why record keeping is a farce

[TonyMLake]

Originally Posted by traynor

I think that records comprised of small samples of races--no matter how "accurately" those records are kept--are pretty much worthless. Attempting to define reality by the results of the last 100, 200, or 300 races is an exercise in futilty. So Capper Al is right on that point--"record keeping" as ordinarily defined by the average bettor is a farce. Lots of pretty numbers and colorful screenshots, but essentially worthless as predictors.

The "trends" supposedly "discovered" are little more than a sprinkling of data points in a random distribution that "seem" to form a pattern. Sprinkle in a few more races (as in "try to apply the trend in the real world by betting on it") and the trend disappears, while a "new trend" seems to appear. Not worth much.

I agree in the majority, but respectfully disagree for the more grisly of us. Here's why:

If you look at the real statistical data kept since perhaps the 18th century, THERE IS ONLY ONE STATISTICALLY WINNING BET. That's an odds on favorite to place in a graded stakes. Sorry folks. This is what *all* the data kept has shown over the longest sample available.

However, short term "boutique" or "window" or "snapshot" analysis does indeed produce dollar value.

Why?

For one cynical example, if you go to the same track every day for a season, somewhere midway through the season, you'll learn who the cheaters are. Just knowing which jockeys will throw a race can make you a literal fortune. It's a VERY small sample, relatively... and it WONT last... but it is has OH SO MUCH value during its window.

Other boutique data might be "an old nag" who runs like hell on a muddy track. If you happen to know that your horse performs like Seattle Slew of the mud, well, you've got a piece of information that - although limited in duration and application - is exceptional in value.

As an interesting caveat to this: YOU CAN ONLY KNOW THAT YOU HAVE USEFUL SNAPSHOT DATA IF YOU ARE AN OTHERWISE EXCELLENT RECORD KEEPER.

So, go figure.

Just sayin

[Capper Al]

Originally Posted by Cratos

I am coming late to this thread, but I want to say loud and clear that I keep records and I keep them because I don't trust everything to my memory; and another reason I keep records is because I always want to keep my database up to date.

This is a good reason to keep records. The farce is believing that we can improve game by picking wager types.

[Capper Al]

Originally Posted by Tom

1. Handicapping 101
2. Yes, but I put you back on ignore and will stop soon.

That hurt!

You are full of yourself. And I'd appreciate it if you did put me on your ignore list because your contributions are more destructive than positive. I'd prefer that you would stay out of my topics. You just can't handle difficult subjects.

[Capper Al]

Record keeping for your personal use is helpful especially for budgeting and learning about yourself. The farce is in altering your handicapping based on information obtained from your records. If the game was simply picking your winning bets and playing them then there would be no need for anymore discussion.

[davew]

Originally Posted by Capper Al

Record keeping for your personal use is helpful especially for budgeting and learning about yourself. The farce is in altering your handicapping based on information obtained from your records. If the game was simply picking your winning bets and playing them then there would be no need for anymore discussion.

Is this the same guy who started this thread?


Record keeping has helped me by telling me what pools I lose the most in. Without changing my handicapping at all, just the pools I bet into has improved my results.

[Dave Schwartz]

If you look at the real statistical data kept since perhaps the 18th century, THERE IS ONLY ONE STATISTICALLY WINNING BET. That's an odds on favorite to place in a graded stakes. Sorry folks. This is what *all* the data kept has shown over the longest sample available.

I cannot address whether the above statement was ever true but it is clearly not true now.

This goes back 9 years.

Graded Stakes.
201-PubChPLACE BETS
Field1 Field2  Starts   Pays    Pct   $Net     IV    PIV
1/20                6      5   83.3  $1.75   4.75   0.75
1/9                14     13   92.9  $1.96   4.68   0.83
1/5                40     37   92.5  $2.06   5.49   0.96
1/3                79     64   81.0  $1.80   4.90   0.91
2/5               136    107   78.7  $1.82   5.11   0.94
1/2               138    108   78.3  $1.85   5.05   1.00
3/5               333    242   72.7  $1.81   4.71   1.02
4/5               434    288   66.4  $1.79   4.60   1.04

Total           1,180    864   73.2  $1.82   4.80   1.43

[TonyMLake]

Wasn't that the posters point? Even nine years worth of data yields an "incorrect" result when compared to the total population of all horse races ever, and therefore is meaningless (from within the context of the broader assertion).

My point was the opposite, that short aberrations can indeed produce value before reverting to the mean.

As for the "truth" about the quoted statistic: if it is not "true" within your 1188 race sample, I won't belabor it- its completely consistent with my point that sample data might or might not match the "true" population. As for the truth of the assertion itself, I would like to source it to Dr Ziemba, circa 1984 if memory serves. Perhaps it was truer then from within his sample than it is now from within yours ;^)

[Dark Horse]

Record keeping is simply another term for measuring. It's the basis for science. Some people don't like measuring, prefer the old ways, and go around in antiquated horse carts. Others dedicate themselves to precise measurement, insist on the best proportions, and end up zooming around in beamers. Bottomline? To each their own.

[Tom]

My point was the opposite, that short aberrations can indeed produce value before reverting to the mean.

Bingo!
Now is how.

For me, long term is often two weeks.
Often less.

[Dave Schwartz]

As for the "truth" about the quoted statistic: if it is not "true" within your 1188 race sample, I won't belabor it- its completely consistent with my point that sample data might or might not match the "true" population. As for the truth of the assertion itself, I would like to source it to Dr Ziemba, circa 1984 if memory serves. Perhaps it was truer then from within his sample than it is now from within yours ;^)

I am not trying to be argumentative but I am not understanding your meaning. Please explain.

It sounds a lot like, "If I had been wagering for the last 30 years I would have been profitable but now it doesn't work and that doesn't matter because I will still make money with it."

If so, I would say that you are in denial.

You cannot make money on something that worked 2 decades ago but loses now.

[TonyMLake]

No, I'm saying quite the opposite. Your data is a perfect example. One doesn't have to be a statistician to realize it's not quite representative of the whole population of horse racing ever. Why?

Your 1/9 and 1/20 stated odds aren't even mathematically significant. The 1/5 barely is. For the other 5 data points, it's obvious the trend line for $Net doesn't correlate exactly as we'd expect with the stated odds column... note the up and down pricing... and yet you are confident that your model has predictive power... AND SO AM I, because there is obvious predictive meaning behind it even though there is an incomplete data set.

I'm confident that if you look at horse racing and include the ENTIRE population of horse races (every track every surface every condition every type) you will find only a handful of statistics, AND NONE OF THEM CAN POSSIBLY PRODUCE VALUE because this is ultimately a (nearly) zero sum game (not to mention track take out). If you consider EVERY stat, you will always end up back at Zipf's law, which applies to the number of times a horse will win based on the tote board at post time. (the favorite wins twice as often as the second favorite, three times as often as the third favorite...) and not much else.

Indeed... I'll go so far as to say that it is ONLY in the outliers and aberrations where we will find value over time.

But maybe I'm wrong. What do you all think?

[Dave Schwartz]

I think that you are assuming (incorrectly) that racing does not change.

[lansdale]

Originally Posted by TonyMLake

No, I'm saying quite the opposite. Your data is a perfect example. One doesn't have to be a statistician to realize it's not quite representative of the whole population of horse racing ever. Why?

Your 1/9 and 1/20 stated odds aren't even mathematically significant. The 1/5 barely is. For the other 5 data points, it's obvious the trend line for $Net doesn't correlate exactly as we'd expect with the stated odds column... note the up and down pricing... and yet you are confident that your model has predictive power... AND SO AM I, because there is obvious predictive meaning behind it even though there is an incomplete data set.

I'm confident that if you look at horse racing and include the ENTIRE population of horse races (every track every surface every condition every type) you will find only a handful of statistics, AND NONE OF THEM CAN POSSIBLY PRODUCE VALUE because this is ultimately a (nearly) zero sum game (not to mention track take out). If you consider EVERY stat, you will always end up back at Zipf's law, which applies to the number of times a horse will win based on the tote board at post time. (the favorite wins twice as often as the second favorite, three times as often as the third favorite...) and not much else.

Indeed... I'll go so far as to say that it is ONLY in the outliers and aberrations where we will find value over time.

But maybe I'm wrong. What do you all think?

Hi TML,

Your post borders on the incoherent, but as far as I can piece together an argument, it seems to be that it's impossible for anyone to use statistical data (unless the database includes races dating from the time of the prehistoric eohippus) to build a profitable handicapping model. You seem not to realize that there are people who have done this very successfully, a number of them on this site. Unless you're a Bayesian (and I sense the answer is no), small samples are meaningless, as Traynor said. However sample sizes don't need to be infinite - Bill Benter did okay with ca. 2,500 races. That handicapping is a zero-sum game doesn't preclude the existence of winners, however small their number.

Cheers,

lansdale

[TonyMLake]

What I said that was wrong was my application of Zipfs law. The real stat is the favorite wins 1/3, the second favorite wins 1/3 of the remaining 2/3 that's left,and etc. I was typing on lunch before but I believe this is the correct stat.

In any case, simply because a person believes the value is in the outliers does not in any way mean that outliers cannot be systematically identified. That's exactly what underlays and overlays are!

The sum entirety of my point was that I both agreed and disagreed with the OP. The whole population is often not descriptive of current sample trends, and vice versa. But the only way to know is with historical data from which to draw comparisons

[Capper Al]

Originally Posted by Dave Schwartz

I think that you are assuming (incorrectly) that racing does not change.

This could be a thread on it's own.