jeff232323 said:
28 of his last 100 picks have had higher juice than -110 so not even sure -110 is a fair number to assume here. Some has high as -190 or more.
Jeff,
You are correct that a -110 betting line is only approximate and NOT a "fair" metric when Moneyline plays are included in the record.
I updated the record to include picks through April, 24, 2016, which was a bad month. Now the overall win% has dropped to 53.12%, 936-826-35, (excludes "FREE" picks) down from the 53.5% that Andy correctly quoted through March, 2016. I have computed the average of all the breakeven percentages for the Fezzik log plays. It is 53.08% which corresponds to a -113.1 average betting line. As we see overall the picks, when unweighted, have been just about breakeven. When given the prescribed weight, the record shows a profit before fees because the 3 unit plays have done very well.
There is a major flaw in using a full Kelly bet. Sports betting is not like Blackjack where you may have billions of hands simulated. As we have seen here, even with almost 1800 plays since inception, part of a bad month (51 picks) has taken the win% down by almost 0.4%. To achieve a win % confidence interval of +/-1% with a confidence level of 95% we would need roughly 8,000 more plays. This means with 1,800 picks, we are cannot be sure the average win rate so far is all that close to the eventual "long-term" win rate. Note this model applies to most handicappers, but does not apply to scalpers and middlers who are using correlated bets. Almost all cappers using Kelly limit their bets to a fraction such as 1/2 or 1/4 of full Kelly.
As one might expect, using the example of $48K bankroll, 1/2 Kelly with a win rate 53.12% at -113.1 suggests a very small bet and produces almost no profit. Incidentally, if we use the average betting line of -113.1 instead of -110, 1/2 Kelly instead of full Kelly, but $48K bankroll, 1000 picks, and 53.3% win rate, we get an expected profit of $496 rather than $17,200.