How to make money in horse racing

[formula_2002]

Originally Posted by elhelmete

have you met formula_2002?

to meet him is to love him

[MJC922]

Originally Posted by headhawg

Getting the data is easy. Expensive, but easy. Creating a well-designed database is important, and understanding how to do a proper analysis is even more important. Drawing the wrong conclusions is detrimental to the bankroll.

Adding on to that warning from some of my own experiences, I would strongly recommend the OP split a large sample before anything is tested whatsoever. People need to understand that when someone asks the computer, show me all of the combinations of factors which make money over a sample, what is being returned in that subset is honestly nothing more than a series of backfits. It can be a really large sample too, sadly that doesn't matter because there's a ton of noise in payoffs even in large samples.

So we have these combinations with promise in hand. From there those combinations of factors need to be validated on another sample altogether, a clean pristine never tested sample. At this point when 19 out of 20 of those go negative, now we're finally getting somewhere, and frankly I would go ahead and use the newer more recent split for the initial queries and then do the validation on the older sample. The benefit to that is it puts to rest these folks who think large samples go negative afterward because people (whales, the easter bunny etc) are adjusting and taking away all the value that was there. Well if you do it in reverse that puts that myth to rest.

Now what's remaining, that 1 out of 20 that survived trial by fire has good promise, but even then we have to keep in the back of our minds it's not impossible with enough positive ROIs in the first sample that some of these subsets will no doubt be positive due to noise in the payoffs of the second sample too. A combination which 'accidentally' beats two samples is inevitable with enough positive ROIs to retest.

So IMO it's good to do research obviously, WPS% etc it's all worthwhile but taking ROIs by the horns are a serious challenge.

[Triopstor]

I guess rule #3 might be have data “race type” and “race track” specific. I'm more concerned about "Know Your Track" than "race type".

[Triopstor]

Originally Posted by headhawg

Drawing the wrong conclusions is detrimental to the bankroll.

One example could be wager the horse in the race with the most letter "D"s in the name. Northern Dancer has one "D" in it. Could make a profit for 5 months with positive Return On Investment(ROI) then mysteriously beaten when new maidens come to the track.

[Triopstor]

Originally Posted by formula_2002

to meet him is to love him

Hello formula_2002! Peace to you! And your kingdom!

[Triopstor]

Originally Posted by MJC922

So IMO it's good to do research obviously, WPS% etc it's all worthwhile but taking ROIs by the horns are a serious challenge.

I figure that having some longshots in the mix could recover the losses and in the end produce a positive Return On Investment(ROI).

[Robert Fischer]

Twitter

__________________
Preparation. Discipline. Patience. Decisiveness.

[ranchwest]

Originally Posted by Triopstor

I guess rule #3 might be have data “race type” and “race track” specific. I'm more concerned about "Know Your Track" than "race type".

My opinion... that's exactly what not to do. Let the data guide you, don't make assumptions up front about where the data "should" go. It's always possible that race type could be the cornerstone to your best findings.

[ranchwest]

Originally Posted by MJC922

Adding on to that warning from some of my own experiences, I would strongly recommend the OP split a large sample before anything is tested whatsoever. People need to understand that when someone asks the computer, show me all of the combinations of factors which make money over a sample, what is being returned in that subset is honestly nothing more than a series of backfits. It can be a really large sample too, sadly that doesn't matter because there's a ton of noise in payoffs even in large samples.

So we have these combinations with promise in hand. From there those combinations of factors need to be validated on another sample altogether, a clean pristine never tested sample. At this point when 19 out of 20 of those go negative, now we're finally getting somewhere, and frankly I would go ahead and use the newer more recent split for the initial queries and then do the validation on the older sample. The benefit to that is it puts to rest these folks who think large samples go negative afterward because people (whales, the easter bunny etc) are adjusting and taking away all the value that was there. Well if you do it in reverse that puts that myth to rest.

Now what's remaining, that 1 out of 20 that survived trial by fire has good promise, but even then we have to keep in the back of our minds it's not impossible with enough positive ROIs in the first sample that some of these subsets will no doubt be positive due to noise in the payoffs of the second sample too. A combination which 'accidentally' beats two samples is inevitable with enough positive ROIs to retest.

So IMO it's good to do research obviously, WPS% etc it's all worthwhile but taking ROIs by the horns are a serious challenge.

Good summary. I would probably break it down into smaller sessions and see what percentage of sessions were profitable. And look at both medians and averages.

[thaskalos]

It's very easy to create a system that will show a profit over a thousand races that have already been run. The hard thing is to show a profit over a thousand races that haven't been run yet.

[DanBoals]

Originally Posted by thaskalos

It's very easy to create a system that will show a profit over a thousand races that have already been run. The hard thing is to show a profit over a thousand races that haven't been run yet.

This !!!

[Triopstor]

Originally Posted by ranchwest

My opinion... that's exactly what not to do. Let the data guide you, don't make assumptions up front about where the data "should" go. It's always possible that race type could be the cornerstone to your best findings.

I have seen handicapping that didn't work for other tracks. Like Fort Erie, Belmont's post position 1.

[ranchwest]

Originally Posted by Triopstor

I have seen handicapping that didn't work for other tracks. Like Fort Erie, Belmont's post position 1.

Nothing wrong with specializing in a track or circuit.

[DanBoals]

I would like to expand a bit on what thaskalos said earlier...


What works in the past, is certainly not guaranteed to work going forward. The most famous examples I can think of are Robert Saunders Dowst who iirc had a system of betting horses that had 6 wins in their last 10 races. Pretty simple, right? Apparently people were not doing it. He told everyone how well it worked, people watched and saw it worked, everyone started using it, and then the mutuels fell so low it no longer worked. Howard Sartin championed pace handicapping in the 1980s. He used ideas from Ray Taulbot and Huey Mahl and oversaw the creation of software that for the first time really allowed any kind of quick pace analysis. He published his work and gave his followers the formulas, and they were quickly pirated by ex members and others until his pace picks dropped from $11.00 wins to $6.00 wins. As ideas catch on, the mutuels reflect this.


Having a database to test your ideas on is a very powerful tool.


You can't stand still with something like that though, you are going to have to constantly refine what is happening, because everyone else with a huge database like yours is going to be constantly refining their approach. I would be willing to wager that the large betting consortiums have some nice databases and some decent computer guys to mine their data and they get bets from this work that are currently profitable. You will be competing with these outfits. The mutuels determine the profitability as much as the hit rate, they go hand in hand. As the whales bet, the mutuels fall. If you could figure out what the whales were betting on and then figure out when those bets were not profitable, that seems like it would be the most profitable niche.


When Howard Sartin developed his software, he had Bob Purdy and another PIRCO founder type in 10,000 races and their results. He told me the data changed very little after 1200 races and not at all after 2000. The database of the 10,000 races is where he got his percentage win rates for his numbers on early/late/total. Iirc, a horse that was 1-1-1 would win about 53% of the time. This is pretty useful information, since you can be sure a horse that is top early, top late and top total is going to be heavily bet. If you can figure out why those horses lose 47% of the time, you can bet against those false favorites and make some money.


Instead of trying to come up with an instant database, I would suggest you build your database for free over time with result charts. As you are building your database, handicap, handicap, and handicap some more. Come up with ideas on whatever way too small sample you have and try those ideas now. You are going to learn a lot more from handicapping races and testing ideas, than you are from playing with a database for a year. Everyone has a perfect battle plan until bullets start flying, so handicap and learn from your experiences and then when mine your database and start developing ideas, you will be way ahead of the game.


Best of luck to you in this wonderful and difficult game.

[Triopstor]

Originally Posted by DanBoals

I would like to expand a bit on what thaskalos said earlier...


What works in the past, is certainly not guaranteed to work going forward. The most famous examples I can think of are Robert Saunders Dowst who iirc had a system of betting horses that had 6 wins in their last 10 races. Pretty simple, right? Apparently people were not doing it. He told everyone how well it worked, people watched and saw it worked, everyone started using it, and then the mutuels fell so low it no longer worked. Howard Sartin championed pace handicapping in the 1980s. He used ideas from Ray Taulbot and Huey Mahl and oversaw the creation of software that for the first time really allowed any kind of quick pace analysis. He published his work and gave his followers the formulas, and they were quickly pirated by ex members and others until his pace picks dropped from $11.00 wins to $6.00 wins. As ideas catch on, the mutuels reflect this.


Having a database to test your ideas on is a very powerful tool.


You can't stand still with something like that though, you are going to have to constantly refine what is happening, because everyone else with a huge database like yours is going to be constantly refining their approach. I would be willing to wager that the large betting consortiums have some nice databases and some decent computer guys to mine their data and they get bets from this work that are currently profitable. You will be competing with these outfits. The mutuels determine the profitability as much as the hit rate, they go hand in hand. As the whales bet, the mutuels fall. If you could figure out what the whales were betting on and then figure out when those bets were not profitable, that seems like it would be the most profitable niche.


When Howard Sartin developed his software, he had Bob Purdy and another PIRCO founder type in 10,000 races and their results. He told me the data changed very little after 1200 races and not at all after 2000. The database of the 10,000 races is where he got his percentage win rates for his numbers on early/late/total. Iirc, a horse that was 1-1-1 would win about 53% of the time. This is pretty useful information, since you can be sure a horse that is top early, top late and top total is going to be heavily bet. If you can figure out why those horses lose 47% of the time, you can bet against those false favorites and make some money.


Instead of trying to come up with an instant database, I would suggest you build your database for free over time with result charts. As you are building your database, handicap, handicap, and handicap some more. Come up with ideas on whatever way too small sample you have and try those ideas now. You are going to learn a lot more from handicapping races and testing ideas, than you are from playing with a database for a year. Everyone has a perfect battle plan until bullets start flying, so handicap and learn from your experiences and then when mine your database and start developing ideas, you will be way ahead of the game.


Best of luck to you in this wonderful and difficult game.

1. Someone told me that the favourites of the 1970's usually paid 4-1
odds. I'm curious if that's true. I wonder how far a $2.00 ticket
in the 1950's got the handicappers of that age.

2. Doctor Sartin changed the horse racing world.

3. You have given me a big arsenal with " If you can figure out why those
horses lose 47% of the time, you can bet against those false favorites
and make some money." Make money against the super rich.

4. Build a database from DRF Results?

Best of luck to you too Dan!