Posted by Alex Bonilla on May 14, 2019
Starting today, while you browse Baseball-Reference, you can find probabilities of each team to reach the postseason, win the division, and advance to each playoff round including winning the World Series.
To compute these odds, we simulate the rest of the season and the postseason 1,000 times each day. The methodology relies on Baseball-Reference’s Simple Rating System (SRS), which provides a strength-of-schedule-adjusted rating of each team, expressed in runs per game better or worse than an average team.
Prior to going into the details, we should tell you what our goals were for the system. Systems can vary in what they focus on, so having a clear idea of the questions we are trying to answer can add some insight and guide you in how you might use the system. We wanted a relatively simple system that would most accurately estimate the team's end of the year win total. This system could answer questions such as: Should a team go for it at the trade deadline? or Is a team in second place at the All-Star Break likely to fall off or contend for the division? or Is it too early to be certain a hot start will continue? This system is not designed to predict World Series win odds as well as possible since it's tuned with regular season data only. We are assuming that teams are as likely to win in the postseason as they are in the regular season and this is probably a poor assumption given the increased importance to bullpens and superstar starting pitchers.
Additionally, since we wanted a simple system, we are not considering player movement at the trade deadline or individual pitcher matchups which could become relevant during the final games of the season. If you want a more complicated system that considers roster composition, we would point you to the fine system at Baseball Prospectus or FanGraphs.
Typically, SRS is calculated and displayed (for example, on the standings page) based on the season to-date. For the purposes of the playoff odds simulation, though, we are calculating a value of SRS using each team’s previous 100 games, adding in 50 games of .500 ball for regression to the mean. After a lot of backtesting, these are the numbers that provided the most predictive value. Running the simulation as of July 15 and August 15 of each year from 2009 to 2018, the simulation produced a root-mean-square error of 4.63 wins when compared to teams’ actual end-of-season win totals. For example, last season, both the July and August simulations predicted the Atlanta Braves within 1 win of their eventual season total of 90. This error was the lowest of any of the 50 potential inputs we considered. It was lower than a system that used just the current season SRS, any system with no regression to the mean, and, as a sanity check, a system that just flipped a coin for each game.
Of course, using past performance to predict future performance has its quirks, especially early in the season. For instance, look at the Philadelphia Phillies, who experienced significant roster turnover this past winter. The Phillies added Jean Segura and J.T. Realmuto via trade, as well as David Robertson and Andrew McCutchen via free agency (I think that’s everybody). Looking back over their final 100 games of 2018, Philadelphia’s SRS comes in at -0.7. In other words, they were 0.7 runs per game worse than a league average team.
As we get further into the season, the numbers start to shift, as 2019 performance makes up a larger portion of that 100-game population. Through the games of May 12, Philadelphia’s SRS value over the past 100 games is -0.6, boosted by their 0.4 value in the current season.
While teams like Philadelphia have obvious additional context to keep in mind, using a system that takes into account last season’s performance as well as this season’s prevents the simulation from being fooled too early on by a team that’s simply off to a hot start. The result is a more skeptical simulation that needs to be convinced over time that a club’s new success is legitimate.
Check out this season’s current playoff odds for all teams here, and be sure to check out team pages to see how a team’s odds have changed over time.
If you have any questions or suggestions, feel free to contact us through our feedback form.