ERA Projection Battles

What's our best bet at predicting future ERA?

---

Someone in Discord asked a good question today. They were wondering if it’s smarter to draft pitchers based on SIERA, xFIP, and K-BB% last year rather than this year’s projections.

I thought that was worth a test.

I found something pretty bad along the way. I wrote this post a month ago:

In that post, I found that my own pitching projections won the accuracy test last year. That sounded suspicious to me, and I should have dug into it further before publishing. I found out that my projections re-ran in the middle of the season last year, giving them a huge, unfair advantage. And I can’t even locate a saved copy of the pre-season projections (it overwrote the file in my Google Drive) to re-check it. So please, ignore everything I said about the MLB DW pitching projections last year. I’m guessing they would not have won if fairly measured.

And if you became a paid subscriber to get my pitcher projections because of that claim, let me know, and I’ll refund you. Be honest, though!

So we’ll move forward with this post, not including my own projections.

But it will be a super quick one. I compared 2025 ERA with these:

→ 2025 ATC Projection
→ 2025 The Bat Projection
→ 2025 OOPSY Projection
→ 2024 xFIP
→ 2024 SIERA

We end up with a TIE between OOPSY and ATC for the crown. They scored almost exactly the same.

Two takeaways:

1. The Bat doesn’t do nearly as good a job at predicting ERA as ATC & OOPSY

2. Previous season indicators are much worse than projections

For point one, The Bat actually had the most wins. It was closest on the most players in the sample (relatively small, by the way), but the average error was worse. Apparently, Derek is finalizing the Bat X pitcher projections. That’s been in the works for years, so those will very likely be more accurate than The Bat. He’s agreed to come on a podcast with me in a week or two to talk about it.

Point two was what I expected. A good projection system would take into account the same components as what the previous year’s SIERA/xFIP do, and then add on additional stuff (park changes, aging curves, and other things) to improve the projection.

It might feel smarter to use last season’s SIERA or xFIP or some other variation using K-BB% and stuff, but the better way to go is just to use the projections.