To all those who say you can't win using speed & pace figures only...

[DeltaLover]

Originally Posted by PaceAdvantage

If not, then I think we're all in agreement that no measure of the past, in any way, shape, or form, is guaranteed to be duplicated (or even nearly duplicated) in the future when it comes to racing.

What you are saying here is absolutely correct and it is not limited to racing but in any experiment that can be measured empirically. Induction does not and cannot prove anything but merely cultivates our curiosity until we prove it wrong!

[traynor]

Originally Posted by PaceAdvantage

So are you trying to say you have a method that allows you to test/measure/whatever, and is virtually guaranteed to perform similarly going forward, however many number of races?

If not, then I think we're all in agreement that no measure of the past, in any way, shape, or form, is guaranteed to be duplicated (or even nearly duplicated) in the future when it comes to racing.

Not guaranteed to perform similarly. Quite the contrary. Data mining horse races in the real world does little more than suggest possible trends. Applying other filters and processes to the mined data can improve the consistency of (possible) returns. That is a LONG way from "guaranteed to perform similarly going forward."

Data mining does not produce The Answer or Truth-with-a-big-T. It produces things that can be used for a time to generate profit. An equivalent in business is milking the cash cow while it is standing still.

[traynor]

Originally Posted by DeltaLover

What you are saying here is absolutely correct and it is not limited to racing but in any experiment that can be measured empirically. Induction does not and cannot prove anything but merely cultivates our curiosity until we prove it wrong!

Absolutely. Well said.

[thaskalos]

Originally Posted by PaceAdvantage

I bet on almost every race I posted...these were all live plays...nothing backfitted...this was a live test from the start.

In fact, the method of play I used to post all of these plays was never conceived in any way by backfitting anything in a database on past races.

When you started this "live test", was your intention to bet on ALL your overlays in a race...or were you only isolated on your overlaid "top choices"? Because, if your original intention was to bet on ALL your overlays in a race...but then you noticed during the test that your overlaid top choices were the real profit-generators, and you "adjusted" the rules of the test mid-stream...then "backfitting" did indeed play a role in this handicapping exhibition.

If a "live test" is to be deemed legitimate...then the initial rules must be followed during the entire duration of the test...IMO.

[PaceAdvantage]

Originally Posted by thaskalos

When you started this "live test", was your intention to bet on ALL your overlays in a race...or were you only isolated on your overlaid "top choices"? Because, if your original intention was to bet on ALL your overlays in a race...but then you noticed during the test that your overlaid top choices were the real profit-generators, and you "adjusted" the rules of the test mid-stream...then "backfitting" did indeed play a role in this handicapping exhibition.

If a "live test" is to be deemed legitimate...then the initial rules must be followed during the entire duration of the test...IMO.

I did just as you stated in your first paragraph, however, I didn't change anything regarding the actual process of selecting contenders, although I did increase my overlay requirement from 25% to 50% above fair value (thus the value line was crafted differently about 50% of the way into the test...from 25 to 50% above fair value...the fair value line never changed).

I can live with this, as I had no way of knowing if doing that was going to work out or not. And I believe I had a positive ROI on the top choice when I did this. And since I ended with a 4% ROI, it looks like it didn't have much impact, at least as far as this particular test is concerned.

[Tom]

We have no idea how this method will perform in the future.
We do know how it has performed so far, and PA has the proceeds in his pocket.

IMHO, this bets testing several hundred thousand races and then deciding to start making bets. As PA has mentioned, this current method is the result of many tweak to the Excel sheet he uses. Had he started with the first weightings and waited for a quarter mil results to decide he needed to make a correction, he would be freaking 104 years old when he finally gets around to betting.

1 race with $2 bet on it is worth 1,000 test races.

Hats off to PA for doing it the right way - now, live, and with money on it. Now is how.

[Robert Fischer]

Originally Posted by classhandicapper

The problem with the "more bets" equal "more profit" theory is that reality sometimes doesn't match the theory.

For arguments sake, let's say I had a positive 10% ROI over many years being very selective. Some might argue that I could make a lot more money if I just loosened up a little and was willing to accept a positive 5% ROI making a lot more plays and dramatically increasing my handle.

That's perfectly logical.

But if the reality is that I'm breaking even (or even worse losing) on all those extra plays, I'm not really adding anything to the bottom line even though my handle would be way higher.

Most people already bet whenever they think they have a clear edge. Despite that, it's a struggle to win because sometimes we are working with imperfect information and understanding. When we start cutting it even closer, a lot of the time we think we are playing small overlays we are actually playing small underlays. The net could be disappointing to the point where we are just running in place on all those extra bets or even worse losing on them - even if we are still netting positive on everything because of the original standard plays.

Before you start adding bets, I think you have to insist that the standards for the new ones are similar to the standards for ones that are already doing well. That's easier said than done.

I agree with what you are saying.

Frequency vs. ROI
churn ev vs roi
arguments are complex, and depend on perspective.

A big part of that perspective is governed by the player's 'circle of competence' or 'style'.

The golden rule is to stay in your circle of competence.
By all means, find more plays(preparation, applying your models across additional cards, stamina, etc..) within your circle of competence.
By all means, grow(become competent in a wider range of pools, and race-types) your circle of competence.

But the second you start to leave your circle of competence to guess and gamble, your EV is likely to drop and your risk of ruin is likely to rise.


assuming you wish to maximize profits = of course you should make every positive expectation wager that you can find.

bet the cream of the crop
bet an increased amount(that agrees with insight into probability and risk, eg Kelly) on the crème de la crème.

[pandy]

The bottom line is, testing is good. Live testing is the best. Whether or not the results will remain the same in the future is an obviously moot point. The only thing guaranteed in life is death. But testing is the way to go and if your test shows a profit you've got something to work with.

Although live testing is the best, back testing is better than no testing.

[classhandicapper]

I just tested an automated spot play that was profitable for 2016 and so far in 2017 at AQU, BEL, SAR, SA, DMR, CD, KEE, and GP. So I decided to test it against 2015 data at the same tracks and it was a disaster.

Now I wish I had 2014 data.

[cj]

Originally Posted by classhandicapper

I just tested an automated spot play that was profitable for 2016 and so far in 2017 at AQU, BEL, SAR, SA, DMR, CD, KEE, and GP. So I decided to test it against 2015 data at the same tracks and it was a disaster.

Now I wish I had 2014 data.

2015 is irrelevant by now anyway

[classhandicapper]

Originally Posted by cj

2015 is irrelevant by now anyway

Agreed. It's 2018 I'm worried about.

[DeltaLover]

Originally Posted by classhandicapper

I just tested an automated spot play that was profitable for 2016 and so far in 2017 at AQU, BEL, SAR, SA, DMR, CD, KEE, and GP. So I decided to test it against 2015 data at the same tracks and it was a disaster.

Now I wish I had 2014 data.

Maybe you need to rethink your approach and instead of using fixed factors to work out a solution to dynamically change them through time and location. Nothing is going to work globally or eternally, almost everything changes (but also something remains always the same!)

Everything Flows (Heraclitus)

Change is impossible (Parmenides)

[classhandicapper]

Originally Posted by DeltaLover

Maybe you need to rethink your approach and instead of using fixed factors to work out a solution to dynamically change them through time and location. Nothing is going to work globally or eternally, almost everything changes (but also something remains always the same!)

Everything Flows (Heraclitus)

Change is impossible (Parmenides)

What happened was I built a system of reports that I could use to quickly identify a list of races every day that might be more playable and a list of horses that could potentially be underrated or overrated in those races. It mimicked my general thinking as best I could without dedicating a huge amount of time to development.

The idea was mostly to save time in my handicapping process. That way I could go straight to the races and horses that offered the most promise and waste less time.

Once it was done, I decided I should at least take a look at the back results to see if an automated process like that was at least putting me on "live play" horses and "live bet against" horses the way I intended. I was hoping as a group they would outperform the take in both directions and that my handicapping and the odds would dictate my plays from there.

But when I ran the first test I was shocked by the results. With no handicapping at all (no looking at surface, distance, figures, class, layoff etc..) it was very profitable on live plays and the bet against horses were terrible. But the celebration was premature. When I tested 2015 it was terrible.

The net of all of it was still very good as a starting place for the original intent. The "plays" were still mildly profitable and the bet against were still worse than the take, but I don't think I'll be playing any of these horses on automatic pilot.

It was a good test study of this kind of problem. The original test was over a substantial amount of time, covered quite a few horses, and all the major tracks. Yet it was pretty bad in 2015.

[pondman]

Originally Posted by PaceAdvantage

If you've read this thread, you'd see I already acknowledged that my results would probably improve by working in other handicapping factors.

Not my point in creating this thread though. Besides boasting about my positive ROI on my top pick over close to 400 races, the major point of the thread is to show proof that you CAN obtain a positive ROI by ONLY using numbers to the exclusion of everything else. Something a lot of people think is nearly impossible to do in this day and age, when most EVERYONE is using numbers.

If your top pick finishes in the top 4 spot at least 80% of the time, you're on your way to becoming wealthy. But I'd suggest not trying to add factors to this selection, and making a giant super selection. I'd search for a completely different independent variable, and make it another leg of a horizontal. In other words your top selection will always be part of a Tri or Super, and probably should be used through out both.

When we developed our screen for verticals, we found that our number 1 contender was still running at 20% in the 4th position (due to shorter fields in California.) Our 2nd contender is stable enough to consider it for the super. But after the 2nd, it degrades so bad that it's best to look for something else. It's not that you throw the 3rd out, because it could come into play as another method, for example if it's a surprise favorite and gets obvious pounded by a connection or whale. That's really why I'd always make that point of telling people not to get hung up on playing a 1-2-3-4 vertical box from their own super duper selections, if they are only using one method. It will be a long term loser.

[pandy]

Originally Posted by pondman

If your top pick finishes in the top 4 spot at least 80% of the time, you're on your way to becoming wealthy. But I'd suggest not trying to add factors to this selection, and making a giant super selection. I'd search for a completely different independent variable, and make it another leg of a horizontal. In other words your top selection will always be part of a Tri or Super, and probably should be used through out both.

When we developed our screen for verticals, we found that our number 1 contender was still running at 20% in the 4th position (due to shorter fields in California.) Our 2nd contender is stable enough to consider it for the super. But after the 2nd, it degrades so bad that it's best to look for something else. It's not that you throw the 3rd out, because it could come into play as another method, for example if it's a surprise favorite and gets obvious pounded by a connection or whale. That's really why I'd always make that point of telling people not to get hung up on playing a 1-2-3-4 vertical box from their own super duper selections, if they are only using one method. It will be a long term loser.

How would you structure the bet?