There are three facts about baseball that make it by far the best sport for projections:

1. The sample sizes are huge. A standard MLB season has 162 games. That’s about twice as many as an NBA season, and nearly ten times the number of games in an NFL season.
2. Playing time is extremely consistent. A hitter in the MLB will get 4-5 plate appearances in most games. A starting pitcher will get as many innings as he can manage every five games, like clockwork. By contrast, even the best receivers in the NFL will have a lot of week-over-week variation in how many targets they get.
3. Events are mostly independent. A hitter’s success at the plate is a pure duel with a pitcher. It doesn’t really matter how anyone else on his team has done. In Football, you can frame a receiver’s success in terms of a duel with a cornerback, but the outcome is dependent on the talent of all the other players on the field, and the play-calling of both teams.

As a result, baseball projections are an extremely mature science, and their accuracy is the gold standard by which other sports are judged. It is also a very competitive space – there are a lot of very smart people working hard to make the best projections they can. As a result, we get to do our favorite thing here at FantasyPros – figure out who is the most accurate.

Methodology

Fantasy is very different from other applications of baseball projections. If you’re an MLB general manager and are trying to trade for a pitcher, you basically don’t care at all about their Wins stat. You’re looking for talent, and Wins are heavily influenced by factors the pitcher cannot control – how well their team hits. But if you are playing in a fantasy baseball league using the standard 5×5 categories, you care a lot about Wins – that’s 10% of your team’s performance. We’ve developed a method for measuring the accuracy of MLB projections that is designed to reflect how useful they will be for winning your fantasy league. Here’s a rundown of how we do it:

• We look at the standard 5×5 categories – R, HR, RBI, SB and AVG for hitters, and W, SV, K, ERA and WHIP for pitchers. Because the counting stats are heavily influenced by playing time, we add two more categories – at-bats for hitters and innings pitched for pitchers. We then treat all the counting stats as rate stats – home runs is actually home runs per at-bat, for example. That way, if you do a good job projecting a hitter’s home run rate, but are wrong about how much he’ll play, you accurately get a good mark for HR but a bad mark for AB.
• For each player/stat combo, we evaluate accuracy in the context of the field using z-scores. Essentially, we calculate the average error across all the projection sources, and give each source a score based on how much better or worse they were than the average. For example, Valdimir Guerrero Jr. had 604 at-bats in 2021. Projections mostly under-projected this number, with an average miss of 51.4 and a standard deviation of 16.3. The most accurate projection for this was 619 from Harper Wallbanger’s Big Board projections. That’s an error of just 15, which is 2.2 standard deviations better than the average. So Big Board gets a score of -2.2 for Guerrero’s at-bats. (A projection worse than the average would get a positive score – lower is better, since these numbers represent the “error” of the projection.)
• We average together these individual scores across the player pool to get each system’s accuracy for that particular stat. We then add together the stat scores to get overall hitter and pitcher scores, and add those together to get our final overall accuracy scores.

2021 Projections Accuracy

Without further ado, here are the overall accuracy standings for 2021 fantasy baseball projections:

Sometimes excellence is just excellence, and that is the best way to describe Ariel Cohen and his Average Total Cost (ATC) projections. This is its third consecutive year as our #1 most accurate projection system. (And ATC got second place in 2018, the first year of this study.) ATC’s success is a testament to the power of consensus. Baseball projection systems can be separated into two categories: “original” projections like Steamer that are created from scratch, and “consensus” systems which combine existing projections in a (generally successful) effort to find more accurate projections than any single original source can provide. The most simple example is FantasyPros’ own Zeile consensus, which is a straight-up average of every non-paywalled projection source. ATC is on the opposite side of the spectrum – it is a weighted average of several projection sources, and historical data, with different weightings for each stat. You can find more detail on how ATC works here.

Congratulations are also in order for Derek Carty’s THE BAT X, which has now been the most accurate “original” projection system in both years since it was introduced as an improved form of THE BAT. You can find the full detail here, but the short version is that THE BAT X is an upgraded version of the hitters portion of THE BAT which incorporates Statcast data like exit velocity, which has only been available for a few years. Since those upgrades are hitter-specific, the same pitcher projections are considered for both versions for accuracy purposes. The extra layer on THE BAT X is working – as you’ll see in the category breakdown below, it beat the base version of THE BAT in three of the six hitter categories, and tied in two others.

Category Results

Here’s the category breakdown for hitters:

For pitchers:

—

This concludes our 2021 Fantasy Baseball Projections accuracy study. Congratulations again to Ariel Cohen for making ATC the most accurate projection system for three years in a row!

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