How well do cold streaks predict the future performance of pitchers?

June 21 11:58am
Derek Carty

Over the past three weeks, I’ve been looking at how predictive hot and cold streaks are for baseball players. Thus far, I’ve looked at hot and cold streaks for hitters and hot streaks for pitchers, finishing up this week with cold streaks for pitchers.

The Study

I’ll be using the same methodology I’ve been using all along. To study cold streaks, I’ll be comparing the stats from the pitcher’s 3-start streak to his next start. I’ll then compare the pitcher’s preseason Marcels projection to his next start. This will tell us which is more predictive – a full projection for a player or the streak. When comparing these sets of numbers, I’ll be using average error to study the differences.

I looked at all possible 3-start streaks for all pitchers back to 1993 that occurred in either March, April or May (I did this so that the preseason projections would be as accurate as possible – if we’re looking at streaks in August, it’s likely the player’s projection would have changed considerably by that point in the season). From there, I only included a streak if all three starts came within an 18-day window and the fourth start came no more than a week after the third. No context adjustments were made to the data.

“Cold” Pitchers – 3 starts

While I’ll be looking at a number of different stats, I decided to define a “cold” streak based upon FanDuel points accumulated (based on FanDuel’s MLB 35k format). The average pitcher accumulates roughly nine points per game, so to be included among the “cold streak” population, a pitcher must have averaged fewer than 4 FanDuel points over a three-start period. Here are the average errors for a number of different stats, keeping in mind that the lower the number, the better. All stats were scaled to league average before comparison except FanDuel points for clarity’s sake.

As we saw for hot streaks last week, for all of the stats that we look at, the preseason projection is better predictor of the start immediately after a 3-start cold streak than the cold streak itself is. However, for strikeouts and FanDuel points, it’s quite close.

“Cold” Pitchers – 5 starts

If we repeat the study over a five-start period (starts must have come within a 30-day window), we get the following results:

This time, we again see that the projection is better for ERA and WHIP, but for strikeouts and FanDuel points, now the streak becomes slightly better. We should note that the sample size drops quite a bit here, though.

“Cold” Pitchers – 7 starts

Repeating the study for seven-start periods (starts must have come within a 42-day window), our results look like this:

We get very similar results here to the 5-start window with ERA and WHIP preferring the projection but strikeouts and FanDuel points locked in a dead heat between the projection and the streak. The sample gets even smaller here, though, as a number of different categories of players start dropping out: 1) those whose streaks were simple random variation, 2) those who performed so poorly as to be sent to the minors, 3) those who performed so poorly as to be released, 4) those whose performance was injury based and who have been placed on the DL. This makes it difficult to trust these results fully (and even the 5-start results to a slightly smaller extent), but it does lend at least some evidence to the credibility of streaks.

Final results

Just like we found with hitters, cold streaks do seem to have some credence after a little while, but because our sample size dwindles so much as the streaks get longer, it’s hard to really trust them. And even after a 5 or 7-start cold streak, the streak is still essentially tied with the projection.

Also necessary to note is that Marcel is the most basic of projection systems. Using a more complex system like Baseball Prospectus’s PECOTA or THT’s Oliver would certainly narrow the gap. Also, remember that I used a preseason projection. Preseason projections won’t change much by the end of May on the whole, but they will change a little, especially when a pitcher performs so poorly as to be put into the “cold streak” bucket.

Cold streaks are definitely real for some pitchers thanks to injuries, mechanical issues, etc, but today’s shows that it’s difficult to trust cold streaks (even if they’re more reliable than hot streaks) unless we have some outside information about the player that leads us to believe it’s more than just random variation.

Concluding thoughts

That wraps up my series on streaks.  As always, if you have any questions, feel free to comment or e-mail me. Also, be sure to add me as a friend on Facebook and follow me on Twitter.

Derek Carty’s work can also be found at Baseball Prospectus, CardRunners Fantasy Baseball, and DerekCarty.com. He has previously had his work published by The Hardball Times, NBC’s Rotoworld, Sports Illustrated, FOX Sports, and USA Today. He is the youngest champion in the history of LABR, the longest-running experts league in existence, and is a graduate of the MLB Scouting Bureau’s Scout Development Program (aka Scout School). He welcomes questions via e-mail, Facebook, or Twitter.

Over the past three weeks, I’ve been looking at how predictive hot and cold streaks are for baseball players. Thus far, I’ve looked at hot and cold streaks for hitters and hot streaks for pitchers, finishing up this week with cold streaks for pitchers.

 

The Study

I’ll be using the same methodology I’ve been using all along. To study cold streaks, I’ll be comparing the stats from the pitcher’s 3-start streak to his next start. I’ll then compare the pitcher’s preseason Marcels projection to his next start. This will tell us which is more predictive – a full projection for a player or the streak. When comparing these sets of numbers, I’ll be using average error to study the differences.

I looked at all possible 3-start streaks for all pitchers back to 1993 that occurred in either March, April or May (I did this so that the preseason projections would be as accurate as possible – if we’re looking at streaks in August, it’s likely the player’s projection would have changed considerably by that point in the season). From there, I only included a streak if all three starts came within an 18-day window and the fourth start came no more than a week after the third. No context adjustments were made to the data.

“Cold” Pitchers – 3 starts

While I’ll be looking at a number of different stats, I decided to define a “cold” streak based upon FanDuel points accumulated (based on FanDuel’s MLB 35k format). The average pitcher accumulates roughly nine points per game, so to be included among the “cold streak” population, a pitcher must have averaged fewer than 4 FanDuel points over a three-start period. Here are the average errors for a number of different stats, keeping in mind that the lower the number, the better. All stats were scaled to league average before comparison except FanDuel points for clarity’s sake.

TABLE 1 HERE

As we saw for hot streaks last week, for all of the stats that we look at, the preseason projection is better predictor of the start immediately after a 3-start hot streak than the hot streak itself is. However, for strikeouts and FanDuel points, it’s quite close.

“Cold” Pitchers – 5 starts

If we repeat the study over a five-start period (starts must have come within a 30-day window), we get the following results:

TABLE 2 HERE

This time, we again see that the projection is better for ERA and WHIP, but for strikeouts and FanDuel points, now the streak becomes slightly better. We should note that the sample size drops quite a bit here, though.

“Cold” Pitchers – 7 starts

Repeating the study for seven-start periods (starts must have come within a 42-day window), our results look like this:

TABLE 3 HERE

We get very similar results here to the 5-start window with ERA and WHIP preferring the projection but strikeouts and FanDuel points locked in a dead heat between the projection and the streak. The sample gets even smaller here, though, as a number of different categories of players start dropping out: 1) those whose streaks were simple random variation, 2) those who performed so poorly as to be sent to the minors, 3) those who performed so poorly as to be released, 4) those whose performance was injury based and who have been placed on the DL. This makes it difficult to trust these results fully (and even the 5-start results to a slightly smaller extent), but it does lend at least some evidence to the credibility of streaks.

Final results

Just like we found with hitters, cold streaks do seem to have some credence after a little while, but because our sample size dwindles so much as the streaks get longer, it’s hard to really trust them. And even after a 5 or 7-start cold streak, the streak is still essentially tied with the projection.

Also necessary to note is that Marcel is the most basic of projection systems. Using a more complex system like Baseball Prospectus’s PECOTA or THT’s Oliver would certainly narrow the gap. Also, remember that I used a preseason projection. Preseason projections won’t change much by the end of May on the whole, but they will change a little, especially when a pitcher performs so poorly as to be put into the “cold streak” bucket.

Cold streaks are definitely real for some pitchers thanks to injuries, mechanical issues, etc, but today’s shows that it’s difficult to trust cold streaks (even if they’re more reliable than hot streaks) unless we have some outside information about the player that leads us to believe it’s more than just random variation.

Enhanced by Zemanta

Posted 4 years ago on · Permalink

About FanDuel

FanDuel is the market leader in one-day fantasy sports leagues. Play for real money against other people with immediate cash payouts. Up to $1M cash paid out every day.

Read more

How to get started

  1. Signup & use promo code INSIDER100
  2. Choose a one-day contest
  3. Pick any players you want - Salary cap format
  4. Win your league, and win real money
Get Started