For any of you modelling baseball games with Markov chains, I am curious how you approach games that involve call-ups, either as starting pitchers or players in the lineup, since there is either nothing or very little in the way of accrued sabermetric statistical data regarding these players. Do you use "replacement player" data? Certainly that approach would have its inaccuracies, especially in regards to starting pitchers. In light of the impending September call-ups, I'm wondering how this aspect is generally approached. I tend to stay away from these games. However, I suspect that there may be the opportunity for increased edge in these situations due to the public's unfamiliarity with the new players and the unavailability of pertinent performance data.
-
-
I definitely think you can exploit the public's unfamiliarity with minor league call-ups. How do you factor that into statistical data? I'm not really sure. My only guess would be to find as much information on the player since he left high school as you can and then scale it down. There has to be some sort of factor one can use to compare AAA stats to major league stats. I know they do it all the time with Japanese players.
One example would be with Fukadome (sp) on the Cubs. e.g. "Oh he hit .420 last season in Japan. He should be hitting around .300 now."
I would be super conservative with these "interpolations" if you will. Living in Vegas I've seen amazing players on the 51s who turn out below average once they're called-up by the Dodgers. I'll give you some specific proof. The current second baseman for the 51s is great. He was called up by the Dodgers, sucked, and sent back down to AAA where he was great again.
The MLB is the greatest baseball organization in the world. No matter how good a team is in AAA, it won't compare to an MLB team. I think it's fair to say the same goes for Japanese baseball, which is like AAA.Comment