It doesn't involve handicapping except to determine the horses extremely unlikely to finish in the money.
The main theory behind it is that the betting public generally do a pretty good job of determining the chances any particular horse has to win. The problem with handicapping is that you have to overcome a 20% track take for wps bets just to break even -- all but impossible at my skill level.
I use an exotic bet, the quinella. To win a quinella bet, each of two horses you pick has to come in first or second in either order. I dutch the bets, making separate bets in an amount for each combination with one key horse and each other horse, such that if the key horse comes in first or second with any other horse in the race (except the ones I've thrown out), I win the same amount, no matter which other horse, bomb or chalk, comes in.
I wrote a spreadsheet with macros to extract the betting pool data from the net and determine the overall odds of all Dutched bets with any horse combined with all the other runners in the race at any given total bet amount. If the combined odds of the dutched bets is significantly higher than the win odds for the key horse with the 25% kibosh for exotic bets added back in at 1 minute before the race goes off, I will make the bets. The beautiful part of this strategy is that I'll get more back even if my horse places than I would on a win bet if he wins (assuming theres no significant extremely late money on the combination that ends up winning and placing).
I haven't bet with this strategy yet, but I have found from playing with the data that very often there are opportunities where the combined Dutch bet quinella odds are quite a bit higher than the win odds for the key horse at post time. Risk of ruin can be determined by factoring in the odds of the key horse but it's still hard to judge with my level of statistics understanding. I have found in the races I have looked at to my happiness that there is a longshot bias in tvilling bets and the key horses with theoretically positive expectation are often the favourites (lowers variance and therefore risk of ruin). What do you think?
