Team regression modification and back test results

I have not heard back from the fellow that originated this approach to handicapping MLB money lines. Therefore, I can’t verify that my program code replicates his Excel approach perfectly. I decided to move forward and make a couple of modifications to his approach and back test again.

The first is the manor of expressing the ML odds. He subtracted 100 from to ML’s to eliminate the gap between -100 and +100, then did linear regression. This is a technique I hadn’t seen before, so I elected to convert the ML’s from the American style to the European style and do the regression. Converting these does make the favorite odds non-linear, but it is a more traditional approach.

Te second change dealt with how to combine each team’s expected win probabilities into a normalize form (adding up to 1.00). I couldn’t quite figure out how he did this, and he didn’t venture an explanation when asked. I elected to again go a more traditional route and use Bill James’ Log5 technique.

Following are the results when I made both changes and included 2018 data to the All-Star break.

Eur and Log5	Number	Amt Bet	Net	Actual Ret/$
2010	1,960	$246,779	-$2,651	$0.989
2011	1,963	$239,981	-$3,934	$0.984
2012	1,971	$242,239	-$3,164	$0.987
2013	1,971	$249,978	-$9,798	$0.961
2014	1,941	$231,529	-$8,865	$0.962
2015	1,949	$237,962	-$9,495	$0.960
2016	1,942	$239,530	-$21,387	$0.911
2017	1,945	$247,642	-$5,172	$0.979
2018	968	$122,265	-$134	$0.999
Total	16,610	$2,057,905	-$64,600	$0.969

The results appear to be almost random, no better than throwing darts. That will terminate my look at this approach. This is the second system that was described in the “Handicappers Think Tank” forum in www.sportsbookreview.com that I decided to analyze. I probably won’t be chasing anymore of these unless I can see some logic behind why it might work.

I probably will limit myself to systems I’ve developed in future posts. Follow me on Twitter, @old44bill, to know when I post again.