hermes
27th January 2004, 06:23 PM
I notice lots of people on this forum and elsewhere plugging away at their systems without the benefit of a big database. Me too. I want to say to these people that you don't need a big database. You just need some scientific sampling.
The science is the same as that which underlies opinion polling and market research. Opinion Pollsters can quite accurately guage the opinions of a 20 million. population using samples as small as 2000. They have a good record of doing so.
Like others I design and try lots of systems. Most fail. I test them like this:
1. First sample = 150 recent races (or bets) If this shows miserable results, forget it. If it gets near to break even or pays POT continue. This sample will include at least one period of three consequitive days.
2. Second sample = 100 races from the same season in previous years (not recent races) but with NO consequitive days. If its near to break even or in POT, continue.
3. Third sample = 100 random races from all seasons trying only one or two races per day from here and there. If its near to break even or in POT, continue. A system MUST pass this test. (If you only sample consequitive days you can hit distorting cycles and mini-cycles. You have to sample against this phenomenon.)
4. Fourth Sample = carefully tabulate results over the last week's races, going back over some days covered in sample 1 but writing it up on a spreadsheet to examine results more closely.
In all cases under-state long-priced winners. It saves averaging results. Set a rule such as "All results over $20 will be recorded as $15 winners". This is essential. The greatest danger in sampling (not a danger in big databases) is that long priced winners blow out your results - they are not really part of the longer strike pattern but just flukes peculiar to that sample.
Also in all samples of 100 I begin by marking down 4 straight losses. This is an added margin of error.
If a system makes it this far and shows a reasonable result I then try it on paper only over a week's live races, continuing the spreadsheet of Sample 4.
If its still ahead then I try it at 50c units in live races. Slow and careful. I used to lash out and say "I have confidence in my methods!" Not any more.
The risk is that you hit a boom day and think you've found the holy grail and overeact. Experience will teach you that many systems romp through the sampling, perform well in live races for a fortnight and then die.
The thing to look for in live sampling is not POT but the same PATTERN of results as in the past results samples. If it shows the same pattern, persist.
Obviously systems that chase longshots require bigger samples to look at longer cycles and patterns.
Here's a trap for new players:
In sampling systems it is easiest to move through one meeting at a time race 1 to race 8 looking for cases of the configuration you are testing.
But in live racing punters tend to move from meeting to meeting these days, so race 1 of the day is at Mornington and race 2 of the day is at Townsville and race 2 is at Randwick, etc.
In which case you cannot apply the patterns of your samples to live racing. If you test it meeting by meeting you will have to bet it meeting by meeting and not across meetings, especially if you are parlaying or loss-chasing etc or depend on a certain strike-rate.
The limitation of sampling is that you tend to only try systems with simple, easy-to-look-up rules rather than complex, finely-tuned rules. But there's nothing wrong with simplicity, I say.
I find the TABQ records the best to sample from. Its an on-line data base of the last two or so years results.
Sampling can be mechanical and time-consuming, but I listen to music and "multi-task". I'm a chronic insomniac anyway and late night TV just isn't what it used to be.
The great advantage to this way of doing things is that my eye passes over hundreds and hundreds of results and fields. While testing one system new observations arise and new ideas come up. You notice patterns. That doesn't happen if you just click the button on a database. I feel like I'm learning more about the game while sampling.
The thing to remember, I reckon, is that in this country we are not playing against the track or the TAB (except as taxpayers) but against each other. This means that you can only win by one of two strategies:
*Do what other punters do and do it better than them.
*Do something other punters aren't doing, exploiting a niche.
Most of my system ideas are looking for a niche.
Happy punting. May you win, but not at my expense. Pari-Mutuel.
Hermes
The science is the same as that which underlies opinion polling and market research. Opinion Pollsters can quite accurately guage the opinions of a 20 million. population using samples as small as 2000. They have a good record of doing so.
Like others I design and try lots of systems. Most fail. I test them like this:
1. First sample = 150 recent races (or bets) If this shows miserable results, forget it. If it gets near to break even or pays POT continue. This sample will include at least one period of three consequitive days.
2. Second sample = 100 races from the same season in previous years (not recent races) but with NO consequitive days. If its near to break even or in POT, continue.
3. Third sample = 100 random races from all seasons trying only one or two races per day from here and there. If its near to break even or in POT, continue. A system MUST pass this test. (If you only sample consequitive days you can hit distorting cycles and mini-cycles. You have to sample against this phenomenon.)
4. Fourth Sample = carefully tabulate results over the last week's races, going back over some days covered in sample 1 but writing it up on a spreadsheet to examine results more closely.
In all cases under-state long-priced winners. It saves averaging results. Set a rule such as "All results over $20 will be recorded as $15 winners". This is essential. The greatest danger in sampling (not a danger in big databases) is that long priced winners blow out your results - they are not really part of the longer strike pattern but just flukes peculiar to that sample.
Also in all samples of 100 I begin by marking down 4 straight losses. This is an added margin of error.
If a system makes it this far and shows a reasonable result I then try it on paper only over a week's live races, continuing the spreadsheet of Sample 4.
If its still ahead then I try it at 50c units in live races. Slow and careful. I used to lash out and say "I have confidence in my methods!" Not any more.
The risk is that you hit a boom day and think you've found the holy grail and overeact. Experience will teach you that many systems romp through the sampling, perform well in live races for a fortnight and then die.
The thing to look for in live sampling is not POT but the same PATTERN of results as in the past results samples. If it shows the same pattern, persist.
Obviously systems that chase longshots require bigger samples to look at longer cycles and patterns.
Here's a trap for new players:
In sampling systems it is easiest to move through one meeting at a time race 1 to race 8 looking for cases of the configuration you are testing.
But in live racing punters tend to move from meeting to meeting these days, so race 1 of the day is at Mornington and race 2 of the day is at Townsville and race 2 is at Randwick, etc.
In which case you cannot apply the patterns of your samples to live racing. If you test it meeting by meeting you will have to bet it meeting by meeting and not across meetings, especially if you are parlaying or loss-chasing etc or depend on a certain strike-rate.
The limitation of sampling is that you tend to only try systems with simple, easy-to-look-up rules rather than complex, finely-tuned rules. But there's nothing wrong with simplicity, I say.
I find the TABQ records the best to sample from. Its an on-line data base of the last two or so years results.
Sampling can be mechanical and time-consuming, but I listen to music and "multi-task". I'm a chronic insomniac anyway and late night TV just isn't what it used to be.
The great advantage to this way of doing things is that my eye passes over hundreds and hundreds of results and fields. While testing one system new observations arise and new ideas come up. You notice patterns. That doesn't happen if you just click the button on a database. I feel like I'm learning more about the game while sampling.
The thing to remember, I reckon, is that in this country we are not playing against the track or the TAB (except as taxpayers) but against each other. This means that you can only win by one of two strategies:
*Do what other punters do and do it better than them.
*Do something other punters aren't doing, exploiting a niche.
Most of my system ideas are looking for a niche.
Happy punting. May you win, but not at my expense. Pari-Mutuel.
Hermes