from Amazon
Monday, January 22, 2007
Slave Wages
MLB operates on a weird system. They preallocate their MLB payroll based on their revenues (say 55%-60% of revenue goes toward their players). However, according to The Harball Times Annual 2007, the amount they pay their arbitration-eligible players is half what they pay for free agents, on a per-win basis. You would expect that they’d pay the market rate for the free agents, and pay half the market rate for the arbitration players, and peanuts for the slave players, so that overall, they’re paying way under the true marginal $ per marginal win. They don’t.
Anyway, looking at arbitration players:
According to Burger and Walters, the % of MRP for the 3, 4, 5, and 6-yr class players is 31%, 44%, 61%, 64%. This gives us a simple average of 50%.
I don’t have my THT book handy, but the numbers provided there was something like 250,000 per win share (WS) in the 3-yr class, 500,000 per WS in the 4-yr class, 1 million per WS in the 5-yr class, and 1.15 per WS in the 6-yr class. Free agents get 1.5 million per WS. Scaling the THT numbers against the free agent class, where free agents get “100%”, the 4 arbitration classes are: 17%, 33%, 67%, 77%. The weighted average was around 50%.
The Burger/Walters numbers was based on 391 arbitration hearings cover many years, and I have no idea how they determined player value. My guess is that positions were not considered, and perhaps fielding was ignored. The THT numbers were based on the 2006 season, but used Win Shares. While Win Shares has lots of problems, and there is no bigger critic of the metric than me, it’s also pretty useful in group settings, where alot of the systematic biases are overwhelmed by the sample.
I get the feeling that the THT numbers make more sense these days. I think, as a rule-of-thumb, a 20%, 40%, 60%, 80% makes the most sense. When I eventually create the salary calculator for arbitration players and free agents, those are the numbers I’ll use. Unless of course, someone wants to extend the Hardball Times work to past years.