Kelly discussions

As hoped for, there was quite a response to yesterday’s post. Several people responded on sportsbookreview.com in the Handicapper Think Tank forum on the topic Kelly Re-Visited. As expected, most were supportive of the Kelly approach. Following are some of the points made.

• In the first response user “u21c316” pointed out the main flaw in the argument made in the anti-Kelly article. The writer compared the results of Kelly wagers to flat wagers equal to the average Kelly bet. “u21c316” points out that you would not know the average Kelly wager until the end of the season and can’t then go back to make those wagers.
•  In the same response, “u21c316” also points out that the Kelly approach is designed to maximize your bankroll by creating an exponential growth pattern.
•  In another post user “KVB” indicates the approach taken depends on your goals and time frame. He argues that using flat bets and increasing them periodically seems to be a sounder approach.
•  “KVB” also brings up using fractional Kelly. Several concur with this. Using “half Kelly” will result in growth, but with less volatility.
• "Larkman” brings up the issue of the results being a function of how accurate your system is in providing a good predictive edge.
• User “HeeeHAWWWWW” has apparently done some studies and quantifies the risk of using full Kelly.
• Finally, “oilcountry99” summarizes the key issue. “How many people can accurately predict their edge?”

It does come down to what you are trying to accomplish. If you are beginning a season and are willing to risk $1,000 only, then Kelly is a good approach. One thing not mentioned is while Kelly could provide exponential growth in a good season, it limits your loss in a bad one. By only wagering a percentage of your bankroll, your bankroll can never go to zero. A very positive reason for using Kelly.

So, Kelly provides potentially limitless growth and a very specific limited downside, why not use Kelly? My experience leads me to a few concerns.

1. When compared to flat bets of the same average size, Kelly seems to under produce slightly.
2. When you win, Kelly raises your wagers. When you lose, Kelly lowers them. At the end of the season, the average wager on losing bets is higher than the average wager on winning bets.
3.  What impact does long losing streaks have on your results?

To resolve my concerns and address questions like #3, I’ve decided to build a model of wagers using Kelly, partial Kelly, and flat bets. This model will have a known expected return that is precisely accurate. I’ll post more on this as it develops. Again, I’ll tweet at @ole44bill as I make these posts. I’ll also be posting on the Kelly-Revisited forum and look forward to comments and criticisms there.