Keeping records for success at different odds, what to use?

[lansdale]

Originally Posted by MJC922

I'll provide a status on this from time to time. It had a good week, looks like it finally caught a couple of decent prices and got into the black, the totals after two weeks now look as follows:

# of Bets Win % $1 ROI Wagered Payoff Profit/Loss
162 14.81% +0.03 $324.00 $332.90 $+8.90

10 of the 25 (40%) winners went off below the requested 5-1 conditional minimum odds.

Good to hear things seem to be going better. Re the odds drop, I just wanted to make sure you were aware that the odds-dropping winners are +EV that would not have been included in your orignal Model #2 results - aka gravy. It might allay your concern to chart the odds-dropping winners vs. odds-dropping losers - I'm guessing the ROI of the former would at least cancel out, and more likely, exceed the ROI of the latter, which should be few in number.

BTW, assuming that your Model #2 is 'global'', extracted from the original model of all races in N.A., the results suggest to me a strong, low-variance performance that should continue going forward.

No need to post any more than you wish, but it could help players here attempting the same kind of thing, and if you run into problems, the math people here might have some solutions. Go for it.

[MJC922]

Originally Posted by lansdale

Good to hear things seem to be going better. Re the odds drop, I just wanted to make sure you were aware that the odds-dropping winners are +EV that would not have been included in your orignal Model #2 results - aka gravy. It might allay your concern to chart the odds-dropping winners vs. odds-dropping losers - I'm guessing the ROI of the former would at least cancel out, and more likely, exceed the ROI of the latter, which should be few in number.

BTW, assuming that your Model #2 is 'global'', extracted from the original model of all races in N.A., the results suggest to me a strong, low-variance performance that should continue going forward.

No need to post any more than you wish, but it could help players here attempting the same kind of thing, and if you run into problems, the math people here might have some solutions. Go for it.

Thanks, yes it's 'global', and yes I'm well aware the odds-droppers wouldn't be part of the Model #2 results, it summed up my whole concern in a nutshell, i.e. too many horses that wouldn't have qualified by the Model #2 would largely chew down the ROI to the point of insignificance. All plays qualify as 'the selection' from model #1 so I already know if those as a group were negative 3% ROI over a massive sample and they hold at 40% of the current winners of 'this' sample then I might ballpark the ROI as 40% of the negative 3% ROI and 60% of the positive 20% ROI, which amounts to roughly +10% ROI. I'd be happy to provide more detail around the ROI of the odds-dropper group and had planned a deeper dive into all of this as the sample grows. Right now I can understand the thought that the late money would be smart anyway and these droppers actually help matters or at least don't hurt much, I'm rather skeptical of that but again we'll be able to put aside theory eventually and provide hard data.

[lansdale]

Originally Posted by MJC922

Thanks, yes it's 'global', and yes I'm well aware the odds-droppers wouldn't be part of the Model #2 results, it summed up my whole concern in a nutshell, i.e. too many horses that wouldn't have qualified by the Model #2 would largely chew down the ROI to the point of insignificance. All plays qualify as 'the selection' from model #1 so I already know if those as a group were negative 3% ROI over a massive sample and they hold at 40% of the current winners of 'this' sample then I might ballpark the ROI as 40% of the negative 3% ROI and 60% of the positive 20% ROI, which amounts to roughly +10% ROI. I'd be happy to provide more detail around the ROI of the odds-dropper group and had planned a deeper dive into all of this as the sample grows. Right now I can understand the thought that the late money would be smart anyway and these droppers actually help matters or at least don't hurt much, I'm rather skeptical of that but again we'll be able to put aside theory eventually and provide hard data.

I do understand your concern about post-close odds movement. But, as I said, the only subgroup of late movement that would not have been included in your Model #2 that could have affected its ROI would be the winning risers - and I doubt that this number could be more than .01 of the population, if that. It sounds like you're keeping track of the net of all post-close odds movement, which is great - this will provide an accurate idea of its effect on the model going forward.

I still believe this to be a minor factor, and that working from a very large initial sample implies the stability of the model going forward.

One thing that came to mind re the issue of model-making, especially with a large global sample like yours, is the influence of the seasons. Many handicappers have mentioned this, and one very successful model-builder who used to post here, named Rook, discussed this issue in detail. In sum, he said that his ROI was never as good during the late fall and winter as during the warmer seasons - I don't remember all the factors he cited, but certainly the higher frequency of smaller fields was part of it. Maybe some of the long-time modellers here could comment - possibly the elusive Rook himself.

Best of luck.

lansdale

[MJC922]

Originally Posted by lansdale

I do understand your concern about post-close odds movement. But, as I said, the only subgroup of late movement that would not have been included in your Model #2 that could have affected its ROI would be the winning risers - and I doubt that this number could be more than .01 of the population, if that. It sounds like you're keeping track of the net of all post-close odds movement, which is great - this will provide an accurate idea of its effect on the model going forward.

I still believe this to be a minor factor, and that working from a very large initial sample implies the stability of the model going forward.

One thing that came to mind re the issue of model-making, especially with a large global sample like yours, is the influence of the seasons. Many handicappers have mentioned this, and one very successful model-builder who used to post here, named Rook, discussed this issue in detail. In sum, he said that his ROI was never as good during the late fall and winter as during the warmer seasons - I don't remember all the factors he cited, but certainly the higher frequency of smaller fields was part of it. Maybe some of the long-time modellers here could comment - possibly the elusive Rook himself.

Best of luck.

lansdale

Umm, that was the elusive me, not 'Rook'. Cheers.

http://www.paceadvantage.com/forum/s...d.php?t=106830

With no disrespect to Rook, possibly Rook posted something before that. And yes this is a big concern as the 5 months of this model were June through October and from several years ago.

[Rook]

Originally Posted by lansdale

One thing that came to mind re the issue of model-making, especially with a large global sample like yours, is the influence of the seasons. Many handicappers have mentioned this, and one very successful model-builder who used to post here, named Rook, discussed this issue in detail. In sum, he said that his ROI was never as good during the late fall and winter as during the warmer seasons - I don't remember all the factors he cited, but certainly the higher frequency of smaller fields was part of it. Maybe some of the long-time modellers here could comment - possibly the elusive Rook himself.

Hi lansdale,
Your memory that many years ago I wrote about seasonally streaky betting is correct.

In 2005, I only had one year's experience betting in the winter so I didn't have a clear idea about what to expect. Almost a decade later, I can give an update that return wise, October thru February are by far my best months, March thru May are my worst and June thru September are neutral.

The main reason for this is that I consistently prefer states' B tracks over A Tracks (ex. Hawthorne over Arlington; Turfway over Churchill/Keenland; Delta over Louisiana; Aqueduct over Belmont/Saratoga)

[lansdale]

Originally Posted by Rook

Hi lansdale,
Your memory that many years ago I wrote about seasonally streaky betting is correct.

In 2005, I only had one year's experience betting in the winter so I didn't have a clear idea about what to expect. Almost a decade later, I can give an update that return wise, October thru February are by far my best months, March thru May are my worst and June thru September are neutral.

The main reason for this is that I consistently prefer states' B tracks over A Tracks (ex. Hawthorne over Arlington; Turfway over Churchill/Keenland; Delta over Louisiana; Aqueduct over Belmont/Saratoga)

Hi Rook,

Since you rarely post any more, I appreciate the response. Although I got it wrong about which season was better for you, I did remember you mentioning that you got widely varying results according to time of year.

For those who haven't seen your handle, as I remember, you are a very successful pro handicapper, and have been one for a decade or so. A few years ago I believe you moved to Canada for tax purposes. Since there have been some differences between traditional and database handicappers here, maybe worth adding that, you're very knowledgeable about traditional handicapping methods and have a large library of handicapping books, but also have a good grasp of probability and statistics and, are well-versed enough in SPSS to use (or used) it to create your model. There's no innate conflict between these approaches.

Thanks again for dropping by, and would appreciate any comments on the current state of the game you might have - very few reach your level of expertise.

Cheers,

lansdale

[lansdale]

Originally Posted by MJC922

Umm, that was the elusive me, not 'Rook'. Cheers.

http://www.paceadvantage.com/forum/s...d.php?t=106830

With no disrespect to Rook, possibly Rook posted something before that. And yes this is a big concern as the 5 months of this model were June through October and from several years ago.

Happy to see you had this covered. One thing that struck me from this post, and possibly part of why you remain somewhat skeptical, is the timeframe of the first model. If so, I can understand that, but my own view remains positive.

[Rook]

Originally Posted by lansdale

A few years ago I believe you moved to Canada for tax purposes.

Nope, I have always been a Canadian.

[lansdale]

Originally Posted by Rook

Nope, I have always been a Canadian.

Sorry about that. I know a number of people who have done this for the reason mentioned, and wrongly included you.

[Rook]

Originally Posted by lansdale

Since there have been some differences between traditional and database handicappers here, maybe worth adding that, you're very knowledgeable about traditional handicapping methods and have a large library of handicapping books, but also have a good grasp of probability and statistics and, are well-versed enough in SPSS to use (or used) it to create your model. There's no innate conflict between these approaches.

Thanks again for dropping by, and would appreciate any comments on the current state of the game you might have - very few reach your level of expertise.

The interest (obsession) in racing came first. Handicapping was certainly the fuel that drove me to learn about databases and stats.

If I was to limit myself to one comment I would say the most misguided postings on this board are from those who are anti rebate. Realistically, since takeouts aren't coming down anytime soon, rebates are the only thing that give horseplayers a fighting chance.

I have a book full of other comments about the current state of the game but I do my best not to get sucked into those discussions!

[lansdale]

Originally Posted by MJC922

Umm, that was the elusive me, not 'Rook'. Cheers.

http://www.paceadvantage.com/forum/s...d.php?t=106830

With no disrespect to Rook, possibly Rook posted something before that. And yes this is a big concern as the 5 months of this model were June through October and from several years ago.

I've run out of caveats ;-).

[lansdale]

Originally Posted by Rook

The interest (obsession) in racing came first. Handicapping was certainly the fuel that drove me to learn about databases and stats.

If I was to limit myself to one comment I would say the most misguided postings on this board are from those who are anti rebate. Realistically, since takeouts aren't coming down anytime soon, rebates are the only thing that give horseplayers a fighting chance.

I have a book full of other comments about the current state of the game but I do my best not to get sucked into those discussions!

I don't understand why any serious player wouldn't take adavantage of whatever rebates were available to them - they're a major tax on any player who doesn't, as was frequently mentioned in a long thread here some time back.

[MJC922]

Originally Posted by MJC922

I'll provide a status on this from time to time. It had a good week, looks like it finally caught a couple of decent prices and got into the black, the totals after two weeks now look as follows:

# of Bets Win % $1 ROI Wagered Payoff Profit/Loss
162 14.81% +0.03 $324.00 $332.90 $+8.90

10 of the 25 (40%) winners went off below the requested 5-1 conditional minimum odds.

Reporting back on the current status, dreadful numbers for October:

# of Bets Win % $1 ROI Wagered Payoff Profit/Loss
352 12.22% -0.11 $702.00 $624.60 $-77.40

15 of the 44 (34%) winners went off below the requested 5-1 conditional minimum odds. Median profit per winner is down 11% from the original sample and average profit per winner is down 17% from the original sample.

What sticks out so far, probably the biggest single thing triggering the losing month was the WPCT of 12.2% resulting in approximately 7 fewer winners than expected which would project to break-even or slightly ahead for the month. On the other hand wpct being a couple pts low is not at all unusual after just 352 races.

The way I'm tracking this isn't so good either. It's going to require a lot more effort to find out some things that I feel are important to interpreting the results. I see Twinspires doesn't retain a record of all the drafted wagers that were submitted, only the ones that were actually sent to tote.

[lansdale]

Originally Posted by MJC922

Reporting back on the current status, dreadful numbers for October:

# of Bets Win % $1 ROI Wagered Payoff Profit/Loss
352 12.22% -0.11 $702.00 $624.60 $-77.40

15 of the 44 (34%) winners went off below the requested 5-1 conditional minimum odds. Median profit per winner is down 11% from the original sample and average profit per winner is down 17% from the original sample.

What sticks out so far, probably the biggest single thing triggering the losing month was the WPCT of 12.2% resulting in approximately 7 fewer winners than expected which would project to break-even or slightly ahead for the month. On the other hand wpct being a couple pts low is not at all unusual after just 352 races.

The way I'm tracking this isn't so good either. It's going to require a lot more effort to find out some things that I feel are important to interpreting the results. I see Twinspires doesn't retain a record of all the drafted wagers that were submitted, only the ones that were actually sent to tote.

Hi MJC,

You seem to realize this, but you really do have to put these results in perspective. This number of bets is approximately 4 hours of action for the average BJ pro and 1.5 days for the average poker player. Many successful players have found the 2500-3000 race sample size as an acceptable ballpark figure for model validation. Remember that is roughly the size of your Model 2 sample, so I wouldn't have any concern until you start to approach it - about 7 months from now, at the current rate.

A few things to possibly consider:

- seasonal difference - we mentioned. Your original sample was May-Oct. - this one will go throught the winter.

-timeframe - we mentioned. If this sample was pre-2008, you can assume that the ROI will be lower - you mentioned the 6% overall decline in ROI during the post-2008 period.

One important stat you haven't discussed much is that of losers that dropped after the close. I've been assuming that there couldn't be many of these, but if there were, that would obviously explain something. But the most notable difference thus far in this sample is the drop in the odds of the average winner. You can verify, but Model 2 average winners odds appear to have been ca. 8-1, in this sample they're about 6.5-1. That's not much, and it should be easy to see that a slight shift would be enough to put you over the top.


Thanks very much for posting.

Cheers,


lansdale

[MJC922]

Hi lansdale, agree with everything you said.

It's easy to anticipate the average odds on winners to drop in real world play because even on the original sample when I did a search for all of the selections that were within 1 point of 5-1 at final odds it was something like one third or thereabouts. Given that info I suspected there were going to be a very substantial number of plays hovering right around the cutoff at 0 MTP and dropping below it thereafter.

Coming along with that expectation however, I was also looking for (and still am) a slight bump in win percentage to maybe 15% from the 14% -- instead we're tracking at just 12%. I'll post once monthly on it from here on out.

To get a line on the seasonal impact I should re-run the model over the entire year, there's no real reason I didn't do it this time other than to dodge the effort, the day job has me run-down a bit but over the holidays I might look at that. Winter has been brutal on other models I've looked at in the past. It might even make sense to have a different model for each season, or if winter is impossible to drop it altogether.