Am I falling into a trap?

[traynor]

Originally Posted by MJC922

The model that is worthwhile to me has to have inherent value and as such I suspect it should make some money without considering the odds -- not much money but some several pct. Certainly such a model will make more money at certain odds ranges but leveraging that fact in anything but a large pool is well beyond my current capabilities. I agree with you on value capping.

That is a requisite first step that almost everyone fails to take (or to succeed in taking). If one cannot pick enough winners consistently enough to generate a healthy flat bet profit, all the rest is wasted effort. Once that point is reached, leveraging that advantage into a major return is not all that difficult. What I suggest is that one can gain far more leverage by tossing the outliers in the high odds range than can be gained by "value wagering."

[Ray2000]

Thanks for all the comments, it's made me think of the trap in a different way. Good points everyone.

I see now that the 15000 subset of races between 4 and 8 dollar payoffs, by definition has no payoffs outside that range.
But if I make conditional bets on my picks when odds are in the range 1.0 to 3.0 ($4-8) there will be many losing races where the winner paid outside of that range.

So my set of wagered races won't have the same properties as the select test group of races and I shouldn't expect similar performance boost.

I may change my mind again after a few Scotch-n-sodas....Happy New Year, Everybody

[MJC922]

Originally Posted by traynor

That is a requisite first step that almost everyone fails to take (or to succeed in taking). If one cannot pick enough winners consistently enough to generate a healthy flat bet profit, all the rest is wasted effort. Once that point is reached, leveraging that advantage into a major return is not all that difficult. What I suggest is that one can gain far more leverage by tossing the outliers in the high odds range than can be gained by "value wagering."

Seems to me we're in agreement that value wagering in the traditional sense is now dangerous waters. There are cases when one horse gets pounded by some loon and it creates overlays on multiple horses but I don't think it's a scenario which occurs often enough for value wagering to carry the day. Furthermore, if the inputs being used in the model aren't sufficiently divergent from the crowd's at large then all too often these 'value horses' are what might amount to 'dead on the board' horses. Not fertile ground for profits. If one has a model that produces crowd odds that's a nice achievement however it's sure to be a loser. It takes divergence to create inherent value. I think the OP has achieved some divergence as it's losing a fraction of the take, but it would be nice to get a bit more.

[classhandicapper]

Originally Posted by traynor

You may also recall that I mentioned "discovering" (with Katherine Jung's assistance and encouragement) that as soon as I culled my selections going to post at odds OVER 3/1 at Fraser Downs (the $8.00 top end you mention above), my ROI took a major leap up.

Interesting.

Back in the early 90s I was very selection oriented. Most horses were not huge longshots, but not all favorites either. Someone suggested to me I look through my database to see what would have happened if I had only bet on the horses that went off at 3-1 or higher. That changed the way I bet. I stopped betting favorites.

Now I am sort of a combination of selection oriented with a minimum price and "value oriented insight" that puts me all over the map.

[classhandicapper]

Originally Posted by MJC922

Furthermore, if the inputs being used in the model aren't sufficiently divergent from the crowd's at large then all too often these 'value horses' are what might amount to 'dead on the board' horses. Not fertile ground for profits.

I agree.

I've said this in the past. I try to insist on understanding why the horse is being overlooked before I bet. I have to know going in "this horses will probably be in the 8-1 range because the crowd will not appreciate the fact that XXXXX". If I'm expecting the horse to be 3-1 because his favorable qualities are very apparent, but he winds up at 8-1, I'm doing some more thinking before I pull the trigger. I could easily be missing something or someone might know something I don't know.

[traynor]

Originally Posted by classhandicapper

Interesting.

Back in the early 90s I was very selection oriented. Most horses were not huge longshots, but not all favorites either. Someone suggested to me I look through my database to see what would have happened if I had only bet on the horses that went off at 3-1 or higher. That changed the way I bet. I stopped betting favorites.

Now I am sort of a combination of selection oriented with a minimum price and "value oriented insight" that puts me all over the map.

Consistent healthy mutuels were readily available in the early 90s, and "winning" was something one did in between other interesting aspects of life. Not so much anymore. Nothing to do with the "onerous takeout." Lots to do with how people bet, especially online.

Up until about two years ago, I pretty much ignored odds, and bet according to models. Some up, some down, but as long as I stuck to the model selections, it was profitable. Often very profitable.

After the bowhunting trip/betting trip to British Columbia with Miss Jung, I realized that buckets of money were being ignored by the "value bettors"--especially those using online conditional wagers--and that money represented low-hanging fruit ripe for the plucking with minimal effort. It made a major difference in how I view races in general, and has been quite rewarding.

[classhandicapper]

Originally Posted by traynor

Consistent healthy mutuels were readily available in the early 90s, and "winning" was something one did in between other interesting aspects of life. Not so much anymore. Nothing to do with the "onerous takeout." Lots to do with how people bet, especially online.

Up until about two years ago, I pretty much ignored odds, and bet according to models. Some up, some down, but as long as I stuck to the model selections, it was profitable. Often very profitable.

After the bowhunting trip/betting trip to British Columbia with Miss Jung, I realized that buckets of money were being ignored by the "value bettors"--especially those using online conditional wagers--and that money represented low-hanging fruit ripe for the plucking with minimal effort. It made a major difference in how I view races in general, and has been quite rewarding.

Interesting and congrats.

The major obstacle I have now is that over the years I have varied my emphasis from a time based approach, to a comparative based approach, to a hybrid approach. I see advantages and disadvantages to each. When they agree, there is rarely an attractive price. When they disagree, there is rarely clarity of thought. So the only time I feel really comfortable is when I think I have a special insight into a particular horse, trainer, or race that is not generally understood. Those are infrequent in NY.

That's why I've taken to databasing. I am creating tools that may enable me to use the same insights to find races without the workload of following a circuit closely.

I produced a report for a coworker yesterday. He wrote back, "I'm sitting here looking for needle in a haystack and you just gave me a box of nails".

I hope he is right.

[MJC922]

Originally Posted by classhandicapper

I agree.

I've said this in the past. I try to insist on understanding why the horse is being overlooked before I bet. I have to know going in "this horses will probably be in the 8-1 range because the crowd will not appreciate the fact that XXXXX". If I'm expecting the horse to be 3-1 because his favorable qualities are very apparent, but he winds up at 8-1, I'm doing some more thinking before I pull the trigger. I could easily be missing something or someone might know something I don't know.

Certainly even today there are small pockets of specialized knowledge which go largely unnoticed. I think those who are tapped into these areas will continue to hold their own picking spots because of the inherent value in their selections and less due to 'value wagering' in the usual sense of creating a theoretical line and betting above it. As for models that's a different game, I'm no authority but I'd like to think I'm finally getting my senses dialed in. There appear to be some counterintuitive synergies which create just enough divergence with the crowd to be profitable, such horses don't particularly make better 'sense' to anyone eyeballing them. I haven't proved it to myself yet, however if a model can take the game head on then you might suspect it's not always making the obvious 'stand out' horse stand out even more, rather it's taking another sensible horse and elevating it often enough to discreetly capitalize.

[osophy_junkie]

Originally Posted by Ray2000

osophy_junkie

I did use 'Final known odds' in my early runs but it washed out all other factors except Morning Line. Maybe I should revisit.

What is the final number you get from regression? If it's a probability you can use it, plus the final odds in conditional logit to get a new probability. This is what I do.

Ed

[traynor]

Originally Posted by MJC922

Certainly even today there are small pockets of specialized knowledge which go largely unnoticed. I think those who are tapped into these areas will continue to hold their own picking spots because of the inherent value in their selections and less due to 'value wagering' in the usual sense of creating a theoretical line and betting above it. As for models that's a different game, I'm no authority but I'd like to think I'm finally getting my senses dialed in. There appear to be some counterintuitive synergies which create just enough divergence with the crowd to be profitable, such horses don't particularly make better 'sense' to anyone eyeballing them. I haven't proved it to myself yet, however if a model can take the game head on then you might suspect it's not always making the obvious 'stand out' horse stand out even more, rather it's taking another sensible horse and elevating it often enough to discreetly capitalize.

I very much agree. Most of the profit in racing is in the fact that most bettors want to be "right" more than they want to generate profit. That is, the more consensus they find by viewing multiple scenarios, the more certain they are of being right--and the more they tend to do little beyond picking the favorite. Specifically, the more an entry appears to be a standout, the less likely it is to be worth betting at all.

[classhandicapper]

Originally Posted by traynor

That is, the more consensus they find by viewing multiple scenarios, the more certain they are of being right--and the more they tend to do little beyond picking the favorite. Specifically, the more an entry appears to be a standout, the less likely it is to be worth betting at all.

I agree with that. At the same time, when you identify truly legitimate favorites, an even faster path to ruin is playing against them.

Assume favorites as a group lose around the track take (give or take). If you eliminate some of the truly suspect ones, the remainder will outperform the take even if they still produce losses. If one group outperforms the take, betting against them is bucking odds even worse than the take. That's a tough path.

That's what I was saying above. When all the stars line up, I can't get a price. So I generally pass. When the stars don't line up, it's hard to be confident you are assessing the situation properly because various valid methodologies can suggest different things.

[JJMartin]

Originally Posted by Ray2000

Here's the setup

I did a logistic regression analysis on several race factors using about 15000 races from 2013. Derived the coefs and intercepts for each of 20 different tracks and used them in a Computer program for making Win bets. It had an overall performance of 40% Strike, 0.95 ROI. I back tested it on 15000 races from 2012 and have now completed another 15000 from 2014. Very similar results for both testing years.

Here's the trap

If I look at a subset of ~ 5000 races (2012-2014) where the win payout price is >$4.00 but <$8.00 and test my program on just those races then the ROI performance my win selections jumps dramatically into a range of ~ 45% Strike, 1.20 ROI.

OF course, I realize that I won't know the winner's price at the time of making the bet... but if I limit my bets by conditional betting set at odds > 1.0 AND < 3.0 (at-1 MTP), can I expect the jump in ROI performance?

I have an un-easy feeling that my logic isn't sound and I'm mixing apples with oranges here.

Any opinions?

Test again and exclude the m/l fav completely

[traynor]

Originally Posted by classhandicapper

I agree with that. At the same time, when you identify truly legitimate favorites, an even faster path to ruin is playing against them.

Assume favorites as a group lose around the track take (give or take). If you eliminate some of the truly suspect ones, the remainder will outperform the take even if they still produce losses. If one group outperforms the take, betting against them is bucking odds even worse than the take. That's a tough path.

That's what I was saying above. When all the stars line up, I can't get a price. So I generally pass. When the stars don't line up, it's hard to be confident you are assessing the situation properly because various valid methodologies can suggest different things.

I think a very useful approach is avoid (like the plague) any notion of "triangulation" or agreement using multiple methodologies in selecting wagers. I don't consider individual races with much interest (or anxiety). I do all the work in the modeling stage.

In fact, the most profitable wagers I find are those most would consider "suspect" or (if favored) "false favorites." When profit conflicts with certainty and confidence, I will leave the warm fuzzies for others and go for the profit every time.

[Ray2000]

osophy_junkie

The method I use follows Wong's "Mathematical Methods in Horse Racing'

Each track has its own set of coeffecients (and an intercept) for 30 variables (includes 10 post positions)
Each entry in a specific race gets a rating by EXP(SUM(coefs*var+int)) / (1+EXP(SUM(coefs*var+int)))
The probability for each entry is then rating / sum of ratings

I'll look at combining it with final odds to get a new prob. ...,,Bayesian approach?



JJMartin

I've done that using Morning Line Ranks.
11,369 bets on picks with MLRank > 2 gives Strike of 26% and ROI of 1.02
6,306 bets on picks with MLRank > 3 gives Strike of 23% and ROI of 1.04

The image shows the same 44,281 bets sorted in order of increasing ML rank. (Ignore the blue lines)

[thaskalos]

"Lady luck never gives, she only lends"...

...with heavy interest.