Why You Should Focus on Strikeouts (And How to Predict Them)

Pitchers are clearly more consistent than batters from game to game, which is why allocating a large percentage of your salary cap to your two hurlers on DraftKings is smart; you’re increasing your lineup’s floor and upside at the same time.

Even within the pitcher position, though, there are different levels of consistency. Outside of GB/FB rate, K/9 has proven to be the most consistent stat for pitchers—far more consistent than ERA, WHIP, and even BB/9.

That means that the flamethrowers who generally rack up a lot of Ks are also the most consistent types of pitchers. They rely on a consistent stat for fantasy production, not something that’s rather volatile like earned runs or, worse, wins.

And which types of pitchers are the ones who generate a lot of strikeouts? The expensive ones, for the most part. Thus, not only is it smart to pay top dollar for expensive pitchers because they’re more reliable than hitters, but also because they’re more reliable than other pitchers (not just in terms of bulk production, but rather in terms of how consistently they produce at specific rates).

Using the FanGraphs Stat Correlation Tool, I charted the correlation between various pitching stats from one year to the next (with the ‘next’ year being Year Y+1). The correlations that are in boxes are those for how well a specific stat predicts itself. How much does a pitcher’s ERA carry over from year to year, for example? The data suggests not all that much…

Pitcher1

There’s a 0.311 strength of correlation between a pitcher’s ERA from one year to the next, which isn’t very strong. Compare that to 0.703 for K/9, for example, or 0.752 for GB/FB. Pitchers who induce a lot of ground balls or whiff a lot of batters tend to do so every year, while a pitcher’s ERA can fluctuate to a large degree.

I’m analyzing season-to-season numbers because I think they have usefulness on the nightly level. Yes, the odds of an individual hitter going deep or a single pitcher racking up 12 Ks isn’t great in any given game, but if you’re continually giving yourself exposure to the players with the highest probabilities of favorable outcomes—the players who are expected to produce over the long run—the variance will eventually even out. With daily fantasy baseball, you’re just trying to assert those small edges here and there, night in and night out.

Going back to ERA, you can see that the best predictors of it, by far, have been xFIP and (in particular) SIERA. This is the reason I emphasize SIERA when analyzing my arms: it works. When a pitcher’s ERA is up early in the year but his SIERA is low, it suggests he isn’t throwing the ball that poorly and that’s probably a great time to buy low on him. And vice versa, when the ERA is down but the SIERA is up, there’s a good chance that DraftKings has that pitcher overpriced.

Note that K/9 is actually better than past ERA at predicting future ERA. The correlation between K/9 and ERA in Year Y+1 is -0.352. The negative value doesn’t mean anything other than that as strikeouts increase, ERA decreases. So not only are strikeouts extremely consistent and a massive part of a pitcher’s fantasy value, but they also predict success in other areas—like allowing runs—better than you might think. And if you look at batting average, you’ll see that K/9 is actually the best predictor available, i.e. the most accurate way to predict a pitcher’s future batting average allowed is to look at his strikeout rate.

We see similar effects across the board with pitcher stats, with SIERA and K/9 being the shit when it comes to making accurate predictions.

 

Maximizing Strikeout Probability

Okay, we want strikeouts. Got it. Now how the hell do we get them?

One way is to use astrology to target pitchers. I’ve always found that Libras start off the year really slowly, for example, and Pisces are unlikely to go nine full innings (they don’t have long attention spans).

And don’t even get me started on Scorpios!

A slightly more scientific way of predicting strikeouts is to use numbers. It’s a crazy idea, I know; imagine picking your pitchers without even looking up their signs! (LOL, yeah right).

If you’re going to insist on being a total dumbass and bypass astrology in favor of data, you can use the K/9 for both a pitcher and the offense he’s facing to project strikeouts. For pitchers, I simply use their K/9 total from the previous season since the best predictor of future strikeouts is past strikeouts.

Then we need to adjust for the opponent. To do this, I calculate the K/9 for each batter in the lineup versus the handedness of the pitcher. So if we’re projecting Kershaw against the Giants, I’d look at the strikeout rates for the Giants versus lefties—or more specifically, the total K/9 for each player in the lineup versus southpaws.

Then we have two numbers: a K/9 for the pitcher and a K/9 for the offense. Let’s say that both of those numbers are 9. You might think you can project the pitcher at exactly nine strikeouts in that situation, but you’d be way off. We need to account for league baselines when projecting a stat like strikeouts.

Last season, offenses averaged 7.70 strikeouts per game—a number that continues to increase each and every season. Here’s a look at total strikeouts per game from 2005 to 2014.

Pitcher2

Strikeouts per game are up 22 percent since 2005, which is pretty insane. Here’s a look at the increase in strikeouts per game since 2005.

Pitcher3

Based on recent trends, we should expect somewhere around 15.7 strikeouts per game in 2015 (or 7.85 per team). That’s going to be our league baseline for strikeouts—7.85.

In the hypothetical game of a nine-strikeout pitcher versus a nine-strikeout offense, we have two inputs that are both greater than the league mean. So if we have a pitcher who records nine strikeouts per nine innings, on average, then we’d naturally project him above his average facing an offense that whiffs more than the typical team, right?

Of course. So here’s how I do that:

Pitcher K/9 + Offense K/9 – Mean K/9

Add the pitcher’s K/9 to the offense’s K/9, then subtract the league mean K/9. In our hypothetical, that would be 9 + 9 – 7.85 = 10.15.

Note that the only time a pitcher’s K/9 would match our projection for him would be if the opposing offense struck out at exactly the league average rate. This is a method, though imperfect, that I’ve found can help provide a nice baseline to predict strikeouts.

Of course, we’re not actually using the K/9 totals as a strikeout projection since we’d never anticipate a pitcher going nine full innings in any single game. You can adjust the final K/9 total as you’d like based on the number of innings you think a pitcher might throw—obviously a guy like Strasburg and similar players have a tendency to get pulled early in games—but I find projecting innings to be pretty challenging on the day-to-day level.

Ultimately, the K/9 projections are more of a tool in our arsenal than an exact number on which we want to place a ton of emphasis. I care more about using K/9 to visualize upside and to rank pitchers according to strikeout probability than I do creating a fragile projection system.