Why Do Good Streaks Happen to Bad Hitters?

A lot of my job involves spinning a story I don’t completely believe in. I know, I know, you’re shocked! You mean I don’t actually think that the four to five players I highlight every week are each breaking out by doing something they’ve never done before? And I don’t think that each of them is doing it sustainably? What are the odds?
Some of that comes with the territory. If you’re looking across the universe of major league players for something interesting, some portion of what you find interesting will have happened by random chance. That pitcher who’s striking everyone and their mother out? He might just be on a hot streak. The hitter who’s currently smashing high fastballs? There’s some chance he just felt really good for a week and then will stub his toe when walking out of the clubhouse tomorrow.
I know all that. One thing I wasn’t sure about, though, was how often false signals pop up. Even without searching them out, you might end up seeing a breakout around every corner. There’s a famous quote from Nobel Prize winning economist Paul Samuelson: “The stock market has predicted nine out of the last five recessions.” Is the same general idea true of batted ball data? I came up with a simple experiment to investigate. What follows is a breakdown of the exact method I used, but if you’re just interested in the conclusion, it won’t surprise you: When hitters put up hot streaks of a reasonable length, it’s a good but not infallible sign that they will finish the year as above-average hitters.
I took every batted ball from the 2022 season and broke it out by player. From there, I put them all in chronological order and calculated each player’s best stretch of 50 batted balls. I calculated it for a variety of “advanced” metrics: average exit velocity, xwOBA, and barrels per batted ball. Those are some of the most commonly used underlying statistics – if I’m citing someone who’s really hitting, I’d likely use batted ball outputs like this to assess the validity of their performance, so I excluded things like batting average on contact or wOBA on contact, which might be quite noisy in 50-ball samples. Read the rest of this entry »