## Friday, August 18, 2017

### Is It Really A Mystery How The Royals Exceed The Expectation Of Computer Projections? (it might be timing as they perform relatively better in high leverage situations)

That is what a Wall Street Journal article says. See The Baseball Team That Computer Models Can’t Figure Out: The Kansas City Royals have exceeded the expectation of every prominent computer projection over the past five years by Jared Diamond. Excerpts:

"PECOTA, Baseball Prospectus’ sophisticated computer model for forecasting on-field performance, is impressively accurate. Half of teams have, on average, finished within 2.75 wins in either direction of PECOTA’s predictions since 2013, with nearly two-thirds of teams coming within 3.25 wins.

But there’s still one team PECOTA can’t figure out: the Kansas City Royals, the unexplained mystery of the major leagues. The Royals outperformed PECOTA by an average of 12½ games from 2013 through 2016, the most in the sport over that span, and are on pace to beat it by about that much again this season."

"The problem with predicting the Royals, prognosticators say, is that they have consistently won more games than their underlying statistics would indicate. They’ve been, as Davenport put it, “far more efficient at winning games with the number of runs they score and allow.”

That spells trouble for the models. Systems like PECOTA, which stands for Player Empirical Comparison and Optimization Test Algorithm, use piles of statistics and historical aging curves to predict how individual players will fare in a given season. After accounting for a team’s projected depth chart, the systems use that data to predict how many runs a team will score and allow, which results in a predicted record."

The article did mention their bullpen as one reason. If your relievers can come in and stop rallies consistently better than other teams, you will win more than expected.

Here is a regression generated formula to predict winning pct based on team OPS differential High, Medium and Low leverage situations. It is based on all teams from 2010-2014.

Pct = .5 + .306*LOW +.420*MED + .564*HIGH

If applied to the Royals of 2013-2016, it predicts them winning fewer games than they actually did. But not by alot. Here are the differences

-1.88
-2.87
-3.71
-0.76

That averages out to 2.31 wins per season. So over 2013-2016, the Royals won 2.31 more games per season than predicted. Not too big of a difference. The table below shows the OPS their hitters had, OPS their pitchers allowed and the differential in the three situations for each season. Notice how they tend to do better both hitting and pitching in High leverage cases than otherwise. OPSA is OPS allowed.

Their 4 best differentials are all in High cases. Of the 8 Low & Medium cases, 6 differentials are negative.

 Year Split OPS OPSA Diff 2013 High 0.733 0.672 0.061 2013 Low 0.674 0.688 -0.014 2013 Medium 0.693 0.723 -0.030 2014 High 0.712 0.636 0.076 2014 Low 0.660 0.700 -0.040 2014 Medium 0.713 0.696 0.017 2015 High 0.772 0.669 0.103 2015 Low 0.719 0.697 0.022 2015 Medium 0.734 0.748 -0.014 2016 High 0.728 0.653 0.075 2016 Low 0.712 0.778 -0.066 2016 Medium 0.704 0.767 -0.063