Heading into the final week of the regular season, the MLB playoff picture is almost settled. Although a handful of wild-card spots are still unclaimed, every division in baseball is all but locked up. This year has been one of the more imbalanced years in recent memory, featuring the 100-game-winning Cubs alongside the (likely) sub-60-win Twins, and the lack of parity has been good for statistical projection systems. After a rough 2015 season, the forecasters are having their most accurate year since 2013. But they weren’t perfect, particularly at the individual level: In the midst of a leaguewide home run spike, there’s one type of player that algorithms utterly failed to predict.
But let’s start with team-level projections, where the predictors scored a win. As I wrote early this season, balance had been increasing in MLB — until this season, that is. Whether due to chance or planning, several prospect-rich teams (such as the Red Sox and Cubs) got big production from former farmhands this year, which gave the projection systems a wider spread of team strengths to forecast. Why was this beneficial for the projections? Just as it’s easier to predict the outcome of a home run derby between Giancarlo Stanton and a high school player than one between Giancarlo and, say, Todd Frazier, projection systems work best when teams’ performance levels are more varied.
As a result, the correlation between actual and predicted wins is at its highest level since 2013 and the second-highest level since 2007.1 Problems that haunted the projections in 2015 have diminished somewhat this year. While teams were uncommonly good and bad at clustering their offensive output in 2015, that one-year spike in the effects of luck has disappeared in 2016.
So the projections had a good year, at least for the standings. But they didn’t perform nearly as well for individual players. It’s hard to blame them, though: No one knew whether the spike in exit velocity at last year’s All-Star Break would continue into 2016. It did — and home runs are near an all-time high as a result.
Fly balls are uniquely sensitive to the ball’s speed off the bat. Every additional mile per hour of exit velocity adds about five feet2 of batted-ball distance, which can be the difference between a wall-scraper and a warning-track out. As exit velocity jumped last year by about 1.5 miles per hour, many formerly harmless fly balls turned into dingers instead. The ratio of home runs to fly balls is the highest ever recorded since batted-ball types first began to be tracked in 2002.
So it follows that the projections have failed the most on players who hit lots of fly balls. I grabbed FanGraphs’ projections for hitters and pitchers in 2016 and then compared them to players’ actual performances. As expected, for both hitters and pitchers, there was a significant correlation between their fly-ball rate in 2016 and how much their projection deviated from their actual performance.3 Specifically, hitters who smashed lots of fly balls have done better, but pitchers who allowed fly balls have done much worse.
That’s had a real effect on some of the most extreme fly-ball pitchers in the league. Jered Weaver is allowing the sixth-highest fly-ball rate4 among pitchers with more than 100 innings, and he’s been so bad that the most advanced pitching metric, Baseball Prospectus’ deserved run average, thinks he was worth almost six wins less than a replacement-level pitcher.5 Chris Young, who just last year posted a solid season for the championship-winning Royals by allowing almost exclusively weak fly-ball contact, has moved to the bullpen this year after being on track to allow the highest home run rate of all time. He’s also been worth negative wins above replacement, according to FanGraphs, which indicates that a readily available triple-A starter would be better.
The trend is similar on the hitting side, but in reverse. Of the top 10 hitters in fly-ball rate this season,6 nine have seen their weighted runs created plus exceed projections, with the overall average about 14 percent higher than expected. Those top 10 include out-of-nowhere San Diego Padres rookie Ryan Schimpf and unexpectedly resurgent Seattle Mariners catcher Mike Zunino. By coincidence, the leaguewide trend in exit velocity has helped Chris Young the hitter (a fly-ball slugger whose actual wRC+ is 36 points higher than projected7) at the same time as it’s hurt Chris Young the pitcher.
And although the projections nailed the overall state of the standings, a few pitching staffs and lineups have been especially affected by the new home run climate. The Cincinnati Reds pitchers, for whom 38 percent of contact results in fly balls (fourth-highest in baseball), have given up the most homers of any team, ever, and are perhaps the worst pitching team of all time. The Tampa Bay Rays, a preseason darling of some projection systems (including Baseball Prospectus’s PECOTA), have underperformed their ERA projection by two-thirds of a run, largely on the basis of disappointing performances from a handful of fly-ball starters such as Drew Smyly. Meanwhile, the Orioles lineup, which hits the third-most fly balls in MLB, is on the verge of setting home run history.
Projections accurately predicted the dominant Cubs and the embarrassing Braves. But they didn’t do so well for a handful of players and teams. Since the algorithms look at long periods of time (typically several years), huge and abrupt changes to MLB’s underlying environment take a while to be reflected in their code. So with that lag time in mind, be wary of the projections next year. Team records are likely to be more crunched together in 2017 than they were this season, as cellar-dwelling teams such as the Phillies, Braves and Brewers progress toward the mean and escape from their rebuilds. Further, Rob Manfred is considering significant rule changes that could alter the game in ways no projection system can account for. That means 2016 might have been the exception to baseball’s new rule of unpredictability and chaos.