available (i.e. reliability) and standard ERA formulas for those with more data.

Relievers

ERA

65%

75%

85%

Within 5%

19

19

18

20

Within 10%

38

39

39

41

Within 15%

59

62

63

63

Within 20%

79

82

83

83

Within 25%

96

98

101

100

Within 30%

111

111

114

112

Starters

ERA

65%

75%

85%

Within 5%

23

22

24

25

Within 10%

64

62

63

57

Within 15%

79

79

80

74

Within 20%

96

93

95

92

Within 25%

107

107

110

106

Within 30%

122

120

122

121

As you can see by the results, using FIP to project ERA for relievers with less than 85% reliability increased the accuracy of MORPS projections. This was not the case with starters. The best way to project starter ERAs is to use standard ERA projection formulas in lieu of FIP. The same thing holds true with relievers that have a substantial track record over the course of multiple years.