Different Betting Systems
Over the last few weeks, we’ve taken a look at a range of different horse racing form factors that go into successful betting systems.
We first looked at how betting syndicates create complicated systems using multinomial logit regressions to help weigh the different factors.
We also took a look at how you can build your own betting systems and the proper workflow that you need to use to ensure that you’re not overfitting your data.
Now we are going to take a look at some of the mini-systems and factors that I’ve found to be useful when analysing horse races. I’ll also look at some of the factors that appear to be effective on the surface, but didn’t do well in testing.
For the purposes of the exercise we’ll be discussing simple betting systems and basic factors as they are the most user-friendly.
Three step system
For the most part, when assessing horse racing form to develop a betting system you use a three step approach:
(1) Come up with the factors that you want to test for. Importantly you want to make sure that there is a fundamental reason that the factor contributes to the outcome of a race. That way you are effectively testing a hypothesis.
(2) Once that’s been established, you can test for statistical significance. In short we are looking to test to see whether there is a statistical relationship between the factor and the outcome of the race – specifically if that factor contributes to the horse winning the race.
(3) If we’ve found a factor that is statistically significant we can then move on to testing. Again you need to remember that the testing is done on the out-of-sample data after we have built the model on the in-sample data.
That way we avoid overfitting our data. Hopefully you’re already aware that backfitting never ends well.
So with that said, here are some simple factors that I’ve found useful in that they are have been found to be statistically significant and/or performed well in in-sample and out-of-sample testing.
If you’re able to get access to sectional data then you can use those individual 200-400m sectionals to calculate average speed. As you would expect, max speed appears to be a good indicator of the ability of a horse to win a race. This is also an important factor in trials, however it’s much more difficult to quantify, because not all horses necessarily go out to win a trial.
Weight Carried / Distance
It stands to reason that the longer a race, the more that weight plays a factor. Again this factor stands up in testing and can help identify strong performers. While many punters might simply look at raw weight, you’ll get better results by transforming it into a more robust factor.
If a horse gets a bad run for whatever reason, whether it be injury or simply where they settled, this clearly has a significant impact on the result. However if you factor this in you can often find horses that are undervalued in subsequent races. Looking at vet reports or stewards comments can unearth potential winners. This is really the approach that a video analyst would use. In a system we are simply quantifying those different factors in an objective way.
While these were a few simple factors that showed promise not all simple strategies perform well. Quite often the best way to develop an edge is to take common factors and transform them into your own custom factors.
But even still there are a few factors that I’ve felt that should be statistically significant that weren’t.
3-Wide the Trip
One of the first systems that I stumbled across when I started out was looking for horses that finished strongly from a wide run. It makes sense that if a horse finishes strongly and has been wide, then they have actually covered a longer distance. However in my tests I’ve never been able to find a statistically significant relationship, which is quite counterintuitive to what most horse racing form analysts would instinctively believe.
Fast Closing Sectionals
Whenever we see a backmarker charge home and put in a cracking closing sectional, we make a note to ourselves that it’s one to watch. In my tests of raw sectionals, I’ve never been able to find an edge using the fastest closing sectional. Again this might be down to the fact that it’s simply too obvious and we need to transform the factor further. Or it might in fact be a statistically significant factor, however the edge is already well and truly factored in to the price.
Building systems and looking for factors is a process of addition, subtraction and interrogation. It’s about continually coming up with new factors and then testing. It’s not about finding a magic factor, but rather starting with one that shows promise and building over time.