"The Good Day" LTD system

[Ryan]

@lee-woodman Nice list that.

[A Former User]

Quite the list today! Russia Orenburg Ural
Australia Perth Glory Adelaide United
Russia Dynamo Moscow FK Rostov
Singapore Home Utd Tampines
Bulgaria Etar Lok. Plovdiv
Denmark Odense Esbjerg
Romania Din. Bucuresti Gaz Metan Medias
Switzerland Servette Lausanne
Belgium St. Liege Gent
France Clermont Troyes
France GFC Ajaccio Chateauroux
France St Etienne Montpellier
France AC Ajaccio Lens
Spain Granada CF Tenerife
Chile Antofagasta Huachipato

Make sure you are happy the teams have something to play for and that you want to trade that league. Australia & Belgium are play-off games so may be ones to avoid

[A Former User]

@Richard-Latimer thanks - yes it does seem really odd but like you say, all you can do is apply a consistent approach

[Richard Latimer]

Tomorrow starts at 09:30am in Russia then nothing until the evening:

Orenburg v Ural
St. Liege v Gent

[Richard Latimer]

@will-b said in "The Good Day" LTD system:

@Richard-Latimer i’m New to the forum and I’ve been following this strategy along with some others with interest. Forgive me if this has been discussed before but when I try to recreate your selections the expected goals data I download quite often differs from yours. For example for the Mioveni match today you made the selection based on expected goals of 2.97 but when I ran the data earlier today the expected goals for this match were 1.49, looking now this match is still showing as 1.49

I assume the difference arises because some of the data and calculations are updated over night but this seems like a huge movement and obviously in this case it’s the difference between a match appearing as a valid selection or not. Just wondering if you or anyone else can shed any light?

Firstly they are not my seleections. Credit must go to @Chris-Watts for the idea. I have just streamlined what he has, added a couple leagues perhaps and given my own take while he is away. Also like you say, I run my data the day before so I can get overnight games too. Plus it helps me to have things organised so far in advance as every day is busier than the last with a wife, 2 kids, 2 cats and a crazy puppy haha!!

Not sure what to tell you. I raised the very same issue only recently. Something happens overnight and some matches change dramatically. In the interest of consistency I have to keep it so data is all from same time period.

[A Former User]

@james-h as you have identified, the good day system can largely be tested using just the stats downloaded from BTC. Unfortunately it doesn't give match odds so you can calc the draw% but not the profit and loss. To get a fuller picture I have also downloaded stats from http://www.football-data.co.uk It is then possible, albeit a bit fiddly, to auto match the two sets of data and this gives a pretty complete picture for each match. In excel you can then analyse the data using pivot tables which are an enormously powerful way of very quickly testing out different factors in a system.

I think you are correct that testing other leagues will be important to get a full picture. It may very well be that the English Leagues are an anomaly. Only problem is that the amount of data available for each foreign league is a bit limited since I would typically only consider the top league or maybe league 2. As an e.g. in 17/18 & 18/19 in the Bundesliga there have only been 96 matches in total that would have been flagged up by the system and so any results must be suspect at best. Having said that, any data is better than none, so I would be really interested to see whatever you come up with. I have done all the English leagues, Spanish la liga and segunda and the Bundesliga. I'm planning to do Serie A and the main French league. Ultimately what I am hoping to build is a system that allows pretty much any system to be easily(ish) backtested.

[A Former User]

@Richard-Latimer i’m New to the forum and I’ve been following this strategy along with some others with interest. Forgive me if this has been discussed before but when I try to recreate your selections the expected goals data I download quite often differs from yours. For example for the Mioveni match today you made the selection based on expected goals of 2.97 but when I ran the data earlier today the expected goals for this match were 1.49, looking now this match is still showing as 1.49

I assume the difference arises because some of the data and calculations are updated over night but this seems like a huge movement and obviously in this case it’s the difference between a match appearing as a valid selection or not. Just wondering if you or anyone else can shed any light?

[Richard Latimer]

@richard-latimer said in "The Good Day" LTD system:

I've got Viborg and Mioveni tomorrow. Both at 17:00.

Injury time winner for viborg means it's 2 out of 2 for today.

[Richard Latimer]

@james-h said in "The Good Day" LTD system:

@richard-latimer cheers Richard, was having a problem viewing the Copy button appearing as Flash wasn't running correctly. Can see it now

If you look at my spreadsheet I've set formulas for everything so I paste the raw material and then move it along and paste existing formulas down.

[A Former User]

@richard-latimer cheers Richard, was having a problem viewing the Copy button appearing as Flash wasn't running correctly. Can see it now

[Richard Latimer]

@james-h said in "The Good Day" LTD system:

@denis-caunce love how in depth you’ve been going with these posts. Is it possible for you to try something similar with other leagues?
I was under the impression that this strategy was an amalgamation of many, many leagues so there would probably be some unprofitable ones in there.
Perhaps if we can get a league-by-League breakdown we could achieve a better strike rate and ROI?
Happy to work on a few of the leagues myself but I’m struggle to download the ratings - how do you do it? At the minute I’m just copy and pasting 100 at a time from the web page which is not ideal.

Just press copy (there's a tab on the ratings page) and then paste it into a spreadsheet.

[A Former User]

@denis-caunce love how in depth you’ve been going with these posts. Is it possible for you to try something similar with other leagues?
I was under the impression that this strategy was an amalgamation of many, many leagues so there would probably be some unprofitable ones in there.
Perhaps if we can get a league-by-League breakdown we could achieve a better strike rate and ROI?
Happy to work on a few of the leagues myself but I’m struggle to download the ratings - how do you do it? At the minute I’m just copy and pasting 100 at a time from the web page which is not ideal.

[Richard Latimer]

@denis-caunce said in "The Good Day" LTD system:

@richard-latimer It's really difficult to draw conclusions from the data. It does certainly feel that English league odds are pretty accurate (the biggest single indicator of a draw that I have found is the odds being offered) but it also feels as though the very competitive English league games are difficult to predict. All the indicators that you would expect to use to predict a draw (league pos, ELO, attack v Defence etc..) do indeed correlate to draw% but only weakly. So if we try to predict draws from these measures we increase our hit rate a little but our loss more.

E.g. If you look at the table below which shows he ELO difference V draw % and draw P&L for a £10 stake you will see that there isn't any obvious pattern. It is true that the closer matchups do give higher than average number of draws but not appreciably higher (obviously we need to understand that the number of games analysed is still pretty low and so variance is undoubtedly built in). The supposition is that poorer odds then kick in to the extent that there would be a loss of £462 if we bet on the draw for all games where the difference as less than 60.

ELO diff Draw% Draw P&L
10 27.82% -86.2
20 29.27% 101.3
30 28.01% -20
40 27.07% -144.7
50 25.50% -293
60 27.59% -20.1
70 31.82% 354.2
80 25.70% -167.4
90 23.44% -297.1
100 30.20% 214
110 18.18% -541.1
120 25.34% -98.6
130 23.21% -136.1
140 29.76% 95.2
150 20.43% -212.9
160 26.67% 35.7
170 29.31% 92.2
180 23.40% 13.8
190 29.17% 75.9
200 41.94% 256.1
300 17.53% -289.1

I have looked at the difference in the league position between the teams. Again it shows marginally higher than average number of draws where the difference in position is low but a loss if we bet on them. It is interesting that the good day system which tries to identify close matches but where there will be goals (and so less draws) actually seems to predict the number of draws (31.41% in England) better than pretty much anything I have been able to find. Obviously this wasn't it's intention.

If we are betting on draws (I know this thread is about laying them) we seem to be better in England trying to identify where there is a slight mismatch between the teams. It is in this range where we seem to get the best balance between hit rate and profit.

I am now going to move onto looking at the time of each team since their last draw. I don't believe this impacts the chances of them having a draw next but I have seen suggestions that the odds improve in these cases and so give extra value.

I would be interested to hear what others who are doing backtesting or other analysis are finding. Happy to share my data if anyone is interested. Since I haven't yet been able to pin a strategy down it feels a bit like I am ending up in 'paralysis by analysis'.

Lucky we're not completely relying on the English leagues then. Next few weeks and months could be quiet though.

[Richard Latimer]

I've got Viborg and Mioveni tomorrow. Both at 17:00.

[A Former User]

@richard-latimer It's really difficult to draw conclusions from the data. It does certainly feel that English league odds are pretty accurate (the biggest single indicator of a draw that I have found is the odds being offered) but it also feels as though the very competitive English league games are difficult to predict. All the indicators that you would expect to use to predict a draw (league pos, ELO, attack v Defence etc..) do indeed correlate to draw% but only weakly. So if we try to predict draws from these measures we increase our hit rate a little but our loss more.

E.g. If you look at the table below which shows he ELO difference V draw % and draw P&L for a £10 stake you will see that there isn't any obvious pattern. It is true that the closer matchups do give higher than average number of draws but not appreciably higher (obviously we need to understand that the number of games analysed is still pretty low and so variance is undoubtedly built in). The supposition is that poorer odds then kick in to the extent that there would be a loss of £462 if we bet on the draw for all games where the difference as less than 60.

ELO diff Draw% Draw P&L
10 27.82% -86.2
20 29.27% 101.3
30 28.01% -20
40 27.07% -144.7
50 25.50% -293
60 27.59% -20.1
70 31.82% 354.2
80 25.70% -167.4
90 23.44% -297.1
100 30.20% 214
110 18.18% -541.1
120 25.34% -98.6
130 23.21% -136.1
140 29.76% 95.2
150 20.43% -212.9
160 26.67% 35.7
170 29.31% 92.2
180 23.40% 13.8
190 29.17% 75.9
200 41.94% 256.1
300 17.53% -289.1

I have looked at the difference in the league position between the teams. Again it shows marginally higher than average number of draws where the difference in position is low but a loss if we bet on them. It is interesting that the good day system which tries to identify close matches but where there will be goals (and so less draws) actually seems to predict the number of draws (31.41% in England) better than pretty much anything I have been able to find. Obviously this wasn't it's intention.

If we are betting on draws (I know this thread is about laying them) we seem to be better in England trying to identify where there is a slight mismatch between the teams. It is in this range where we seem to get the best balance between hit rate and profit.

I am now going to move onto looking at the time of each team since their last draw. I don't believe this impacts the chances of them having a draw next but I have seen suggestions that the odds improve in these cases and so give extra value.

I would be interested to hear what others who are doing backtesting or other analysis are finding. Happy to share my data if anyone is interested. Since I haven't yet been able to pin a strategy down it feels a bit like I am ending up in 'paralysis by analysis'.

[Richard Latimer]

I got nothing for tomorrow.

[Richard Latimer]

@james-h said in "The Good Day" LTD system:

@richard-latimer I wonder if English leagues are less successful because the prices are more accurate and hence offer less value? I’ve found the same as you - anything that I try on the English leagues (as well as the top Spanish and German divisions) seem to offer very little value.

I'm not sure it can be just prices as this wouldn't account for a poor strike rate.

[A Former User]

@richard-latimer I wonder if English leagues are less successful because the prices are more accurate and hence offer less value? I’ve found the same as you - anything that I try on the English leagues (as well as the top Spanish and German divisions) seem to offer very little value.

[Richard Latimer]

@richard-latimer said in "The Good Day" LTD system:

@denis-caunce said in "The Good Day" LTD system:

@chris-watts I am new to this trading malarky (I am ex match better and poker player) but am really keen to find a system that works. Yours sounds really interesting so I thought I would run some independent backtesting. I notice from the thread that some testing is going on but figured that another pair of eyes wouldn't hurt.

I have focused on the UK leagues (Prem, championship, L1, L2 & national league premier). I have downloaded all the results from those leagues for 17/18 and 18/19 so far, downloaded all the available ratings from BTC for those leagues and matched the data. There are a few gaps in the ratings so I am missing data for some fixtures, but still have 3896 matches. Of these we have 357 that meet the criteria (ELO diff < 100, Home att stronger than away def, away def stronger than home attack, Goal expectancy > 2.5).

Now the interesting bit....My data shows that these matches had a draw % of 31.37% which is considerably higher than the average for those seasons of 26.48%. It also shows , based on B365 odds, BETTING on the draw would have given a profit of £536.6 based on a £10 stake (ROI of ~15%) and LAYING the draw would have resulted in an equivalent loss. Counter-intuitively the sweet spot for catching draws in English leagues seems to be those matches where the teams are quite closely matched but not very. E.g. betting on the draw where the difference between the odds on the home win and the away win is between 1.4 and 2.2 would give a profit of £1086 (11.4% ROI).

I have also tested the Spanish (La liga & segunda) and the Bundesliga but there isn't enough data on those to draw any firm conclusions. They are showing a broadly net zero profit/loss for the system.

Also interestingly, those data items that you would expect to be a good indicator of a draw (e.g. goal expectancy) actually only appear to have a weak correlation. E.g. you would think that matches with a goal expectancy of 2.5 (low scoring games definitely give more draws) but the opposite is true. I suspect that much of the data we might rely on as part of a formula should be viewed cautiously.

I know from the thread that you are showing good profits from the system, albeit over a limited time/number of matches, so not sure what the discrepancy is. There are some options:

• I have made a pigs ear of the analysis
• My analysis of the English leagues is correct but the system still works on the non English leagues
• The system doesn't work and the results so far are a blip.

I'm really interested in your thoughts and those of other testers on this thread. Now I have it all set up I can pretty easily run different analysis so happy to explore further.

Hope it all makes sense.

Interesting.

Did you look at league position no ing of teams and the differential? Also as mentioned I have found 2.61+ on the goal difference to be best so far.

Lastly, almost any system I seem to look at doesn't give great results where the English leagues are concerned. Don't know why. Just seem too inconsistent from 1 week to the next.

Sorry for typos. I'm on phone. Meant goal expectancy.

[Richard Latimer]

@denis-caunce said in "The Good Day" LTD system:

@chris-watts I am new to this trading malarky (I am ex match better and poker player) but am really keen to find a system that works. Yours sounds really interesting so I thought I would run some independent backtesting. I notice from the thread that some testing is going on but figured that another pair of eyes wouldn't hurt.

I have focused on the UK leagues (Prem, championship, L1, L2 & national league premier). I have downloaded all the results from those leagues for 17/18 and 18/19 so far, downloaded all the available ratings from BTC for those leagues and matched the data. There are a few gaps in the ratings so I am missing data for some fixtures, but still have 3896 matches. Of these we have 357 that meet the criteria (ELO diff < 100, Home att stronger than away def, away def stronger than home attack, Goal expectancy > 2.5).

Now the interesting bit....My data shows that these matches had a draw % of 31.37% which is considerably higher than the average for those seasons of 26.48%. It also shows , based on B365 odds, BETTING on the draw would have given a profit of £536.6 based on a £10 stake (ROI of ~15%) and LAYING the draw would have resulted in an equivalent loss. Counter-intuitively the sweet spot for catching draws in English leagues seems to be those matches where the teams are quite closely matched but not very. E.g. betting on the draw where the difference between the odds on the home win and the away win is between 1.4 and 2.2 would give a profit of £1086 (11.4% ROI).

I have also tested the Spanish (La liga & segunda) and the Bundesliga but there isn't enough data on those to draw any firm conclusions. They are showing a broadly net zero profit/loss for the system.

Also interestingly, those data items that you would expect to be a good indicator of a draw (e.g. goal expectancy) actually only appear to have a weak correlation. E.g. you would think that matches with a goal expectancy of 2.5 (low scoring games definitely give more draws) but the opposite is true. I suspect that much of the data we might rely on as part of a formula should be viewed cautiously.

I know from the thread that you are showing good profits from the system, albeit over a limited time/number of matches, so not sure what the discrepancy is. There are some options:

• I have made a pigs ear of the analysis
• My analysis of the English leagues is correct but the system still works on the non English leagues
• The system doesn't work and the results so far are a blip.

I'm really interested in your thoughts and those of other testers on this thread. Now I have it all set up I can pretty easily run different analysis so happy to explore further.

Hope it all makes sense.

Interesting.

Did you look at league position no ing of teams and the differential? Also as mentioned I have found 2.61+ on the goal difference to be best so far.

Lastly, almost any system I seem to look at doesn't give great results where the English leagues are concerned. Don't know why. Just seem too inconsistent from 1 week to the next.