Wednesday, January 16, 2008

Myth Buster - Dosage

It seems that every year the Dosage figures of all the Derby contenders get bandied about. Some get drawn into the notion that only certain horses with a certain Dosage profile are bred to win the Derby. I beg to differ, a look at the numbers suggests that Dosage and the Dual Qualifier system has no validity at all. I think that it is flawed to the core. For those of you who aren't aware what Dosage and Dual Qualifiers are here is a brief explanation.

Dosage is a points system that essentially tries to predict a horses most effective racing distances by their breeding. The rule of thumb is that no horse with a Dosage over 4.00 has the breeding to be a Derby winner. A Dual Qualifier is simply a horse with Dosage less than 4.00 and a rating on the Experimental Free Handicap that is within 10lbs of the division leader. That is to say the two components are breeding and a solid 2yo season.

Dosage is useless as a handicapping tool. It was created to be useful with past data and has a poor record going forward. Its effectiveness is decreasing with each passing generation because new sires come onto the scene and don’t get added to the Chef-de-race list until their progeny have already proven themselves. Another reason why its useless is because it eliminates virtually nobody. In the past 12 years 89% of the Derby starters have had a dosage of 4 or less. So essentially on average 18 of the 20 starters have the requisite Dosage to win the Derby. The cumulative record of the non-Dosage qualifiers is 23-3-1-1. Ironically if you focus on the 23 horses who didn’t have a low dosage you might have done very well indeed.

In 12 years you've got just 23 horses who didn’t have a dosage of 4.00 or less from 214 starters. A straight win bet on each prospect would have cost $46 and returned $186. A tidy little 328% ROI. Also the non-Dosage qualifiers boast a sturdy 2.34 Impact Value, meaning that each horse with a dosage of over 4.00 has won the Derby almost two and half times more than they would be statistically expected to do.

The problem with the dosage system is that its based on stallion progeny statistics which of course are only relevant once the stallions progeny record is complete. That is why this system struggles on a going forward basis. It can only tell 10 years after the fact who the chef's-de-race for that period actually were. It is a lagging quantifier instead of a leading indicator.

The reason Dual Qualifiers doesn't really work is because not only is Dosage fundamentally flawed but the Experimental Free Handicap is also flawed. First of all its a ranking compiled based on the opinions of handicappers. Like Timeform or the International Classification ratings its a subjective rating. Secondly the way 2yo's and 3yo's are being campaigned differs greatly from how they were in the 70's and 80's. 2yo's are maxxed out in order to capitalize on their precocity and are washed up by the time the Derby preps roll around. Top Derby candidates are often brought around slowly and often aren’t cranked up until they're 3yo's.

Street Sense did manage to win the Derby as a Dual Qualifier last year. However he was actually just 1 of the 8 Dual Qualifiers in the race so betting on them was again a losing proposition. Dual Qualifiers since 1996 have compiled a Derby record of 42-2-2-2 and have yielded a shabby -69.3% ROI. Compare that to the figures for the non Dosage qualifiers above then decide which one you think looks better.

The performance of the Dual Qualifiers in the Derby itself has been declining every single decade since it's inception.

7 races
7 winners (100.0%)
19 DQs (19.8%)
36.8% DQ winners
96 starters
IV = 5.05

10 races
8 winners (80.0%)
33 DQs (19.3%)
24.2% DQ winners
171 starters
IV = 4.15

10 races
5 winners (55.6%)
39 DQs (25.7%)
12.8% DQ winners
152 starters
IV = 2.16

8 races
1 winners (12.5%)
28 DQs (18.9%)
3.8% DQ winners
148 starters
IV = 0.66

The one thing you ought to note is that the DQ system has a declining winning pct for every successive decade of data provided. To get a better picture you should remove the years 1973 to 1983. Those years are of course perfect, because the system was created on historical data and was not officially released until 1984. The real test of a system is not using back tested data but from a certain point going forward.

Since 1983 there have been 24 Derby's and just 10 Derby winners fit the system. What it means for the Dosage theory is that its most successful period was prior to its actual creation, which of course leads one to assume curve fitting was an integral part of the research. On a going forward basis the entire system has been under performing. Even look at the system on a 5 year basis

4 of 5
2 of 5
3 of 5
0 of 5
1 of 4

The system has been irrelevant since 5 years after it was created. Up to that time it only missed 1 of 16 Derby's. Since then its gone 6 for 19.

This very much flies in the face of conventional wisdom, pretty much everyone accepts that Dosage is a handy tool to help narrow down Derby contenders, but the real test of every system is how well it does going forward. Even my own Triple Crown 20-20 is a system in its infancy based primarily on historical data. If it fails to perform over the next 5 years it will be a useless fraud.

Be very careful about the Statistics you accept. Just because they seem good at a glance doesn't mean they're useful at all. Among the horses on this years Triple Crown Trail with a Dosage greater than 4.00 are War Pass and El Gato Malo. It remains to be seen if either will make it to the Derby but as of now both look like decent plays. It may be yet another tough year for the Dosage system.

No comments: