Welcome to the FTS Forum - Read the introduction here

Avatar
Please consider registering
Guest
Search
Forum Scope


Match



Forum Options



Minimum search word length is 3 characters - maximum search word length is 84 characters
sp_Feed
Topic Rating: +6 Topic Rating: +6 (6 votes) 
sp_TopicIcon
Data testing-how much is enough?
sp_NewTopic Add Topic
Avatar
Beginner
Members
Forum Posts: 10
Member Since:
August 2, 2013
sp_UserOnlineSmall Online
1
October 31, 2019 - 9:25 am
sp_Permalink sp_Print
+1
sp_QuotePost
sp_ReportPost

I think this section is the most relevant for the purpose of this post so apologies if not (it would be related to FDD but not exclusively which is why I’ve put it in here). My thoughts are it’s mainly related to excel and reporting of data so fits well here.

Anyway my subject for debate is the following:
How much of a sample size is needed to make people feel comfortable with a potential betting system and how do you go about creating your own? (using excel presumably). I’m not asking this as a newbie looking for advice as I’ve used soccer stato (though in hindsight nowhere near to the full potential it had) and have now built up a database of over 20,000 matches from FDD since it’s been available back in late 2017. Obviously on top of that there’s the Ultimate and Poisson databases too.
What interests me most is people’s interpretation of data. We all have pretty much the same info available, though for me the biggest eye opening moment was FDD having actual Betfair prices which removed the “estimated” approach to what odds were available for most systems-particularly stuff like Ht CS Lays.
The general consensus is always the bigger the sample size the better but what if your method only throws up as an example 4 bets per month due to whatever filters you have. Over a season you have say 45-50 bets for that one method. People will usually pluck a figure like 500 samples minimum before betting for real money (again this is just generalising for the purpose of a debate). Using this way of thinking you’d need 10 seasons’ worth of data which is never going to happen. Now I appreciate this scenario would probably only happen due to a very specific set of fiilters and there will of course be other methods which will have 30-40 possible bets each month. But the reality is, in a robust system or method, you might need a very long (and perhaps unrealistic) amount of testing to get a large sample size. Interested to know people’s opinions on this.
And likewise, how people decide to bet methods based on their interpretation of the data. We all know back testing is a minefield as Ian has said, the tendency can be to keep filtering until nearly all the losses are removed but of course we need to back test to gain an initial idea of what “might” work. Forward testing ideas then in real time (without betting them which can be very tough) is obviously the next step before betting for real. Or is it? How many people back test data be it on FDD/Ultimate or another source of research and base their methods on the historical data? What’s people’s experiences been with this approach?
Hopefully this thread can provoke some good ideas and debate

The following users say thank you to marko25 for this useful post:

jamiep, bettyblogger
Avatar
Beginner
Members
Forum Posts: 36
Member Since:
October 23, 2019
sp_UserOfflineSmall Offline
2
October 31, 2019 - 10:12 am
sp_Permalink sp_Print
+1
sp_QuotePost
sp_ReportPost

You can pull statistical significance into the argument based on an expected yield and average odds which will give you mathematical comfort.

The key for me is live forward tested samples. You can have 50 years of FDD but if you sit and filter a few filters to fudge a profit, 99 times out of 100 that will regress back to the mean. It is data mining. Trust me I have done it and seen systems crash after 2-3 years.

If you think you’ve spotted a system, forward test it for a year at least to see its validity, IMO.

The following users say thank you to Lord Dio for this useful post:

bettyblogger, Baldwinsdad
Avatar
Beginner
Members
Forum Posts: 10
Member Since:
August 2, 2013
sp_UserOnlineSmall Online
3
November 1, 2019 - 11:49 am
sp_Permalink sp_Print
+1
sp_QuotePost
sp_ReportPost

Yes obviously in an ideal scenario you’d be able to live test every method for a year or so but in reality who does actually do that with every single idea? Presumably most ideas for systems these days come from some form of excel database research and testing historical results without filtering it to death to remove all losers.
Ultimate methods are quite heavily reliant on back tested data (albeit going back a few seasons more than the current FDD data would) and I am fairly sure most who have devised their own Ultimate methods used the data as Ian has shown in his videos to look for profitable systems based on past results. Obviously the old “past results do not guarantee any future returns” phrase should make us all err on the side of caution but as I say above, at what point do people feel comfortable betting for real money?
Very interested to hear what FTS members experiences have been with this side of things…..Has anyone developed systems using Ultimate/FDD/Soccer Stato or any other data source and began betting based on historic data which seemed robust only for the method to go completely belly up over the course of a few seasons?
And of course, the other old chestnut would be does a method appear to collapse when in fact it’s just one bad season when you go “live” and in fact if you continue to monitor the method gets back on track? Again, I know Ian has always said he doesn’t concern himself with one season and sees things as more of a continuum of results but of course many of us would see something do well on paper, trial it and when it appears to have a bad few weeks or months, bin it and move on. Just wondering how many potentially valid systems have been ditched over the years….
It’s all quite interesting to me as the excel database approach and being able to collate your own databases has really only taken off in the last 5-6 years. Before that it would have all been labour intensive manual work but obviously people like Ian made that work but presuming without years of previous data or certainly to the extent we have available to us these days. Would be good to get Ian’s views eventually when he has some spare time.

The following users say thank you to marko25 for this useful post:

bettyblogger
Avatar
Beginner
Members
Forum Posts: 36
Member Since:
October 23, 2019
sp_UserOfflineSmall Offline
4
November 1, 2019 - 1:34 pm
sp_Permalink sp_Print
+1
sp_QuotePost
sp_ReportPost

I’ve had a different sort of approach with my other data source, away from FDD.

I made some filters around late 2017 maybe, for late goals specifically 80+, I recorded around 500-600 samples going forward for about 12 months and identified a slow decline in results, as expected really.

I made 1 simple tweak and forward tested them since March 2019, I’ve now got 386 samples and every single month has been very promising results with no sign of decline. The opposite of data mining really. If only it was always so simple.

My FDD late goal sheet started at around 52% (80+ goal is typically 40-42%) then slowly went, 50%, 49%, 48% etc and is now at 44% after 18 months. Data mining at its finest. Not a waste of time as such as it was a massive eye opener/learning curve.

The following users say thank you to Lord Dio for this useful post:

bettyblogger
Avatar
Beginner
Members
Forum Posts: 4
Member Since:
August 4, 2013
sp_UserOfflineSmall Offline
5
November 1, 2019 - 9:51 pm
sp_Permalink sp_Print
+1
sp_QuotePost
sp_ReportPost

I spent literally months during the summer of 2015 working with Soccerstato devising systems and it was very time consuming, as the data wasn’t presented in the same way it now is in FDD. Sure enough they all went downhill very fast during forward testing (and foolishly some real money betting). The issue with Soccerstato was the recorded Betfair price, which I believe was a harmonic mean. This made it almost impossible to identify accurate qualifiers going forward. FDD is a game changer, as every price is 5 mins before KO. I’ve now started forward testing some new stuff again and will see where I am next summer. It boils down to mindset again, as the number of false dawns you will have using back tested data can crush your spirit, but every one is a learning experience and the nuggets are there if you look hard enough and don’t give up.

Spookily I also have an 80+ goal filter at around 52% for the previous two seasons, so will be interested to see where this goes during the next 12 months after reading Dio’s comments.

The following users say thank you to danny2 for this useful post:

Lord Dio, bettyblogger
Avatar
Beginner
Members
Forum Posts: 36
Member Since:
October 23, 2019
sp_UserOfflineSmall Offline
6
November 1, 2019 - 10:20 pm
sp_Permalink sp_Print
+1
sp_QuotePost
sp_ReportPost

danny2 said
I spent literally months during the summer of 2015 working with Soccerstato devising systems and it was very time consuming, as the data wasn’t presented in the same way it now is in FDD. Sure enough they all went downhill very fast during forward testing (and foolishly some real money betting). The issue with Soccerstato was the recorded Betfair price, which I believe was a harmonic mean. This made it almost impossible to identify accurate qualifiers going forward. FDD is a game changer, as every price is 5 mins before KO. I’ve now started forward testing some new stuff again and will see where I am next summer. It boils down to mindset again, as the number of false dawns you will have using back tested data can crush your spirit, but every one is a learning experience and the nuggets are there if you look hard enough and don’t give up.

Spookily I also have an 80+ goal filter at around 52% for the previous two seasons, so will be interested to see where this goes during the next 12 months after reading Dio’s comments.

  

You need to up your game my son, my latest one that actually seems to work is 55% 😉

The following users say thank you to Lord Dio for this useful post:

danny2, bettyblogger
Avatar
Beginner
Members
Forum Posts: 21
Member Since:
March 19, 2019
sp_UserOfflineSmall Offline
7
November 18, 2019 - 11:11 am
sp_Permalink sp_Print
0
sp_QuotePost
sp_ReportPost

Lord Dio said

danny2 said

I spent literally months during the summer of 2015 working with Soccerstato devising systems and it was very time consuming, as the data wasn’t presented in the same way it now is in FDD. Sure enough they all went downhill very fast during forward testing (and foolishly some real money betting). The issue with Soccerstato was the recorded Betfair price, which I believe was a harmonic mean. This made it almost impossible to identify accurate qualifiers going forward. FDD is a game changer, as every price is 5 mins before KO. I’ve now started forward testing some new stuff again and will see where I am next summer. It boils down to mindset again, as the number of false dawns you will have using back tested data can crush your spirit, but every one is a learning experience and the nuggets are there if you look hard enough and don’t give up.

Spookily I also have an 80+ goal filter at around 52% for the previous two seasons, so will be interested to see where this goes during the next 12 months after reading Dio’s comments.

  

You need to up your game my son, my latest one that actually seems to work is 55% 😉

  

Hi Dio, 

With regards to building a late goal system what fdd sheet do you use the halftime correct score or the goal times?

Thanks

Avatar
Beginner
Members
Forum Posts: 36
Member Since:
October 23, 2019
sp_UserOfflineSmall Offline
8
November 18, 2019 - 3:30 pm
sp_Permalink sp_Print
0
sp_QuotePost
sp_ReportPost

jamiep said

Lord Dio said

danny2 said

I spent literally months during the summer of 2015 working with Soccerstato devising systems and it was very time consuming, as the data wasn’t presented in the same way it now is in FDD. Sure enough they all went downhill very fast during forward testing (and foolishly some real money betting). The issue with Soccerstato was the recorded Betfair price, which I believe was a harmonic mean. This made it almost impossible to identify accurate qualifiers going forward. FDD is a game changer, as every price is 5 mins before KO. I’ve now started forward testing some new stuff again and will see where I am next summer. It boils down to mindset again, as the number of false dawns you will have using back tested data can crush your spirit, but every one is a learning experience and the nuggets are there if you look hard enough and don’t give up.

Spookily I also have an 80+ goal filter at around 52% for the previous two seasons, so will be interested to see where this goes during the next 12 months after reading Dio’s comments.

  

You need to up your game my son, my latest one that actually seems to work is 55% 😉

  

Hi Dio, 

With regards to building a late goal system what fdd sheet do you use the halftime correct score or the goal times?

Thanks

  

Hi mate it wasn’t actually derived from FDD data at all. I did try and use the Goal Times sheet for a late goal method but badly regressed after 2 years sorry. 

The following users say thank you to Lord Dio for this useful post:

jamiep
Forum Timezone: Europe/London
Most Users Ever Online: 92
Currently Online: marko25, ArtTest, hirdyx, Mark, bladesmanstan, Neil, Beeble
7
Guest(s)
Currently Browsing this Page:
1 Guest(s)
Forum Stats:
Groups: 4
Forums: 22
Topics: 99
Posts: 510

 

Member Stats:
Guest Posters: 2
Members: 3587
Moderators: 1
Admins: 2
Administrators: Ian, Geeky Boy
Moderators: ArtTest