The Fantasy Baseball Sherpa

It happens every year, just like the swallows returning to San Juan Capistrano. A previously undistinguished hitter gets off to a fast start, and sportswriters speculate on whether a major league hitter will ever be able to hit .400 for an entire season again (I believe the answer is “no”, but that’s another topic for another day). If the same hitter were to hit .400 for 15 games in the middle of the season rather than at the beginning of the season, most of us would barely raise an eyebrow. However, due to what I referred to in a previous post as “Tuffy Rhodes Syndrome“, baseball fans tend to give a disproportionate amount of weight to events at the start of the season.

This year’s fast starters include A.J. Pierzynski (.421 as of this morning), Jason Kendall (.405), Angel Pagan (.385), Nate McLouth (.383), Luke Scott (.375), and Kurt Suzuki (.370). Obviously, none of these hitters will have a batting average anywhere near .400 when the season ends. But how many of them will finish with even a .300 average? Again, the answer could very well be zero.

So, how should you go about forecasting a batting average for the remainder of the season? Let’s use Pierzynski as an example. Suppose that going into the season you expected Pierzynski to hit .270 for the season. How should that expectation be combined with the .421 he’s hit through approximately the first 10% of the season? I’ve read a number of fantasy sportswriters’ articles on this subject, and their approaches usually fall into one of 2 categories: (1) expect Pierzynski to finish the season with his expected average of .270 (which implies that his average for the remaining 90% of the season will be .253); (2) expect Pierzynski to hit .270 for the rest of the season (which implies that his batting average for the season will be .285).

I disagree with both of these approaches. The first is an example of what statisticians refer to as the Gambler’s Fallacy, which means that (supposedly) independent events (such as future at-bats) are entirely dependent on past events. Andy Behrens, a very thought-provoking and entertaining fantasy sportswriter for Yahoo, had a great description of the Gambler’s Fallacy in a post he made yesterday. The second approach goes too far in the opposite direction, assuming that what a hitter has done season-to-date has zero predictive value in forecasting what he’s likely to do for the remainder of the season.

I suggest a third approach that combines what the hitter was expected to do with what the hitter has actually done in order to forecast what he’s likely to do for the remainder of the season. There are several possible weighting schemes, but for the sake of simplicity, I’ll go with a linear weighting scheme (i.e. - if the season is 10% complete, the hitter’s actual results should receive 10% weight, and his expected results should receive 90% weight). Applying this approach to the Pierzynski batting average example suggests that a reasonable forecast for Pierzysnki’s batting average for the rest of the season is .285 (which implies that his batting average for the season will be .299).

Some may still argue that .270 is a better forecast than .285. Let’s look at another example, this one from last season. If you expected Andruw Jones to hit .260 for the season, but he’s hitting just .211 at the All-Star break, would you still expect him to hit .260 for the remainder of the season? Probably not. Since the All-Star break occurs after roughly 55% of the season has been played, I would have forecast a rest-of-season average for Jones of .233 (= 55%*.211 + 45%*.260). Jones actually hit .236 for the rest of the season. I realize that one cherry-picked example doesn’t prove my argument, but hopefully, you get the idea.

How can you use this information to your advantage in your fantasy leagues? People often talk of wanting to “sell high and buy low” with respect to making early-season trades, but do you actually have the backbone required to do so? If so, congratulations - you’re probably well on your way to scooping up some above-average players at below-average prices. If not, re-read the above, pick some real-life examples from the current season, and follow them.

Others may have an easier time selling high on a fast-starting player than buying low on a slow-starting player. Who are some of this year’s “slow starters” who may be ideal buy-low candidates? C.C. Sabathia and Roy Oswalt come to mind immediately on the pitching side, while Carl Crawford, Alfonso Soriano, Robinson Cano, and Ryan Braun are among the hitters off to sub-par starts. A savvy team owner will rebuff your attempts to trade for one of these players, but some may be willing to part with these players for a below-market offer.

I’ll leave you with an example I witnessed last season. A friend had Alex Rodriguez on his team, but was struggling in the pitching categories. His league required that all trades be balanced from a position standpoint (i.e. - you couldn’t trade a Third Baseman straight up for a Pitcher). In late May/early June he took advantage of a fellow owner’s willingness to sell low on Garrett Atkins and buy high on the fast-starting Boof Bonser, trading A-Rod and Boof Bonser in exchange for Garrett Atkins and Johan Santana. As you might expect, my friend was able to climb a number of places in his league’s standings after pulling off that trade.

Until next time,

The Sherpa