We have moved into the month February, the Super Bowl is over and like a large number people I have switched my main attention over to baseball. But as Spring Training hasn’t even started yet so my attention is more drawn to the world of fantasy baseball.
I, like many baseball fans, participate in multiple fantasy MLB leagues; points and roto, daily and weekly. These all reward players performance differently but a main stay stat in most fantasy leagues for pitchers is the win. It has been replaced in some by the quality start but it something that fantasy fans will care about.
Each year this leaves me ambivalent to the win. As a sabermetric fan I despise the win as a method of determining a pitcher’s performance because it isn’t a predictable or descriptive stat as it is influenced by many other factors. But as a fantasy fan, I care about them because they have value and therefore I want to try and predict them, so I can win my leagues.
This ambivalence also applies to runs, saves and runs batted in, but all to a lesser extent.
My general method for fantasy baseball involves taking all the projections I can find, combining them together and then adding in the scoring method of the league to see who I should draft. The projection models all include a win projection for pitchers but I am always more sceptical of these numbers than I am of any other, as it is much less predictable factor. So, let’s look at these win projections and talk through why I have an issue.
Issue: Not enough big win totals
If we look at pitchers who are project to qualify come the end of the season, throw 162 innings. All of these four projection systems they have just one pitcher reaching 17 wins and all bar one of them has 5 for less pitchers reaching 15 wins.
At the time of writing the PECOTA predictions don’t think that anyone will get more than 15 wins, this is mostly due to their game started projections which limits all starters to 31 or less. This seems off to me given that 31 of 56 qualified pitchers from 2018 started 32 or more games.
These numbers are off and we only have to look at the last two seasons to see why. There were 56 qualified starters in 2017 and 2018 and they combined for 702 wins both season, which is an average of 12.54 wins. There hasn’t been a season since 1995 which the average wins of qualified starters was less than 12.
Looking at the table above, you can see that much larger numbers of pitchers reached the higher win totals. In fact there was 30% getting 15 or more wins and 17% getting 17 or more wins. So, the ATC projections look the best with comparison to the last two season.
A lot of projection systems are not going to handle extremes well due to lack of data, but these aren’t extreme values so this just smacks of caution on a category they have less confidence over. For me it would acceptable for the SDs to be smaller which would reduce the high end values but for the means to be significantly different it suggests that the models are slightly off.
The next question for me is can I do any better.
As stated before, sabermetricians don’t like the win as a metric for determining pitcher performance because there are other factors than just the pitcher. The graph below shows the win rate compared to the ERA of all qualified pitchers from 1995 to 2018. You can see that while there definitely is correlation between a high win rate from a low ERA, pitchers always stand a chance to be Jacob deGrom from last season and have a much lower win rate than expected.
The factors, outside of the pitcher, that we generally think off affecting the win are the performance of the batters and the relievers on the pitchers team. Imagine a stud pitcher that can throw 7 innings of 1 run baseball every game. If he gets either no run support from his hitters and/or the relievers always give up multiple runs in the last two innings, he is not likely to pick up many wins as either the in the end team doesn’t win or when he is replaced the team isn’t winning.
Let’s look at the two to see how much they affect win rate when in conjuncture with pitcher ERA. To do this I will take the average runs scored by a team and the ERA of relievers and see their impact on the win rate. I have bucketed the pitchers into 0.20 ERA buckets and the average runs scored into 0.20 run buckets.
The graph below shows the impact of starter ERA and team average runs, the colouring is green for a high win rate and red for a low win rate.
This shows that both ERA and runs scored by the team have an impact on starter win rate. If you follow any of the ERA buckets left to right you can see it getting less red and more green.
It shows that a pitcher with a sub 3 ERA on a team that doesn’t hit well can have the same win rate as a 5 ERA pitcher on a team that rakes.
I have also highlighted when the ERA matches the average team runs. For this scenario the win rate is very similar for all values with a starting pitcher winning on average just over a third of their games.
So their runs scored by teams definitely impacts how about the reliever ERA. The chart below shows the impact of starter ERA and team relief ERA. As before, the colouring is green for a high win rate and red for a low win rate.
This doesn’t show the same impact. While there does look to be some impact it isn’t as clear as the runs scored graph.
So, now we are now armed with that knowledge, I am going to make a simple model for projecting wins based of a players projected ERA and the projected run production of their team. Let’s use the Depth Chart ERA and RS projections as an example.
After applying my win model to the Depth Charts projections there is a much larger number of pitchers getting 15 wins and the mean is much similar to the last two season. Interestingly this model only decreased the number of wins for one qualified pitcher, Corey Kluber. It removed one win from him taking him from 15 wins to 14 wins.
Below is the list of pitchers I now have as getting 15 wins, from the DC projections. You can see that there are a couple which have had quite big jumps, Walker Buehler and Aaron Nola both go from 12 wins to 15 wins.
In a standard 15 teams 5×5 Roto league, there is on average only 2-3 wins difference between each place. So while these differences seem small they can make a reasonable difference to a players projected worth and therefore draft rankings, adding 3 wins to Walker Buehler moves him up 1 pitcher place and 3 overall places in my rankings.
The way you win in fantasy leagues is finding value in players that other people don’t and this for me is hopefully the first of many such finds.