selection service performance statistics..

[formula_2002]

Can someone tell me what is the value in these selections results?

Did I miss something?

"Michael E. Nunamaker's selection service performance statistics"

http://www.minnow.com/racing/pstat1.html

[dav4463]

Bet only Mdn 3YU Claiming turf routes !

[Jaguar]

Formula, you undoubtedly are aware- along with the rest of us- that these public guys are no good- no better than the DRF Consensus.

2 reasons for this:

1. These methods are not optimized, nor are they track-specific. By that I mean: They don't measure the relative strength- or impact value- of various data as it pertains to a particular track, or a particular type of race.

So, these guys take values from Davis or Nunamaker and use it as a universal template- applied equally to all tracks and all races.

2. Today the trainer/jockey nexus is so crucial that the corny methods used by the touts- which are based upon pace and a few generalized algorithms(few or none of which measure the relative trainer/jockey incidence at that particular track, for that particular race)- will always lead the bettor astray, except in the case of stickouts- and the bettor can't make money playing horses at 6-5.

Moreover, the good A.I. programs have superb trainer/jockey algorithms- and these programs are the most powerful handicapping methods available. All-Ways and TB5 are examples of this.

All the best,

Jaguar

[GR1@HTR]

Heck, I give him credit for posting his results. At least he has the guts for posting his loses...

[hurrikane]

jag, I could be wrong but I believe all those points you made pertain to formula too.

Formula, have'nt looked at your page in a while. Don't remember, do you list your results in a detailed format?

[formula_2002]

hurrikane
You are correct, I list my system results in several reports that can be viewed on my web page.
They are reports sorted by ;
odds
track
race type
and several others;

My e-mailed picks are almost always follow up the next day with
horses by horse results.

Some days I use my "good day at the track" avatar and some days I must use my "bad day at the track" avatar.

The major difference in my web page and most, if not all, with the exception of Numamaker's "selection service" web page,
is the detail reporting of results (which to date I am fortunate to report the results are profitable) and an effort to demonstrate the basis for the selections (picks). In my case the data reviewed is over 120,000 horses.

Numamakers is the most forth right reporting , with the exception of mine, that I have seen anywhere.

[formula_2002]

here is a site with plenty of picks...but I can't find any results..

http://www.trackmaster.com/cgi-bin/axtmtls.cgi?tmf

[Rick]

formula,

I find those statistics to very useful in deciding what the upper limit on win % would probably be with a good selection method. He used some of the most advanced methods of statistical analysis available and only came up with about 29% (at all tracks). It's possible that nobody could do any better under those circumstances. But, most of us don't play every track. I play 3 tracks over the course of a year. If I played every race at those 3 tracks I'd approximately equal his win % but beat his ROI by a lot. He may very well be over-analyzing the data too, but it's hard to tell since he's taking on the nearly impossible task of playing every track.

[formula_2002]

Rick.. I have not done a thorough review of his results.

But the results appear to be attainable without any handicapping.
That is his $ loss, overall, appears to equal the track take-out+-.

The gauge of long range upper limit win % is, 1/(final track odds+1)..

Joe M

[Aussieplayer]

Joe,

Not sure I understand your little forumula:

"The gauge of long range upper limit win % is, 1/(final track odds+1)"

.....To me, this says that (you're) saying that breakeven is the best we can do??

eg. 1/5 = 0.20

20% on 4/1 shots sees you breaking even.....?

Cheers
AP

[formula_2002]

AP

Playing most any one horse, with out handicapping will cost about 17 cent on the dollar (or whatever the track take out is).

Favorites will cost you less, long shots cost you more.

I'm re-stating something that has been calculate many times . The % win probability is determined by the final odds (adjusted for take-out).

Our job is, 1st..beat the track...2nd beat each other.
Most pople are stuck in item 1!!!

If you want to know the upper and lower limits on a bet...just look at the final odds and add and subtract a few % points.

Joe M
(losing day avatar showing)

[Rick]

formula,

If I remember correctly, I think Nunamaker's overall results show about a 9% loss, which is certainly significantly better than the track takeout but not very good as far as selection systems go. My own results for playing every race are about 29% wins and 1% profit. But I was able to win money even when I was playing a method that showed a 5% loss on every race. I think you could even win at the -9% level but you would have very few (maybe 1 per track per day) plays that were overlays.

[formula_2002]

Rick, you have my attention

"But I was able to win money even when I was playing a method that showed a 5% loss on every race."

How do you do that?

Joe M

[cj]

Just a guess...he didn't bet every race?

CJ

[MichaelNunamaker]

Those performance data are from
1996. Things have improved since then. Much more up to date
performance statistics are included next to every race in my
selection sheets. I suppose that's why I never updated the web
stats. I figured no one was looking at them. I clearly was
wrong<G>. I should either update the stats on the web or delete
them.

In any case, to address performance, the data you were looking at
was the data on the top horse in all races at all tracks. If you
restrict wagering to overlays, it is easy to show a profit. For
example, currently, on my entire database my top pick, when played in all
races and all tracks from 1996 to the present time has 431,508 plays,
31.578% top pick winners and a ROI of -9%. This is roughly half of the
track take. If you only wagered on my top pick when it was going off at
4-1 or greater odds, then there would have been 60511 plays, 15.46%
winners, and a ROI of +3%. There are other games one can play to boost
ROI. For example, only bet on top picks going off at 10-1 or more gets an
ROI of +17% but at the price of only having 4911 plays in seven years!

Now, to return to your original question, "What is the value in
these selections results?"

They do different things for different people. The most common
thing by far is to simply to lose less at the track. About 99.9% of the
people who bet lose money. A 9% loss when betting every race cuts the
track take in half. This allows loads of action while cutting one's
losses in half. This is a great help for the 99.9% of the population.

I also have a few patient customers who only bet on overlays and
claim to show a profit (a couple of guys say they make their living
betting my picks).

[Exactaman]

Originally posted by formula_2002
here is a site with plenty of picks...but I can't find any results..

http://www.trackmaster.com/cgi-bin/axtmtls.cgi?tmf


That's the page of yesterday's picks, they leave them up for a day for free download.

[Rick]

Mike,

It was my understanding that you used different combinations of factors depending on the type of race but the same across all tracks for the same type of race. Is that still the case? One would think that pace related factors, at least, would be vastly different at some tracks.


formula,

I was referring to playing only overlays with a method that overall lost 5% if played in every race. I only got 2 or 3 plays per day typically at a given track. And, I only played a few tracks that had the best results. My current method has much better results, but I usually play only one track. I think track selection is a big factor that is frequently overlooked.

[MichaelNunamaker]

Hi Rick,

You wrote "It was my understanding that you used different combinations of factors depending on the type of race but the same across all tracks for the same type of race. Is that still the case?"

I use the same general equation for a given type of race, but there are track specific inputs for that equation. For example, I have adjustments for post position at every track, surface, and distance combination. Likewise for several other track specific things.

However, none of them are pace related. That's a great idea. I think I'll add some things like that.

[Rick]

Mike,

Based on your vast amount of research, what's your opinion about what the most important handicapping factors are? And, what's your opinion of the claims made by the pace handicapping gurus of 60% or more wins playing two horses in every race? Also, have you found that the weighting of various factors is relatively constant over time or are some becoming relatively less or more important? Finally, have you looked into seasonal factors or weather related factors?

[MichaelNunamaker]

Hi Rick,

You wrote "Based on your vast amount of research, what's your opinion about what the most important handicapping factors are?"

That one is easy. I'll probably get flamed for this, but here is goes. In my opinion, the most important handicapping factors are speed, speed, and more speed. Everything else is frosting on the cake of speed.

Another way to say this, you could outperform a serious percentage of public selectors simply by taking the average of the last three races speed rating and picking the horse with the highest average. There is no other single factor rating that I'm aware of that can accomplish that feat.

You also wrote "And, what's your opinion of the claims made by the pace handicapping gurus of 60% or more wins playing two horses in every race?"

Well, I've never found _anything_ that can win 60% of the races when playing two horses in every single race. That's not to say it is not possible. But it is a task that is beyond me.

Now, if you let me pick and choose my races... Heck, I can give you 60% winners on my top pick alone. My top pick when predicted to win by 6.0 to 6.99 lengths wins 67.4% of the time with a ROI of +5%

You also wrote "Also, have you found that the weighting of various factors is relatively constant over time or are some becoming relatively less or more important?"

Weightings have remained _shockingly_ constant. The way people bet have changed dramatically, but the horses don't appear to notice any of this<G>. In other words, the IVs and win percentage in Modern Impact Values haven't change much. The NETs have changed greatly.

You also wrote "Finally, have you looked into seasonal factors or weather related factors?"

I did many years ago. The changes for off tracks were smaller than I ever would have imagined. Prediction equations for off tracks also performed worse than my standard equations. So I stopped looking at surface condition. I've got about 50 times as much data now as I did the last time I looked. Perhaps I should revisit the issue.

[Rick]

Mike,

I'm not at all surprised that speed would be most important in selecting winners. However, $ Net must be pretty poor these days though. What are the biggest changes that have occurred in $ Net over the years? Thanks for answering all of my questions. You're the only one who has the knowledge as far as I know.

[trying2win]

Mike,

In your book Modern Impact Values, I thought those were interesting NET dollar figures for tests you did, for some of those William Quirin running line angles he wrote about in some of his old books. I'm assuming the declines were mainly caused by formerly profitable angles becoming public knowledge, and thus the prices on the angles were pounded down accordingly. What do you think?

It was good to see some stats in MIV, that hold some promise for possible profitable angles (sounds like an alliteration here!) Anyways, I'd have to check them out through some more research.

T2W

[MichaelNunamaker]

Hi Rick,

You wrote "However, $ Net must be pretty poor these days though."

You are most certainly correct.

You also wrote "What are the biggest changes that have occurred in $ Net over the years?"

The pools are more efficient in general. For example, if you took my top two public pick horses and purchased a cold exacta on every race in North America your profit/loss by year would be:

1996 $5414 profit
1997 $3180 profit
1998 $5048 profit
1999 $1672 profit
2000 $3122 loss
2001 $706 profit
2002 $4630 loss
2003 $12104 loss

I'd bet you can notice a trend<G>. It is like this on virtually everything I track. The pools are so much more efficient now.

For example, I was looking at consecutive increases and decreases in speed rating last week. If a horse has had five consecutive increases in SR, its IV is 1.29 and the NET is $1.65. If a horse has had five consecutive _decreases_ in SR, its IV is 0.78 but the NET is $1.66! Situations like this were almost unheard of in the early 1990s. They are the norm now.

Hi trying2win,

You wrote "I'm assuming the declines were mainly caused by formerly profitable angles becoming public knowledge, and thus the prices on the angles were pounded down accordingly. What do you think? "

I think either Quirin's positive results were caused by a small number of longshot winners in a small sample or exactly what you said, it became public knowledge and people look for it now.

[GameTheory]

I recently did my own study of the efficiency of the public odds in the win pool for the past 5 years (1999-2003). I will post my full results and methodology used sometime in the near future, but I want to expand the study (I only looked at a subset of tracks) and track some other variables before I report my findings.

Anyway, what I found was that yes, the pools are becoming more efficient. But the interesting thing was that no clear trend was indicated at the big circuits -- SoCal, NY, & FL (CRC & GP in my study). (Remember, I looked at win pool odds only.) All of the increased efficiency seemed to be coming from small tracks like MNR & the Texas circuit. (My results showed that in 2003 the tracks Lone Star, Retama, & Sam Houston taken together were the most efficient of all that I looked at over the whole five-year period, although most of the small tracks still lagged far behind the big circuits even though they were catching up.)

One thing I left out which I want to look at is how the pool sizes have changed over time -- it is my guess that as the pools at these smaller tracks have grown, their efficiency has increased; while at places like Santa Anita & Aqueduct where the pools were already big no clear increase in public efficiency could be found.

My study looked at the public odds/probabilities only as related to a horse's winning chances; and equalized for field size so that the public betting 1/5 of the pool in a five horse race was equivalent to the public betting 1/8 of the pool in an 8 horse race. As I say I'll discuss the full methodology later.

But what you (Mike) said about the pools becoming more efficient based on your "cold exacta" profit/loss is not quite measuring overall public efficiency, is it? I say that because your picks are based on certain X factors that you look at; so it could just be that the public is betting down those factors now and they aren't necessarily getting more efficient OVERALL (although they probably are, but not as drastically as your profit/loss figures indicate). Actually, with the exacta pools it cannot be precisely measured unless you have access to the amounts bet on each combination rather than just the ultimate payout (as I'll explain later). The other factor that must be taken into consideration is field size -- the public "automatically" becomes more efficient in terms of average payout as field size shrinks. In my study, I assigned a zero score to what would be achieved by random guessing. A positive score was positive predictiveness and a negative score was non-random but negative predictiveness. I wanted to know, "Is the public REALLY getting smarter?" so you have to look at their performance relative to random performance. I will show all the scores for various tracks when I do the full study.

[Rick]

Time has definitely changed things. In the 80's I was able to win consistently using a method just based on early speed, recency, finish last race, and class change from last race. You certainly couldn't do that now. Everyone has access to power ratings (like BRIS Prime Power for example), not just speed ratings, and anything obviously good in recent races seems to get noticed and bet down accordingly. As to small tracks being less efficient, there may be some that have unusual pace characteristics that are less efficient because the public bets more like what's true at the average track. Just a theory, can't prove it.

[MichaelNunamaker]

Hi GameTheory,

You wrote "But what you (Mike) said about the pools becoming more efficient based on your "cold exacta" profit/loss is not quite measuring overall public efficiency, is it?"

Certainly it is true that my cold exacta results do not prove the pools are more efficient. It may simply demonstrate that many more people buy my picks (which I know for a fact is true!) However, I believe I did not say I believed the pools were more efficient because of my exacta results. To reiterate, I believe the pools are more efficient because I see this trend in almost everything I look at. Before 2000, it was trivially easy to identify betting factors that were profitable if the horse was going off at 10-1 or more. Would you believe as late as 1999 you could consistently make money by simply betting the top SR horse if it was going off at 10-1 or higher? It is incredible to me that angle worked that recently. But those days are over. I had a list of about 30 factors that would consistently produce profits if wagers were restricted to horses at 10-1 or higher. _NONE_ of these work anymore. And they all stopped working around 2000-2001. It was like someone threw a switch.

You wrote " I wanted to know, "Is the public REALLY getting smarter?" so you have to look at their performance relative to random performance"

Can't one simply evaluate how the public did against the public's own estimate? In other words, if the public sent the horse off at 10-1, how accurate was that?

[Tom]

Welcome to the board, Mike.
I am a customer of yours from way back and a very satisfied one.
I made countless small methods from your stats and used them to make odds lines like you showed in the handout that came with your book.
You'll find some sharp guys here, and some great conversations.

[MichaelNunamaker]

Hi Tom,

I glad you are happy with my stuff!

Hi again GameTheory,

I wanted to add one thing to my last post. I've not approached the subject of the efficiency of the pools with any mathematical rigor. Therefore, I want to pre-concede that, whatever your conclusions, they almost certainly have a greater probability of being correct than my impressions.

[Niko]

I spoke a few times with someone that claimed to be very succesful by making his own speed figures and he re-iterated what Mike had said. Speed is the single most important factor and his edge came from his adjustments that were based on post-position and the horses trip. He then factored in form cycles to make his picks. I haven't been able to figure out how to best adjust speed based on post (haven't spent much time on it) yet but with Mike's confirmation of speed and post/surface adjustments I'll have to work on it.

Rick,
You asked Mike what was most important and he said Speed. You seem to be very succesful (I don't know anyone else who can show a profit or break even on all races over time at even one track or show up to 20% profit being selective which truly impresses me) so I'd like to know your opinion on what's most important. If speed is most important can you tell us what you feel is second most important?
I'd have to say things such as trainer changes, jockey changes or form changes but I can't pcik one over the other. Now I might have to add post-position changes.

[GameTheory]

Originally posted by MichaelNunamaker
Can't one simply evaluate how the public did against the public's own estimate? In other words, if the public sent the horse off at 10-1, how accurate was that?

Yeah, but even though 10-1 will always represent the same mathematical probability (assuming equal takeout rates), it represents a different assesment of the strength of this horse in relation to the field in a five horse race than it would in a twelve horse race. See what I mean?

If we want to determine how predictive the public is, we've got to give them more credit for giving the eventual winner a higher probability (lower odds) than if they gave that winner a lower probability (higher odds). But the average of all the public probabilites for a given race is always 1/field size, and that average is what you'd achieve just by random guessing. So you can't just look at something like win percentage without considering field size. If you have an average field size of 10 and you guess randomly, you're going to have a win% of 1/10 = 10%. But if you have an average field size of only 5 and you guess randomly, you're going to have a win% of 1/5 = 20%. But you couldn't say that you're being "smarter" (or more efficient) in the second case even though you're winning twice as often -- you're just picking randomly in both cases. So any "predictive quotient" must measure performance relative to this moving benchmark, see?

What I'm talking about will all become clear in a long, boring post coming soon to this message board...

[MichaelNunamaker]

Hi GameTheory,

You wrote "What I'm talking about will all become clear in a long, boring post coming soon to this message board..."

Your "long boring posts" are the ones I like the best! I am very much looking forward to it.

You wrote " If you have an average field size of 10 and you guess randomly, you're going to have a win% of 1/10 = 10%. But if you have an average field size of only 5 and you guess randomly, you're going to have a win% of 1/5 = 20%."

Certainly. And I agree that 10-1 is most certainly a different relative assement varying with respect to field size. In a five horse field, the public is saying the horse is half as likely as a random horse to win. In a twenty horse field, the public is saying the horse is twice as likely to win as a random horse.

The more I think about this, the more I am unsure about how this should really be treated. Perhaps different field sizes should be treated disjointly. For example, say the world is getting much better at field sizes of 12 and lower but is totally screwed up on larger field sizes because the large fields are infrequent. The only I am know of that would allow us to see this is if we measure efficiency of large and small fields independently of each other.

It seems to me that any measure that would bring field sizes together would obscure or lose that kind of information.

Man oh man, this is getting way more complex than I thought it would. That's pretty much the story of my involvement with horse racing<G>.

[GameTheory]

When breaking it down by field size, I found that there is a significant increase in public predictiveness with fields of 7 horses or less (I think it was). But I didn't find much difference in the 5-7 horse range, or in the 8-12 horse range -- only between those two ranges. We use the term "efficiency" a lot, but that is actually the wrong term for what I'm talking about because the oft-talked about "market efficiency" is something slightly different, dealing less with the overall "smartness" of the public and more with their biases. For instance, if the public bet the fraction 1/field size on every horse in every race, that would be a perfectly "efficient market" because the returns on each odds range (or ranking, even though they're all ties) would be 1-takeout rate. This equalizing of returns across all horses as ranked by the public is what the econometric analysis done in a number of published studies is all about. (Cratos might be better able to comment on that than me -- I might be talking out of my ***.)

[Rick]

Niko,

I didn't say that speed was most important for winning money, only for getting a high win %. You could probably use speed only to win money, but it wouldn't be anything obvious like last race speed or average of last three. Pace and class are also useful factors, but again you can't use anything too obvious. And when you combine factors, you must try to make them as independent of one another as possible. However, I do agree that you can't do much better in win % than what you get by just using some simple speed method. Public handicappers who don't have their ROI monitored would pobably be well advised to use a lot of speed in making their predictions.

[Rick]

GT,

I don't know if this is related, but my ROI has always been better in fields of 8 or more. Of course, the larger the field the more chance there is of finding a really good overlay.

[Rick]

Niko,

I should point out that I use two entirely different methods. One of them is a rating than gets both a higher win % and ROI. The other is based on jockey and trainer factors and gets a lower win % and lower ROI. That's how I'm able to get plays in most races. If one method doesn't find an overlay, the other usually will. I throw out a lot of high win % underlays and when there is an alternate play it's usually at much longer odds. The rating method plays have a win % of 25-26% and the jockey/trainer plays get 17-18% wins. Overall, it works out to about 21%.

Anyway, I can't put a relative ranking on those other factors since I haven't tried to combine them with the horse ratings. Maybe I will in the future though. I only use post position as a tie breaker now, but I'd admit that it probably would also be a useful addition to my ratings.

[Rick]

Nobody asked, but here's my advice on building horse racing models using statistical techniques. Simple trial and error may provide better results as I've explained elsewhere but since it's a lot of fun to play with statistical tools, I'll pass along what I've found helps. First, you want to include post time odds as one of your independent variables (Log(odds+1) usually works best). Also, number of horses should be one of the variables because it will improve the prediction. Now, you can predict finish position or lengths using multiple linear regression. The key is finding variables that are STILL significant after adding odds as a variable. A lot of that recent speed garbage will fall out after you've done that. Even better would be to predict wins using a logit model but that requires more data. You might also need to convert odds into a probability or use the log of that probability to get the best prediction. Once you get your weightings, you can just use everything but the odds variable in your rating method. The bottom line here is that you're trying to predict winners better than the odds do, thus hopefully resulting in a better ROI. If anyone tries this, I'd be interested in knowing how well it turns out for you. I intend to go back and do some more of this with my own rating variables when I have the time.

[Niko]

Thanks for all the helpful info Rick!
Read an interesting book on commodities by one of the "turtles". The "turtles" were a group of people (12?) that were taught the trading methods of two of the most succesful traders ever. Just about everyone of them went on to considerabel success but they recently went through a tough time for a couple of years and now are rebounding nicely. A couple of things from the book I'm trying to relate to horseracing.

I'm not sure this concept will apply to horseracing but one area that really has me thinking lately concerns the following. The originators firmly believe that although the different commodities markets (aka different tracks) have different characteristics you have to use a model that is the same for all of them. They say if you use different models for different markets you are curve fitting the data based on what has happened in the past to a particular set of data and the method is not robust enough to work over time. This is only one of the reasons that they were able to achieve so much success. Anyone can make a model or system that works by back-fitting data to a particular period of time or market. This is the basics, if you want something more detailed I can pull the book.
This makes sense to me if you are relying on specific individual past data from software or trainer method percentages to hold up in the future such as winning post positions, trainer stats, race shapes for a particular meet and/or distance (which I found they rarely do but some others may have success with it).
In your research have you found that you would be curve fitting if you seperate distances?, tracks?. Would this mean you should use the same method for all speed favoring tracks and another for tracks distances that favor pressers? Or should you do all turf routes, dirt routes and sprints together? Or are the best models the ones that are robust enough to work over all tracks and distances? What would be the best way to divide the races while maintaining a robust enough model to work over time? I haven't gotten track or distance specific with my handicapping yet and I'm wondering if I should.

[sjk]

Niko,

You have hit on a point that I find very interesting. I developed my handicapping program back in 1993-96 starting with one circuit.

You might think that it would only apply to that one circuit. My experience is totally the opposite. I play new tracks with confidence because the fundemental elements of thoroughbreds performing and competing are consistent from track to track.

Looking at my records, I see that I have made bets at 42-45 tracks during the course of each of the past 3 years. My overall results are as good or better than my results on the circuit I started with.

There are track specific elements which need to be accounted for as you go from track to track. But there is an underlying consistency which is almost suprising.

When it is all said and done, all of these animals are close relatives.

[Rick]

Niko,

Well, it's true that you don't want to have a lot of different models for different tracks if you're only considering things that probably aren't different. I know that sounds confusing but, most factors probably won't be different but some almost certainly will. In my opinion, the most likely to be different would be the effect of early speed and post position biases. It's possible that even these factors wouldn't be significantly different for 95% of the tracks. That's why I asked Mike about whether he had studied differences in pace at individual tracks. If it isn't that much different I'd really like to know and only someone with a really large database could answer the question.

There also may be some differences relating to how long the season is for a particular track. I'd expect to have better consistency of performances if the horses, on average, had run more races at the current track. But, being more consistent or predictable might not translate into better profits.

By the way, what's the name of the book you read? I've found that just about everything about investment applies to horse racing.

I haven't found it necessary to have totally different methods according to distance on the dirt. Although I do take distance into account by emphasizing performance at the approximate distance, the weightings are exactly the same. Turf races are a totally different thing though and I've found that they're very track specific and very different than dirt races most of the time.


sjk,

What you've said is very encouraging. For the last year or so, I've concentrated mostly on fine tuning my method at one track because I think that's the best investment of my time. But, eventually I'll probably go back to trying the same method at multiple tracks. Congratulations if you've found something that works at all or most tracks though. You'll probably make a lot more money than I will.

[sjk]

Rick,

If someone had asked me what the most prominent track specific elements where as mentioned in my previous post, I would have answered speed bias and post position bias. It is good to see that you concurr.

I utilize track specific constants for these elements which are data based and which differ considerably from track to track.

[Rick]

One more thing. I know it's been said before but I really appreciate how great people are at sharing ideas here. And here's to PA for making it all possible. I know people don't want to give away their most closely guarded secrets (I know I don't), but the truth is that most of the really good ideas I've had were inspired research or things that people have said here that are freely available to everyone.

My philosophy about sharing ideas is that I won't give out all of the details about what works for me. But, if you're willing to do the research, you can easily use what I've said to duplicate my results. I'm not willing to help lazy people because it took a lot of hard work for me, but I will help anyone who is willing to work hard at it. That's not much of a threat to my profits though because almost everyone is lazy. Just my opinion, I could be wrong.

[Niko]

The book is called The Original Turtles Trading Notebook. It's published by Trade Wins Publishing. I've lost their phone number but I believe they have a web site. If you'd like I'm sure I'll see another ad in Investers Business Daily for it and I can let you know what their number is.
I'd also check out The Market Wizards books which I found fascinating and you can find in any bookstore. I've discovered more solid advice in investment books than I have in many handicapping books. The similarities between handicapping and investing are uncanny in regards to developing winning methods, money management, discipline and psychology. Like horseracing there's a lot of bad investment systems though and they cost a whole lot more! If you read about the documented superstars you learn a lot. And that's the great thing, you know if they're actually winners or not.

Rick you might be curious to know that the Turtle method, although one of the most succesful approaches of all time is very simple in it's main premises. Your posts have mentioned the same thing, concentrate on a 2-3 main factors and tweak it from there.

Being you both mentioned post position and track bias I found out something real interesting on turfday.com when they had a free preview of their site. Although I didn't find any long-term post-position ideas that would help one thing really stuck out for me. In short fields I suspected that front runners and early pressers would have an advantage due to less pressure in the race. In many circumstances the results were the opposite. The closers actually fared better in short fields than large fields. It appers that getting a clean trip may be more important in short fields than the lone pace horse. Maybe jockeys ride different?
I've always wanted to do a study on running style in conjuction with particular post-positions, odds and field size to see if I could find something. Maybe that's my next project after learning how to print my database results in descending order.

I'd also like to thank people for sharing their ideas. I don't expect exact methods or systems because of the nature of the game but the ideas and general recommendations on this board are great. I know there's a few people here that are doing a whole lot better than I am but I try to pass on what has/hasn't helped me and try to save others some time.

[Niko]

Rick brought up another subject that me and a friend had a lengthy discussion about. When is the best time to get long-shots at a track and is it also the best time to make the most profits? The beginning of the meet with horses coming off lay-offs and from different tracks seems to be the best time for long-shots. Also the end of a long meet when horses are tiring and fresh horses may be able to step up and beat them. But it may be more profitable to bet when the horses are a little more formful if you can't predict which long-shot will come through. That's all the further we got. Never did a study on it.

[Rick]

Niko,

Thanks for the investment info. I'm sure I can find something on them just by doing a search.

I think front runners do better in large fields just because the closers have so many other horses to go around. Even a front runner in an outside post doesn't lose much ground if he gets a good start. In fact, I think about 95% of the "early speed bias" at most tracks just has to do with avoiding wide trips.

[yak merchant]

Rick,

Your post got me pondering on some things. I'm a big believer in handicapping by science first. While I know that tracks have biases and faster tracks may help early speed save energy, I'm also a huge believer in the early speed = optimal path = perceived bias theory. Even if a track had some type of true early bias I feel eventually I will come up with some round about way to quantify that bias, and seeing how it affects all runners, shouldn't be a problem .

Now I'm not anywhere near getting my "time sucking project" off the ground, and about half of my crazy ideas are still theory. But the one thing I'm having issues with is quantfying rail bias. If there really is a rail bias at a track, every calculated number whether it be speed, pace or class is pretty much junk. I know some people track rail bias, especially Kenwoodallpromos. But the standard post position analysis just doesn't cut it when it comes to what I believe is probably the undoing of just about every, pace handicapper, sheet player and number freak. And not that I'm a big conspiracy theory guy, but a couple of extra passes with the tractor or watering truck and a track super can do whatever he or she wants. I know for a fact, at certain greyhound parks when the mutuels get a little low, they go change the bank in the first turn a couple of degrees, and bingo, dogs that used to be flying into the parking lot start dragging in some monsters. So in no attempt to hi-jack a thread on Service Statistics, just wondering about thoughts on what variables (not human or intent related) ie rail bias, are thrown into our wonderful puzzle that might not apply to all horses.

Regards,
YM

[Rick]

YM,

Well, if they do in fact make changes to the track surface deliberately as a result of some perceived bias, then you ought to be able to use the results of recent races to be able to predict those changes. What little research I've done on short term track bias (on Southern California tracks) seemed to show that they try to eliminate any radical bias, at least on a day-to-day basis. But, the effect was so small as to be relatively insignificant. All of the pace handicapping books claim that you should go WITH a track bias based on relatively small samples, and I found it to be just the opposite.

Another thing that might be worth tracking is how many favorites have lost in a row. The public apparently bets more on favorites after many losses as if they were "due".

Another unusual betting anomoly has to do with odds being lower on horses whose names contain people's first or last names.

If you really want to look into some different things to study, you should get the book "Trackfacts", by Dan DiPleco, probably available at Gambler's Book Store in Las Vegas. The last two things I mentioned were from that book. He also has a large amount of information on various "body language" characteristics. It's an interesting read, but I haven't really looked into any of his findings.

[Rick]

Mike N,

A couple of more questions if you're still around. First, I think I read that you changed data providers from BRIS to somebody else, I guess Trackmaster. Anyway, BRIS Class Ratings seem to be a pretty unique data item for which an equivalent wouldn't be provided by anyone else. Did you find that they didn't provide any additional information over speed ratings and therefore could do without them? Also, what do you consider the most important factors for first time starters and foreign horses with no US record?

[MichaelNunamaker]

Hi Rick,

About BRIS class ratings, I found that they were essentially the BRIS speed ratings. For example, looking at Modern Impact Values, BRIS class had an IV of 2.14 with a NET of $1.77. BRIS speed had an IV of 2.15 with a NET of $1.78. The top horses correlated extremely well.

When I switched to Trackmaster, my software still had all these references to bris class ratings, so I just mapped them to speed ratings. That change had essentially no impact on the accuracy of my picks.

So, my experience with them is that people who use BRIS class ratings are really speed handicappers. They just don't realize it<G>.

As far as first time starters, my analysis (which I admit could be wrong as I've spent about 1% as much time on first time starters as other horses) shows that the workouts tell the story. Trainers, jockeys, sire stats, and all the other things people love simply appear to have a full order of magnitude lower significance than the workout record.

As far as foreign horses with no US record, I haven't ever investigated this. For all I know, my software is doing really stupid things with these horses, so I should look into it.

The problem is there are about a billion other things I should look into and the amount of time I have is quite finite.

[Rick]

Mike,

I was pretty sure that was the case for the class ratings. I did a multiple regression study a few years back using speed rating and some other things (I think they were finish position, finish lenghths, and number of horses) to predict class rating and got a very high r squared. As to workouts for first time starters, would it be dates (or spacing), distances, or time that would be most important?

[Rick]

Anyone remember the Gibbs Composite Odds Line? What ever happened to that? It was pretty impressive in testing for a while anyway.

[MichaelNunamaker]

Hi Rick,

You wrote "As to workouts for first time starters, would it be dates (or spacing), distances, or time that would be most important?"

I really don't remember relative importance, but all the things you mention mattered quite a bit. Specifically, I looked at how fast the workouts were, how long the workouts were, how many workouts were there, what was the frequency of the workouts, and how recent the last workout was to today. They all mattered quite a bit.

As to your other question, what is the "Gibbs Composite Odds Line"?

[Rick]

Mike,

The Gibbs Composite Odds Line was an attempt by a guy from the Las Vegas area (Christopher Gibbs was his name) to create an accurate estimate of odds based on past performance data. His numbers were originally tested by Sport Stat (Jim Bayle) on a sample of more than 12,000 races and showed 29.9% wins and 12% profit playing every top choice. A subsequent test on a smaller sample showed only 26.3% wins and 3% loss. It's possible that the original test included races that were used to develop the method since there was some collaberation between Gibbs and Bayle. Still, even the second test seemed pretty impressive. Gibbs sold a service for a while with odds for most tracks and limits on the number of customers at each track. I thought you or someone else may have heard of him and how he wound up doing over the years. I'm not sure what year it was that year started it, but it must have been at least 10 years ago.

Another question for you. Did you use odds in last race as a variable in your models and, if so how did you handle entries? Some others, Jasper in "Basic Betting" for one, have claimed that it was a significant variable. I might as well drop another name while I'm at it. Have you read Benter's paper on how he developed his Hong Kong Syndicate horse racing model?

[MichaelNunamaker]

Hi Rick,

You wrote "A subsequent test on a smaller sample showed only 26.3% wins and 3% loss."

Well, I'm not particularly impressed. My own odds-line has a top pick winning percentage of 31.6% although the ROI is -9%. The ROI is +3.5% if one actually pays attention to the odds-line and only bets a good (50%) overlay. Actually, it used to be higher than that, the last three years it is just break even on overlays. I really should stop selling this stuff and bet it through a rebate shop. Amazingly, my ROI is a full 2% better at the tracks that Trackmaster does not sell my selections. Someone is putting some serious money down on my picks.


You also wrote "Did you use odds in last race as a variable in your models"

I only added that in the last six months. Nothing using that is live yet with my selections. It is surprising to me. Yes, it does have some predictive value. As far as entries are concerned, ooops. I forgot about them. How do you think they should be treated?

It's a small world. Jasper actually recruited me into the position where I spend most of my time these days (I primarily write computer software). Regarding Benter, no I've never seen his paper. I would _love_ to get my hands on a copy.

[Rick]

Mike,

William Benter's paper, entitled "Computer Based Horse Race Handicapping and Wagering Systems: A Report" appears in "Efficiency of Racetrack Betting Markets", edited by Hausch, Lo, and Ziemba. This book is usually very difficult to obtain and expensive. I got my copy from Gambler's Book Store in Las Vegas but don't know if they're selling it any more. It's also probably possible to get a copy at places like eBay but again, it may be very expensive. If you're able to get a copy though, it's definitely worth reading. I think that, with your technical background, you could get a lot of good ideas from it.

As far as entries go, I don't have a clue except to flag them as such and thereby downgrade the significance. It's possible that you could go back and look at the past performances before that race to try to decide how to weight each of the horses but I think that would just be too much trouble.

What is Jasper doing these days? Even though his books were not widely available, I was impressed by his way of tackling gambling problems. I used a modified form of his NFL football method to win money for many years. And, his general advice on money management has held up better than most of the theory that is always talked about. I've learned over the years that it's a really dangerous thing to ignore the assumptions (which may not be met) behind the mathematical techniques that you're using.

You make a good point about rebates. You don't even have to win any more, just come close to breaking even.

The 3% loss for Gibbs was betting every top selection, not just overlays, so I wouldn't say it's all that bad. But 26% is not a very good win % unless you're only looking at overlays. It's hard to say how to interpret the claims. Nobody else has any further information on him so he probably didn't do too well in the long run.

Occasionaly, I get people asking me what my ROI is, and I'm not sure what to say because it varies so wildly. This year it's positively amazing, but when you look at the last few months of last year it's considerably less impressive. So what really will be the truth in the future? I'm sure you totally understand what I'm talking about.

[MichaelNunamaker]

Hi Rick,

Someone is selling "Efficiency of Racetrack Betting Markets" at amazon for a grand. If I had more time I would pick it up, but I'm overwhelmed as it is these days.

I haven't heard from Jasper in a few years. He was extremely comfortable and retired spending his winters out west and the rest of the year here in Illinois where his kids and grandkids mostly live. When I first moved to the Chicago area, he tried to beat the races again, but failed. I think it kind of sucked the joy out of racing for him. He had always dreamed of electronic data making incredible things possible. But by the time he got it, the game was so much harder he couldn't turn a profit. He did quite a bit of analysis on my picks and I was able to get rid of a few weak spots because of his analysis (my main weakness at that time was that I wasn't paying enough attention to field size in previous races).

As an aside, if any of the database guys here want to do free consulting for me<G>, I would be happy to produce an electronic record of all my picks so anyone can tell me what factors I'm missing.

You wrote "Occasionaly, I get people asking me what my ROI is, and I'm not sure what to say because it varies so wildly. "

That's one advantage to being a public handicapper.<G> Since I pick every race in North America, things average out pretty darn quickly (to a very consistent -9% sigh).

[Rick]

Mike,

You should be able to get the book for much less than that somewhere. I got my copy for something like $90, but it's a reprint and just happened to be available at GBC for a while several years back. Also, it's possible that the individual paper is out there somewhere either on the internet or at a good library.

As to ROI, I think there's a natural tendency to notice it more when it's at it's best even when you're not intentionally trying to be misleading. Of course, if you're handicapping tens or hundreds of thousands of races it wouldn't vary much.

Interesting about Jasper. We all have a lot more enthusiasm and energy when we're younger. A 25 year old with a huge racing database and nothing better to do should be able to come up with some amazing ideas.

[Rick]

Come on all you database gurus and mathematical geniuses out there. Let's get a lively discussion going on this. Surely this is really interesting stuff to more people than we've heard from so far.

[Rick]

And all you pace gurus out there, here's your chance to tell Mike what he's missing that allows your heroes to get those amazingly high win percentages.

[MichaelNunamaker]

Hi Rick,

You wrote "Surely this is really interesting stuff to more people than we've heard from so far."

Perhaps the most depressing aspect of all the books I published was I received a huge amount of feedback that the books were "too advanced". Gambler's Book Club wouldn't even carry them because they were "too advanced". Essentially every return of a book cited that.

Those books were simple statistical observations. There were no formulas, nothing tricky at all. When I first wrote them, I was thinking of writing one book devoted to some of the interesting things I've done. One book reseller told me he thought the audience for a book like that would be about ten people. <Sigh>

Now that I'm on this topic, I'll mention something I was always fond of. You can make an odds line out of impact values by multiplying them together, and the result is the horse's score. Add up all the scrores of all the horses and the percentage of any one horse his its chance to win. Of course, this is horrible in practice because the factors are not independent of each other. But it can work OK if you pick a reasonable set of factors.

I wrote some really cool software inspired by the Fast Fourier Transform's butterfly that could test millions of combinations of factors per hour (on a 486!) across my entire database. I believe those were probably the best impact value based systems the world has ever seen.

It was all pointless as it turned out that even relatively simple regression analysis beats even the most finally tuned impact value based system. But it was wildly cool to see my computer evaluting these systems at a rate millions of times faster than anyone else I was aware of.

[Derek2U]

I've already posted how to maximize ROI. #1 start with a reliable
system that brings you close to break-even. Its more important
for this to be (a) a puter system with NO capping & (b) it must
be RELIABLE with a know range of results. EX: after ~100-150
races I KNOW, with total certainty, that my NY system gives
+1 to +4 ROI. NO Tweaking etc. #2 ... Use my proggy to ASK,
"should I bet my $100 to WIN on the system's pick or should
I spread that $100 on EX combos? (This alone will boost your
systems ROI by ~~3%. and Lastly, bet through Pinnacle or
whatever for a Rebate. Another ~3% plus? (as of now I cant
get myself to do this cause i dont trust them 100%.) ***
Also, RE my #2 point its a dammn that I cant trust the payoffs
on both Win & EX returns. If the tracks had RealTime Odds
then my use of my proggy would yield a steady +5%!!!
anyways, all data is incomplete so .... RE, IV's .... yes the
regressions works better because many IVs are silly random
noise AND are too Inter-Related whereas the Inter-Correlations
are reduced with Regression. However, if you have solid data
I can estimate the correction for that inter-correaltion within a
a point or 2 with chi square. PS: Point #2 means this: DO I PUT
MY PICK ON TOP in Exactas, NOT Exacta Boxes.

[GameTheory]

Benter's paper is reproduced in the back of Nick Mordin's latest book (I think), or one of Mordin's books anyways. Check with the member "swetyejohn" here on PA about that. You could probably order that one from the UK if you can't find it locally. Of course, you won't find anything in Benter's paper that hasn't been covered in this message board at one time or another...

[Rick]

Mike,

I understand that even simple statistical tabulations might be too complex for the average person to appreciate but this group has some surprisingly knowlegable people. Anyway, another problem with your publications is that some people just don't like seeing large data samples confirm how hard it is to beat the game. While I looked at things with a $1.80 net as having a huge potential when combined with other factors, some would be disappointed that they didn't see anything like the unsubstantiated claims they always read about in the typical horse racing book.

There are still a huge number of people out there who believe that you can get 60-70% winners betting two horses in every race using pace ratings because somebody said you can in a book and gave a bunch of seminars. They just won't believe any differently no matter what real data you show them. Mostly, they think that they can somehow pick just the right races subjectively and do that much better than a computer can. Well, as you've mentioned, it's possible for a computer to try millions of combinations of rules and it seems unlikely that a human could do better than that in the long run. If there's some secret information out there that they're using that's not in your database, then I say "let's test it and settle the argument once and for all".

Derek is difficult to understand as usual, but I think he's correctly pointing out in there somewhere that there are inter-correlations between horse racing variables that can't totally be sorted out in regression models. Maybe it's nonlinearities or mixing too many variables with non-normal distributions, I don't know for sure, but it's really true. Ironically, you can develop a model with just a few variables that gets only a slightly lower win % but a much better ROI. That's what I finally went to after being frustrated so many times with the many-variable regression approach.

Here's a link to a place you can get Nick Mordin's books:

http://www.pjmracing.com/

The main things about Benter's approach, to me, are that he used multinomial logit regression and got better results and that he advocated using actual odds in his model along with other variables for his predicted odds. I thought you might find these things interesting.

[MichaelNunamaker]

Hi Derek,

You wrote "EX: after ~100-150 races I KNOW, with total certainty, that my NY system gives +1 to +4 ROI."

I see dramatically larger swings in ROI with samples that small. For example, my Trackmaster selections for all North American tracks on February 14, 2004 had 189 top picks that won 38% of the races with an ROI of +8.5%.

But look what happens as sample size increases. If I take all North American racing from February 1, 2004 to February 14, 2004, my top picks winning percentage falls to 31.98% (only .5% away from the long term average!) and the ROI drops to -8% (only 1% off from my long term average).

[GameTheory]

You can get Nick Mordin books at Peter May's site?

Nick has a site, actually:

http://www.nickmordin.com/

I don't he's updated it in a while though. Amazon.uk would probably have Nick's stuff.


Rick -- did you ever read that Peter May neural network book?

[Derek2U]

POINT #1 ... i am NOT difficult 2 understand at all. Its just YOU
rick who seem intent on NOT reading carefully. of corse, almost
all racing factors are related math wise & so what: YOU CAN
correct for this math wise. This seems truer with IV's than
regression cause most Regression models have "built in" ways
2 correct for this whereas just multiplying IVs leads you astray
unless you correct for "set intersection" -- which is EZ as I
said (so clearly). Now, Mike, my more stable ROI even with
small samples came about because (A) we used small sample
statistics ON PURPOSE and (B) we have a huge DB & we VERY
carefully used it and (C) we DONT bet ~18% of the RACES.
All this is done by computer although we do allow for sub-routines. We use a max of 5 factors but they do differ for
type of race: distance/surface/class. Now, this NY system
does NOT work at California for ex. The factors are different.

[MichaelNunamaker]

Hi Game Theory,

Thanks for the heads up on where to find Benter's paper.

Hi Rick,

You wrote "Ironically, you can develop a model with just a few variables that gets only a slightly lower win % but a much better ROI."

I am all ears. Anything I've ever tried with just a few variables does not work as well as my many variabled equations. Clearly I'm missing something important. Any pointers you could give would be _GREATLY_ appreciated!

You also wrote "he used multinomial logit regression"

I mostly use simple least squares regression. I've installed R and intend to use it to investigate more sophisticated techniques. However, those darn time constraints get in the way. R isn't exactly the most user friendly thing in the world, but it can do everything except wash your dishes.

You also wrote "and that he advocated using actual odds in his model"

I've often thought this would be true. However, I've never tried it because I was unwilling to sit around to bet. With all this on-line automated betting, I should revisit this decision.

[GameTheory]

Originally posted by Derek2U
POINT #1 ... i am NOT difficult 2 understand at all. Its just YOU
rick who seem intent on NOT reading carefully.

Of course you're hard to understand -- people say so all the time. You're speaking a different language and use lots of words that don't exist. Did you know that there are NO real words that have numerals in them? I usually just skip over your posts since they give me a headache to look at. To blame Rick for not reading carefully is outrageous when it is you who are being lazy for not writing carefully and clearly. I can just imagine you telling your English teacher who gave you an F on a writing assignment -- "You didn't read it carefully enough!" :)

[Derek2U]

You should have left that time you threw a temper tantrum
and declared you were leaving 4Ever. So litenUP cause nothing
you type, no matter how good english it is, is interesting.

[GameTheory]

Originally posted by Derek2U
You should have left that time you threw a temper tantrum
and declared you were leaving 4Ever. So litenUP cause nothing
you type, no matter how good english it is, is interesting.

I never said that. How come people are always blaming me for stuff I never said?

[Rick]

Derek,

If you use heiroglyphics it's much harder to read what you're saying. I understand that it may be really important for you to appear to sound cool but that impresses stoners more than it impresses people here who are trying to understand you. No disrespect intended, I think you're an otherwise intelligent guy. Just don't make us wade through all of that garbage to get to the point.


GT,

"Forecasting Methods for Horseracing" is probably the book you're referring to. Yes, I did read it. Worthwhile reading for some different points of view but it didn't change anything I'm doing. I should go back and look at it again sometime though.


Mike,

Logit or probit models used to predict the winner do better than linear regression models to predict finish position or lengths for some reason. And remember, linear regression just minimizes the sum of the squares of the error, which may not be what you really want to accomplish. Nobody pays you for doing this, for sure. There are models that minimize the absolute value of the differences from the median, for example, that might do better. And non-parametric models, using ranks for example, might work better because they don't rely on any assumptions about the distribution. I have a three factor model that was developed mostly by trial and error. I've tried going back to using regression with the same data to fine tune the coefficients for the variables but it comes up with an inferior prediction, much to my chagrin! I know it's not supposed to be that way, but I keep finding it to be true. So, the answer for me is "whatever works". My advice to you is to just try adjusting the coefficients by trial and error on sets of just a few relatively independent variables. I'll bet that you can come up with something that does better than -9% even when retested on an independent sample.

The value in using actual odds in the prediction is to eliminate or reduce the effects of variables that are already incorporated into the public odds. I'd first do the analysis with odds included, then remove it and just use the remaining variables for your prediction. That would surely improve your ROI.

[MichaelNunamaker]

Hi Rick,

What you are saying is almost identical to what Jasper told me. He wrote a program he called "weightmaker". Rather than using regression to find the proper relative values of variables, his weightmaker program simply tried many different values. He said he got much better results than he did from regression.

OK, I surrender. I now have had two people I respect suggest the same bloody thing. I'll give it a whirl and see what it comes up with!

[chrisg]

Originally posted by Rick
Derek,

If you use heiroglyphics it's much harder to read what you're saying. I understand that it may be really important for you to appear to sound cool but that impresses stoners more than it impresses people here who are trying to understand you. No disrespect intended, I think you're an otherwise intelligent guy. Just don't make us wade through all of that garbage to get to the point.


:cool:

Roll another one...just like the other one...you've been holding on to it...but I sure would like a hit...Don't...bogart that joint my friend, pass it over to me... [L.Feat]

Eye N Eye C nO prablem n reeedn wat Derek sehz. n-E-weighz, hee looz all hiz coolness wen hee stoP tIpin leyek dat.

nO worries, der b no point 2 dis mon. Eye just seh dis, c?

Seen.
:cool:

[Steve 'StatMan']

I sense Text Messaging is starting to cause the next 'Generation Gap'

[Derek2U]

Both Rick & GT almost NEVER NEVER post any picks. And Rick, like so what? I think you need an eye Exam. And all that math, like what do You really know anyways? And GT, ur post just shows how removed you are from what you type. Just look over your 658- page-rants against poor little defenseless VetScratch --- and how many times did you PM PA to get rid of her? Both you guys got fragile egos and hate me cause Chris likes me.

[Rick]

Derek,

I don't usually post selections for the simple reason that, in many cases, it would be very easy to reverse engineer what I'm doing. I know, I've done it myself with other's selections in the past. One of the pro's I worked with in the past had all of his selections marked in the racing forms that we used for research. It was a simple matter to come up with a set of rules that selected the same horse at least 99% of the time.

All that math? What? You probably know at least as much as I do about mathematical techniques. At least you use the terminology in a lot of your posts anyway.

Thanks for answering this time in English though.

[Rick]

Mike,

Try it, you'll like it. I'd really like somebody to give me a good answer for why this might be true though.

[Rick]

I should also mention one thing I had a total LACK of success with that I thought would work. It involved adjusting coefficients on a race-by-race basis on pace variables to try to capture the effects of short-term track bias. I tried a multitude of different approachs that I thought were pretty creative and NONE of them did better than just using constant coefficients. That was another thing that I was totally surprised to see.

[Rick]

Mike,

Speaking of pace, an argument comes up from time to time here about Total Pace Rating (TPR). A few people swear that it works really well. All of my tests show that it's not very effective. If I recall correctly, you did some testing in the past on that rating, so I'm wondering what your opinion is on it.

[MichaelNunamaker]

Hi Rick,

About TPR, it appears to work OK. To elaborate, I have a pool of about 350 variables, mostly cooked data that I've put through some kind of manipulation. My regression software then starts trying regressions using small numbers of variables from these pools. The software can run for months looking for ideal combinations of variables to get the most accurate finish positions. Huge sample sizes (hundreds of thousands of horses) seem to do a pretty good job of preventing overfitting the data (although it makes things _SLOW_).

So, with this background, I view the most important factors as the ones my software includes and then giving the highest weight. TPR wins on the first count. TPR related variables are _always_ included in any equation I produce. The bad news is that they always earn significantly less weight than speed ratings.

[Rick]

Mike,

Interesting. Then they do provide additional information not present in ordinary speed ratings. I'd expect an early pace variable to be more significant than TPR (when combined with speed ratings), but I'll admit that I haven't tested it on hundreds of thousands of races and most of my data is limited to just a few tracks.

[Niko]

Rick,
If it's easy to reverse engineer your picks I assume you never bet over the net?
I've always wondered what succesful players who bet off-shore or online do to cover their picks? How easy is it for someone at these companies to copy the bets of the winners or reverse engineer what they're doing?
I've tried to do it looking over some contest results but the winners were always different and I wasn't able to come up with anything yet. So obviously I understand your concern because I've already tried to do what you mention.
Great conversation. I love stats but never got beyond the beginning level with them.

[Jeff P]

Rick,

You struck a chord when you said you use whatever works. I've tried many times to apply regression analysis and other statistical techniques to improve my results- but always without success.

I've had the most succeess using the following approach:

1. Run simple queries on single factors to discover win rate, avg win mutuel, and roi.

2. Then take single factors that look promising (those that come closest to breaking even) and combining two to three of these into a new unique single number. I often find the resultant new number generally outperforms each of the single factors that it is composed of. I suppose this might be because it is a composite of things that the public isn't betting in the first place. Also, if you consider such a resultant new number as a single factor, it certainly is something that the public doesn't quite have the ability to see.

I haven't found a new single number that has a positive roi across all races by itself. Yet. But I have found new single numbers that are very close to break even (playing all races.) And certainly, by paying attention to the odds and demanding a little value before betting- and by categorizing races by distance, type of race, surface, track, and rider, I have been able to isolate fairly large subsets of races where the top choice consistently shows a healthy positive roi.

Derek does make a couple of valid points. If you take the time to look, you'll find there are times, by dutching exactas on top of the rest of the field, when you can get more money back if your horse wins than you can if you had bet the horse to win through the win pool. You can also do the same thing using the double pool. I currently have an account at Pinnacle and I can tell you that the 7% rebate they offer on all wagers is huge.

[Rick]

Niko,

Well, you bring up a very good point. Off-shore wagering sites don't have to reverse engineer your selections but they could just overbet your selections in the parimutuel pools and nullify your advantage. I haven't seen any evidence of this happening so far with the service I use, but I don't bet big bucks anyway so maybe I'm under the radar. That being said, they could also lay off bets ON your selections to other off-shore bookies and get a little extra profit too. This is not really so far fetched and I have heard of people being paid by bookies for their selections. Usually on sports, but it could happen on horse racing as well.

[Niko]

I've also been doing a lot of study on pace figures and speed figures in the last 6 months--my research takes a lot longer because it's done manually. I never did any better using TPR than speed and race shapes but I figured I just didn't get it.
It's interesting because you have the speed guys (beyer, ragozin, brown) saying you can't make accurate pace figs and the figs they've come up with aren't any better than their speed ratings. Then the pace guys say you can't win using only speed. Who's right?
I've found that speed alone is enough in about 90% of the races but I keep pace in there because there's always a race or two won by a longshot that comes up on pace. I've tried different weightings to the segments of the races and one day I'd like to weight them based on track profiles but I may never get to that.

[Rick]

Jeff,

I've been experimenting with two and three factor models for at least five years now and each time I get a little better ROI. I started out at about -6% and now I'm at about +3% (playing the top choice in all races). I've also experimented with various ways of identifying overlays without knowing the final odds. My current method is to eliminate both the morning line favorite and the horse with the highest last race speed rating. You might find that a worthwhile filter if your method doesn't depend too much on recent races.

Derek really is a pretty smart guy, but he's so much fun to tease because he takes himself so seriously.

[GameTheory]

Originally posted by Derek2U
Both Rick & GT almost NEVER NEVER post any picks. And Rick, like so what? I think you need an eye Exam. And all that math, like what do You really know anyways? And GT, ur post just shows how removed you are from what you type. Just look over your 658- page-rants against poor little defenseless VetScratch --- and how many times did you PM PA to get rid of her? Both you guys got fragile egos and hate me cause Chris likes me.

That's right, I don't post picks. I don't look at anyone's picks that are posted either. Doesn't interest me. What does that have to do with anything? I never PM'd PA to remove anyone -- where does this stuff come from?

And I don't hate you either, I agree with Rick -- why you make your posts hard to read on purpose is what I don't understand. They look like this:

jfdksaj a;kljf;sdalkjak kfjkdsj fa;lkdjsk akjfiweurojm ;aoesjr;lkv oejra;lkjf;lk ejf;oiefa;lijeoium ;aleifj; kalejrlak f;lkjfe;irua;lkf jf;alwieur ;aljk eurpioaukje;iua;lkmeiuaipfa;jkjekj qiefuakejpoqm q;lieurkljfeiuq; puqreiojfl;qkewj rpoiqur mfkquerpoiqjflo qjofquefliqjfoi efjqopieurjvakefjqoimeirjqfmlkerqe;woijfi jiuerq;jfoiefeoijroirjelimf rpoqieikmflkejlkej ei kfjekljrkjek kejrekje;k


Actually, looking at it now the above is a little clearer.

Don't you see that you're showing contempt for everyone here by writing like that? Are you only talking to yourself? You post on a public forum, but you don't think enough of the people here to bother typing out a few extra letters to make a legible word. If you don't think anything of us why should we bother to listen to you?

[Rick]

Niko,

Well, here's the thing. To a large extent, speed=class, class=pace, and pace=speed, so when people say the most important factors are A, B, and C it really doesn't mean all that much. What really matters is finding relatively independent measures of performance. That could be using different ways of measuring it or it could be by getting it from different races. There are a lot of ways that are effective, but none of them involve highly correlated variables.

[GameTheory]

I think regression doesn't work that well because the error functions try to minimize overall error, and you end up with an average that can't make money. You can never get that accurate with horse-racing anyway -- 30% wins is pretty good, after all. It is better to be dead on correct with 30% of your numbers and way off on the other 70% than it is to minimize overall average error.

[PaceAdvantage]

Originally posted by Derek2U
You should have left that time you threw a temper tantrum
and declared you were leaving 4Ever. So litenUP cause nothing
you type, no matter how good english it is, is interesting.

You're confusing GT with someone else

[PaceAdvantage]

Originally posted by Derek2U
tch --- and how many times did you PM PA to get rid of her?

Ummm...never!

And anymore off topic notes in this thread will be deleted. I was getting a lot from this discussion, and don't wish to see it mucked up with anymore off-topic discussions....

[Rick]

GT,

You're right. It's far more important to select winners at overlaid prices than it is to be really accurate about predicting finish position. As I said before, you just don't get paid for doing that. And the jockeys who are on horses further back than 5th place in any race don't care (and aren't required to care) about where they finish.

[Rick]

But, if you can predict finish position better than the odds predict it, you might have something. There is some evidence that it might be best to categorize finishes as 1, 2, 3, 4, 5, 6th or worse, or something like that. I'd experiment with things like that all the way up to 1, 2, 3, or worse and see what works best. Then there's the Quirin way of categorizing with the "front half" and "rear half" designations. I never found that useful but I'm sure he had a reason for preferring it.

[Jeff P]

Rick,

Thanks. I'm always looking for and trying out new filters. Especially ones that enable me to bet my selections ahead of time and not be a slave to the toteboard all day.

One good filter that I use- as it applies to my own selections- is to simply throw out horses with morning lines of 9-2 and lower. This just about always includes the morning line favorite.

Another filter that I use- again, applying it just to my own selections- is to simply throw out horses that finished 1st 2nd 3rd and sometimes 4th in their most recent starts. This just about always incudes the horse with the highest figure (Beyer or otherwise) in its most recent start.

I'm beginning to suspect that I am doing things very similar to what you are doing- just going about it in a slightly different way.

I find that applying these and other contrarian thinking filters to my own selections lets me bet them ahead of time and frees me from being a slave to the toteboard all day. Whenever I check, I find that my "ahead of time" bets essentially end up mirroring what I would have done had I sat and watched the toteboard all day betting only those horses with odds at or above my own odds line.

[Rick]

Jeff,

I probably don't throw out enough horses using my liberal rules but I like to continue playing high percentage low positive ROI horses in order to minimize losing streaks.

The first time I ever noticed that rules like this would be useful was when I was playing Turf Paradise and had a good method that would select the top two horses at a small loss if I bet both of them (around 1997-1998 I guess). Well, I discovered that if I calculated average earnings for both of them that the LOWER average earnings horse showed a profit. That particular method suffered from a common problem. The second choice actually got a slightly better ROI than the first choice because I was using factors that were too highly inter-correlated. But, I solved the problem by using the low average earnings filter. Now, I've mostly eliminated the inter-correlation problem so I don't need to do that anymore, but I'd bet that a lot of people here have exactly the same problem and would be helped by trying something like that.

By the way, I still don't see any value in hanging around watching toteboards all day. My opinion still is that you don't care about all of the money as much as you care about "dumb money" when it comes to finding overlays.

[Rick]

I might add that I've been surprised several times about what I considered "dumb money". Sometimes it was me who was "dumb", so I suggest that you study it thoroughly first before you decide that you're smarter than everyone else.

[Rick]

Actually, if anyone from DRF is here, that's how I won the big Pick 3 at the DRF Christmas party back then. My wife worked for them at the time and I still think they were, and probably still are, a really great bunch of guys.

[Tom]

Interesting thread so far.
I have always found TPR to be of use in contender selection.
the thing htat has been eleuding me is the very good ealry rated horse who quits so his SR is bad, his TPR is bad, but he is a key horse in the pace.
I have been trying a new variation on ratings by using EP+EP+SR/3 where EP is the Quirin style ealry pace number and SR is the Quirin speed rating.
Not enough testing so far, but early results are leading me to think this figure might be important when coupled with a class drop or cut back.

[Rick]

Tom,

It seems to me that early pace works for frontrunners and overall speed works for closers, so it's always a challenge to try to find a single rating that works for both.

[Tom]

Rick,
Agreed. Adding twice the EP to a closer doesn't help him at all, in fact, it penalizes him unjustly. And Quinn's PDQ pace ratings which add in the pace of race really don't properly rate early and late horses under the same system. I am looking at this idea for an early only rating, a way to look at horses that have an edge early but noting late, recently.
I think andicap uses a multiple rating method that looks ar early and late horses differently.

[Rick]

Tom,

Exactly. I don't use two ratings myself now but I think it might be a better way to go.

[cj]

I rate horses by factoring the pace in accordance with the position of the horse at the first call. So, a horse running 10th of 12 will have very little pace factored in his rating, a horse running first will. Its not perfect, but it works well for me.

[Rick]

CJ,

Good idea. I hadn't thought of that.

[formula_2002]

Originally posted by formula_2002
Can someone tell me what is the value in these selections results?

Did I miss something?

"Michael E. Nunamaker's selection service performance statistics"

http://www.minnow.com/racing/pstat1.html

after going through some recent replys, it looks like the answer is no. I didn't miss a thing.

[Rick]

formula,

I don't understand. You post statistics like that all of the time yourself.

[formula_2002]

Originally posted by Rick
formula,

I don't understand. You post statistics like that all of the time yourself.

exactly!

[Rick]

formula,

Oh, I understand now.