Monday, October 19, 2009

Session Handicapping

From an economics viewpoint (expected utility theory), the prescribed way to approach a series of bets is to focus on expected utility (for example. wealth). This leads invariably to Kelly staking and maximizing the long-term expected growth rate of one's bankroll. With this approach, the handicapper is advised to play every race where he has an edge (namely, "bet your beliefs"). By contrast, from a psychology viewpoint (prospect theory), the prescribed way to approach a series of bets is to focus on loss aversion (pain of losses far outweighs joy of wins). This leads invariably to session handicapping and minimizing regret.
If for example, we define session handicapping as a day's wagering, then it is possible to adapt Belgian mathematician Thomas Bruss’s Odds Algorithm ( to determine when is the optimal time to stop betting to enhance the probability of ending the day in profit. In other words, the odds algorithm works out after which race during a day's session you should quit assuming you are ahead!

Tuesday, October 06, 2009

Trade Selection (Trailing Stop Loss / Take Profit)

I would like to propose a betting wizard enhancement for betting exchange tools that would be of great value to both Pre-Event Scalpers and In-Play Traders. The new wizard, Trade Selection, would include options to set both trailing stop loss and take profit thresholds for all bets, as standard.

Friday, October 02, 2009

Sample Size: Confidence Interval and Confidence Level

There are two key factors to consider when estimating sample size for testing a betting rule: confidence interval and confidence level. Confidence interval refers to the range within which you expect the correct strike-rate to fall and confidence level refers to how certain you are that this range holds the true value. For example, with a confidence interval of +/-5% and a confidence level of 99%, using these values and without going into the exact calculations [=POWER(PRODUCT(PRODUCT(NORMSINV(PRODUCT(SUM(ConfLevel,1),1/2)),1/SQRT(2)),1/PRODUCT(SQRT(2),ConfInterval)),2)] gives a sample size of approximately 664 (663/664). What this means is that testing a betting rule on 664 races drawn randomly from, for example, 2008 (i.e. different population from that used to discover betting rule), you can be 99% certain that the strike rate your betting rule generates from this new random sample will be within +/- 5% of the true strike rate.