For those of you who spend time at the track, I am curious, what kind of mathematical model, let us say even crude model, do you use to sit and work out the probability of any possible result in a given upcoming race? We can assume we have all the old Racing Forms and past performances. There should also be some logistic regression and so on?
Unless I’m completely misunderstanding your questions, what you’re asking about is called Handicapping. Basically the punter considers a variety of factors, all there for you to see in black and white, to predict the winner. Past performances, Beyer Speed Rating, the track’s normal bias, the pace set by the pace-setting horse, the length of the race, etc. Somehow one assigns value and/or derives meaning from all of these stats in aggregate and makes the decision based on that. I don’t claim to understand it, mainly because I’ve never had the gumption to pay to learn handicapping, either online via a “system” or via the track, where I hear tell that they will offer handicapping classes from time to time.
There are as many handicapping systems as there are handicappers. The cynic in me would say that the only people who make money with handicapping “systems” are the people who sell them. But at the same time, people have been handicapping horses for centuries, and some people are actually good at it.
NB: A guy named Bill Benter basically beat horse racing. He used his innate math skills, a computer program that he developed, and an obscure academic paper (“Searching for Positive Returns at the Track: A Multinomial Logit Model for Handicapping Horse Races”) to develop a program that would predict the races. Long story short, he went to Hong Kong, lost his shirt, tweaked his program, lost more, then turned things around and soon was making so much money that at one point he won $13 million on a parlay and never got around to claiming it. He probably made hundreds of millions, if not a billion or two.
Paging @Spoons , whom I believe is fairly serious into horse racing and wagering.
I mean, speaking completely hypothetically, if you can calculate the “real” odds, via some kind of handicapping or other statistical model, then in principle you might be able to identify bets that are likely to make you money, and that will add up in the long term.
Thanks for the heads-up. I’m happy to try to help, though I’m not entirely sure what the OP’s question is. Is it “How do you handicap a horse race?” or is it “Is it possible to devise a formula that will predict winners?” or is it “How are the odds set for horse racing?” Or is it something else entirely?
I could answer all of the above, but I’d like a little more clarity as to what the OP is looking for. OP, maybe you could put a finer point on your question(s)? Like I said, I’d be happy to help—I’ve been following racing for over forty years, studied a lot of books, attended seminars, and yes, done my share of wagering, so I should be able to answer your questions.
Let us start with, suppose there is a race coming up. There will be a number of horses running; let us call them A, B, C, etc. Not counting ties, there are various possible results: the horses could finish
- A, B, C, … (A wins)
- A, C, B, …
- B, C, A, …
There are a lot of permutations, and the result is in any case uncertain because performance is variable. My question was, could one assign probabilities to the possible outcomes? Based on past performance? I don’t care if it is a complicated calculation, either, because we can always run it through a computer.
We can come up with models for the situation, and then apply mathematical calculations to the models. Which doesn’t say anything about how good the model is.
Okay, I think I get it. You’re looking at the probability of the horses finishing in a particular order. The short answer is No.
The long answer is No, because there are any number of possibilities that can happen between when the race field is set, and the race itself. Note that the field is set usually four days or so in advance of the race itself, and during that time, horses might be scratched from the race, stewards might change the distance of the race, they might change the surface of the race (say, from turf to dirt), and similar.
On race day itself, any such changes between then and now will be announced over the PA, displayed on the track’s CCTV, and posted on the Change Board. But there are things that can happen on race day too: weather might cause late changes in distance or surface, the track vet might scratch a horse that appears to be unhealthy in some way, there might be a late jockey change, and other last-minute changes.
I’ve even seen gate scratches, where the horses are loading, but one is having none of it, and refuses to cooperate. In one particularly memorable example of that, I once saw the horse throw the jockey, take off on his own, jump the rail, and go prancing around the infield (note that this was at Woodbine, where the infield is beautifully landscaped, and hand ponds and flowerbeds). The race was delayed fifteen minutes, while they tried to get this out-of-control horse under control, and out of the infield. Needless, to say, that horse was immediately scratched.
During the race itself, a horse might develop a problem of some sort, causing the jockey to slow him down (“pull him up” as we say). Or the horse might suffer a fall. Or the jockey might falll off, immediately disqualifying the horse, no matter what he does without the jockey.
Speaking of individual horses, it is important to remember that they are living, breathing animals, just like you and me. And sometimes, just like us, they wake up in the morning, and don’t feel like going to work. So they put in a half-hearted effort that you cannot predict from looking at the horse’s past performances.
So in the end, you cannot analyze a Form and determine an outcome ahead of time. There are too many variables that can happen between when the races are set and published, and the actual race itself. What I like to do is to get my Form a couple of days early, and make preliminary selections then. Then, on race day, I’ll finalize selections based on any late changes. But there is no way anybody could do any kind of mathematical analysis, with the aid of a computer or not, that could accurately predict the outcome of a horse race.
Does that help answer your question?
“All models are wrong, but some are useful.”
–British statistician George E. P. Box
I am not a horse racing fan, but as a statistician, I’d note that one of the issues with such a model would be sample size.
AIUI, most horses don’t run a large number of races over the course of their careers – it appears that a typical horse might race five or six times a year, over the course of a career that might be three to five years long. And, across the races that a particular horse runs over its career, there will be variables like its age, its overall health in a particular race, jockey, track conditions, race length, etc.
Unlike a human sport, like baseball, football, etc., where an individual athlete might have hundreds, even thousands, of discrete performance data points to feed into a model (at bats, passes, shots, etc.), the sample size for a particular athlete in horse racing is small, and not statistically robust.
@Spoons how do you feel about teaching a handicapping class? This is something I’ve always wanted to learn.
Is a horce a horse?
Of course, of course.
@HeyHomie, I could try. I’ve certainly done it before, although that was in person. Here, we’d be doing it in writing. But I’d be happy to give it a try.
I do want to address this …
… which is an excellent point. Unlike the baseball player, who plays almost every day; or the football player, who plays once a week; or the hockey player, who plays two or three times a week; a horse may race weeks or months after its last race. It is not unusual, for example, especially in the spring, to find horses that have not raced since last fall.
Most of the owners and trainers that I know will not race their horses more often than once every two weeks, at least. And even that is cutting it short; some like to go longer. Of course, the horse does get some exercise out on the track during that time, including timed workouts (with a jockey up, the horse runs a measured distance, often 4 or 5 furlongs, under various conditions meant to approximate race conditions). Workouts, their distances, and conditions, are reported in the horse’s Past Performances, so you can see them, by the way.
But a very good point. Is this horse coming off a six-month layoff with only one workout six weeks ago; or did he race three weeks ago, with a workout just last Tuesday? That can make a difference in my selections.
I think I am beginning to understand. So we have to contrast with pure games like chess or certain types of multiplayer/team games, where it is assumed that the strength or skill of each player may be estimated with a rating— albeit one which may not be known precisely (error bars) and moreover may fluctuate and be erratic— but nevertheless may be estimated from the axiom that players with equal strength (respectively, a known difference) have a quantifiable probability to beat the other. Or with the factor-analysis rationale that observations may be explained as a function of a small number of latent variables.
Yes, I think that you are getting there. Remember, a horse race is not a chess match, a baseball game, or a football game, where there are only two competitors and one must win. No, it has multiple competitors, each of whom, on paper and in theory, should have an equal chance to win. You’d never put a world-class grandmaster chess player up against P.S. 53’s Grade 5 Chess Club Champion.
But that can happen in horse racing. Sometimes, owners and trainers experiment. I recall asking one trainer I know, why he would put that horse in this race. “He’ll be going farther than he ever has and he never did well then,” I remarked. “Yeah, this is a new distance for him, but I want to see if he can handle longer distances against a field of others; maybe he needs the additional distance to make a stretch drive” came the reply. An experiment, in other words, with no expectations; nice if the horse can win, of course, but if he does not, we learn a lot anyway.
Then there is the question of class. Horses settle into a class eventually, but trying to find what class a given horse belongs in can be a challenge. This is done mainly by looking at its past performances in claiming and allowance races. A horse that last ran in a claimer for a $15,000 claim price is running in a claimer today for $7500? That’s a big drop in class. What’s going on? Is the owner trying to get rid of him? Is there something wrong with him? Meanwhile, another horse that had been running in $5000 claimers is moving up to the $7500 level, because he’s been slaughtering his opposition at $5000. But there are bigger purses at the higher level. Can he handle the jump in class?
I should mention that Equibase’s “Class Rating” in its PPs is no indicator of class; it is more an average of the horse’s speed figures. A horse’s class cannot be distilled down to a single number; it is something you can only learn by looking at the horse’s racing history.
Note that class handicapping is an actual thing, and what I learned when I began. There is also speed handicapping, and I’ve actually met a couple of people who practice pedigree handicapping. But class and speed are the most commonly used. I like to combine the two of them myself, though I’ll lean more toward one or the other as I finalize my selections.
Hope I haven’t confused you too much. But at any rate, horse racing is not a chess match, or a football or baseball game. The number of competitors—anywhere from two to twenty—makes it difficult from the outset, and whether or not those competitors fit the conditions and class of this race is something that only the horseplayer can determine.
Come on:
“Of cource, of course”
Andy Beyer has called handicapping the world’s hardest way to make an easy living.
One factor is that you not only need to figure the odds of a horse winning – every person on the planet figured Secretariat had a lock on the 1973 Belmont – you have to find one who has been overlooked so that the profit on your bet is higher than the odds you get the payout. Over 5,600 tickets for that race were never cashed, people preferring a two-dollar souvenir over their 20-cent profit.
The favorite in a race wins one time in three but between the rakeoff and over subscription typically pays only 2 - 1.
Recently read a good piece on Bill Benter and the value of the useful phrase “horses for courses” here.
The rule is that you can’t statistically model horse racing in a way that gives you a single optimal prediction, because there is as said too many dimensions and too little data. Bill Benter’s success doesn’t disprove this, because Hong Kong is a unique environment: there are two race courses, and a more or less fixed cast of horses (because it’s expensive to move horses around, once you’re in HK you’re there for a while). So in fact the same horses race each other over the same course often enough that you minimise dimensions and multiply data. This is unlike horse-racing anywhere else in the world.
This matters, because “horses for courses” isn’t just a fun rhyming thing to say – it’s absolutely fundamental to horse race gambling. Some horses are good at shorter distances, some are stayers. Some of them have the brains to adjust their pace on a hill, others will try to keep sprinting and exhaust themselves. Some of them have a gait that suits muddy conditions, others run better on hard ground. Et cetera …
Racecourses are very different. In the UK, for example, they aren’t even consistent as to whether the horses run clockwise or anticlockwise. The course at Chester is almost circular, which means that racing there is a test of how good a horse is on its front left foot. Perth has a steep slope up to the finish which always catches out front-runners. A sharp bend (like the one at Catterick) will confuse and unsettle a horse that hasn’t seen it before, but not one that’s expecting it. Et cetera …
The piece goes on to say that nonetheless people can make money from betting on horses, but not through data crunching to produce a flawless algorithm. It is as @Spoons says, all about having a big depth of knowledge and knowing when to apply it.
The data issues I’ve outlined make it more or less impossible to get the kind of model that Moneyballers and poker players want, which gives you a single statistically optimal prediction. But that doesn’t mean the formbook is useless. There are all sorts of rules of thumb and statistical regularities that can be established. (For example “The sharp turn at Catterick will always catch out a horse that hasn’t raced there before”, as I said a couple of paragraphs ago).
Really good racing analysts can systematically beat the odds by knowing a lot of these little rules of thumb, being really familiar with the form and having enough experience to know which statistical regularities are most salient for any given race. They are also good at spotting which races are easiest to analyse and most likely to offer attractive odds, and good at not betting on the other ones.