This is the first blog post, in a continuing series, where I’m going to investigate the Race Advisor ratings and show you how they perform under different conditions.
Some will perform well and make insane horse racing profits on their own, others will not perform as well but will indicate strong horses and work excellently when combined with other factors.
I’m going to start with the PFP rating, for no reason other than I feel it is considerably under-used.
The PFP rating was one of the first horse race ratings I ever developed, and it’s based on the Glicko (both versions) chess rating systems.
It’s a collateral form based rating which has a base level of 1500. When the results of a race have come in, we look at the finish positions and distance to winner, and determine how well a horse has performed.
This is done for every horse in the race.
However, just doing this for a single race may tell us how the horses performed against each other in that race, but if one horse was significantly better than the rest we don’t take that into account. This would skew the results.
So.. we look at all the horses a horse has ever raced against and how they performed to determine the level of ability the horse has.
But even that’s not accurate enough.
So… we go look at all the horses those horses have ever raced against!
That gives us a level of ability for our horse.
By doing this we can accurately rate the ability of each of the horses in the race, and now when we look at the finish positions and distance to winners, we can determine if a horse ran better or worse than expected.
We then increase or decrease the horse’s PFP rating based on this.
And of course, like most Race Advisor ratings, we have a ranking and difference from top version.
The ranking version tells the top rated (1), second top rated (2), third top rated (3) etc.
The difference from top tells us the difference in score between the top rated horse in the race and the horse we’re considering. If the top rated horse had a score of 1500 and our horse had a score 1450, then the difference from top version of the rating would give the horse 50. This enables us to see at a glance how far behind the top rated horse each of the runners in a race are.
WHAT THE STATS MEAN
We use a few pieces of information that help us to understand the performance of horse race ratings. Below you will find an explanation of these for reference.
SELECTIONS
The number of selections that have been found
PROFIT
The profit, or loss, of the selections to Betfair SP with a 2% commission. Where Betfair SP isn’t available we have used the SP odds.
WINS
The number of winners found in this sample of selections.
STRIKE RATE
The percentage of winners found from the selections.
ROI
Return On Investment. The percentage return made on turnover, based on betting 1 unit per selection.
A/E
Actual/Expected. Also known as the PIV (Pool Impact Value). This shows how many winners were found compared to the number of expected winners (based on odds). If this figure is greater than 1 then we have an edge, less than 1 and we don’t!
CHI2
The chi square is a statistical test which tells us how likely the results are due to chance or skill. 100% means the results are 100% based on chance, we want them to be as low as possible.
DOES THE PFP RATING MAKE HORSE RACING PROFITS?
Using data from 2018 to the present day, I’m going to start by looking at the top four rated horses in every race.
This gives us the following results:
RANKINGS
TOP RATED
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 25941 | -892.34 | 4840 | 19% | -3% | 0.99 | 62.49% |
2nd TOP RATED
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 28132 | -763.58 | 4469 | 16% | -3% | 0.97 | 2.38% |
3rd TOP RATED
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 28025 | -1838.32 | 3861 | 14% | -7% | 0.98 | 14.09% |
4th TOP RATED
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 27030 | 916.38 | 3352 | 12% | 3% | 1.01 | 57.90% |
A pretty good start, the top ranked across all races has a 19% strike rate. It’s not making a profit, but it’s only losing 3%, so we should be able to push it into profit fairly easily.
We get a decreasing strike rate, which we would expect, and a decreasing ROI.
The 4th top rated have made a huge profit of +916 points, but with a Chi2 of 57.90% we can see that this is very likely due to chance. We’d need more data to investigate this further.
DIFFERENCE FROM TOP
LESS THAN 1 POINT DIFFERENCE
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 33256 | -714.20 | 6037 | 18% | -2% | 0.99 | 62.59% |
MORE THAN 1 POINTS AND LESS THAN 2 POINTS DIFFERENCE
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 9257 | -232.37 | 1312 | 14% | -3% | 0.99 | 65.93% |
MORE THAN 2 POINTS AND LESS THAN 3 POINTS DIFFERENCE
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 11405 | -286.70 | 1346 | 12% | -3% | 0.95 | 3.94% |
MORE THAN 3 POINTS AND LESS THAN 4 POINTS DIFFERENCE
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 11233 | 568.09 | 1327 | 12% | 5% | 1.02 | 49.36% |
MORE THAN 4 POINTS AND LESS THAN 5 POINTS DIFFERENCE
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 11267 | -386.32 | 1194 | 11% | -3% | 0.97 | 21.41% |
All horses that were within 1 point of the top rated horse have produced a 2% loss. It’s a small improvement on the top ranked, but not huge.
Between three and four points from the winner has made a whopping profit of +568 points. We have a small edge of 2% here, with a large Chi Square that we would want to reduce, but again this is a point where more data could provide us a fairly easy angle to profit.
RAW RATING
GREATER SCORE THAN 1530
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 3182 | 163.72 | 361 | 11% | 5% | 1.13 | 1.15% |
GREATER SCORE THAN 1515 AND LESS THAN 1530
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 23953 | -642.27 | 2882 | 12% | -3% | 0.97 | 4.58% |
GREATER SCORE THAN 1500 AND LESS THAN 1515
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 145159 | -1290.75 | 18716 | 13% | -1% | 1.00 | 82.22% |
GREATER SCORE THAN 1475 AND LESS THAN 1500
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 91991 | -7180.70 | 5943 | 6% | -8% | 1.01 | 63.33% |
SCORE LESS THAN 1475
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 841 | 95.95 | 29 | 3% | 11% | 1.13 | 49.09% |
This shows us some very useful information. First, if you’d bet on every horse with a score of 1530 or higher, you’d have made +163 units profit in the last two and a half years, with a huge edge of 13%.
It also shows the the majority of horses raw ratings falls between 1500 and 1515, and less than 1500.
We have a pocket of horses with scores of less than 1475 who are making a profit. Again the Chi Square shows us that we’d need to do more work on this approach to bring this score down, but there’s an angle here which can be exploited.
This gives us a strong enough base to move on.
WHAT IMPACT RACE CONDITIONS HAVE
Moving forwards from this point, I’m going to look at different race types to see how our most profitable elements have performed under different race types.
Our most profitable are:
4th RANKED
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 27030 | 916.38 | 3352 | 12% | 3% | 1.01 | 57.90% |
MORE THAN 3 POINTS AND LESS THAN 4 POINTS DIFFERENCE
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 11233 | 568.09 | 1327 | 12% | 5% | 1.02 | 49.36% |
GREATER SCORE THAN 1530
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 3182 | 163.72 | 361 | 11% | 5% | 1.13 | 1.15% |
Taking this information, and breaking it down by race type gives us the following results:
4th RANKED
| Race Type | Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| NH Flat | 1240 | 4.03 | 113 | 9% | 0% | 0.91 | 30% |
| Chase Turf | 3654 | 88.48 | 545 | 15% | 2% | 1.05 | 22% |
| Flat Turf | 9333 | 1207.53 | 1162 | 12% | 13% | 1.03 | 28% |
| Hurdle Turf | 6035 | -334.70 | 708 | 12% | -6% | 0.98 | 60% |
| Hunter Chase | 169 | 34.31 | 19 | 11% | 20% | 0.90 | 63% |
| Flat AW | 6575 | -81.69 | 802 | 12% | -1% | 0.99 | 86% |
| Chase AW | 0 | 0.00 | 0 | 0% | 0% | 0.00 | 0% |
| Bumper AW | 24 | -1.60 | 3 | 13% | -7% | 1.29 | 65% |
| Missing | 0 | 0.00 | 0 | 0% | 0% | 0.00 | 0% |
MORE THAN 3 POINTS AND LESS THAN 4 POINTS DIFFERENCE
| Race Type | Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| NH Flat | 1249 | -155.68 | 108 | 9% | -12% | 1.00 | 98% |
| Chase Turf | 990 | 38.36 | 148 | 15% | 4% | 0.97 | 72% |
| Flat Turf | 4389 | 955.48 | 484 | 11% | 22% | 1.01 | 78% |
| Hurdle Turf | 1906 | -22.13 | 297 | 16% | -1% | 1.11 | 6% |
| Hunter Chase | 39 | 9.95 | 7 | 18% | 26% | 1.00 | 100% |
| Flat AW | 2637 | -252.09 | 281 | 11% | -10% | 0.97 | 65% |
| Chase AW | 0 | 0.00 | 0 | 0% | 0% | 0.00 | 0% |
| Bumper AW | 23 | -5.79 | 2 | 9% | -25% | 1.16 | 82% |
| Missing | 0 | 0.00 | 0 | 0% | 0% | 0.00 | 0% |
GREATER SCORE THAN 1530
| Race Type | Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| NH Flat | 3 | -1.21 | 1 | 33% | -40% | 1.69 | 55% |
| Chase Turf | 373 | -66.06 | 65 | 17% | -18% | 1.04 | 71% |
| Flat Turf | 1637 | -342.30 | 153 | 9% | -21% | 0.92 | 30% |
| Hurdle Turf | 407 | 13.61 | 57 | 14% | 3% | 0.96 | 74% |
| Hunter Chase | 20 | -6.60 | 4 | 20% | -33% | 1.17 | 73% |
| Flat AW | 742 | 17.02 | 72 | 10% | 2% | 0.91 | 39% |
| Chase AW | 0 | 0.00 | 0 | 0% | 0% | 0.00 | 0% |
| Bumper AW | 0 | 0.00 | 0 | 0% | 0% | 0.00 | 0% |
| Missing | 0 | 0.00 | 0 | 0% | 0% | 0.00 | 0% |
IN SUMMARY
Based on this investigation, the PFP rating used under the following situations produces the best results.
- The fourth ranked horse in Flat races has found 12% winners, made a +1207 unit profit since 2018, and has a 3% edge.
- Horses between 3 and 4 points of the top rated horse in Flat races has made a +868 units profit since 2018, found 11% winners and has a 1% edge.
Considering that we’re focusing on just a single rating, with very few conditions, this is a strong performance.
Each of these also have enough selections for more conditions to be put in place should you prefer a higher strike rate.
We could add some more simple factors to improve this performance. For example, if we took the fourth ranked horse and focused on those runners outside of the top four in the market (under the assumption the market odds on these runners will offer minimal, if any, value), we see this:
| Selections | Profit | Wins | Strike Rate | ROI | A/E | Chi2 |
| 4293 | 1262.73 | 219 | 5% | 29% | 1.10 | 13.38% |
An increase in profit and edge, the edge is up to 10% now, and a reduction in the Chi2 which indicates this is becoming a stronger approaching.
AND… there’s still enough selections to add more conditions if required.
Whether you choose to include more conditions or more ratings in your analysis, I always recommend that you confirm a horse has proven it can perform under the current race conditions. Whilst this means you may miss some runners, it also means you will cut out a lot of losers and keep a higher strike rate.
In fact, doing this on the above selections should enable you to remove the horses with no chance immediately and increase the strike rate to a reasonable level.
If you’re not yet a member of RA Pro, you can register for a free account here.
Even though there’s no racing at the moment, we’ve pioneered a virtual racing platform that’s unique.
The races Aldermist, our virtual race course, are all based on real data, the horses running are real horses that you will recognise, and the trainers and jockeys are real.
This means that not only does every horse, jockey and trainer have full form and ratings for the race, as well as an entire detailed form history, exactly as you would expect from live racing. The results are also based on a horses true chance of winning the race.
Any system or strategy that is profitable at Aldermist will also be profitable on live racing.
Whilst there is no live racing currently, Aldermist has a race every ten minutes, allowing you to test your systems and strategies in a live environment far faster than has ever been possible before.
If you’re an RA Pro member, I’d love to hear whether you’re already using the PFP rating in your analysis, or if you’re now going to consider using it.
If you’re not yet an RA Pro member, then please let me know what ratings you currently use, and whether you’d like to have a rating like this in your portfolio.
Stay safe and healthy.
** This article has been updated after discovering there were some issues with the datasets used in previous examples.
Hi Michael – this was a VERY interesting analysis. I actually tried once to develop something similar using Excel but I didn’t manage to make it work. I believe this kind of rating is very useful and your figures definitely confirm that. Have you considered combining it with going? I know that Heavy going can sometimes throw historic performances up in the air for instance.
You could absolutely combine this with going. As you say, the going extremes can make a big difference to how ratings perform. I have made a note to look at how you can use going in conjunction with the PFP in tonights Facebook Live event 🙂
How does ths system deal with the ‘newcomer’? Those lightly raced horses who have little in the way of previous races, race somewhat green but may well be improving for the experience. Would a fast improver blow the stats>
An interesting question Graham. I wouldn’t call this a system, more an analysis on one rating in our stable and a bit more of an inside look at how it works 🙂
In terms of newcomers, this rating will give them a base level of 1500 in their first race. It does take a few races to start to get an accurate read on what level the horse should be at, with this in mind with have an upper and lower 95% confidence level version of this rating. This shows, with a 95% confidence, the range we expect the true rating to be in. Generally speaking, the more races a horse has had the smaller this gap becomes as we can become more confident in their true rating.
Michael
It seems almost too good to be true.
Do you know if when using this score in the Monte Carlo Simulator whether there is any difference in using the Raw Score, The Ranking or the Difference from Top? Part of me says there should be a difference between the ranking and the rest but part of me says as they are measuring the same thing there should be no difference.
There will be a difference when using them in the Monte Carlo Simulator. The difference is because of how the simulator works. We use a variance range for each rating in the simulation. This is defaulted to 15% (although you can adjust it). The variance will change depending on whether you’re using Raw, Ranking or DiffTp (Difference From Top). As a general rule, using a continuous rating (such as Raw or DiffTp) will produce results which are less skewed to the rank of horses for a rating. I tend to use Raw ratings in the simulator most of the time.
A few things that are worth bearing in mind regarding this analysis. It’s an analysis of historic results, we never have any guarantee that they’re going to continue this way. One of the biggest impacts in making this rating, and a number of others, perform better out of the box is the ability for most people to be able to get a 2% commission on most exchanges now. The first two elements would be break-even or in a small loss at a 5% commission. Horses with scores of 1600+ don’t come around that often.
There’s something you could do to improve your performance on the 1600+ scores. If a horse has a hugely high PFP rating in a race where all the other horses are significantly lower (within reason as 1600+ is a very high rating). When we can confirm that this horse has been declining in performance with other ratings, we can consider these to be very poor horses to bet on. Usually they are trainers/owners trying to squeeze out some more profits before a horse gets a layoff.
Hi Michael
I plan to join RA and use the lock in to research options. Can members use the software to analyse past results as you have done? The 35% strike rate is excellent when considering the average odds but statisticly could still have a losing run sometime of 16 which I would find mentally hard if it occurred early on. I estimate place/ew would reduce this to 8 or less and would like to see if this is a viable option
Absolutely, as an RA Pro member you can export past data, as well as use Aldermist to test your strategies in a like real environment. Please note the above stats included some horses that shouldn’t have been as I pulled them straight from the database and our developers had changed the way some things worked I wasn’t aware of (because they didn’t think I still delved straight into the database!). I am going to be re-working the entire article and updating it to confirm that all the figures are 100% accurate, and there may be some changes to the criteria
Hi does this work better to bookies BOG or is this something you cant run,
This will work better to BOG, generally if you can get BOG you will beat all other odds.
you got 4,293 selection ,and 219 went on to win meaning 4,072 lost so what was the draw down on the betting bank as only 5% strike rate must mean a very long loosing run so what was the betting bank and staking you was using to get the profit
The strike rate is low, and I am definitely not suggesting this is a system or strategy that should be used exactly as is. The aim of these posts is to look at some of our ratings to show how they perform stand-alone, with no other factors and just some basic race conditions. This drives a greater understanding of them, and whether you may want to use them as a base factor. I would always recommend using them with at least a few other factors or with some form analysis. A
Am I the only one to notice this?
Greater than 1530 bets= 3182 profit +163.72
When you split these up by race type and add up the runners you still get 3182 bets BUT the profit has gone and its MINUS 385.54
How did that happen?
You’re absolutely right, I must have copied in the wrong table by mistake. I’ll need to dig through the files to find the right one. Thank you for spotting it.
Hi Michael,
In addition to the query raised by Paul above there is a problem that affects the DSLGR+PFP figures
On PFP RATING ONLY we have 3182 selections and 361 winners
On DSLGR& PFP we have 1911 selections and 383 winners
As all 1911 selections would also be in the larger group of 3182 it is impossible for the number of winners to increase
I’m not sure what I’ve done with these tables. I’ll go through every single one and re-calculate them all and re-upload them once we’ve migrated to our new databases next week, to be sure that all the numbers and every table is correct.
Did these tables ever get sorted. It is nine months later now?