Most Accurate Fantasy Baseball Projections (2019)

It’s March, and that means we’re all furiously preparing for our fantasy baseball drafts, and nothing else is happening in the world. The foundation of a good draft strategy is a good set of projections. Fortunately for us, baseball projections are the crown jewel of sports analytics. No other sport embraced statistical analysis as early and as fully as baseball did. That means that in general, baseball projections are really good. That also means there are a lot of options. So which projections should you use? Who’s the most accurate?

Methodology

Our system for evaluating projection accuracy is tailored to the way we use them – for fantasy. We evaluate the ten standard 5×5 categories: R, HR, RBI, SB and AVG for hitters, and W, K, SV, ERA and WHIP for pitchers. We also evaluate playing time projections using AB for hitters and IP for pitchers. Projecting how well a player will play and how much he will play are very different tasks. In an effort to isolate those skills, we evaluate counting stats on a per-plate-appearance basis. So our scores for HR, for example, are actually based on HR/PA.

To calculate accuracy for a stat, we start by calculating absolute errors for each player and projection system. For example, Mike Trout had 470 AB last year, so a projection of 490 would have an error of 20. Direction doesn’t matter – a projection of 450 would also have an error of 20. We calculate errors for each projection system, and then find the average and standard deviation for each player. Suppose we had just five projection sources, and they projected Trout to have 460, 470, 480, 490 and 500 AB. Then the errors for Trout’s ABs are 10, 0, 10, 20 and 30. The average of those is 14, and the standard deviation is 11.4. Then we calculate z-scores for each projection source, which tells us how many standard deviations their error was away from the average. The source that projected 460 AB for an error of 10 was better than the average of 14 by 4. Since the standard deviation is 11.4, they beat the average by 4/11.4 = 0.35 standard deviations. Thus, the z-score for that source was -0.35. (Projections better than the average have negative z-scores, and projections worse than the average have positive z-scores.)

Now that we have each source’s projection error for each player translated into z-scores, we average those together across the entire player pool to get category scores. The Steamer projection system got a score of -0.20 for AB, which means that for the average hitter, Steamer was more accurate than the average projection system by 1/5 of a standard deviation. Then to calculate overall accuracy, we simply add together these scores across each of the 12 categories.

Results

Now that you know how they’re calculated, here are our 2019 fantasy baseball projection standings:

 
Congratulations to Fangraphs for their Average Total Cost (ATC) projection system, which took the #1 spot for both hitters and pitchers. This is an excellent year for Ariel Cohen, the creator of ATC, who was also our most accurate fantasy baseball draft rankings expert for 2019. ATC and our own Zeile projections are both consensuses of several projection sources. While Zeile treats each of its component systems equally, ATC uses unequal weighting to give sources a bigger say in categories where they have a track record of accuracy. This added layer of complexity paid off, making ATC the most accurate projection source for 2019.

Consensus systems like ATC and Zeile couldn’t exist without the rich variety of projections that are built from scratch. Mike Podhorzer’s Projecting X deserves recognition for being the best of them – it was 2019’s most accurate non-consensus system.

Category Results

Here’s the category breakdown for hitters:

 
And for pitchers:

 
—

Jacob Herlin is a Senior Data Analyst for FantasyPros. You can find him on twitter @jacoblawherlin.