change in long to short bias

[formula_2002]

In one of my threads I poinetd out something I though worth noting on it's own righ here;

Many articles have pointed to the bias in the short to long odds.
Short price horse lose less $ for their bettor then long priced odds.

"There is of course a short to long odds bias.
Fabricand's ("Horse Sense"), table on page 34 was based on play from 1955 to 1962 .



he looked at 93011 horses.
25,044 horse in the odds range 20-1 through 100-1 had a dollar loss of 54% .
3-1 lost 11% and
1 to 1.15 lost 2%.

In todays market, my data form 2003 to 2004 shows the following;

average 1-1 odds, dollar lost =15%
3-1 , dollar loss= 16%
>=20-1 dollar loss = 34%
that's base on over 100,000 horses.

I guess I could say there is less of a differential in the bias between short to long horses today then there was 40 some years ago.

I think the math for all odds ranges could possibly show that there is less of a standard deviation (a measure of risk) in todays market then 40 years ago.

If that was true, It would mean there is more certainity (less risk) to lose money today then 40 years ago. :)

[kenwoodallpromos]

So do you have an opinion on why this may be or are you just putting out the numbers?
My wild guess is nowadays those betting on short-priced horses believe more the BS about speed ratings and early speed and lay off the longshots.
If you could produce it I would like to see stats on the % of the total betting pool going to longshots and even money or less horses as compared to the past.
What effect does a higher % of low-priced horses losing have on those stats?

[Jeff P]

Do you really think that overlays are easier to find with short odds horses as opposed to longer odds horses? If so, why?

[formula_2002]

Based on those numbers it appears to me that the bettor is making better judgements about a horse's ability.

It may be that the factors derived from the past performance data is more accurate.

Perhaps the game is a bit more honest.

[kenwoodallpromos]

1962: 1 to 1.15 lost 2%.

In todays market, my data form 2003 to 2004 shows the following;

average 1-1 odds, dollar lost =15%.
Today's chalk bettors are losing 10% more according to youe post; they are much lousier at picking low-odds horses.
They stupidly accept 8' lengths at 1/5 second or 40' per second or 16 1/2 seconds per furlong for comparing speed; while at the same time accept 52-54' per second when comparing the same horses by pace figures.
In 1962 bettors did not have Beyers numbers to screw them up when picking favorites.
A One Rocket got a 100 Beyers then was DQed for a baking soda milkshake and his trainer arrested. What would you adjust the Beyer number to?

[Jeff P]

posted by Ken-
1962: 1 to 1.15 lost 2%.

In todays market, my data form 2003 to 2004 shows the following;

average 1-1 odds, dollar lost =15%.
Today's chalk bettors are losing 10% more according to youe post; they are much lousier at picking low-odds horses.
They stupidly accept 8' lengths at 1/5 second or 40' per second or 16 1/2 seconds per furlong for comparing speed; while at the same time accept 52-54' per second when comparing the same horses by pace figures.
In 1962 bettors did not have Beyers numbers to screw them up when picking favorites.
A One Rocket got a 100 Beyers then was DQed for a baking soda milkshake and his trainer arrested. What would you adjust the Beyer number to?

It is for these and other reasons that I think trying to develop wagering models targeted to low odds horses is a bad idea in general. With low odds horses, you have to be more accurate in your assessment than you do with longer odds horses. None of us is able to create the perfect number or the perfect odds line. In reality, all any of us can really do (even the very best of us) is create a good odds line and then bet accordingly. Along these lines I have found it far easier developing wagering models targeting winning horses 9-1 and up than horses going to post at 3-1. I'm not saying it can't be done (it can and it has.) I'm just saying that races have a certain amount of inherent unpredictability and that it's easier to take advantage of that unpredictability by targeting longer priced horses than shorter priced horses.

[GameTheory]

1962: 1 to 1.15 lost 2%.

In todays market, my data form 2003 to 2004 shows the following;

average 1-1 odds, dollar lost =15%.
Today's chalk bettors are losing 10% more according to youe post; they are much lousier at picking low-odds horses. They aren't worse than they used to be, they are better (meaning smarter about picking winners), which is why prices are down. The market is more efficient.

[GameTheory]

posted by Ken-


It is for these and other reasons that I think trying to develop wagering models targeted to low odds horses is a bad idea in general. With low odds horses, you have to be more accurate in your assessment than you do with longer odds horses. None of us is able to create the perfect number or the perfect odds line. In reality, all any of us can really do (even the very best of us) is create a good odds line and then bet accordingly. Along these lines I have found it far easier developing wagering models targeting winning horses 9-1 and up than horses going to post at 3-1. I'm not saying it can't be done (it can and it has.) I'm just saying that races have a certain amount of inherent unpredictability and that it's easier to take advantage of that unpredictability by targeting longer priced horses than shorter priced horses.Interesting. I have found exactly the opposite in my work. Even more interesting is that I do better at tracks with a high percentage of winning favorites. It probably just depends on how you build your model.

[Jeff P]

posted by GameTheory-
Interesting. I have found exactly the opposite in my work. Even more interesting is that I do better at tracks with a high percentage of winning favorites. It probably just depends on how you build your model.

I think a great deal depends on what tracks and race types you focus on and how you build your models. Given your statement and that we take opposite approaches, my take would be that we could both probably learn a great deal from each other.

About midway through 2004 I discovered something that led me to profitable selections involving lower odds horses. The model itself was a long time in coming and resulted from hundreds of hours of testing time spent in front of my own "Data Window." A recap from calendar year 2004 for this model's plays looks like this:

Dirt (All*) SPRINTS (From Index File: F:\FALL_2004\pl_JRating_1_ytd.txt)


Data Summary Win Place Show
Mutuel Totals 1358.40 1156.10 1081.00
Bet -1100.00 -1100.00 -1100.00
Gain 258.40 56.10 -19.00

Wins 187 286 365
Plays 550 550 550
PCT .3400 .5200 .6636

ROI 1.2349 1.0510 0.9827
Avg Mut 7.26 4.04 2.96


By: Odds Range

Min Max Gain Bet Roi Wins Plays Pct
-999.00 0.00 0.00 0.00 0.0000 0 0 .0000
0.00 0.49 -13.00 18.00 0.2778 2 9 .2222
0.50 0.99 -10.40 120.00 0.9133 32 60 .5333
1.00 1.49 -31.20 136.00 0.7706 24 68 .3529
1.50 1.99 15.10 166.00 1.0910 34 83 .4096
2.00 2.49 28.60 142.00 1.2014 27 71 .3803
2.50 2.99 2.90 108.00 1.0269 15 54 .2778
3.00 3.49 27.00 82.00 1.3293 13 41 .3171
3.50 3.99 17.20 44.00 1.3909 7 22 .3182
4.00 4.49 59.00 46.00 2.2826 10 23 .4348
4.50 4.99 49.40 52.00 1.9500 9 26 .3462
5.00 5.49 17.80 32.00 1.5563 4 16 .2500
5.50 5.99 -20.00 20.00 0.0000 0 10 .0000
6.00 6.49 -16.00 16.00 0.0000 0 8 .0000
6.50 6.99 13.00 18.00 1.7222 2 9 .2222
7.00 7.49 -10.00 10.00 0.0000 0 5 .0000
7.50 7.99 -4.00 4.00 0.0000 0 2 .0000
8.00 8.49 15.00 22.00 1.6818 2 11 .1818
8.50 8.99 7.40 12.00 1.6167 1 6 .1667
9.00 9999.00 110.60 52.00 3.1269 5 26 .1923


Recaps from each of the past four years look remarkably similar.

But during live play, I seldom find myself playing this model's selections unless they are going to post at 8-1 or higher. Sure, profits are available. But it's just not my style of play. There's a certain something I get from cashing a $21.00 winner that just isn't there from cashing three $7.00 winners.

Most of my live plays come from a model targeted at race types that produce higher payoffs. I developed the basics for this model several years ago and it has held up well for me since. The basic premise of the model is twofold. First, throw out races that are likely to produce chalk winners. Second, purposely look for hiorses that are "ugly" on paper that actually do have a chance to win. Last year's recap is posted below. No, I didn't make all 3200 plays. Actually, I made less than half of them but did proportionately well with those I did play.

My point in responding is that it was much easier for me to develop the model targeting longer odds horses. Much of what has been published to date tells us to avoid horses that are "ugly" on paper. What I found was that the public makes many errors when ignoring these horses. Included among those that are tossed out by the public at first glance are lots of horses that can and do outrun their odds.


Data Window Settings:
AProbOL Divisor = 999
Filters Applied: 2HLFC-
Surface: (ALL*) *ANY Distance* (From Index File: F:\FALL_2004\pl_JRating_123.txt)


Data Summary Win Place Show
Mutuel Totals 8536.80 7164.80 6324.40
Bet -6588.00 -6588.00 -6588.00
Gain 1948.80 576.80 -263.60

Wins 392 822 1235
Plays 3294 3294 3294
PCT .1190 .2495 .3749

ROI 1.2958 1.0876 0.9600
Avg Mut 21.78 8.72 5.12


By: Odds Range

Min Max Gain Bet Roi Wins Plays Pct
-999.00 0.00 0.00 0.00 0.0000 0 0 .0000
0.00 0.49 0.00 0.00 0.0000 0 0 .0000
0.50 0.99 0.00 0.00 0.0000 0 0 .0000
1.00 1.49 0.00 0.00 0.0000 0 0 .0000
1.50 1.99 0.00 0.00 0.0000 0 0 .0000
2.00 2.49 6.20 20.00 1.3100 4 10 .4000
2.50 2.99 -11.00 72.00 0.8472 8 36 .2222
3.00 3.49 -12.00 148.00 0.9189 16 74 .2162
3.50 3.99 51.20 156.00 1.3282 22 78 .2821
4.00 4.49 -53.40 262.00 0.7962 20 131 .1527
4.50 4.99 47.40 246.00 1.1927 26 123 .2114
5.00 5.49 4.20 268.00 1.0157 22 134 .1642
5.50 5.99 -22.20 290.00 0.9234 20 145 .1379
6.00 6.49 -0.20 288.00 0.9993 20 144 .1389
6.50 6.99 12.20 278.00 1.0439 19 139 .1367
7.00 7.49 38.00 342.00 1.1111 23 171 .1345
7.50 7.99 -64.80 230.00 0.7183 10 115 .0870
8.00 8.49 -116.00 264.00 0.5606 8 132 .0606
8.50 8.99 13.20 268.00 1.0493 15 134 .1119
9.00 9999.00 2056.00 3456.00 1.5949 160 1728 .0926

[GameTheory]

Your first box shows you doing very well indeed around the 3-1 area, so you are doing ok in what I would consider the lower odds ranges. Before we get into these results, let's explore oddslines a bit (by which I really mean assigned probabilities).

Do you assign a probability to each horse or just rank them?

If the former, do this experiment for me (if possible) and let me know the results. First, remove races with betting entries so that odds only apply to one horse (and other oddball races, like if a horse is declared a non-betting entry but still raced).

Now, for each race, let the variable N = 1 / field size. In other words, N is the "natural" chance of winning for each horse given no other information. So the N is the same for every horse in the race.

And for each horse, let P be your assigned probability (your P's in each race should sum to 1.0) and O be the public's assigned probability (normalized so that the O's in each race also sum to 1.0).

Throw out all horses where P <= O. We only want to look at horses where we think the horse has a better chance of winning than the public because this is the group where overlays come from. Because we normalized O, these may or may not be true overlays betting-wise, but we think the public is underestimating their chances in any case.

Of the remaining horses in the sample (all where P > O), break them up into three groups:

P > N and O > N

P > N and O <= N

P <= N and O <= N

Now, for each group calculate total runners, expected winners based on P (the sum of P), expected winners based on O, and actual winners based on what actually happened. And let's see those results for as big a sample as you can muster.

Anyone else who makes full oddslines (assigns a specific probability to every single horse) is welcome to duplicate the experiment. We'll discuss the implications of the results (if any) when we get them...

[formula_2002]

They aren't worse than they used to be, they are better (meaning smarter about picking winners), which is why prices are down. The market is more efficient.


I don't want to miss an opportunity where we agree!

[GameTheory]

I don't want to miss an opportunity where we agree!I agree with many of your premises. It is your conclusions where you take flying leaps that "do not follow" that I have trouble with.

Did you examine the first box that Jeff posted? He's got a 29% hit rate in the 2.5 - 3.5 odds range....

[Jeff P]

GameTheory,

I do indeed create an Assigned Probability for every horse. From this, I derive an Odds Line. And the sum of the assigned probabilities for all horses in each race adds up to one.

I'll play along.

It may take me a day or two to find time to segment a sample into the categories you suggested. I'll post it when I have it done.

I'm curious to see where this goes.

[GameTheory]

GameTheory,

I do indeed create an Assigned Probability for every horse. From this, I derive an Odds Line. And the sum of the assigned probabilities for all horses in each race adds up to one.

I'll play along.

It may take me a day or two to find time to segment a sample into the categories you suggested. I'll post it when I have it done.

I'm curious to see where this goes.It may not go anywhere, but I'm sure we'll learn some things in any case. I have some observations that may apply just to me or they may be more generally applicable. I'd like to see some results before I do too much theorizing...

[formula_2002]

I agree with many of your premises. It is your conclusions where you take flying leaps that "do not follow" that I have trouble with.

Did you examine the first box that Jeff posted? He's got a 29% hit rate in the 2.5 - 3.5 odds range....

A few questions
1. are these results back fitted ?
2. what was win% of all horses (system and non-system) in the odds ranges 2.5 - 2.99 and 3- 3.49 ?

[formula_2002]

If you were to go to my web page, I have the results of my study of over 100,000 horses.

I developed a system that produced a 24% profit in 1584 plays in the 7-20 odds range.

That was the back fitted data..


Then I tested it against another segment of my data base.
It produced a 28% profit in 551 plays.

Then I tested it live. It cost me $1000 for the data (ALL-Ways) and in over 10000 horses I had 90 plays and lost 6 1/2 cents of the dollar.

I gave up my Bris Gold membership account.

However, I just noticed I would have made a 50% profit if I had played the place.

Given that information, a few years ago I would have re-instated my Bris account and probably would have lost 50 cents on the dollar in the next 90 place bets.

See I'm a $1000 ahead already :)

[GameTheory]

Of the remaining horses in the sample (all where P > O), break them up into three groups:

P > N and O > N

P > N and O <= N

P <= N and O <= N
Actually, while we're at it let's go ahead and tabulate results for the other groups too (the underlay groups). They will be useful for comparison:

So we've got six altogether:

P > O and P > N and O > N

P > O and P > N and O <= N

P > O and P <= N and O <= N

P < O and P > N and O > N

P < O and P <= N and O > N

P < O and P <= N and O <= N


Hopefully this won't be a big waste of time...

[sjk]

GT,

Here are some results along the lines of what you suggested. I have added a third group where p>1.5o which might have a greater bearing on actual play.

group a b count exp p exp o no wins
p>o p>n o>n 16143 5368 3449 4291
p>o p>n o<=n 12956 2289 1105 1477
p>o p<=n o<=n 13427 1183 674 853
p<o p>n o>n 31023 7027 8013 8070
p<o p<=n o>n 11694 1096 2091 1715
p<o p<=n o<=n 16267 1307 1534 1444
p>1.5o p>n o>n 3042 1149 531 781
p>1.5o p>n o<=n 7720 1475 571 841
p>1.5o p<=n o<=n 5619 487 197 288

What conclusions do you draw from this?

[sjk]

Game Theory and Jeff P,

On the chance that you have not lost interest in reading my ill-formatted results I have had the computer running nearly non-stop since yesterday to add to the dataset posted above. It now represents the past year of dirt racing (less a few odd distances I don't track and races with numerous firsters and/or layoff horses).

Unfortunately I don't have any better way to create lines that to go one day at a time so it takes the computer quite some time to do so (and for me to test any proposed programs modifications). I have done quite a bit of this sort of thing the past month (as I seem to do each winter) and have made several changes that I think will be a benefit going forward. I have brought a second computer into the mix to keep from bogging me down on betting races etc.

I feel pretty good about these results and am interested to see how they measure up to your lines and others (on the off chance anyone else is still reading).

group a b count exp p exp o no wins
p>o p>n o>n 27413 9196 5902 7376
p>o p>n o<=n 22037 3919 1897 2545
p>o p<=n o<=n 22801 2022 1156 1452
p<o p>n o>n 51720 11814 13501 13529
p<o p<=n o>n 19727 1857 3549 2875
p<o p<=n o<=n 27543 2219 2608 2476
p>1.5o p>n o>n 5230 1990 919 1334
p>1.5o p>n o<=n 13048 2507 971 1458
p>1.5o p<=n o<=n 9522 829 338 480

Maybe I finally got the formatiing right!

[formula_2002]

GT,

Here are some results along the lines of what you suggested. I have added a third group where p>1.5o which might have a greater bearing on actual play.



sjk, if you say p>1.50 do you really mean "odds 1.50-1",
P must be <=1 ?

[sjk]

Formula,

The groups in those rows are where the p probability is at least 1 1/2 times the o probability. The probabilities themselves are spread over the full range of possibilities with only the restrictions shown in the column headings (and also that I have excluded those with p<.05 since I don't play those horses on top).

If that is not clear you may want to go back to Gametheory's post where he defines p, o, and n.

[formula_2002]

Formula,

The groups in those rows are where the p probability is at least 1 1/2 times the o probability. The probabilities themselves are spread over the full range of possibilities with only the restrictions shown in the column headings (and also that I have excluded those with p<.05 since I don't play those horses on top).

If that is not clear you may want to go back to Gametheory's post where he defines p, o, and n.


I got it.. well I tabulated your results in a simple manner, looking for those conditions where actual winners exceeded
expected winners, as you can easily do your self.

Where aw/ex>=1 is a good thing. Since the odds are normalized I would think >=1.20 is a profitable thing. That happend in 6 of the 16 cases. Try it you'll like it.
Given your information I'm curious how GT see's this.

[formula_2002]

SJK,
In the two cases where P>(1.5 x 0) you appear to have a large profit.

The question in my mind is "where are all the profits coming from"? Is there still a profit in the range,
p>(1.5 x 0) and p<(2 x o)?

Say, Profit=aw/ew >=1.2.

I also think that a incremental odds line analysis would help clarify the value of the results.

[formula_2002]

further, you may want to isolate your top ranked odds line.
In my experience, I have found it to be the most efficient..

[sjk]

Formula,

In practice I play horses where p>1.8* (unnormalized o) which probably relates to something like p>2*o where o is normalized as in the above table.

I would show a modest profit in the 1.5 to 2 range but I sacrifice some total dollars won to improve the win percent because it is more fun if you can cut out some of the barely better than breakeven play.

[Jeff P]

GT, SJK,

I am still reading and very much interested. Haven't had the free time yet to do the same with my own data but will as soon as time allows.

J

[formula_2002]

sjk, did gt get back to you on the
"P > O and P > N and O > N

P > O and P > N and O <= N...." analysis?

Thanks

Joe M

[sjk]

No. I was kind of hoping for some feedback but I guess no one thought it was worth the time to pursue this thread.

[Foolish Pleasure]

ROTFLMAO GT I would retire 1000 times over if there were more of you around.


ANy more insight into what already happened?

And right will always put my money where my mouth is any time you wanna go mano v mano for 10 dimes that you cannot beat post time favorites over a four digit sample size-just tell me when you are ready. Beat as in bet on and win.

[Foolish Pleasure]

BTW the avg field size shrunk dramatically=less error=not efficiency created by external forces, yea I know is still ME but damn if it woulda taken a miner yrs to figure out.

[formula_2002]

BTW the avg field size shrunk dramatically=less error=not efficiency created by external forces, yea I know is still ME but damn if it woulda taken a miner yrs to figure out.

Well, at an average fav odds of 1.66-1 and a 1% profit or loss, it would only take 450 bets to prove the point (2.5 standard units)

If someone said the could make a 6% profit. it would take even fewer plays. About 200...so lets get it on!!

[formula_2002]

BTW the avg field size shrunk dramatically=less error=not efficiency created by external forces, yea I know is still ME but damn if it woulda taken a miner yrs to figure out.

Actually that may be a good point. Not certain it's the entire answer.. What was the average field size at that time? Regardless, you would still have to say today's races are more efficient.

[JPinMaryland]

FIeld size would seem to be an important factor in all that.

ALong the same lines, what about this? Isnt it possible that in the old days there were more races w/ single favored horse and nowadays there are more 3 horse horse races.

E.g.: Race circa 1955: Sightseek 1-10; Strom 30-1, Flag 60-1, Flying 100-1.

Race circa 2000: Nick's Hot young thing 2-1; Bobby's Favorite: 5-2, Dewaynes Baby: 4-1....

Wouldnt you have to correct for this sort of thing in the two data sets before any conclusions can be drawn?

[Jeff P]

Foolish Pleasure's statement about taking on race favorites, IMHO, holds water.

Results for a set of selection criteria aimed at short priced horses that I posted earlier turned a nice profit when applied to all starters in 2004. Instead of applying it to all starters for 2004, here's what happens when it's applied only to post time favorites.

Nothing new for me here. I've always been a proponent of doing one of two things: 1. Finding and betting value or 2. Sitting on my hands.

Dirt (All*) SPRINTS (From Index File: F:\FALL_2004\pl_Favorites_1_ytd.txt)


Data Summary Win Place Show
Mutuel Totals 407.70 451.50 482.90
Bet -478.00 -478.00 -478.00
Gain -70.30 -26.50 4.90

Wins 91 146 189
Plays 239 239 239
PCT .3808 .6109 .7908

ROI 0.8529 0.9446 1.0103
Avg Mut 4.48 3.09 2.56

[formula_2002]

FIeld size would seem to be an important factor in all that.

ALong the same lines, what about this? Isnt it possible that in the old days there were more races w/ single favored horse and nowadays there are more 3 horse horse races.

E.g.: Race circa 1955: Sightseek 1-10; Strom 30-1, Flag 60-1, Flying 100-1.

Race circa 2000: Nick's Hot young thing 2-1; Bobby's Favorite: 5-2, Dewaynes Baby: 4-1....

Wouldnt you have to correct for this sort of thing in the two data sets before any conclusions can be drawn?

It's not so much the "cause", which in of it self is an interesting subject, rather the "effect".. All these "causes" seem to make the current betting market more efficient then yesteryear and there by decreasing the possibility for profit.

[formula_2002]

Foolish Pleasure's statement about taking on race favorites, IMHO, holds water.

Results for a set of selection criteria aimed at short priced horses that I posted earlier turned a nice profit when applied to all starters in 2004. Instead of applying it to all starters for 2004, here's what happens when it's applied only to post time favorites.

Nothing new for me here. I've always been a proponent of doing one of two things: 1. Finding and betting value or 2. Sitting on my hands.

Dirt (All*) SPRINTS (From Index File: F:\FALL_2004\pl_Favorites_1_ytd.txt)


Data Summary Win Place Show
Mutuel Totals 407.70 451.50 482.90
Bet -478.00 -478.00 -478.00
Gain -70.30 -26.50 4.90

Wins 91 146 189
Plays 239 239 239
PCT .3808 .6109 .7908

ROI 0.8529 0.9446 1.0103
Avg Mut 4.48 3.09 2.56



Jeff, are all figures you have been posting backfitted?
Or are they groups that you have tested against a model?
Are they live, go with the real money, stuff?

Also, and very important, how do all these figures compare to the public's results in the same events.

[Jeff P]

Backfitted?

Yes and no. I divide my data into two sets. 2/3rds Development data and 1/3rd Validation data. When I develop a model I use Development data only. Here, I do tweak things to try and get a good fit. So yes, when working with Development data during the creation of a model I may backfit a little. Once I have what I feel is a good working model, I then confront it with races it hasn't seen by testing it against Validation data. If the model performs well here, if it "validates," then I move on to live play. If the model performs poorly when confronted with fresh races, I don't try to re-fit. Instead I simply scrap it as a bad idea.

The results of the set of plays I posted earlier showing a profit is something I discovered in mid 2004. After validating the model using fresh data I moved (very timidly) into the area of live play. Like I said before, my bread and butter are horses paying 9-1 and up so I'm a little out of my element laying out money on 3-1 and 5-2 shots. But, like I said, the model has held up well since its discovery so I'm playing it, perhaps with a little more caution than I should.

About half the plays from the original set would fall into the live play category. As would everything for 2005. So far in 2005, here are the results from the same model:

Dirt (All*) SPRINTS (From Index File: F:\2005\Q1_2005\pl_JRating_1.txt)


Data Summary Win Place Show
Mutuel Totals 234.40 212.60 203.20
Bet -212.00 -212.00 -212.00
Gain 22.40 0.60 -8.80

Wins 35 52 63
Plays 106 106 106
PCT .3302 .4906 .5943

ROI 1.1057 1.0028 0.9585
Avg Mut 6.70 4.09 3.23


By: 21 TheOdds

Min Max Gain Bet Roi Wins Plays Pct
0.00 0.00 0.00 0.00 0.0000 0 0 .0000
0.00 0.50 1.60 4.00 1.4000 2 2 1.0000
0.50 1.00 -7.40 18.00 0.5889 3 9 .3333
1.00 1.50 3.40 24.00 1.1417 6 12 .5000
1.50 2.00 6.40 36.00 1.1778 8 18 .4444
2.00 2.50 -6.80 20.00 0.6600 2 10 .2000
2.50 3.00 -3.20 26.00 0.8769 3 13 .2308
3.00 3.50 11.20 22.00 1.5091 4 11 .3636
3.50 4.00 -5.00 14.00 0.6429 1 7 .1429
4.00 4.50 11.00 10.00 2.1000 2 5 .4000
4.50 5.00 27.60 6.00 5.6000 3 3 1.0000
5.00 5.50 0.00 0.00 0.0000 0 0 .0000
5.50 6.00 -4.00 4.00 0.0000 0 2 .0000
6.00 6.50 0.00 0.00 0.0000 0 0 .0000
6.50 7.00 9.60 6.00 2.6000 1 3 .3333
7.00 7.50 -6.00 6.00 0.0000 0 3 .0000
7.50 8.00 -2.00 2.00 0.0000 0 1 .0000
8.00 8.50 0.00 0.00 0.0000 0 0 .0000
8.50 9.00 0.00 0.00 0.0000 0 0 .0000
9.00 9999.00 -14.00 14.00 0.0000 0 7 .0000

[GameTheory]

ROTFLMAO GT I would retire 1000 times over if there were more of you around.Why? What'd I do?

ANy more insight into what already happened?

And right will always put my money where my mouth is any time you wanna go mano v mano for 10 dimes that you cannot beat post time favorites over a four digit sample size-just tell me when you are ready. Beat as in bet on and win.I don't know what you are talking about. Is this addressed to me? Who said anything about "beating post time favorites"?

BTW the avg field size shrunk dramatically=less error=not efficiency created by external forces, yea I know is still ME but damn if it woulda taken a miner yrs to figure out.True, and obvious. Although I think that WOULD be an external force, the internal force being the skill of the public. The skill increasing, or the difficulty of the problem getting easier, same result. Another reason for changes in efficiency over time is the size of the pool -- handle is increasing most places. Still, I've seen that some tracks are getting less efficient...

[GameTheory]

No. I was kind of hoping for some feedback but I guess no one thought it was worth the time to pursue this thread.I will get back to that. Been very busy and was hoping to hear something from Jeff.

[sjk]

Look forward to hearing from you and Jeff P.

[formula_2002]

FIeld size would seem to be an important factor in all that.



I just checked my data base for fields <10 and fields>=10
I find no substancial difference in the average odds ranges of 1,2,3, >=10 and <20 between the two sets.

But oddly, in the field<10 for odds>20, there was a loss of 43% and in the fields>10 the dollar loss was but 28%.

all the above was tested for over 83,000 horses.

Perhaps another data base user can run off a set of numbers corresponding to the above odds.

[sjk]

Here are returns for all horses >20-1 based on field size:

field size return
5 -0.58
6 -0.43
7 -0.44
8 -0.38
9 -0.34
10 -0.33
11 -0.32
12 -0.33
13 -0.51
14 -0.17

[formula_2002]

Here are returns for all horses >20-1 based on field size:

field size return
5 -0.58
6 -0.43
7 -0.44
8 -0.38
9 -0.34
10 -0.33
11 -0.32
12 -0.33
13 -0.51
14 -0.17



Thanks SJK.
It seems to confirm what my data showed.

So it would “appear” when the field size increases so does the efficiency.

I’m often amazed what empirical data can do to popular thought..

However;
“The only thing we have to fear is…” the interpretation on the data.
(That’s probably what FDR meant anyway)

[osophy_junkie]

In a paper Benter published he described a way to integrate the public's odds and prior correctness into an odds line. Is what your doing along these lines?

Ed

[formula_2002]

In a paper Benter published he described a way to integrate the public's odds and prior correctness into an odds line. Is what your doing along these lines?

Ed
I thought that's what GT had in mind to do,
however, Benter used incremental odds analysis.

We all await GT's analysis