Anytime Goalscorer System

[harry_rag]

10 points on Pukki (Finland) at 7/4 Sky Bet (or 2.76 exchange) (U)

[harry_rag]

10 points on Gnonto (Italy) at 3.45 (U)

[harry_rag]

No joy with the last 4 bets so 125 winners from 390 bets (32.05%) for a loss of 100.69 points.

[harry_rag]

With thanks to @Valiant Thor for explaining the concept of AE, I decided to calculate it by price band for my sample of anytime goalscorers. This is what I came up with.

#	Range	Players	Expected	Actual	AE	P/L	E P/L
1	2.00	81	42	37	0.88	-£9.91	-£9.64
2	2.25	237	109	95	0.87	-£30.84	-£30.44
3	2.5	414	172	166	0.97	-£15.77	-£14.44
4	2.75	312	117	119	1.02	£4.25	£5.33
5	3	312	108	108	1.00	-£0.38	£0.00
6	3.25	122	39	39	1.00	-£0.50	£0.00
7	>3.25	97	28	30	1.07	£7.05	£6.93
		1575	615	594	0.966	-£46.10	-£53.78

Price ranges are up to evens, above evens up to 2.25 etc. Prices are "best price" from Oddschecker (I ignore "funny" bookies who I've never had an account with and seem to have a tendency to try and get "bold type" for their prices). P/L calculated to £1 level stakes. Expected and actual number of players who scored 1 or more goals.

Early conclusions would be:

• AE is reasonably accurate measure for calculating expected returns given how close the expected P/L is to the actual.
• There's a clear line between the returns from selections priced at 2.5 or lower (negative returns) and those priced >2.5 (neutral or positive) that suggests it may be a good idea to focus more on the bigger priced selections.
• It poses the question, if one wanted to keep it simple, might the best approach be just to try and back all players with a best price >2.5 with a 10% edge added! 

I'm off to run a few more checks.

 

[Valiant Thor]

1 hour ago, harry_rag said:

It poses the question, if one wanted to keep it simple, might the best approach be just to try and back all players with a best price >2.5 with a 10% edge added! 

If >2.75 is beating the market then where/what are you going to add this 10% extra edge  too.
If  2.75 top price has your added 10% then its 3.03 (2/1) , you cant get from a market whats not there
This is the problem with frequentist thinking , by trying to force another 10% onto 2.75 (36%) more money will be made , it will not .

The better option would be to take your above example and run it through all the bookmakers you use and find the Book with the worst return and use this as your metric, then if you are still getting a positive return @ 2.5> then you know another book will be giving a better price and therin lies your inbuilt 'value'.

ATB
VT

Edited July 6, 2022 by Valiant Thor
layout

[harry_rag]

3 minutes ago, Valiant Thor said:

If >2.75 is beating the market then where/what are you going to add this 10% extra edge  too.

Betfair! It's quite possible to beat the best bookies price (from Oddschecker) on the exchange, especially where I'm ignoring the "funny" bookies (you will often get their prices or better on the exchange).

This system has become quite sophisticated/complicated over time (delete as appropriate) in terms of the inputs that determine my view of fair and target prices. There is a category of bets where taking the best available bookies price is proving profitable so I see no reason to change that. (Sometimes I will be getting even better on BF). For all bets where my target price is higher than the best price (so it's BF or no bet) I've tweaked it to make sure I'm always looking for a minimum edge of 10% over the best price.

There are 2 things I'm currently trying to look closer at in the light of what I posted above:

• Might it be worth setting a minimum price point, e.g. stop backing the 2.5 or less selections?
   
• Should I set the maximum edge as "best" price +10%? There are examples where my target price is looking for an edge greater than 10%. IF matched, such selections would have an ROI of 17% compared to "only" 11% with a 10% edge. So I'm thinking I'll get matched more if I'm less ambitious in terms of the edge I'm looking for on the exchange but still be taking bets that have a decent AE!

[harry_rag]

One fairly simple change made. Should be back in action again with the Women's Euros starting today.

• Might it be worth setting a minimum price point, e.g. stop backing the 2.5 or less selections? - Maybe, there's an argument to stop backing at 2.25 or lower as they're borderline profitable to my current target odds. 2.26 to 2.5 is ok though. I'll carry on with these for now but with no change in terms of capping the maximum edge (see next point).
   
• Should I set the maximum edge as "best" price +10%? There are examples where my target price is looking for an edge greater than 10%. IF matched, such selections would have an ROI of 17% compared to "only" 11% with a 10% edge. So I'm thinking I'll get matched more if I'm less ambitious in terms of the edge I'm looking for on the exchange but still be taking bets that have a decent AE! - I've applied this change in respect of all players where the best bookie's price is >2.5 which seems logical given the AE numbers I posted earlier. It should mean I'm trying to get matched at a more realistic and attainable price.

[Valiant Thor]

1 hour ago, harry_rag said:

Might it be worth setting a minimum price point, e.g. stop backing the 2.5 or less selections?

If that is your base starting point then Yes

1 hour ago, harry_rag said:

Should I set the maximum edge as "best" price +10%? There are examples where my target price is looking for an edge greater than 10%. IF matched, such selections would have an ROI of 17% compared to "only" 11% with a 10% edge. So I'm thinking I'll get matched more if I'm less ambitious in terms of the edge I'm looking for on the exchange but still be taking bets that have a decent AE!

Where is this mythical 10% coming from, are you saying 9% or 8% would not be good enough or 11% would be too much?

An 'edge'/'value' can only exists after the event/events have occurred , you can not manufacture it by attaching some numerical significance to it, Its like gravity everyone knows its there but no one can explain how it works or manufacture it.

I have my bot set @ estimated price >1% to<30% anywhere between the 2
Why 1% thats because my estimated price is what Im willing to look to invest in and anything above that should create profit over time
Why 30% top , Its because from past calculations my algo has deduced that profitability reduces significantly when above this mark. ( on my personal selection/pricing methods)
ie book prices are rarely that far out of sync with reality or its just variance (sh*t happens)

All I know is that at this moment in time its running @ 16% yield with a numerical edge of 18%
I cannot say which had the most edge or whether they were down to variance or even from which bets the edge derived, all I can do is group them together to give it a number.

Edited July 6, 2022 by Valiant Thor

[harry_rag]

10 minutes ago, Valiant Thor said:

Where is this mythical 10% coming from, are you saying 9% or 8% would not be good enough or 11% would be too much?

Thanks for your patience and the input which is helping me understand my own data a bit better (I think).

The basic premise of this thread is to use the spread prices for player goal minutes to identify "true" odds for each player in the anytime goalscorer market. My target odds are essentially the true odds with a 10% edge added. Sometimes I can get an acceptable price with the bookies, other times I have to try and get matched on BF. So far, the system is theoretically profitable (in that you would make money backing every player at my target odds) but drifts in and out of profit in reality (as I only get matched on some of the selections).

What I decided to do today was look at the AE for the various price bands based on best bookies price which has shown that, at 2.75 or bigger, the selections are break even at worse even if you just took that price.

Because of that I have decided to cap my target price on BF as best bookies price +10%, so I will no longer be trying to get matched at, say, best price +15%. That should increase the frequency with which I get matched whilst retaining the expectation of a positive return.

Essentially there are two 10% figures at play; the original edge I applied to my attempt at framing "true" odds in the hope of returning a profit and the cap I have decided to apply to the prices I request on BF, i.e. the maximum will be 10% on top of best bookies price.

I suppose it boils down to why try and get matched at a price that is 15% above the best bookies price when the data suggest that 10% above would be profitable and is obviously more achievable. There's nothing magical about 10%, it could be 8% or 12%, but I've alighted on 10% for now.

[Valiant Thor]

1 hour ago, harry_rag said:

What I decided to do today was look at the AE for the various price bands based on best bookies price which has shown that, at 2.75 or bigger, the selections are break even at worse even if you just took that price.

I think this is the part you may be getting what I said mixed up.
We'll cut your data down to 3 imaginary games A,B,C with 2 winning

You go through all your data with each individual Bookie and find the worst one (lowest A/E) this will then be the book which have offered the lowest odds overall. (hardest to beat, Most accurate)
You would then use this book as your datum (take all your odds from this book only to make your A/E)

 

If you are still getting positive returns similar to the one you did with best odds, then great.
You then take the teams which in this case is PP have selected @ 2.75 and look up the prices in the other books which you know by picking the worst book will have better prices
Your getting +EV of approx 3% just by constructing your model from the worst book.

The idea is to try to get as near to break even as possible using the worst performing book as the datum or your model
so your virtually guaranteeing + EV on most bets.

If the worst book doesnt produce a near break even or +EV then move up to the next one etc
NEVER make your model using best prices
Where are you going from there , nowhere , your trying to add a %age that doesnt exist
You go on BF and pinch an odd point or 2 extra, if your model is from the worst odds your getting the 2 points your pinching on top of the EV 3% from the others

 

Edited July 6, 2022 by Valiant Thor
grammar

[harry_rag]

46 minutes ago, Valiant Thor said:

I think this is the part you may be getting what I said mixed up.

I don't think I'm getting what you're saying mixed up, I'm probably just not doing a great job of explaining what I'm doing! What you've suggested is an interesting idea but it would be a substantially different system to the one I'm operating if it used the fixed odds prices as a primary input. I also don't record each individual bookmaker's price in the data I'm gathering. Think of it this way:

• I have my own model to calculate "true" odds for any given player to score a goal that isn't based on the prices in the fixed odds market (that's a grand way of putting it; I use the spread betting prices and convert them to an anytime fixed odds equivalent)
• I add a 10% edge to my "true" price to give my "target" price, the one I am willing to back that player at
• My true odds are fairly accurate. If I was able to back every player at my target price it would be profitable but, obviously, I can't always get my target price.
• Sometimes my target price will be >10% greater than the best price being offered by any of the fixed odds firms. The more "greedy" I am in that sense, the less likely it is that I will get matched at my target price.
• Based on my analysis of the AE for the various price bands I have decided to cap my target price at best bookie odds +10%

Apologies if I've made it sound like my approach is fundamentally based on the fixed odds prices, it's not. They're just used as a reference point.

[harry_rag]

A few thoughts to wrap up yesterday's analysis, probably of no interest to anyone else but no harm in collecting my thoughts and posting for future reference before the new season starts.

• My data sample now stands at 1577 players. The criteria for inclusion is player goal minutes mid-point is between 20 and 35. That roughly equates to anytime goalscorer "true" odds between 2.03 and 3.16
• Backing every player at my current "target" odds would show a profit of 136.3 points to 1 point level stakes with an ROI of 8.64% (with 595 players scoring or 37.73%).
• My current actual return (trying to get £10 on) is 125 winners from 390 bets (32.05%) for a loss of £100.69 with an ROI of -2.65%
• 317 of the bets have been on BF with 100 winners (31.55%) for a loss of 202.48 points with an ROI of -6.59%
• 73 of the bets have been with bookies with 27 winners (36.99%) for a profit of 101.79 points with an ROI of 14.05%
• Selling every player's goal minutes for 1 point stakes would have returned a profit of 5183 points. 

Bottom line is I need the actual strike rate to increase to something closer to the overall data strike rate; 36% or more should be enough for profit. The question is whether it's currently lagging through bad luck or because the market is efficient enough to turn a theoretical profit into an actual loss. I'll aim for 500 bets and see how things stand then but I'll review if the loss hits -200 points.

[harry_rag]

At last, a bet! It's been over a month.

10 points on Diani (France) at 9/4 with Sky Bet.

[harry_rag]

A quick reminder of where we're up to before the new season gets underway and the recent "formula" tweaks get put to the test.

125 winners from 391 bets (31.97%) for a loss of 110.69 points.

I also started tracking the returns for some less likely scorers (goal minutes in the range 15-19 as opposed to the original 20-35) and have been backing them "offline". I've decided I might as well post those as well. So we have the original "Variant A" bets and the new "Variant B" bets which I'll record from now on. There won't be that many of the latter based on experience so far. 

All bets will be of the original "A" variety unless clearly stated.

[harry_rag]

10 points each on Salah at 3.422 and Mahrez at 4.6 (Mahrez is a "B" bet).

Unlikely to get the target 2.6 on Haaland. On the basis of today's game, the spread firms are putting him in for 32% of City's goals, I think he'll do well to achieve that. Though it is similar to their take on Salah for Liverpool.

[harry_rag]

German Super Cup:

10 points each on Nkunku at 3.4 and Mane at 2.8

Muller a B bet if I get matched at 4. 

[harry_rag]

5 hours ago, harry_rag said:

10 points each on Salah at 3.422 and Mahrez at 4.6 (Mahrez is a "B" bet).  

2 hours ago, harry_rag said:

10 points each on Nkunku at 3.4 and Mane at 2.8  

10 points on Duran (Chicago Fire) at 11/4 Uni (B bet)

[harry_rag]

3-0 on the "A" bets and 0-2 on the "B" bets since the last update.

128 winners from 394 "A" bets (32.49%) for a loss of 45.79 points.

0 winners from 2 "B" bets (0%) for a loss of 20 points.

[harry_rag]

10 points on Atik (FC Magdeburg) at 9/2 Sky Bet ("B" bet).

[harry_rag]

10 points on Dennis (Watford) at 3.7

[harry_rag]

After a few hours number crunching today I’ve reached the conclusion that it may be time to admit defeat with this system because a much simpler and more profitable approach has been hiding in the data! I’ll have to recheck my maths tomorrow but just wanted to set my thoughts down while it’s going round in my head.

• The current system is theoretically profitable in that if you backed every one of the 1616 players at my target odds you’d be 137.17 points up with an ROI of 8.5%. In practice though, I’m around 5 points down from 394 actual bets (based on 1 point level stakes where I’m trying to bet £10 per point).
• Obviously it’s easy to come up with a theoretically profitable system (e.g. back Harry Kane to score in every game at 5/1 or better would be a belter) but harder to realise an actual profit. So far the evidence suggests this system is likely to fall just short of break even.
• I decided to see how a fellow punter’s “Rule of 70” system would have performed with my data. To cut a long story short you do a simple sum and if the result is 70 or more you have a bet.
• It would have resulted in 463 bets for a profit of 12.29 points with an ROI of 2.65%. That’s actual rather theoretical profit as it’s based on best bookie’s price. It might have been possible to get better still on BF.
• Having reviewed a range of numbers I alighted on using 72 rather than 70. This reduced the number of bets to 334 but increased the profit to 31.2 points with an ROI of 9.3%.
• As this was loss making at the shorter end of the price spectrum I looked at various options for a minimum price to apply and decided on 2.3 which reduced the number of bets to 287 but increased the profit to 40.8 points with an ROI of 14.2%. This supports the observations when I looked at AE for the different price ranges.
• So, to sum up, I’m trying to see if there’s any reason not to switch from my system (complex and theoretically profitable but not in reality) for another system which would appear to be simpler and profitable in reality. Just tweak the “magic number” from 70 to 72 and apply a minimum price of 2.3.

[harry_rag]

22 hours ago, harry_rag said:

I looked at various options for a minimum price to apply and decided on 2.3 which reduced the number of bets to 287 but increased the profit to 40.8 points with an ROI of 14.2%. This supports the observations when I looked at AE for the different price ranges.

Worth observing that the current system flagged 205 of these bets (71.4%) so it's not as if following the "rule of 72" will flag up an enormous number of additional bets I wouldn't have taken anyway, but fair to observe that the 82 bets I ignored had an ROI of just shy of 10%. So it does seem like a no brainer to back all the qualifiers regardless of whether they meet the existing target odds criteria.

Also worth pointing out that these bets are still theoretical to an extent, in as much as the best price could be with a bookie who I can't get on with to the desired stake. Less of a problem in games where there's a market on the exchange but that won't be the case in "lesser" competitions.

Case in point (and a bet should anyone wish to follow) is De Wit at 13/5 with PP for AZ Alkmaar v Dundee U. Would be a bet under old and new approach but I'm not bothering when I'm allowed a paltry 77p.

No bets for me today, review ongoing.

 

[harry_rag]

On 8/1/2022 at 7:46 PM, harry_rag said:

10 points on Dennis (Watford) at 3.7

That was actually a "B" bet (where I've just started seeing how the less likely scorers perform). Results so far:

128 winners from 394 "A" bets (32.49%) for a loss of 45.79 points.

0 winners from 4 "B" bets (0%) for a loss of 40 points. 

Post to follow regarding the recent review of the bet selection process.

[harry_rag]

On 8/3/2022 at 8:41 PM, harry_rag said:

After a few hours number crunching today I’ve reached the conclusion that it may be time to admit defeat with this system because a much simpler and more profitable approach has been hiding in the data!

On reflection, it's more of a change to the selection process than a completely new system so it seems better to make the change and carry on, to see if that can get the thread back into the black.

On 8/3/2022 at 8:41 PM, harry_rag said:

So, to sum up, I’m trying to see if there’s any reason not to switch from my system (complex and theoretically profitable but not in reality) for another system which would appear to be simpler and profitable in reality. Just tweak the “magic number” from 70 to 72 and apply a minimum price of 2.3.

This will now be the primary criteria for a bet; all selections where the "R" number is 72+ and best bookies price is at least 2.3. As stated this has given 287 bets showing a profit of 40.8 points to prices available with bookies. 205 of them would have been selections based on the current criteria anyway but that leaves 82 bets (+8.18 points) that I wouldn't have been on unless I got my target odds on BF.

Where "R" is <72 I'm going to persevere with the existing criteria which will mean trying to get matched at my target odds on the exchange. It may well prove to be the case that it's not worth continuing with this approach but it's worth seeing how things unfold over the next 100 bets or so to give the tweaks I made at the end of last season a chance to prove any possible worth.

I doubt anyone will be rushing to follow these selections but I'll post along the following lines so it's clear which type of bet it is:

10 points on Kane to score at 6/4 with Hills (74A)

Bottom line, if you only wanted to follow the more proven bets then stick to A bets only that are rated 72+.

That's it for the navel gazing for now, let's see how it plays out now the "major leagues" are getting back underway.

[harry_rag]

10 points on Mane at 27/20 with Uni (80A) and 10 on Gnabry at 2.8 (62A)

Both for Bayern with Muller pending at 3.5.

[harry_rag]

10 points on Lacazette (Lyon) at 2.76 (64A) 

[harry_rag]

3 winners last night making it 6 out of 6 for the last half dozen "A" bets.

131 winners from 397 "A" bets (33%) for a profit of 2.6 points.

0 winners from 4 "B" bets (0%) for a loss of 40 points. 

[harry_rag]

10 points on Diaz (Liverpool) at 3 on BF (74A). Fine at 2.95 with Uni

[harry_rag]

10 points on Firmino at 2.94 (<72A)

Not got spreadsheet open so will confirm precise number later but it’s sub 72.

[harry_rag]

55 minutes ago, harry_rag said:

10 points on Firmino at 2.94 (<72A)

Not got spreadsheet open so will confirm precise number later but it’s sub 72.

It was 60A.

To clarify the number, it's a measure of the value in the best available bookie's price. 72+ is a bet, so the higher the price the bigger the edge should be. <72 just means it's no bet at the bookies prices but doesn't reflect on the value in the price taken on the exchange.

Firmino was 13/8 best price and the 60 reflects that being poor value, it bears no relationship to the 2.94 taken on the exchange.