Invite for complex discussion regarding the Comback Ratings (Filter & Criterias)
-
Intro to the reason behind this research topic.
There are three key factors why I will use Lay The U-dog with ELO Rating & Comeback Rating.
First, the bot service Pouncebet has successfully used a similar setup and approach with several years of long-term success. Secondly, I succeed with fewer criteria and solid filtering with my own solution to get a positive ROI.
Third, Richard's ELO strategy is an excellent approach with good stats to back up the decision to embrace the method as the main core solution.Now I understand that certain formulas will not be revealed and the reason for this topic is to reach cancellation method and reverse engineering to find out the true value behind the Comeback Rating criteria or find a similar working filter solution with similar expectations and results.
So any thoughts ideas or suggestions are more than welcome to improve my project.
Key Factors to work with: Blue Print for inspiration and developing the final solution for Comeback Rating!
-
Maybe only focus on the Away Team not winning more than 40% or less to get a 60% Home Win or Draw as final results for some degree of previous games.
-
How would I calculate the comeback ratings manually?
The past 20 games
a) Do I check the last 20 where the favourite was under one or two goals, and then check how many times they did an equalizer or won?
b) or do I check the last 20 where the favourite won or made an equalizer, no matter if the underdog took the lead or not?
c) or do I check just the 20 past results and see if they made a win or not, no matter if they were behind or not?The past H2H games
a) Do I check the last 10 H2H where the favourite was under one or two goals and then check how many times they did an equalizer or won?
b) or do I check the last 10 H2H where the favourite won or made an equalizer, no matter if the underdog took the lead or not?
c) or do I check just the 10 H2H past results and see if they made a win or equalizer, no matter if they were behind or not?- Clear Hints from Adam
Quote Adam:
Hi Patrik
There's a description of each of the ratings underneath the ratings table on the webpage. The comeback rating is based on each team's last 20 matches (home @ home and away @ away), and their last 10 matches against the same opposition type (top, middle or bottom of the league table).
It's based on how often they have come back from a losing position and either equalised or taken the lead.
Quote Adam:
Hi Patrik
We don't publish the formulae for the ratings unfortunately but what I meant by "same opposition" earlier wasn't H2H. We basically divide the league into 3 sections, top middle and bottom, and look at the match history where each team has played the same opposition type that they're playing today.
So if the home team is currently playing against an away team that's currently at the bottom of the league, the "same opposition" match history that we use will be all the matches where the away team was at the bottom of the league at that time.
- Backtrack Comeback Ratings with 60+ and see what patterns and similarities match with the above assumptions and hints to finding the final solution.
Note
This is how I step by step learn and share for the community and see it as a puzzle a process towards a successful way to complete and fully understand and master a trading strategie from all angles.
Cheers Patrik
EDIT
Here is the Rating Site
https://new.betfairtradingcommunity.com/ratings/?wdt_column_filter[1]=08/06/2024|11/06/2024
Also, add a picture for an alternative for setting up filter using following function:
-
-
@Patrik-Mellqvist great work, how has the comeback rating strategy been going so far?
-
I use the pay version of GPT Chat and have made the same effective comeback algorithm as Adams.
My version gives me moderate or good indication with estimation parameters values that give 93 with Adams.
So I only need to backtrack the scale and fine-tune the scaling, not the value and analyzing model.
That seems to be over-effective and not the other way around, that is a good sign.Here is a hint - you can code in Python and GPT Chat shows you how you can load BTC export filters.
To get the comeback values.But I only pick game candidates with stats software and then name the game with a tailored blueprint checklist that tells GPT Chat to crack the game into the comeback value on my algorithm model.
That was what I wanted - my algorithm for a comeback filter as effective as Adams.
Because strategies are using it that have been proven to work since 2014.The next project is to transform the ranking system page values and parameters into similar towards Tennis.
And then create the same filter for Tennis - so with football, you have 1-0 win 1-1 win 0-0 win and 0-1/0-2 loss and crack the same situations or similar with Tennis and you have created the winning method working for decay.Cheers
-
@Patrik-Mellqvist I often look at both and take a view on how closely they agree with each other. For example if I'm estimating goals and the estimate is 2.1 goals based on the last 10 matches, but the estimate based on the last 20 matches is 5.6 goals, I'm less likely to trust it than if they were more closely correlated.
In any ideal world, you'd look at every match and consider what's changed for that team / league in the last 10 / 20 matches. Obviously that would be a lot of work to do for 200 leagues across the world...
-
@Patrik-Mellqvist I tend to use last 10, because last 20 can go back a few months, so I feel some of the data won't be as relevant
-
I want opinions on the following simple question:
Should I take stats from 10 previous or 20 previous games, what is fine and what is overdoing an analyzing situation?
Reflection 1
The 10 is the fresh sample with current performance in recent times that is more true in reflecting team shape and condition in present or close past time. Feels like a solid and common way to use filters.Reflection 2
A larger sample as the past 20 is more significant trustfully in the long term perspective.
But feel that odds occurrences skew the overall estimation of statistical results so unnecessary selection vanishes even with good overall performance.Summary Options 1 & 2
Option 1 - Do I aim for the present short term with more fresh and time-relevant stats?
Option 2 - Do I aim for the long term with a more significant estimation with the risk of over-analyzing the situation and missing out on selections?
Note:
Also consider options 1 & 2 where 1 is quantity in getting selections and 2 is quality in getting selections!
The difficulty with this assumption is the difference in the Risk & Reward ratio in winning and losing trades that also shows in the overall profit.Detail about the filter
I set 100+ ELO for H or A
I set a specific selection filter that gives almost the same value as the Comeback Ranking Filter
I set league table positions where I cancel certain position combinations.
I set odds between 2.5 to 11.0 for the current U-dogCheers Patrik
-
I will update this topic with results as I have data from the past, so I know that even less advanced filtering gives small and smooth ups and downs.
Martin mentioned that he thinks is wrong to scale up in the beginning and stake larger amounts during the learning curve, but I don't remember the exact video.I personally feel that is the truth if you are new with limitations, but as I have experience from other parts of the gambling arena I feel that my gambling IQ is pretty solid for risk mitigation decisions.
So I will take a shot on this with calculated risk management.But before I reveal how I will stake using Maria Staking Plan suggested in the ELO Strategi I will explain what I do.
First I will only use Home Favorite priced at 1.75 or less.
Secondly, I will only use Home Favorite which has an ELO with 100+ opposite to the Away Team.
Third, I will only lay and relay if the first goal is scored within the first 20 minutes to the Underdog as I know this timeframe affects the match performance at the Home Favorite Pressure Points or in other words, create time-value (opposite to get involved during any time that create unpredictable variance and expectation), my opinion.
Fourth, I set the match not ending as -1 or -2 with around 25% or less with at least 15 previous matches checked. (I might fine-tune and tweak this last option)
Also adding a following note that I checked one set up with this setting and the Home Team with several seasons back in 2016 to 2024 won 15 and lost 3 to give you a hint of the power and simplicity of using such a filter opposite to the Comeback Rating, where I feel is an overdue task to check and separate top, middle, low rankings specific towards -1 or -2 loses, so my solution generalized.Now to the staking - as I have been cracking stats with CGM-Bet statistical software with goal-timings with the setting 0-1 to underdog within first 20 min and now the results from many different leagues and seasons - I can make solid and valid estimations based upon that knowledge.
That is why I decided that there will be quality trades opposite to quantity trades with fewer qualifiers so I need to squeeze the market or milk the market as it would be a cash cow.
So £100 at 1.01 to 3,5
£60 at 3,5 to 7,4
£40 at 7,4 to 11,0EDIT
I will reflect on this and simulate different options.
Before deciding how to and what I find to be the best trading angle.In the past, I have tested the following, Dutch both Fav/Draw
Lay U-Dog
But with some short testing with the relay option I don't find it good.Cheers Patrik
