Looking for opinions

[tackleberry]

Hello all,

Recently I have been looking at handicapping only maiden races.

My current attempt uses several factors to assign each horse a numeric value based on how they stack up to the rest of the field. My first look at how my approach performed showed promise, betting horses that showed either the highest or second highest possible rank showed a flat bet profit for win and place, with just above break even for show bets. About 1/3 of all races I looked at would include a horse that fell into this category.

I have no interest in blindly betting horses without taking into consideration the price I am getting on the horse. So what I did was for every race each horse of every rank I would attach a value for how often this horse should win by random chance. For example, in a 10 horse field each horse would be 0.1. I then looked at every horse that raced in my data sorted them together by rank value and summed the values for how often they should win if everything was random.

My results from that looked pretty good to me.

1st rank won 3.1 x as often as expected if random
2nd rank 2.7
3rd 1.9
4th 2.0
5 0.9
6 0.8

The numbers looked really good to me since they show what you would expect to find, the horses that should be the best, perform the best and it drops off as you go down the line.

My plan would be to then use these numbers to find overlays. For example for a horse that gets the top rank in an 8 horse field I would expect random chance to give a horse a 12.5% chance of winning, so 3.1x12.5= 38.75% which would make 2/1 or better an overlay on the horse.

Really just looking for feedback on this thus far and where other people would go next. What else can I do aside from get fresh data and put the same approach to work there and see how the results compare?

Thanks

[fmolf]

Originally Posted by tackleberry

Hello all,

Recently I have been looking at handicapping only maiden races.

My current attempt uses several factors to assign each horse a numeric value based on how they stack up to the rest of the field. My first look at how my approach performed showed promise, betting horses that showed either the highest or second highest possible rank showed a flat bet profit for win and place, with just above break even for show bets. About 1/3 of all races I looked at would include a horse that fell into this category.

I have no interest in blindly betting horses without taking into consideration the price I am getting on the horse. So what I did was for every race each horse of every rank I would attach a value for how often this horse should win by random chance. For example, in a 10 horse field each horse would be 0.1. I then looked at every horse that raced in my data sorted them together by rank value and summed the values for how often they should win if everything was random.

My results from that looked pretty good to me.

1st rank won 3.1 x as often as expected if random
2nd rank 2.7
3rd 1.9
4th 2.0
5 0.9
6 0.8

The numbers looked really good to me since they show what you would expect to find, the horses that should be the best, perform the best and it drops off as you go down the line.

My plan would be to then use these numbers to find overlays. For example for a horse that gets the top rank in an 8 horse field I would expect random chance to give a horse a 12.5% chance of winning, so 3.1x12.5= 38.75% which would make 2/1 or better an overlay on the horse.

Really just looking for feedback on this thus far and where other people would go next. What else can I do aside from get fresh data and put the same approach to work there and see how the results compare?

Thanks

Last time i checked the numbers at saratoga the favorite in msw races was winning 48% of the races!....I always look for the horses taking money.I play mostly nyra tracks and usually the winner is bet down from the morning line.Cold on the board usually means underlay!

[Some_One]

My POV for what you are up against:
-The fact that Win ROI is > Show ROI makes me believe you have a fairly high average price 5-1+, is that so? My rule of thumb for testing is 200 X average price sample size needed to start ruling out variance.

-On the overlay part, let use a factor like Speed Rating last out, below is the results of non-mdn races for the rank LR SR in my HSH database
Field1 Starts Pays Pct $Net IV
----------------------------------------------------
1 25,734 6,471 25.1 $1.70 1.95
2 24,249 4,385 18.1 $1.66 1.41
3 23,486 3,307 14.1 $1.59 1.10
4 23,017 2,647 11.5 $1.52 0.89

So the best LR SF will win 1.95X vs random, but will you be better off just chosing the higher prices as per your calculations? No

WIN BETS
LR SF Odds Starts Pays Pct $Net IV
---------------------------------------------------------------
1 2 9,429 3,845 40.8 $1.72 3.02
1 4 7,801 1,741 22.3 $1.71 1.75
1 8 5,273 691 13.1 $1.68 1.05
1 16 2,284 162 7.1 $1.65 0.59
1 99 946 32 3.4 $1.71 0.30

The above is the LR SF performance split by odds, the hit rate on a 2-1 horse was much higher than a 16-1 horse, but the ROI was about the same throughout. So what I'm trying to say is you can't turn a -EV into a + result by just focusing on 'price horses'.

BTW here is the result when looking at show betting, you can see why I made my first point about your current test results

SHOW BETS
LR SF Odds Starts Pays Pct $Net
---------------------------------------------------------------
1 2 9,425 7,139 75.7 $1.83
1 4 7,801 4,591 58.9 $1.75
1 8 5,273 2,295 43.5 $1.65
1 16 2,284 679 29.7 $1.57
1 99 946 155 16.4 $1.37

I think if you are using a DRF/Brisnet factor, I wouldn't bet on your success, sorry. If you are doing some crazy Dr Roman dosage thing (or something else not on the DRF), then I'll give you a shot.

(Edit- I selected the last race speed figure at random for demo purposes)

[tackleberry]

The average winning price was 3.43/1

Variance is certainly a concern, which would be one of the reasons to test further with new data.

You example about speed figures last race is well taken.

And no need to say sorry for not betting on my success, neither am I at this point lol

[cashmachine]

Originally Posted by tackleberry

The average winning price was 3.43/1

Variance is certainly a concern, which would be one of the reasons to test further with new data.

You example about speed figures last race is well taken.

And no need to say sorry for not betting on my success, neither am I at this point lol

I would say that your computations of "overlay" are very crude in the sense that in order to come up with probability of a horse to win you use bin average, not actual probability of win computed specifically for this horse. If even such crude method is profitable, then definitely your system has huge potential because you would get much better results simply by changing the way how you compute probability of win.