View Full Version : Chi test help
thorns
5th October 2009, 11:35 AM
Have been playing with a system that has shown some great results so far, so have been lookig at a few of the tests that can be done to see if the results are luck or not. Now, I grabbed this chi test from an old post that angry pixie put up.
http://www.propun.com.au/racing_forums/showpost.php?p=162274&postcount=30
Can someone check that I have done this right? As I intend using betfair, for normalising the market I used 107%, as the markets generally run at around 102%, plus then adding on 5% for commission (not sure if thats a correct way to do it or not?)
Here are the results to date.
System base stats:
Bets:176
Winners:46
SR:26.7
Av div:4.94 (betfair)
Av div:4.33 (supertab)
And my attempt at the calculations:
Observed wins:47
Observed loss:129
expected wins:33.2
expected loss:142.7
Chitest result: 0.008374369
Running the same calcs for the TAB divvy I get:
Observed wins:47
Observed loss:129
expected wins:38.74
expected loss:137.26
chitest result:0.132905506
So if my calcs are correct, that would suggest that the chance that the results are down to luck is 0.8% (betfair), and 13.2%(tab)?
Could one of the gurus check if these are correct?
Cheers
thorns
5th October 2009, 12:43 PM
Thinking about this more, I think I may have made an error with the betfair results. The divivdend shown there is the raw price, not with price with a 5% commission deducted. Getting a bit confused on whether using the raw divi, should the normalised market be 102 or 107%?
thorns
5th October 2009, 01:21 PM
Hmmm, looks like I may have used the inccorect divis as it seems it should be teh average of all slections, not just winners. In this case the av divis should be:
Betafair:5.79 (raw price)
SuperTAB:5.04
Using the supertab price I get a Chitest score of: <table x:str="" style="border-collapse: collapse; width: 54pt;" border="0" cellpadding="0" cellspacing="0" width="72"><col style="width: 54pt;" width="72"><tbody><tr style="height: 11.25pt;" height="15"> <td style="height: 11.25pt; width: 54pt;" x:num="5.4616704963651536E-2" align="right" width="72" height="15"><table x:str="" style="border-collapse: collapse; width: 54pt;" border="0" cellpadding="0" cellspacing="0" width="72"><col style="width: 54pt;" width="72"><tr style="height: 11.25pt;" height="15"> <td style="height: 11.25pt; width: 54pt;" x:num="5.4616704963651535E-4" align="right" width="72" height="15">0.000546</td> </tr></table>
</td> </tr></tbody></table>This seems far to small though so am thinking I have stuffed up somewhere.
jfc
6th October 2009, 11:16 AM
Hmmm, looks like I may have used the inccorect divis as it seems it should be teh average of all slections, not just winners. In this case the av divis should be:
Betafair:5.79 (raw price)
SuperTAB:5.04
Using the supertab price I get a Chitest score of: <table x:str="" style="border-collapse: collapse; width: 54pt;" border="0" cellpadding="0" cellspacing="0" width="72"><col style="width: 54pt;" width="72"><tbody><tr style="height: 11.25pt;" height="15"> <td style="height: 11.25pt; width: 54pt;" x:num="5.4616704963651536E-2" align="right" width="72" height="15"><table x:str="" style="border-collapse: collapse; width: 54pt;" border="0" cellpadding="0" cellspacing="0" width="72"><col style="width: 54pt;" width="72"><tr style="height: 11.25pt;" height="15"> <td style="height: 11.25pt; width: 54pt;" x:num="5.4616704963651535E-4" align="right" width="72" height="15">0.000546</td> </tr></table>
</td> </tr></tbody></table>This seems far to small though so am thinking I have stuffed up somewhere.
That is plain wrong.
It should be average price of winners only.
jfc
6th October 2009, 11:17 AM
Simply using a Binomial Confidence Limits calculator would show:
46/176 reduces to a strike rate of 19.81%.
So even at an average div of 4.94 your system would be considered unprofitable long term.
47/176 would be ~break-even.
thorns
6th October 2009, 11:46 AM
Simply using a Binomial Confidence Limits calculator would show:
46/176 reduces to a strike rate of 19.81%.
So even at an average div of 4.94 your system would be considered unprofitable long term.
47/176 would be ~break-even.Cheers JFC,
So given that, would you bin the system, or run more trials to get a better picture of what the actualy long term srtike rate would be?
Using the java stat page that I beleive you have linked to previously, if for example i enter 460/1760, I get a range of 24-28% (roughly) which would indicate a long term profit, if the divi holds up.
jfc
6th October 2009, 03:37 PM
Cheers JFC,
So given that, would you bin the system, or run more trials to get a better picture of what the actualy long term srtike rate would be?
Using the java stat page that I beleive you have linked to previously, if for example i enter 460/1760, I get a range of 24-28% (roughly) which would indicate a long term profit, if the divi holds up.
While the actual strike rate suggests profitability there's no point in rejecting the system. So just keep testing.
But don't think of betting until it clearly passes the Confidence Tool.
thorns
6th October 2009, 05:17 PM
Thanks JFC, will keep on with it and see how I get on.
Going back to the Binomial confidence, if it passes that test after just say 150 bets, would you actually consider it to be proftable long term just from that, and have the confidence to put real money on it?
I was always under the impression (possibly mistaken) of needing a much larger sample to really be able to back with confidence.
lomaca
6th October 2009, 06:39 PM
I was always under the impression (possibly mistaken) of needing a much larger sample to really be able to back with confidence.Thorns, if your system is based on statistics, then you have to use stats. and use as many races as you can to see if it's profitable .
On the other hand, if you base your system on form and handicapping, then it's easy to check if it was coincidence-luck whatever.
Your selections based on your system have to be consistently in the first four or five placegetters. ("consistently" doesn't, mean in every race, just most) that would assure you, to get a first selection strikerate you are talking about.
jfc
7th October 2009, 07:36 AM
Thanks JFC, will keep on with it and see how I get on.
Going back to the Binomial confidence, if it passes that test after just say 150 bets, would you actually consider it to be proftable long term just from that, and have the confidence to put real money on it?
I was always under the impression (possibly mistaken) of needing a much larger sample to really be able to back with confidence.
Number of winners is more relevant than number of runs, but the Confidence Tool is still superior.
Esentially you should start betting as soon as the system passes the confidence test - providing you are prepared to stop whenever it starts failing.
Imagine an extreme example where you need a S/R better than 20% to be profitable.
Then 3/3 would allow you to start betting.
But if you then lose, that 3/4 means you should stop.
While the Tool will correctly tell you when to switch between actual and theoretical betting, the vast majority of players would not be able to cope with such mood swings. Typically they prefer loss-chasing (under some fancy euphemism like "non-reducing" ) and ruining their lives.
vBulletin v3.0.3, Copyright ©2000-2025, Jelsoft Enterprises Ltd.