Our Approach to Predicting Horse Races

There will always be an element of randomness in an event such as a horse race, many other sites will tell you which “nailed on” horse is going to win the race, but in reality that is not always what happens. A multitude of things can happen just before or during a race which means that the horse doesn’t win. The way we have approached this is to look for races where the model is more confident than the random element in the race and we base our predictions on those. There will still be times when the horse does not win the race, but the way we have developed the system is to try and ensure that over time the winners will be more numerous than the losers and that they should generate a profit.

This often means that the system will pick out the favourite or second favourite in races – this however is the reality of racing – the best horses are the best horses and they aren’t going to be offered at 50/1. Having said that, our system analyses each horses on its own strengths, we have often seen a 50/1 or 100/1 horse being predicted by our system to finish in the places – and, they often do just that and finish in a place position.

LP, MP and UP

The model as part of its output tells us how confident it is in its estimate of the probability of the “event” happening (i.e. of the horse winning, or the horse finishing in a place position etc). This confidence is explained by the LP (lower probability) and the UP (upper probability), these are the figures which represent the range of probabilities which the model believes with a 95% certainty that the actual figure lies within.

Obviously the best prediction is where the model has a stronger value of the event happening (the MP) compared to the other values of MP in the same race and where the model is highly confident of its estimate (i.e. the range is very narrow).

F-Odds and M-Odds

We take the odds which the horse is forecasted to be at the time of the race (F-Odds) and compare this to the odds which the model believes offers a true reflection of the horses ability (M-Odds). Unsurprisingly you will often find that the bookies have built in a nice profit for themselves for the majority of the horses in a race! Another benefit of the M-Odds is that it allows you to hunt out “value” in the market, for example being able to back a horse at odds above the M-Odds or lay a horse at lower than the M-Odds offers a “value bet”.

Place Model

You will often find that the place model will select a horse predicted in second place to be “placed” and the horse we have predicted first is not suggested. This is due to how we have built the system – with profit in mind. We test whether a selection is likely to be profitable over time, an odds on favourite for example, once you have taken into account the reduction factor for a place bet (a fifth of the odds for example) it is no longer profitable to place a place bet.

Lay Model (Multiple)

The lay multiple model starts by selecting the horses which it believes will not win the race. The odds then have a factor applied to them for a guide as to what the horse could be layed at – the rules we use are:

Odds of 10/1 or less can be layed at OddsToOne*1.1 (i.e. 10% higher)
Odds of over 10/1 and 20/1 or less can be layed at OddsToOne*1.2 (i.e. 20% higher)
Odds of over 20/1 and 30/1 or less can be layed at OddsToOne*1.3 (i.e. 30% higher)
Odds of over 30/1 and 40/1 or less can be layed at OddsToOne*1.4 (i.e. 40% higher)
Odds of over 40/1 and 50/1 or less can be layed at OddsToOne*1.5 (i.e. 50% higher)
Odds of over 50/1 can be layed at OddsToOne*1.8 (i.e. 80% higher)

(This is a simplistic description - the percentages we actually use are more accurate)

You will often find that other sites quote their lay figures based on SP – but this is simply not what happens in the real world, we believe that these factors (although crude) represent something closer to what you will experience the prices to be when you come to lay a horse.

The next step is crucial; the model then only selects horses as a possible lay opportunity those which are showing a price anomaly.

For example 2 horses may be expected to lose the race, horse A is priced at 9/1 so we calculate that it can be layed at 9.9/1 (OddsToOne*1.1 which we call the LayOdds), horse B is priced at 6/1 (we calculate it has LayOdds of 6.6/1). We know from the M-Odds (see above) what a “fair” price is for that horse to win, say we have horse A with M-Odds of 7/1 and horse B has M-Odds of 8/1. The model doesn’t think that either will win, but if we put everything together in the following table:

We can see that it is horse B which should be layed as the odds being offered for a lay bet (6.6/1) are shorter than the odds the model has calculated (8/1), this therefore represents a good value betting opportunity (unlike horse A).

The final piece of the lay multiple selection is that it removes any horse which has a LayOdds of over 6.75/1 changed from 10/1 on 02/04/07), this is here purely to stop any major loss in any one bet when the random 100/1 (which we could calculate has a 500/1 chance) defines all logic and wins and therefore wipes out your balance. Our simple rule is only lay a horse if you can lay it at 6.75/1 or less.

Lay Model (Single Selection)

This version of the lay model simply takes all of the selections from the multiple lay output and selects the one with the lowest odds, if there are two or more horses with the same lowest odds (i.e. both priced at 5/1) the selection will be the one with the highest M-Odds (i.e. the least likely of the two to win according to the model).