Mark Cramer
C & X Report

EDITORIAL
BRING BACK ARTISAN RESEARCH
OR, WHAT DO TOMATOES HAVE TO DO WITH HORSE RACE RESEARCH?
This month we’re featuring the early speed / pace factor, with an interview with handicapper and researcher Steve Klein, plus an article on the Moss Pace Figures.
I had been waiting to pass on a vital question to Steve Klein, a question I’d received in different forms from several of our readers. Now I had the chance.
In his book, The Power of Early Speed, Steve included enormous samples of races, in the hundreds of thousands. It would take me a lifetime to accumulate a sample of such magnitude.
But Steve had researched the performance of early speed from the results charts. My readers wondered whether the early speed horses in the results charts were the same as those that figured to be early speed before the race, as seen in the past performances. There’s often a significant difference between early speed in a race, and early speed as projected in the pps. Most of the time we find the clear dominant speed from the pps, we get 6/5 in return, but when there’s a 15-1 wire-to-wire winner, such a horse did not often project to be on the lead in the pps.
In all fairness to Steve, his handicapping shows a great skill in finding such hidden early speed horses. But the research is what I was concerned with. Steve furnished us with the type of honest answer that he is well known for. He said:
Unfortunately, although the database used by Daily Racing Form is capable of doing some clever things with horses who fall into any of a large number of defined categories, it also has a few significant limitations. One of them is that it was not designed to recreate the way specific past performances looked for each horse in the field on the day any particular race was run that is more than a few days old.
Now here’s where C&X, over the years, has been maligned by the numbers crunchers. We have actually done the laborious type of research that is impossible to do with sophisticated databases. It’s like we are baking our own bread instead of picking up Wonder Bread at the supermarket.
I say “we” because it’s not just yours truly. Other members of the C&X community have done heroic feats of research, most recently professor John Sember from Hastings Park. Some of Sember’s research is patiently carried out over years.
It seems that today, great technological achievements draw a boundary for us as to where or where not we can intrude. In some instances, we can learn amazing things (as with Steve’s original research) from the most sophisticated databases. In other situations, we are deprived from learning amazing things, as Steve has testified, because the database has serious limitations for the user.
I believe that this research predicament can be seen as a metaphor for the quality of life that we are promised today. Let’s use food as an example. Do any of you honestly believe that industrial or intensive agri(culture) has a prayer for producing food that can come anywhere near the quality of that which is produced by traditional agriculture.
I once had the good fortune to raise a few pigs and a squadron of chickens. I fed them good stuff. You should have experienced the beauty of watching my pigs dig through leftovers from restaurants and the excess apricots and peaches from my trees. The quality of the resulting pork chops and bacon is memorable even 30 years after those dinners and breakfasts. Nothing since then can compare.
Just as I would not be happy living in a society where industrial needs determined my diet and eating habits, I cannot accept that the presence of high tech databases can determine the boundaries of racing research that we do. Industrial agriculture depletes biodiversity by eliminating foods that require more complex production; industrial databases exclude diversity of handicapping ideas by excluding research methods that require more complex tallying procedures.
If databases cannot reconstruct the original data of the pps, then my friend and colleague Barry Meadow has no right to make deterministic pronouncements that his 20,000 race samples prove that no factor can result in a flat bet profit over the long run. Clearly, if you feed limited, constrained and constipated data into the high-powered machine, the results will not have the depth of what can crafted by the research artisan.
So you get breadth but not depth from high tech research.
There’s a simple way to test what I have asserted. Go to a farmers’ market and choose tomatoes from a pile where no two tomatoes have exactly the same size or shape. Compare the taste of those tomatoes with ones from the supermarket that are all identical in size and shape. You be the judge.
PS. Kevin’s letter near the end of this issue would seem to implicitly challenge the points I’ve made in this editorial. I suppose it boils down to the choice of factors fed into the database. The more inventive that choice, the more likely that the database will produce bettable results. Kevin, do you agree?

STEVE KLEIN GETS GRILLED
(Steve Klein works for the DRF and is based in Kentucky. He is one of the best public handicappers in the nation. He’s one of the few who can pick longshot winners with regularity, and has come up successfully as “informed minority” on several occasions. You’ll hear about Steve’s book and “Speed Points” in the text of the interview.)

MC. To get started I'm asking you a couple of tough, maybe combative questions because that is what's expected of me by my readers. Your pioneering work, The Power of Early Speed, provides formidable evidence that this factor is dominant in racing, especially on the dirt. But from a handicapping perspective, many a horse that wins thanks to early speed did not figure to show that early speed from the past performances. Having read and enjoyed the book, I know you have covered many nuances, including how a trainer or jockey switch can awaken early speed in a horse. However, there seems to be a disjunction between your empirical results-based stats after the fact, and more anecdotal evidence about before-the-race selection process of the right early speed candidate. Could you please comment.

Steve responds:
First, for anyone who is still unfamiliar with the book, The Power of Early Speed, the single most exciting thing about it is the number of horses I studied. Using the same database used by Daily Racing Form, I studied a total of 1,671,627 starters who ran in 201,237 dirt races, and 32,347 turf races. That is a tremendous sample size, bigger than any other study I have ever heard of. The conclusions based on this research are rock-solid.

I was especially pleased to see that this massive amount of research fully confirmed the results I had seen over the years both anecdotally, and in the numerous smaller studies I had done on my own. Personal experience counts for something, after all.

The book was published in December of 2005. Looking back on the process of writing it, there are three things I would have liked to have been able to do differently. For reasons I will explain, it wasn't possible to do them the way I wanted to in the book. But all three of these issues have been solved and turned into benefits in the information provided on my powerofearlyspeed.com web site.
The first issue pertains directly to your first question. I would have liked to examine the results of the Klein Speed Points formula in the same way, and in as much detail as I studied the other data in the book.
Unfortunately, although the database used by Daily Racing Form is capable of doing some clever things with horses who fall into any of a large number of defined categories, it also has a few significant limitations. One of them is that it was not designed to recreate the way specific past performances looked for each horse in the field on the day any particular race was run that is more than a few days old.
For example, if you ask the database today to recreate the past performances for the first race at Turfway Park on January 1, 2006, it can't. That information isn't stored because there is no point in doing so for the purpose of putting out future editions of the newspaper. The data is constantly updated because there is a continual need for the product to be current. There is no reason for a function that would subtract data out of the system to recreate the way the past performances actually looked for each horse's 10 most recent races on any given day six months ago, or two, or five, or 10 years ago.
But some information from those same horses that ran in that race can still be used more generically in numerous other statistical categories. For example, if one of the horses in the first race on January 1, 2006, was claimed that day, his win or loss in his next race would still become part of the statistical history of his trainer in that category for the next year or two, and would be part of his career stats.
Even if back-testing had been available, the track-bias stats used to place those numbers in context, as they are now used on powerofearlyspeed.com, would not have been available, so the testing would have had only limited value.
Since back-testing was not an option, I did the next best thing, which was to use the research from the book, and about 40 years of personal experience in handicapping races with early speed in mind, to tell me which factors were most crucial to include when I created the Klein Speed Points 1.0 formula (the one used in the book).
Now, on powerofearlyspeed.com, Klein Speed Points 3.0 can be seen and tested in real time on a daily basis. I have done so since October, and the results have been impressive. The proof is in the website product, and I'm going to give you and all of your subscribers unlimited free access through the end of the month of March so they can experience the power of it. More about that later.
That brings me to the second thing I am able to do on my powerofearlyspeed.com web site that I wasn't able to do in the book.
I couldn't go all out in making the Klein Speed Points 1.0 formula as sophisticated or comprehensive as I wanted to, because the goal was to make it so simple that an average person, who might be math-phobic, could use it during the 20 to 30 minutes between races to figure the ratings for an entire race, with help from nothing more than a pencil and a calculator.
Of course, I have with no restrictions on the complexity and sophistication of the Klein Speed Points ratings available on powerofearlyspeed.com. I have made two generations of improvements to the formula since the book was written. The biggest leap was between generations 2.0 and 3.0. And if I can find a way to improve it even more, then that updated formula will be used on the web site.
Regarding your insight about horses showing surprise early speed, you are right to say that they contribute to the big return on investment shown in the results in the book. But the results I have seen in real time, before the races have been run, show that there are still considerable benefits to be gained from using Klein Speed Points 3.0.
The third thing I have been able to do a good job of on powerofearlyspeed.com that isn't covered in the book is making money with synthetic racing surfaces. The book was written prior to the installation of synthetic racing surfaces, so they weren't discussed. The good news is that the powerofearlyspeed.com web site brings everyone up to date by measuring the value of early speed, and even more important, the value of all running styles, on all running surfaces on a daily basis at all major tracks. You'll have to see it to realize exactly how valuable this is.

MC. How do your original early speed stats compare and contrast with subsequent results on synthetic surfaces? Visually, it seems that there are many more races like last year's Blue Grass, in which the speed can't hold even with incredibly slow fractions. Doesn’t this throw a wrench into the foundation of your theory?

Steve responds:
Early speed has been less effective on Polytrack and other synthetic surfaces, but that doesn't mean profits can't be found on synthetic tracks. For example, Polytrack has become an advantage for well-informed handicappers, because it can play quite differently from its reputation for long periods of time. Just one example: in one-turn races at Keeneland in October, early speed was an advantage. The first-call leader provided an ROI of $3.19, which is actually higher than the $3.12 overall ROI expected on dirt tracks. Horses with tactical speed returned $2.37. And the only group that lost big money was closers, who returned just $0.85 for each $2 bet. Those results at Keeneland were earned without help from hard-to-find huge longshot winners, who were subtracted from the sample to minimize the distortion they would have introduced.
What makes this particularly fascinating, enlightening, and profitable, is that there is a very good chance that you and your subscribers were totally unaware that Keeneland was playing that way in sprints in October. But it wasn't a secret to the handful of friends, relatives, and handicappers I allowed to help me test my powerofearlyspeed.com web site. I'm sure that a number of track-bias oriented handicappers pounded the closers in their bets in those sprint races while assuming the closers had the preferred running style.
For the record, route races at Keeneland in October were different. There was only a slight edge to being the early leader. Horses in mid-pack fared best, but not by much, and weren't profitable bets.
There have been a number of other profitable situations to exploit with early speed on other synthetic tracks. But that misses a very important point. I keep track of four categories of horses: First-call leaders, tactical speed, mid-pack, and closers. Whichever style is working best is the one powerofearlyspeed.com subscribers are betting on. And there are details provided. The names of each specific horse who is likely to benefit from these track biases is listed in each race. Having four different track bias options quadruples the opportunities for spotting a valuable trend.

MC. Essentially (without getting into proprietary information), what's the difference between your speed point method and that of William Quirin? The Quirin Speed Points were certainly easy to use.

Steve responds:
Quirin Speed Points don't come close to the quality of Klein Speed Points because there are so many really important predictive factors that are either left out of Quirin entirely, or are measured in units that are grossly imprecise in Quirin's method.
For example, Klein Speed Points use field size as a crucial factor. Quirin ignores that factor completely. Common sense should be all that is needed to understand that a horse who is a distant third early in a three horse route race is probably much slower than a horse who is third early in a 20 horse Kentucky Derby field. One horse is in last place, ahead of nobody, while the other horse is faster than 17 other horses, and is in the first 15 percent of his field. Klein Speed Points understand this factor. Quirin Speed Points don't.
Quirin and I both believe the running position of a horse at the first call of a race is an important factor. But Quirin Speed Points are needlessly imprecise on this factor. With only the exception of seven furlong races, Quirin gives the exact same rating to horses in third-place at the first call as he gives to horses who lead at the first call. Once again, common sense should be all that is needed to understand that horses in first place are usually faster than horses in third-place, but for anyone who is still wondering, the massive computer study in my book shows that first-call leaders won 28 percent of the time, while horses who were in third-place at the first call won only half as many races, 14 percent. Klein Speed Points recognize the difference between these two categories. Quirin Speed Points don't.
Another serious flaw with Quirin Speed Points is that they give the same rating to a horse who is two or three lengths behind the leader as they give to the leader. Obviously, that makes no sense. Klein Speed Points take both the margin of a horse's lead, and the margin a horse is behind into account, and give both a sensible rating. Quirin numbers ignore the margin of a horse's lead entirely. Again, that is needlessly imprecise.
It should suffice to say that there are a number of other important factors that Quirin either ignores completely, or measures so imprecisely that they are needlessly inaccurate. Although there are more factors I could mention, including the proprietary information in Klein Speed Points 3.0, this sampling should give you a pretty good idea of how the two rating methods differ.

MC. On your website, you are offering version 3.0 of the speed points method. Does this mean that version 1.0 in your book is obsolete? The reason I ask is because certain artisan handicappers prefer to do their own calculations rather than depend on time-saving numbers from a third party, for the reason that intense involvement in the process
can contribute to more confident betting decisions.

Steve responds:

As I mentioned earlier, the version of Klein Speed Points used in the book was restricted by two constraints. It had to be simplified to make it unintimidating for use by handicappers who might be fearful of math. And it had to be simplified again to enable the user to get an answer for each horse during the time between races. Given those constraints, I achieved my goal.
By way of comparison, Klein Speed Points 3.0 has no such restrictions. And by the time I designed 3.0, I had accumulated years of experience in using 1.0 (which was brand new when the book was written), followed by 2.0. I had all of the freedom I wanted to make 3.0 as complete as I wanted it to be. So the answer is yes, the book version has been made obsolete by the 3.0 version on powerofearlyspeed.com.
Beyond the comparative superiority of version 3.0, the practical side of saving time has to be considered. If you assume about 25 minutes of work per race on a 10 race card for version 1.0, you will have to spend about 4.2 hours to get the obsolete numbers. For about $3 for an entire race card, you save more than four hours of work. That is a cost of only a few pennies, 71 cents per hour, for work that you would otherwise have to do on your own. It would be foolish not to use the superior information from the web site. And that doesn't take into consideration all of the other valuable information on powerofearlyspeed.com that puts the speed points in a context that enables you to bet on, and bet against specific horses, listed by name, with win percentages and ROI's based on the results of recent races run on that surface at that track.
Handicapping is both an art and a science. Since there are only a limited number of hours in any day that can be devoted to handicapping, I want the best of both worlds. I use objective science, the Klein Speed Points 3.0, to give me the competitive advantage of valuable, unique information, which is essential to have in any parimutuel betting situation.
Once Steve the scientist has obtained this information, and has been shown which two or three horses in a race to choose from, and which to leave out of the exotic pools, Steve the artisan handicapper is free to take over and enjoy the subjective, artistic craft of applying the numbers to the races that I am handicapping.

MC. Exacta and tri places face a special dilemma. Many experts have claimed that the same bias that favors a winner may not kick in for the second horse in the exacta and the third horse in the tri. Are you saying that the early speed favoritism of a track will also include the backholes in exotics? That’s hard to believe.
Steve responds:
The broad answer to this question is yes, the running styles that are good in the win slot are pretty good, although not quite as strong, in the place and show slots in the exotics, and the running styles that are bad in the win slot are usually are also at a disadvantage, although not quite as strong, in the place and show positions in the exotics.
The exception is if the preferred running style is early speed, and the Klein Speed Points 3.0 numbers for all of the horses in the field with early speed are very low, or are negative numbers. That indicates the possibility of a hot pace. An early speed horse can still win, but the second and third horses might rally from off the pace to pass the speed horses the winner discouraged. Look first for contenders in the "Mid-pack" group, as they will be best placed to capitalize. Then look to the closers if the mid-pack
horses don't seem good enough.
If there is a closer at low odds on a speed-favoring track who you are trying to beat in the exacta and trifecta, the bigger the field size, the better your chances are of shutting him out of those bets, not only because there are more legitimate contenders in the race to fill those slots, but also because battling a speed-favoring track, and a very wide trip, or
traffic problems in a crowded field maximizes the chances that the closer will run a bad race.

MC. Final question Tell us about your website. What services and information does the
user receive and for what cost?

Steve responds:
Of course, they get Klein Speed Points version 3.0 for every horse who has raced. They also get the up-to-date win percentages and ROI's for recent races for each of four different running styles (leaders, tactical speed, mid-pack, and closers) in one turn and two turn races at each track, featuring data for every different racing surface offered at that track, including synthetic surfaces and turf races. The data gets very specific when you see, by name, which horses running each day at each track are mostly likely to fit into each group.
Please remember that these aren't selections. The idea is to focus your handicapping on the small number of horses in a field (usually two to four, depending on field size) with the highest ROI's, preferrably $2.50 or higher, who have the best chance of being profitable.
Eliminate the longshots who seem overmatched, since they don't have the ability to capitalize on their favorable running style. But if a longshot looks good to you, and is unfairly being ignored by bettors, he can be your play. Your goal is to choose the best horse from the profitable group as your key horse. If none of the horses have high ROI's, pass the race. Or play against a low-odds favorite with a really low ROI ($0.60 and lower). Some tracks will offer more betting opportunities than others, but the win percentages and ROI's do change each day and week, based on recent results.
If the profitable horses in a race are the ones with early speed, don't focus only on the ROI for the first-call leader. Make sure that the "Tactical Speed" category also has an ROI of at least $2.50 before betting on the horse that is listed as the possible "Leader", regardless of how high the ROI of the potential leader is. That way, if the horse is close up in second or third early, rather than in front, he is still in a profitable category.

The best scenario looks like this:
1. A high ROI for a contender you like ($2.50 and higher), and either, or both of the following:
2. The other horses in the race with high ROI's from having the preferred running style are hopeless longshots who look bad on paper, and seem to have little chance. It will be tougher for your selection to win if the other horses with high ROI's also appear to be strong contenders.
3. Low win percentages and ROI's showing on one, and preferrably more of the horses in the race who are being well-bet by the public (the lower the ROI on the low-odds horses the better, with $0.60, and lower preferred). That improves the chances of your key horse, and opens up possibilities in the exacta and, in larger fields, the trifecta, and superfecta.

Testing anyone?
Here’s the way you can enter Steve’s service during the month of March at no charge, in exchange for your critical reactions. Looks like a fair exchange.mc

Steve says: The free offer begins right now. I want all of your subscribers to enjoy and
profit from Klein Speed Points on Powerofearlyspeed.com, so I am going to give them unlimited free access to every track available on the web site, every racing day, starting immediately, and lasting all of the way through the end of March. All I am asking for in return is feedback from them. I want all of them to email me at powerofearlyspeed@gmail.com with their comments, and success stories about the winning bets they made using this information.
Here is how to access all of the tracks:
1. Go to www.powerofearlyspeed.com
2. Register on the site.
3. Go to the User Menu and click on "Purchase Tokens"
4. Go to the bottom of the page where it says "Input coupon code here if you
have been given one" and type in the code: feelthepower
5. Click on "Use Coupon Code"
6. Type in the number of tokens you want- you might as well take the maximum
number of 50. You get one race card per token. You can also go back for more
free tokens later if you run out.
7. Click on "Purchase Tokens"
8. Click the OK button and the tokens are now yours. The race cards are
available the night before the races are run, usually between 7:30 pm and
9:00 pm local time for the track you are downloading.
9. To download race cards, go to the top of the page and click on the
"Download Race Cards" tab. Click on "Klein Speed Points", then download the
tracks you want. Print them out for your use.
Now go download, and enjoy using them!


MOSS PACE FIGURES:
MORE TESTING

I’ve been fiddling around with the Randy Moss pace figures, searching for a way to use them effectively. Unlike the Klein Speed Points, we know exactly how the Moss pace figs are derived. Essentially, they are fractional Beyer figs. None of us would have the time to pile this stuff on our own, so as with the Klein Speed Points, we need to depend on the purveyor.
In the beginning, I had to deal with the figure overload. The past performances already have too many numbers. It’s bad enough on paper, but on a screen, too much of it can provoke a headache. Add the Moss pace figs and the screen becomes much busier that Morningstar fund pps.
As a narrative type of guy, I appreciate when the “race shape” figures are “translated” into words. Moss does this, on the right hand side of the pps.
I’d like to zoom in on one particular race, the February 16 Barbara Fritchie Handicap, Grade II for fillies and mares, at Laurel. We’ll first eliminate the two most unlikely horses and now take a look at the contenders. I’ve extracted the most representative race shape for each horse, using only the word shapes of Randy Moss.

1-1 CONTROL SYSTEM: fast early, slow mid, slow late
4-1 GOLDEN DAWN: fast early, average mid, average late
9-2 BABY BIRD: fast early, slow mid, average late
5-1 OPRAH WINNEY: fast early, slow mid, slow late
10-1 KARAKORUM STARLET: fast early, slow mid, slow late
13-1 FOR KISSES: slow early, fast mid, fast late / or / slow early, slow mid, fast late

At first glance we see five horses going fast for the first quarter of this seven furlong event. Then, only one of them gets the jump in the middle fraction, which is the 4-1 Golden Dawn. In the end it may be too late for Baby Bird, who’s lost ground with a “slow mid”, but For Kisses could possibly keep close enough to unleash his one-run stretch move.
For me three horses popped off the page. First there was Oprah Winney, since she won the same race last year, coming from the same previous Big A race, and with an identical layoff. However, going back a year to this mare’s win in the same event, she also showed a fast-early/slow-mid/slow-late, not a pace advantage in today’s field since there are so many others with exactly the same running style.
The second horse that attracted my attention was the horse-for-course For Kisses. She always fires at Laurel, and had the very different pace shape. However, she might have to make up too much ground on Golden Dawn because GD advances in the second call, and she shows more places than wins in her performance box.
Now we come to the 4-1 Golden Dawn, who had a “big win” of 6 lengths at Aqueduct in her last race, and figures to control the running in the second fraction, which means she’ll be breathing easier than her rivals during that segment of the race. If the Moss figures are right, this could represent a huge advantage for Golden Dawn.
The pace shape of GD, “fast early, average mid, average late”, is impressive since, in the usual energy distribution of horses, a “fast early, average mid” is followed by a “slow late”.
At even money, the favorite CONTROL SYSTEM looks vulnerable because she has the same pace style as three of her main rivals, and quite similar to a fourth rival. Just how vulnerable is difficult to say, since we cannot demote her too much if we consider that she’s five wins for seven starts lifetime, and she’s defeated similar foes.
What I like about the narrative Moss comments is that I can actually see the projected race before me. Using the numbers seems far too precision for me, given that horses are not precision animals. As the economist E.F. Schumacher has noted, in his classic Small is beautiful: a study of economics as if people mattered (1973), quantification may be at the expense of understanding qualitative differences.
In short-term forecasting …a refined technique rarely produces significantly different results from those of a crude technique.
In later lines, Schumacher warns that “endless multiplication of mechanical aids in fields which require judgment” will bring us deceptive findings.
Don’t get me wrong. I like to see that Moss has done the numerical homework. In fact, thank you Randy for sparing me the overwhelming routine of cranking out the figs. I happen to find Moss’s narrative translations especially useful. In the above race, I could visualize that my pattern match for Oprah Winney (won this race last year and has the same prep race as last year) would not be enough to get the job done, and that the heavy favorite, Control System, was made vulnerable by the shape of the race.
The results of one race do not prove or disprove a handicapping technique, but they do offer encouragement for further study.
Golden Dawn won, once more by 6 lengths, with a $10.00 payoff for the win. Second was Control System, with the exacta returning $26.80. Third was the one-rush closer For Kisses, only a half length behind the favorite. The tri paid $160.
Oprah Winney ran out of gas. After capturing the lead midway, she stopped completely and finished last, distanced by the rest of the field.
For those of you interested in researching the Moss figs, I’d love to have a long-term flat bet record for all sprinters with a fast early, average mid, and average late: in other words, early speed horses that do not run out of gas at the end.

C&X CAFE

Dialogue with Jeff

Jeff wrote:
First, in your harness example, introducing the PP filter did not "increase your advantage". It decreased your dis-advantage. You're still losing.
My response:
If there were one simple mechanical filter that reduced the track takeout from 17% to 9%, that would be to our great advantage.
Technically you are absolutely right. It's still a disadvantage. But relatively to what we had before, it's advantageous. Besides, the point of the article was to answer a question about whether or not a factor well-known to the public can still be used in any way to help our percentages.
Jeff wrote:
Second: in re Beyer and the Pimlico bias. Beyer was "...the only one..." who noticed? Horseshit on roller skates. How in God's name would you, me, Beyer, whomever, know who else might have picked-up on the bias? You think dedicated players go around telling everyone when they've picked up on something with serious profit potential?
My response:
Perhaps you've misunderstood. Which of my words mentioned that dedicated players should be telling everyone about an advantage. I thought my point there was that when you find something good and you are in a distinct minority among your fellow investors, that you should bet on it. Sure there must have been others who noticed. But Beyer must have been the only big player to notice. Otherwise, he’d have lost his odds advantage. The point made by the Wall Street Journal, which I reported but which was not my opinion, is that there was a possible conflict of interest, since Goldman touted people to play with the subprime and not against, while Goldman used its own money to play
against. Most public handicappers I know (because they are "public" handicappers) do indeed announce their discoveries to their constituents.
Jeff wrote:
C'mon Mark, really. What to do with racetrack/slots conumdrum? I have grave doubts that there's a realistic "solution" to stopping the slide of this thing of ours into obscurity. But to the extent there might be a plausible longshot, in my mind it revolves around a massive reduction in the takeout, and a concommitant severing of the unholy umbilical which tethers racetracks to the State. I mean, the State shouldn't get a penny from racetrack wagers until and unless the track turns a profit. Almost like real businesses. It sounds kinda funny to look at the slots roughly 5% "rake" on random plays, compare it to
the tracks, roughly, 24% take on random wagers (probably more on exotics), and decide that the slots players are the "dumb" money. I know that's not exactly what you're getting at but if I was a "slots guy" and knew these numbers, I'd be thinking "Yeah, right, I'm the
dumb one".
First, the slots odds are unchangeable. No amount of player analysis can change the outcome. At the races, it’s different. The odds are determined not by the house but by the players. So if the crowd is in some degree of error in its analysis of a race, then suddenly the thoughtful player can have an advantage. You can see this by going back to the time of separate pools and seeing the different odds for the very same horse in the same race. Sunday Silence paid off at 7/2 in New York and only 2-1 in California. Belmont winner Colonial Affair paid $44 in California and only $28 in New York. So in racing, even with a higher takeout than slots, the odds are variable. In slots, the odds are fixed. No one can win in the long run, so it’s a dumb bet.
On the other hand, I agree with you on the tax injustice. A race track is a tax paying business. If it shows a loss, it shouldn't have to pay. And even if it shows a profit, its taxes should be on the profits and not on the customer's wagers. I'm thinking out loud, in reaction to what you wrote. Does it make sense?
From professional researcher: Kevin
Hi Mark,
I haven’t been in touch for some time because I was busy starting a new company, which just celebrated its 3rd anniversary.
I just finished reading C&X 42, which is great. I e-mailed you several years ago about my use of median mutuels, in addition to average mutuels, in order to assess the degree to which outlier payoffs were influencing ROI in a research series (I have pasted in my letter from 2003 for reference). I noticed that you mentioned median mutuel in this issue, which I applaud.
Dr. H.R. has a question about a statistician who might help him (or her) test some combinations of factors. In fact, my own handicapping these days is almost 100% of the database variety. I have a database of more than 20,000 races in which I have used a series of filters to identify combinations of factors that show a flat bet profit. I don’t have much time to handicap, so these days I simply identify 3-6 bets daily using some of the best sets of filters and make my bets online early in the morning. For me, the most satisfying part of the process is the data analysis and testing handicapping hypotheses (explaining why I went into epidemiology). I would welcome the chance to help another C&X reader test some of his (or her) hypotheses. Please feel free to pass on my contact information on to Dr. R.H.
Also, I am going to strongly recommend a book to you that, in my view, has many useful lessons for handicappers. The book is called The Way of the Turtle by Curtis Faith. It is about a group of commodities traders who were trained in the 1980s as part of an experiment that grew out of a bet between the legendary trader Richard Dennis and his business partner, leading to the quote “We are going to grow traders the way they grow turtles in Singapore” (you will have to read the book to find out what that means). My own “database handicapping” approach is very similar to what the turtles did to earn millions of dollars in trading profits (although my bets are on a much smaller scale). Reading C&X (and C&O before that) has been the stimulus for many of the hypotheses I have tested over the years and I am thankful that you are continuing to publish your findings.
Best,
Kevin
Kevin C. Maki, PhD
President
Provident Clinical Research & Consulting, Inc.
Illinois: 489 Taft Ave.; Glen Ellyn, IL 60137
October 27, 2003
Kevin’s previous letter:
Dear Mark,
I was very happy to learn C&X was resurrected with RPM as the new publisher. Your research has often led me to avenues of investigation that I would never have thought of without your insights as catalyst.
I think it is highly commendable that you apply criteria to minimize the risk that a reported positive return is a chance finding, such as excluding the top few mutuels and looking at the ratio of winners to second place finishers. I take a somewhat different approach, with the same goal, in my own research.
I have several databases of race outcomes that have been created to test various hypotheses. When running queries, I often find combinations of factors that show positive returns. Not having the time or energy to test them all, I try to find those few that show the greatest potential for positive ROI to test prospectively. One of the statistics I have found to be very useful is the median ROI.
To calculate median ROI, the median odds on winning bets (value that splits the top and bottom halves of the winners’ odds) is substituted for the average odds on winning bets in calculating the ROI.
ROI = Average Odds on Winning Bets x Probabilitywin – Probabilityloss
Median ROI = Median Odds on Winning Bets x Probabilitywin – Probabilityloss
As an example, I recently ran a query in my database looking at top picks of a computer handicapping program. I was looking at top picks that had a specific combination of factors that I suspected might show a profit. In a group of 828 races with the combination of factors in the program’s top pick, there were 222 winners (26.8% hit rate) and a total return of $1867.00, based on $2 bets. The average payout was $8.41, producing a ROI of 1.127 (12.7% profit). The median payout was $7.20 (median payout is almost always smaller than the mean). Therefore, the ROI and median ROI are calculated as follows:
ROI = 3.205 (0.268) – 0.732 = 0.127 (12.7% profit)
Median ROI = 2.6 (0.268) – 0.732 = -0.0352 (3.5% loss)
I prefer plays with:
A hit rate of at least 20%
1. Positive ROI
2. Median ROI that shows no more than a 10% loss (preferably 5% or less)
If the ROI is positive but the median ROI shows a loss of >10%, it is likely that a few high payoffs accounted for the profitability (which may not occur in subsequent samples). Plays with hit rates below 20% are difficult to test because of the very large number of races required to reach a reasonable level of statistical confidence in the ROI. In addition, the long losing streaks associated with lower hit rates test one’s resolve.
For plays that meet these criteria, I run a prospective test of enough races to reach a 90% level of confidence that the true ROI is positive. To calculate the number necessary, I use a method described in Dick Mitchell’s book Commonsense Betting. He provides a formula for calculating number of races needed for confidence testing.
N = Probabilitywin x Probabilityloss x (Z/E)2
N is the number of prospective test races needed. Z is a constant based on the level of confidence desired, with values of 1.645 (90% confidence) and 1.960 (95% confidence) being used most often. E is the acceptable error in the win percentage. In this case I have chosen 3% (expressed as a fraction) as the acceptable error because this is the hit rate that would result in breaking even, given an average pay of $8.41 (i.e., a hit rate of approximately 23.8% with an average pay of $8.41 would be needed to break even). Plugging these into our formula, we get:
N = 0.268 x 0.732 x (1.96/0.03)2
N = 837
Therefore, a prospective test of 837 races should be sufficient to establish, with 90% confidence, that the ROI is positive. Note that the number of races needed is approximately the same as the number in the dataset from which I generated the original hypothesis. For this particular play, I can identify roughly 75 races per week that meet the criteria. This means it will take me about 12 weeks to run the test needed to determine whether or not I am willing to bet real money.
You can see why I need to narrow down the number of tests to those that are most promising!
Best,
Kevin Maki
From Ken

GP 6 today had a good example of your bet-down overlay method. Horse was dropping from MSW to MC (biggest drop in racing, in my opinion) but still went off at over 10-1 and paid $23.80. A shipper from TP. Costs a lot of money to ship a horse from Kentucky to Miami.
Connections seem to be gamblers to invest that money. Unbelievably, with the drop, trainer had good overall record and 22% win rate in 1st on the turf and this is what that was, as well as good record with 2nd time Lasix. But crowd still let the horse go at great odds.

Ken
Mark replies:
Even with so much information out there, or perhaps because of it, the public is capable of stumbling. Ken came up with a play that fit many of the C&X criteria, adding creative analysis of his own, regarding the shipping costs. In various research projects of the past, when everything else was equal, horses in the “shipper” category had a higher return on investment.mc
Subjects: Testing Samples and Inefficiences

Hi Mark,
I have noted through the years the great efforts you take to discount the effect of longshot winners when running test samples. I wondered if you, when running test samples, have ever considered using proportional betting (wager the amount needed to win a desired payoff - $5 bet to win $100 on a 20:1 horse; $20 bet to win $100 on a 5:1 horse and so on based on odds). Proportional betting would seem to render the discount of longshot winners by capping their payoffs, while affording statistical significance in the small sample. Also, by retaining longshot winners in the sample its integrity would be preserved. I assume that you have considered this betting method and have chosen not to use it and wondered why.
One other question, in the last issue of C&X you wrote about the phenomenon of maximum confusion in identifying market ineffiencies when a race has several contenders. Years ago I attempted to read "Horse Sense" by Burton Fabricand (calculus was a snap compared to Fabricand's Rules on contender selection). He had done exhaustive research on races where there were two top contenders whose odds were close together. His research showed that the shorter priced horses won a disproportionate number of races, such that automatic bets would result in significant profits. It seems like you are looking at similar circumstances, but are headed toward longer, rather than shorter, priced contenders. I would appreciate your comments re: Fabricand's work as it relates to yours.
Thanks for many years of stimulating reading and enjoyment.
Rick P
Rick,
Great to hear from you again, and thanks for the good words.
There are two reasons that I have not used proportional odds in research. First, researchers who know much more than I am (my learning comes from trial and error) have continually supported my exclusion of the highest-priced longshot winner, and in fact, recent stuff from people who know more than I do about research, suggest eliminating the two highest-priced mutuels.
Of course, it's not so simple. If you graph the mutuel payoffs, you discover that there will be clusters of most-frequent payoffs. If the clusters occur in higher payoff ranges, then I'm less concerned about eliminating the big ones. But even in calculating my own returns, for self-critical research purposes, I would have to eliminate an Arcangues.
That said, I have been tempted to use some sort of proportional tally, and the only thing that has really stopped me is my own mathematical inadequacy. The more finessed I try to be, the greater likelyhood that a flaw can emerge.
Thanks to your letter, I am going to toy with a way of doing proportional tallies. There, the difficulty is at the low end of the odds spectrum more than the high end. In most profitable mechanical samples, the real strength comes from between 5-1 and 9-1, and then, even the 10-1 through 19-1 range usually has a better roi (but lower hit rate) than below 5-1. I guess that if I were to try something proportional, it would be in something where I would eliminate all horses going off at odds with a payout of less than double figure.
As for Fabricand, his stuff is too "macro" for me. I don't think his stuff would hold up if we had a good subset with unique traits. Having two low-priced horses is not, to me, maximum confusion. The confusion model needs to be more confusing than that.
So, tell me what you're up to. What tracks are you playing and what types of races?
Cheers,
Mark