The authors argued that the most effective and mathematically principled way of calculating a horse's chance in a given race was to analyse a robust and representative set of statistics using Bayes' theorem. The book carried a listing of a pilot Bayesian program based on a small statistical sample.
Computer hardware has come a long way since 1987, and it is now possible to perform Bayesian analysis on far larger data sets than were viable at that time. The only problem is that of ensuring that the different pieces of data used are genuinely independent. The current version of HOOF only uses in its evaluation pieces of evidence which are not correlated.
Thus a horse with an adjusted rating of, say, 145 is 3 pounds"better in" than one with a rating of 142. This in turn means that in theory, the horse rated 145 should finish about a length in front of a horse rated 142, depending on the distance of the race.
The words "in theory" are crucial here, because if handicap ratings always worked out, bookmakers would cease to exist, as would betting. The best horse on form would always win. In practice life is not so simple. In non-handicap races, only around 40% of top-rated horses win, while in handicaps the figure is around 20%.
A further limitation on the usefulness of handicap ratings is that they are difficult to interpret. For example:
The scale of probabilities ranges from 0 (no chance at all), to 1 (absolute certainty). It is relatively simple to convert probabilities into odds against using the following formula:
Odds against = (1 / Probability) - 1Thus a probability of 0.2 is equivalent to odds of 4-1 against; a probability of 0.25 is equivalent to 3-1 against, and so on.
The essence of value betting involves backing only those horses whose odds are longer than they should be. All professionals use a version of this strategy, since it is the only one which offers any hope of long-term profitability.
Thus a horse which is a true 2-1 shot is good value if offered at 11-4 or better. If one can consistently find such bets, one must win in the long run. Even so, this is no short cut to easy riches. Finding value is hard work, and any backer who can make 10% on turnover after the iniquitous betting tax, is doing very well indeed.
The HOOF v5.0 value ratings are designed to make the business of spotting value considerably easier.
The advantage is that one never backs favourites at short odds. In fact it is rare to back any horse at odds of less than 3-1. Coping with the longer losing runs entailed by the lower strike-rate, is simply a matter of sensible staking with a larger bank, typically 60 points, compared with the 20 point bank which is normally recommended.
The output for a typical race looks like this:
HORSE TRUE PROB TRUE ODDS VALUE ODDS (30%) _____ _________ ________ ________ sunley bay 0.317 85-40 7-2 special account 0.217 7-2 11-2 jumbeau 0.157 11-2 8-1 wise customer 0.107 17-2 12-1 windy ways 0.063 14-1 22-1 some day soon 0.036 28-1 40-1 synderborough lad 0.03 33-1 50-1 mighty falcon 0.029 33-1 50-1 strong beau 0.024 40-1 66-1 keep talking 0.016 66-1 100-1 10 runner handicap at wincanton : 3.15
Any horse whose odds are equal to or greater than the value odds quoted is worth serious consideration.
There may well be two or more value candidates in a particular race, however it is not suggested that all such candidates be backed blindly. This strategy would entail too many wasted bets, and violate the prime aim of the program which is to enable selective, intelligent, value betting.
Furthermore, while horses at extremely long odds may warrant serious consideration, it should be borne in mind that the longer the odds, the greater the chance of a loser even if the horse is a value bet.
For example, suppose a true 20-1 shot is offered by the bookmakers at 40-1. Obviously the odds on offer represent good value, but it would be folly to expect to collect on such a bet every other day. Indeed, the chances of an extremely long losing run, if one were to confine one's betting to such horses, would be dauntingly high.
One possibility is to confine bets to small fields using the selective betting strategy which I have used over the last two seasons with a strike-rate of 20% winners at average odds of 11-2, but which has proved disappointing so far this season.
Aintree, Ascot, Ayr, Cheltenham, Chepstow, Doncaster, Haydock, Kempton, Lingfield (turf), Newbury, Newcastle, Sandown, and Wetherby.
In fact, backing the top-two rated (odds permitting) in all races on these courses (ie without restriction on the number of runners, and without restriction on the true odds) has shown a profit in each of the last three seasons.
There are obvious reasons why it makes sense to avoid the "gaff" tracks and concentrate on high class racing:
Given that HOOF works with form and value, it's not surprising that the program fares better with quality racing. So anyone thinking of following HOOF systematically, could do a lot worse than restricting betting to the better courses.
Since 1st December 1995, these selections have been proofed to the independent monitors at Racing Information Database. They are also proofed to the specialist magazine SMARTsig as part of the SMARTsig Computer Challenge competition. This means that all potential selections are faxed to both RID and SMART well before the start of racing.
However, the early price market is worth serious consideration for a number of reasons:
Where horses are withdrawn the ratings will be affected, negligibly if the withrawn horse is a rank outsider, but considerably if it has an obvious chance on form. Obviously it will not be possible to take account of very late withdrawals, but every effort will be made to update those races in which withrawals are communicated up to one hour before the start of racing. It is important to check the "Today's Ratings" page for late updates..
The Past Results page is also well worth consulting, since it gives a good idea of how the ratings have performed in practice.
Finally, a rather obvious point, but one that needs stating. It is imperative that those accessing the ratings show the utmost discretion. This means that the ratings are strictly for personal use, and that anyone choosing to have a bet, should do so to small stakes. The ratings are acutely price-sensitive, and the edge that they offer will disappear if irresponsible individuals start rattling the bookmakers by lumping on large sums of money, or pirating the ratings in order to run their own services. If there is any evidence of abuse, the ratings pages will be removed without prior warning.
Until further notice the ratings are free of charge. This will enable anyone interested in the concept, to "dry run" the ratings extensively, and to form his/her own judgement of them without risking a penny. A small shareware charge to cover the cost of maintaining the web space and the time involved in generating the ratings may be made later, but in any event will amount to no more than the cost of a daily racing newspaper. In the mean time, any comments and suggestions will be welcome.
Bill Hunter's Soccer Pages
For those eclectic souls interested in betting on both racing and soccer. Software, statistics and prediction, with both pools and fixed odds in mind.