Detailed Recordkeeping / Perfect Example of Why We Need to do this - Pace and Cap

RichieP · 03-20-2006, 11:39 AM

I am currently testing a spot play method in the selections area called "Soccer Guy". This test will run for 7 CONSECUTIVE days and will end with the conclusion of racing this Wednesday.

Anyone interested in the title of this thread please go back to the opening "Soccer Guy" thread and LOOK at my STATED feeling that making win bets at 5/1 and up on the horses was what I would be looking for if I could get the method working profitably. That is what I ASSUMED would be the profit area. ok

If you care to fast forward to yesterday's conclusion and 4 day cumulative totals.Of the four WIN betting categories my INITIAL assumption category is ranked WORST by quite a bit compared to the other 3 when looking at ROI.

While all four categories are extremely profitable right now I would say that

24% compared to 55% - 80% - 99% is noticably NOT the best category.

Detailed RECORDS do not lie. Self thought OPINIONS can and do deceive every horseplayer including ME.

Want another example relating to the Kentucky Derby?

Doc's contender guidelines for the Derby Contenders have had the winner in its final 5 horses like 18 out of the last 19 years ok? Nothing on the PLANET even can come close to approaching it. Yet people that would KNOW this as fact if they actually kept track of what is going on constantly are entering horses based on hype and this and that.

While it's great to sit around and hear everyone "wax poetic"

about his or her "Sleeper" Derby horse year in and year out the FACT is that if the horse did NOT finish 1st or 2nd in a grade 1 or grade 2 prep race the horse is OUT of consideration.

Detailed recordkeeping can help ALL of us get better.

Richie

traynor · 03-21-2006, 01:05 AM

Example of detailed recordkeeping. All selections are one per race, all wagers are flat bets to win, with the exception of races indicated (all of which, by the way, lost!). It would be easy to make a case for place betting, with a number of double digit place mutuels on the top selections. However, because it is a flat win bet model, it only considers selections for win position.

The point of all this is that outliers are strictly controlled. My dataset interquartile range mean (the average mutuel of the middle two quarters of the sample, excluding the lowest and highest quarters) is roughly $8.00. That means that any mutuel over $12.00 is automatically truncated to $12.00 (one and a half times the mean of the interquartile range--basic statistics).

The result is a much more coherent, predictive model. Of all the techniques I use to develop models, that is the most basic, and anyone developing detailed records would do well to make a serious attempt to control for outliers to avoid distortion by aberrant data points.
Good Luck

All selections were posted on www.paceadvantage.com Selections forum the night before the races. All can be verified. Results are normal for our data models, and results for thoroughbreds and harness are similar.

March 20 Results (13 wagers)
#1 Luanda won at RT at $3.00
#4 Czarbarz won at RT at $5.40
#6 RocknRoll Roxie won at RT at $7.20
#2 Kiki Koko won at BM at $13.20 [truncated to 12.00]
#3 Mulberry Gus won at SL at $3.00
#8 Easy Brenton won at VL at $14.80 [truncated to 12.00]
#3 Kb's Speed Chip won at VL at $9.00
#7 Bget Jim Jim Jim won at VL at $11.40

Current Status +$27.60 (model)
Actual Return $268.00 - 218 = +$50.00

March 21 Selections
Bluffs Run - Matinee
Race 5 - #5 Thisbee's Gotyou
Race 7 - #5 B's Knott Head
Race 12 - #4 Hilan Dodge

Birmingham - Evening
Race 9 - #7 Tnt Show Em
Race 11 - #8 San Tan Chev

Southland - Evening
Race 11 - #4 Tnt Vegas

Previous Results
March 1 - 5 Races -$10
March 2 - 5 Races +$ 3.80 (8.00 + 5.80 - 10)
March 3 - 5 Races -$ 4.40 (5.60 - 10)
March 4 - 7 Races +$ 7.20 (8.20 + 7.80 + 5.20 - 14)
March 6 - 2 Races -$ 4.00
March 7 - 4 Races +$ 6.20 (8.00 + 6.20 - 8)
March 8 - 6 Races +$ 7.00 (33.60 [12.00] + 7.00 - 12)
March 9 - 5 Races 0$ (10.00 - 10.00)
March 10 - 5 Races +$ 2.00 (14.00 [12.00] - 10.00)
March 11 - 6 Races -$ 2.60 (5.00 + 4.40 - 12.00)
March 12 - 1 Race [no race]
March 13 - 3 Races +$ 6.00 (16.00 [12.00] - 6.00)
March 14 - 6 Races -$ 1.80 (6.40 + 3.80 - 12.00)
March 15 - 11 Races -$10.00 (16.40 [12.00] - 22.00)
March 16 - 7 Races -$ 6.40 (4.40 + 3.20 - 14.00)
March 17 - 10 Races +$ 5.60 (14.00 [12.00] + 5.40 + 8.20 - 20.00)
March 18 - 5 Races -$ 5.60 (4.40 - 10)
March 19 - 2 Races[2]-$ 8.00
March 20 - 13 Races[2]+$27.60 (3.00 + 5.40 + 7.20 + 13.20 [12.00]
+ 3.00 + 14.80 [12.00] + 9.00 + 11.40 - 26.00)

All selections based on positive expectation model; minimum
33% win percentage, minimum 125% ROI over time.

Sample Selections:
for Greyhounds March 1 - March 31
for Thoroughbreds and Harness April 1 - 15
Good Luck

bobbyb · 03-21-2006, 07:07 AM

Hi Traynor,
I have a question for you. Would it not be more accurate to truncate the actual final odd(s). Eg: 4.7 or 3.6. since your model(s) are reliant on after the event entry and slightly raise your ceiling +-(% ) to incoporate additional breakage that may be received, but not expected.
I can appreciate your sound approach, but I have a problem in my model with what to do with unexpected breakage, which at times can represent upwards of +10 to15% of a return which is unexpected.
Your view with regards to breakage would be appreciated.

bobby

traynor · 03-21-2006, 11:38 AM

Breakage is not a consideration, because we only use real world odds, rather than morning lines or projected odds. There is never a conflict, because what you get at the mutuel window is what you get.

Similarly, because we use actual mutuels, we are able to build more accurate data models, that in turn are based on results in the real world. In a given sample of races, there will be an accurate interquartile range mean that is used to eliminate outliers in subsequent races within the same sample group.

The simple average--with or without breakage--is extremely misleading. Using a simple average tends to distort reality enough to make the handicapping of most racing fans "off" far enough to eliminate most of the available profit.

The point is that a very minor step--correcting for outliers--will enable most fans to make a quantum leap in profit. Or conversely, convince them to stop betting an aberrant "model" that is distorted by one or more unusual payoffs.
Good Luck

traynor · 03-21-2006, 11:45 AM

Another consideration in the breakage area is that the illustration given for greyhounds is not the model itself--it is an illustration of how we create models. SImilarly, the [12.00] truncating point is close, but not exact. That is, our algorithms keep a detailed record of results within each data layer, and re-calculate the mean of the interquartile range dynamically. Subsequent mutuels are truncated in accord with the actual results, rather than projected results.

That re-calculation may change the overall ROI a bit, or the expected win percentage a bit, but those areas are never exact, anyway. There is no "statistical pressure" that "forces" reality to conform to a model, regardless of how carefully crafted that model may be. The key point is that the more accurate the model is, the better the abilty of that model to predict future events.
Good Luck

03-20-2006, 11:39 AM	#1
RichieP Grade 1 Join Date: Jun 2005 Posts: 7,019	Detailed Recordkeeping / Perfect Example of Why We Need to do this I am currently testing a spot play method in the selections area called "Soccer Guy". This test will run for 7 CONSECUTIVE days and will end with the conclusion of racing this Wednesday. Anyone interested in the title of this thread please go back to the opening "Soccer Guy" thread and LOOK at my STATED feeling that making win bets at 5/1 and up on the horses was what I would be looking for if I could get the method working profitably. That is what I ASSUMED would be the profit area. ok If you care to fast forward to yesterday's conclusion and 4 day cumulative totals.Of the four WIN betting categories my INITIAL assumption category is ranked WORST by quite a bit compared to the other 3 when looking at ROI. While all four categories are extremely profitable right now I would say that 24% compared to 55% - 80% - 99% is noticably NOT the best category. Detailed RECORDS do not lie. Self thought OPINIONS can and do deceive every horseplayer including ME. Want another example relating to the Kentucky Derby? Doc's contender guidelines for the Derby Contenders have had the winner in its final 5 horses like 18 out of the last 19 years ok? Nothing on the PLANET even can come close to approaching it. Yet people that would KNOW this as fact if they actually kept track of what is going on constantly are entering horses based on hype and this and that. While it's great to sit around and hear everyone "wax poetic" about his or her "Sleeper" Derby horse year in and year out the FACT is that if the horse did NOT finish 1st or 2nd in a grade 1 or grade 2 prep race the horse is OUT of consideration. Detailed recordkeeping can help ALL of us get better. Richie

03-21-2006, 01:05 AM	#2
traynor AlwNW1X Join Date: Mar 2006 Posts: 13	Detailed recordkeeping Example of detailed recordkeeping. All selections are one per race, all wagers are flat bets to win, with the exception of races indicated (all of which, by the way, lost!). It would be easy to make a case for place betting, with a number of double digit place mutuels on the top selections. However, because it is a flat win bet model, it only considers selections for win position. The point of all this is that outliers are strictly controlled. My dataset interquartile range mean (the average mutuel of the middle two quarters of the sample, excluding the lowest and highest quarters) is roughly $8.00. That means that any mutuel over $12.00 is automatically truncated to $12.00 (one and a half times the mean of the interquartile range--basic statistics). The result is a much more coherent, predictive model. Of all the techniques I use to develop models, that is the most basic, and anyone developing detailed records would do well to make a serious attempt to control for outliers to avoid distortion by aberrant data points. Good Luck All selections were posted on www.paceadvantage.com Selections forum the night before the races. All can be verified. Results are normal for our data models, and results for thoroughbreds and harness are similar. March 20 Results (13 wagers) #1 Luanda won at RT at $3.00 #4 Czarbarz won at RT at $5.40 #6 RocknRoll Roxie won at RT at $7.20 #2 Kiki Koko won at BM at $13.20 [truncated to 12.00] #3 Mulberry Gus won at SL at $3.00 #8 Easy Brenton won at VL at $14.80 [truncated to 12.00] #3 Kb's Speed Chip won at VL at $9.00 #7 Bget Jim Jim Jim won at VL at $11.40 Current Status +$27.60 (model) Actual Return $268.00 - 218 = +$50.00 March 21 Selections Bluffs Run - Matinee Race 5 - #5 Thisbee's Gotyou Race 7 - #5 B's Knott Head Race 12 - #4 Hilan Dodge Birmingham - Evening Race 9 - #7 Tnt Show Em Race 11 - #8 San Tan Chev Southland - Evening Race 11 - #4 Tnt Vegas Previous Results March 1 - 5 Races -$10 March 2 - 5 Races +$ 3.80 (8.00 + 5.80 - 10) March 3 - 5 Races -$ 4.40 (5.60 - 10) March 4 - 7 Races +$ 7.20 (8.20 + 7.80 + 5.20 - 14) March 6 - 2 Races -$ 4.00 March 7 - 4 Races +$ 6.20 (8.00 + 6.20 - 8) March 8 - 6 Races +$ 7.00 (33.60 [12.00] + 7.00 - 12) March 9 - 5 Races 0$ (10.00 - 10.00) March 10 - 5 Races +$ 2.00 (14.00 [12.00] - 10.00) March 11 - 6 Races -$ 2.60 (5.00 + 4.40 - 12.00) March 12 - 1 Race [no race] March 13 - 3 Races +$ 6.00 (16.00 [12.00] - 6.00) March 14 - 6 Races -$ 1.80 (6.40 + 3.80 - 12.00) March 15 - 11 Races -$10.00 (16.40 [12.00] - 22.00) March 16 - 7 Races -$ 6.40 (4.40 + 3.20 - 14.00) March 17 - 10 Races +$ 5.60 (14.00 [12.00] + 5.40 + 8.20 - 20.00) March 18 - 5 Races -$ 5.60 (4.40 - 10) March 19 - 2 Races[2]-$ 8.00 March 20 - 13 Races[2]+$27.60 (3.00 + 5.40 + 7.20 + 13.20 [12.00] + 3.00 + 14.80 [12.00] + 9.00 + 11.40 - 26.00) All selections based on positive expectation model; minimum 33% win percentage, minimum 125% ROI over time. Sample Selections: for Greyhounds March 1 - March 31 for Thoroughbreds and Harness April 1 - 15 Good Luck

03-21-2006, 07:07 AM	#3
bobbyb NewBegining Join Date: Mar 2006 Location: Ontario Posts: 15	Hi Traynor, I have a question for you. Would it not be more accurate to truncate the actual final odd(s). Eg: 4.7 or 3.6. since your model(s) are reliant on after the event entry and slightly raise your ceiling +-(% ) to incoporate additional breakage that may be received, but not expected. I can appreciate your sound approach, but I have a problem in my model with what to do with unexpected breakage, which at times can represent upwards of +10 to15% of a return which is unexpected. Your view with regards to breakage would be appreciated. bobby

03-21-2006, 11:38 AM	#4
traynor AlwNW1X Join Date: Mar 2006 Posts: 13	breakage Breakage is not a consideration, because we only use real world odds, rather than morning lines or projected odds. There is never a conflict, because what you get at the mutuel window is what you get. Similarly, because we use actual mutuels, we are able to build more accurate data models, that in turn are based on results in the real world. In a given sample of races, there will be an accurate interquartile range mean that is used to eliminate outliers in subsequent races within the same sample group. The simple average--with or without breakage--is extremely misleading. Using a simple average tends to distort reality enough to make the handicapping of most racing fans "off" far enough to eliminate most of the available profit. The point is that a very minor step--correcting for outliers--will enable most fans to make a quantum leap in profit. Or conversely, convince them to stop betting an aberrant "model" that is distorted by one or more unusual payoffs. Good Luck

03-21-2006, 11:45 AM	#5
traynor AlwNW1X Join Date: Mar 2006 Posts: 13	breakage Another consideration in the breakage area is that the illustration given for greyhounds is not the model itself--it is an illustration of how we create models. SImilarly, the [12.00] truncating point is close, but not exact. That is, our algorithms keep a detailed record of results within each data layer, and re-calculate the mean of the interquartile range dynamically. Subsequent mutuels are truncated in accord with the actual results, rather than projected results. That re-calculation may change the overall ROI a bit, or the expected win percentage a bit, but those areas are never exact, anyway. There is no "statistical pressure" that "forces" reality to conform to a model, regardless of how carefully crafted that model may be. The key point is that the more accurate the model is, the better the abilty of that model to predict future events. Good Luck