Kelly decision re-visited

In my previous post, I had concluded that a star-based approach was the best solution for me for the remainder of the MLB season. But, when I looked at the results experienced thus far this season, it was impossible to assign values to the breakpoints for the stars. I thought it might be because there were only about 400 data points and went back to the testing results for the past 8 years which had over 3,000 data points. The problem is that as the expected returns increase, the actual results don’t increase evenly. Following is a chart with these bets divided into groups by expected return.

Group	Bets	Net	Ret/$
1.00 to 1.05	965	$8,044	$1.08
1.05 to 1.10	827	$8,275	$1.10
1.10 to 1.15	704	$3,776	$1.05
1.15 to 1.20	343	$4,948	$1.14
1.20 to 1.25	234	$1,962	$1.08
1.25 above	156	$2,646	$1.17
Total>1.00	3229	$29,651	$1.092

If the actual results grew like the expected results, then a progressive wagering scheme would be appropriate. But they don’t so flat bets make more sense. I do have a secondary system that is better behaved.

Group	Bets	Net	Ret/$
1.00 to 1.05	702	$3,585	$1.05
1.05 to 1.10	285	$3,164	$1.11
1.10 to 1.15	74	$3,249	$1.44
1.15 to 1.20	13	$275	$1.21
Total	1074	$10,273	$1.10

This one will work well with a progressive system like Kelly. In fact, when I applied Kelly to the previous data, the overall return per dollar increased from $1.10 to $1.13. So, I’ve switched to using Kelly for this system.

That concludes the project of looking at different betting schemes and it’s time to move on to other projects. Follow me on Twitter, @ole44bill, to know when I resume posting.