The projection system OOPSY makes its debut in the major leagues this year. So how does it do it?
OOPSY’s methodology is similar to that of other FanGraphs projection systems, but with a few changes – most notably, the inclusion of batters’ bat speed and pitchers’ Stuff+. However, the predictive system is made up of many different components, including aging curves, major league equivalence that takes into account minor league and foreign league performance, recency weighting, mean reversion, league operating environment and park factors. In addition to incorporating specific variables like bat speed, there are many ways a projection system can stand out or lag behind its peers. Forecasting systems are made up of hundreds of small methodological decisions. Given all the decisions OOPSY has made, can it hold its own relative to its more established peers in 2025?
To review these predictions, this article follows industry best practices outlined by MLBAM Senior Data Architect Tom Tango. I’ve gone through this review process with pitchers before, as my pitching predictions were covered by Eno Sarris in Competitor Not since 2023, but this is the first year I’ve released a full set of top predictions. This review focuses on batters’ wOBA and pitchers’ wOBA (an alternative to ERA, defined further below). These metrics are often the focus of projection system reviews and are the most important batting average and pitching average statistics a projection system can obtain from a “real life” perspective. Both are catch-all rate statistics that measure a player’s offensive and pitching value respectively. Various component forecasts such as K% and BB% are incorporated into these overarching metrics.
As a guide for anyone looking to replicate the predictive review process, I will draw on my 2023 review experience to walk you through the steps I took.
First, I grabbed all the stats for 2025 from the FanGraphs leaderboard. I removed the batter who pitched the ball from the pitching data and vice versa (although I kept Shohei Ohtani in both data sets). I also downloaded all the preseason predictions and merged them with the 2025 stat rankings. I leave any missing projections blank and return to them later.
Next, I rescaled all projections so that they assume the same league operating environment. Other reputable projection reviews may sometimes skip this point, but sabermetricians consider it a non-negotiable step. If you fail to do this, your forecasting performance may be affected by your assumptions about the environment in which your league operates, which are affected by decisions outside the typical scope of your forecasting system, such as decisions about how baseballs are manufactured. If you’re still curious about the environment in which the league operates, for example, and you want to understand which predictions best explain the impact of a new rule change, then it’s best to save that for a separate exercise.
For demonstration purposes, here is how I rescale OOPSY’s wOBA forecast. I looked at all the major league hitters in 2025 using OOPSY predictions. I took the average of their wOBA projections weighted by their actual appearances in 2025; for OOPSY, it was 0.316. Next, I rescaled all projections to assume a league average wOBA of 0.313 in 2025. To rescale OOPSY, I subtracted 0.316 from each wOBA projection and then added 0.313. To convert ERA to wOBA (Tango’s preferred composite metric for pitchers), I divided each pitcher’s ERA projection by their projection system’s league-average ERA, took the square root, and then multiplied by the 2025 league-average wOBA, 0.313.
Next I had to choose a strategy to solve the missing player problem. Some projection systems cover a wider group of players (e.g., ZiPS), while others focus on relatively more mature players (e.g., THE BAT X). You can project all missing players to be league average (MARCEL method) or slightly worse than league average, which is generally more accurate. I chose a combination of these strategies, which I detail below.
Finally, I calculated the root mean square error (RMSE) of each statistic for each projection system. I could also calculate the mean absolute error, but RMSE is more typical. RMSE represents the standard deviation of the residuals and is a measure of the typical distance between predictions and actual results. I weighted the RMSE based on 2025 bat appearances and total number of arm hitters. Following instructions from my conversation with Tango, I also calculated the difference between each projection and the naive projection, which assumes everyone has the same projection. To calculate the difference between the projected and naive projections, I square the two, subtract the smaller number from the larger number, and then take the square root.
Following these steps, the table below shows the RMSE of preseason projections for the full sample of hitters and pitchers in 2025 (the lower the RMSE, the better). Naive forecasting assumes everyone has a league average forecast. The combined forecast is the average of four independent systems (OOPSY, THE BAT X, Steamer and ZiPS). For missing players, my projections are either league average, 5% below league average, or 10% below league average. I chose the option with the lowest error rate per system. Hitters without steam engines were given league average projections. For other systems, hitters with missing projections had projections that were 10 percent lower than the league average. Pitchers without THE BAT or ATC projection will receive the league average projection. For other systems, the projection for absent pitchers is 10% lower than the league average.
Table 1. Prediction accuracy for the full sample.
Across the entire sample, Steamer was the most accurate independent projection system for wOBA vs. pitchers — generally. For hitters, ZiPS is clearly the best system for wOBA. Steamer’s pitching projection is 0.0216 wOBA better than the naive projection. ZiPS’s hitting projection is 0.0238 wOBA better than the naive projection. OOPSY holds its own, ranking second among independent systems for pitchers and hitters. It’s worth noting that all systems have tight groupings for hitters and pitchers. The composite projection (the average of four separate systems) was the overall winner for the arm and bat. This is not surprising since averaging the “experts” is often a winning approach when it comes to forecast accuracy.
Next, let’s take a look at how the rookie predictions fare. Note that for rookies, naive projections give everyone a projection 5% below league average.
Table 2. Prediction accuracy of rookies.
Among independent systems, Steamer leads the way in predicting 2025 rookies, followed by OOPSY. (OOPSY’s latest list of top prospects based on long-term projections can be found here .) Steamer doesn’t publish long-term projections, but if they decide to do so, I’d love to see them. Once again, integrated projection is the clear winner.
Next, we’ll exclude the rookies and focus only on the veterans.
Table 3. Prediction accuracy among veterans.
For pitchers, the results were similar to the full sample. For hitters, ZiPS remains No. 1 among standalone systems, but THE BAT X jumps to No. 2. ATC also improved when focusing only on veterans. THE BAT Last year, THE BAT X also had best-in-class batsman performance overall. In my early exploration, it appears that THE BAT X may include variables such as optimal position and sprint speed that OOPSY does not yet include.
Finally, as a robustness check, the table below compares OOPSY to each system but excludes missing players – the RMSE below only focuses on players predicted from both systems in each comparison. The results are similar to Table 1.
Table 4. Full sample, predictive accuracy of alternative methods.
Overall, I was hoping for the best in terms of accuracy, but OOPSY’s debut season did not disappoint. I’m excited to be able to use it with other FanGraphs projection systems. I’ve had the pleasure of having some conversations with each of these forecasters over the years, and I’ve learned a lot from them. I still rely heavily on their system. In the pursuit of small improvements in forecast accuracy, there is always more to learn. I still need to study the accuracy of my component predictions and see where I can make the biggest improvements. I already plan to consider additional Statcast metrics to help project hitters, potential sweet spots, sprint speed, angle of attack, and horizontal jet tendencies. There are less low-hanging fruit for pitchers because my projections have longer history, but I may revise the recency weighting and incorporate pitcher-level metrics into the projections. However, I’m not the only one who makes adjustments each offseason. No matter what improvements any of us make, users of projection systems will likely be best served by simply averaging the predictions from different systems.
