Finding different ways to evaluate performance

 

As a racing researcher, I get often find myself analyzing stats in the same old way I have always done, without actually thinking is there anything else I can do with this data? For example – when analyzing trainer data I tend to do the following:

 

a) get the relevant data for each trainer – eg. runs, wins, strike rate, returns, roi:

 

b) look at the number of wins / runs first to determine whether the data set is big enough;

 

c) look at the win strike rate;

 

d) look at the roi (return on investment);

 

e) if the figures look worth further investigation at this point then look at the placed stats to see if they correlate with the win stats;

 

f) break the figures down into years to see how consistent the data is – then go through the whole process again (points b to e) on a year by year basis.

 

At this point I decide whether a trainer stat is worth noting or not. Of course it is not an exact science, but hopefully experience will help determine which stats (both positive and negative) are worth noting down.

 

I find this method works quite well for trainer research, but of course many researchers will follow a similar path and hence most of the “interesting data” will be found by many. Hence, in a constant drive to ‘stay ahead of the crowd” I decided to see if I could find a different method to evaluate trainer performance.

 

Here are a couple of my ideas.

 

1. One way to see how well a horse has run is to compare its finishing position to its market position. Hence, using this idea, you could take all the runners from a specific trainer and compare the two factors to see how often a horse ran a) up to expectations (eg. 3rd in betting finishing 3rd; b) above expectations (eg. 3rd in betting finishing 1st or 2nd); and c) below expectations (eg. 3rd in betting finishing 4th or worse).

 

It may seem a little rudimentary, but I thought I would investigate some trainers with their records in 2009. The trainers chosen all had performed reasonably well during the period of study with the following win strike rates:

 

J Gosden 27%; J Boyle 21%, B Hills 26%, T Barron 18%, M Jarvis 27%.

 

4 of the 5 had shown a blind profit on all their selections, while the other had only a small loss of 5%.

 

The table below looks at the comparison of finishing position against market position. The columns should be self explanatory but to give one example, the “Above” column shows the percentage of runners that finished in a position better than their market position.

 

Here are the findings:

 

Trainer	Runners	Above	Same	Below
J Gosden	67	19.4%	28.4%	52.2%
J Boyle	104	33.7%	24%	42.3%
B Hills	75	32%	21.3%	46.7%
T Barron	72	30.5%	20.8%	48.6%
M Jarvis	47	21.3%	17%	61.7%

 

The most striking aspect of the results looks to be the low percentages for Gosden and Jarvis in terms of horses that have performed above market expectation. Of course both trainers have a large percentage of runners that start favourite or near the top of the betting, which almost certainly has a bearing on these figures – for example, one problem with this method of evaluating trainer performance is that any horse starting favourite cannot do any better than equaling their market position with their finishing position. Hence, one could argue that we should combine the “Above” and “Same” columns, especially for trainers that send out a fair proportion of favourites. Here are the figures if we do combine them:

 

Trainer	Runners	Above/Same	Below
J Gosden	67	47.8%	52.2%
J Boyle	104	57.7%	42.3%
B Hills	75	53.3%	46.7%
T Barron	72	51.3%	48.6%
M Jarvis	47	38.3%	61.7%

 

From here, it still looks for this particular data set, that runners from the Michael Jarvis stable have under performed more often than not. On the other hand Jim Boyle’s runners seem to have run especially well.

 

Having never tackled trainer research in this way before means it makes it difficult for me to give a confident analysis of any findings. Hence, in terms of how best to use the data generated from this idea is something I am still struggling with. Of course I think using this idea with recent trainer performance (last 14 days for example) may give you a better indication of stable form then purely a wins to runs ratio.

 

I do think there is mileage in this type of approach although it almost certainly needs refining in some way. One idea I have had is that this type of idea/method may be useful for people who spread bet on certain races or trainers. As I have yet to go down the spread betting avenue with hard cash, this is simply a theory.

 

The next idea I want to share with you is a similar one. The comparisons are the same – market position versus finishing position. However, this time I simply want to add up all the finishing positions and add up all the market positions and do a comparison. For example imagine these 10 results:

 

Market position	Finishing position
1	1
1	3
2	2
2	5
3	1
3	7
4	5
4	3
5	8
5	6

 

10 results from where we can add up both columns to compare the market position to finishing position.

 

The market positions added up equal 30 (1+1+2+2+3+3+4+4+5+5)

The finishing positions added up equal 41 (1+3+2+5+1+7+5+3+8+6)

 

Hence the horses in this sample have performed below expectations as the total of finishing positions is higher than the total of market positions. Now my idea is to give this performance a figure by dividing market position by finishing position. In this case we would get 0.73 (30 divided by 41). Hence we would now have performance figures (or ratings) to compare the results of different data sets (eg. different trainers).

 

This idea is one that would need more data to make it more accurate, as imagine the following scenario - a favourite finishing 30th in a 30 runner handicap – in a small data set a result like this will skew the figures markedly.

 

What I have decided to do is to compare two trainers discussed earlier – I have chosen John Gosden and Jim Boyle. I have looked at data from 2005 to 2008.

 

John Gosden had a win strike rate of 18.5% during this period and actually backing all his runners would have yielded a small profit. Looking at his finishing position and market position totals – all his finishing positions added up to 8206; his market positions added up to 6094. Hence dividing 6074 by 8206 we get a “trainer performance figure” of 0.74.

 

Jim Boyle had a win strike rate of 11% and backing all runners would have produced a loss of 16%. Looking at his finishing position and market position totals – all his finishing positions added up to 7390; his market positions added up to 6892. Hence dividing 6892 by 7390 we get a “trainer performance figure” of 0.93.

 

Hence we have a slight conundrum – Gosden has the better returns in terms of backing all selections; Boyle’s runners tend to run better on average than Gosden’s when you compare their finishing position to their market position. As with the first idea, I am at the stage of – ‘how can I use this data effectively?’ and ‘is this type of data actually useful?’ The first question I cannot answer as yet, but my hope is that “yes” is the answer to the second question!

 

Clearly, these ideas have only been formulated recently, so any help from PunterProfits members with thoughts on how to move this forward would be gratefully received.