Post Draft, Pre-week 1 completion: how should we be building our roster?

 

Arthur’s big smile here might be about Juan Soto finally playing a game today…or maybe he just yelled, Yo Adrien!

After the draft, the question was posed, how should we be building our roster, both during the draft and then throughout the year?

In 2018, I dove into analyzing what is replacement level for starting pitching?

https://flooredfbl.blogspot.com/2018/06/week-9-what-is-replacement-level.html

I concluded that that only about the top 20 starting pitchers in ERA and k/9 were above replacement level. Below that level, the difference from one pitcher to the next were all about the same and were therefore replaceable: that your roster spots were best left to try and find breakout players (hitters or pitchers) that could elevate to become above replacement level.

So what has changed? Well, for one, our league rules. Back then, we had Earned Runs as a scoring category, so it was discouraged to throw a lot of innings, but now with Quality Starts being a category instead of Earned Runs, we are incentivized to throw more starting pitchers and therefore more innings and because we have limited pickups each week (by design to minimize streaming), we need to own more pitchers full time.

So lets look at what the data to try to figure out where Replacement Level is now from two perspectives: fantasy team build strategy and MLB player talent.

Team roster construction can fluctuate, so let’s make some assumptions to make an analysis possible. Assuming at least one bench spot is required for a hitter to cover off days or Day To Day injuries, and assuming at least 3 Relief Pitchers are needing to be held to stay competitive in Saves Plus Holds, that leaves 7 spots available to hold Starting Pitchers whether via full time rostered or streamed pitchers. With 10 teams in the league this would leave 70 pitchers to be owned at any one time therefore the best replacement pitcher available being SP71. Easy enough.

Now lets talk from an MLB player talent perspective: at what point does pitcher valuation enter a ‘glob’ and there isn’t much difference between them, therefore making them replaceable?

We know from our projections analysis that players are projectable to a certain accuracy, when you at the player pool in general. As in, Max Scherzer’s ERA could deviate by 50%, but the data has shown for three years now that, in general, a pitcher’s end of season ERA will generally fall within 30% of his projected ERA.

So what does that look like? Here is the data from 2020. In the shortened season we had pitchers with ERAs in the 1s and ERAs in the 7s. What is interesting is what happens when you put the 30% error band on it:

With the black line you can see that the worst case 30% error band for SP30 is at the same ERA as the best case error band of SP120. What happened in the past won’t exactly be how players are projected for the next year, so let’s get some more context.

Here is the top 150 drafted started pitchers’ ERAs projected for this year, and an error band of 30% added for the ERA when you sort by ERA from lowest (Jacob DeGrom at 2.75) to highest (Antonio Senzatella at 5.36). I have also added the actual ERAs from 2019 and 2020 from the top 150 pitchers in Innings Pitched.

Messy, right? Here, I have put the 30% error band on the 2021 Projected ERA, the orange line. What this shows is that almost all of the data from the last 2 years is within that 30% error band…other than the extreme outliers. The one thing we can’t do here is match error bands. SP30’s worst case error band is completely within the range of outcomes all the way up and through SP150’s error band. Scary to think that SP30 could end up as SP150, but as we all know, busts happen. SP30 does sometimes become SP150…hello Matthew Boyd.

The black horizontal line here represents the fantasy league’s average ERA in 2020 (representing how the fantasy league utilized the gray ERA line). The black vertical line is at SP70 where the above approximation identifies may be the league-strategy replacement level. It should be noted that the league average ERA is better than the ERA of SP70 (closer to ERA of SP40), so you need to stock up on players above SP70 to beat the league average ERA.

The other observation here is that there is no dramatic fall off in data, There is a normal distribution of ERAs from SP1 to SP150, with some slight change in slope at the beginning and end of the plot: there is a slight change in slope from the top 20 pitchers to the pack to the bottom 20 pitchers, but not dramatically.

Noting that ERA is not the only pitching category that matters, I analyzed WHIP and k/9 in the same way and saw nearly identical patterns, so let’s put the data together and see how pitching plays out.

I ranked the top 150 SP in projected in innings pitched in three categories: ERA, WHIP, and k/9. For example, Jacob DeGrom gets a 1 in ERA for having the best projected ERA and Senzatella gets a 150 for having the worst. Across three categories what we might hope for is a grouping of players such that we can find tiers and therefore a replacement level.

The first thing we see here is that ADP didn’t perfectly fit pitcher value. The blue scatter plot shows a general increase in rank vs ADP, but not a straight line. This just illustrates that some pitchers were overvalued and undervalued in the draft compared to their statistical projection. Anyway…

The next thing we see here is a fairly linear correlation when you sort the ranks from low to high. There are no easily visible tiers that would be visible by changes is plot slope between SP1 and SP150. As in, the difference from SP5 to SP10, is basically equal to the difference between SP55 and SP60. This leads us to conclude there is no ‘level’ at which pitchers become replaceable from a statistical analysis perspective.

For comparison, how do hitters grade?

Using the same methodology, I ranked the top 450 hitters in our 8 fantasy categories. What you see here is a bit more of a change in slope from the top 40 hitters to the pack, and then another change in slope over the bottom 30 hitters. This tells us that there is a slight tier around hitter 40 compared to the pack, as in the difference between Hitter 5 and Hitter 10 is slightly more than Hitter 65 to Hitter 70, but not dramatically.

So what does this tell us? Well, generically, its slightly better to stock up on hitting early….however, you still need starting pitching better than SP70 to beat the league average ERA. There is no statistical drop off from any tier to the next in starting pitching, though we do see a slight one among hitters.

This leaves us with the conclusion that replacement level for pitching would be projected to be at Starting Pitcher 70. However for our league that has really only been owning 50 to 60 SPs at any one time the last few years, this raises the bar for needing to produce better pitching statistics than if the league held 70 Starting Pitchers, because SP61-SP70 will not produce as good of numbers as SP1-SP60, of course.

This also generally tells us that there isn’t much harm in taking ‘shots’ on pitchers in the lower ranks, because chances are pretty good that if they don’t work out, you can just go grab another one and hope they break out without much cost to you.

This methodology does have its limitations. For example, it does not give a pitcher like Jacob DeGrom credit for his projected ERA being significantly better than the number two SP, whereas the ERA for SP10 will be very close to SP11. The reason why this methodology was chosen over more of a weighted-statistical method is that any weighted statistical method requires setting baselines for when stats level out and become replaceable and then comparing how good DeGrom's ERA is compared to that replacement level...and as we've seen, there is a steady climb throughout and not a plateau.

In the end, though we weren’t able to find a statistical replacement level using a player-rank methodology, we were able to see that higher tier hitters are a bit more valuable than the pack, and that can be used when making moves throughout this season and drafting next year.