Thoughts on dynamic filtering techniques

Pages: [1]
Author Topic: Thoughts on dynamic filtering techniques  (Read 1235 times)
VidaBlue
Has experience
**

Karma: 35
Posts: 137


View Profile

VidaBlue

Thoughts on dynamic filtering techniques
« on: October 02, 2019, 10:40:32 AM »

Barnstorm once wrote "If Value Betting is the Holy Grail of betting, then filtering your Value Bets is the Holy Grail of Value Betting."

Regardsless of whether value betting is the holy grail of betting, I believe indeed that filtering is the holy grail of value betting. This post is not about whether value betting is better than arbing or whether filtering works, so please leave those comments for another thread. This post is about techniques for employing our filters.

Pinnacle lean is one reference, which works well overall, but not always, sometimes some asians or betfair is better. Referencing softs vs other softs is another, but which softs are sharper? By analysis of past events, it will get clearer which reference is better for the true odds.

In my shop betting endeavour, I track my own record with regards to mainly:
  • sport
  • bettype
  • bets placed at bookmaker (only 3)
I could also have taken into account which bookmaker(s) I used as reference, edge (only that is high), time before game start, etc., but I don't. It simply gets too data-intensive for me and instead I have decided to prioritize high production(turnover) on the streets. Also, I cannot wait for track records of hundreds of bets in each combination of all of the above before starting to employ filtering, because then I would never achieve a statistically significant sample size.

So, I start employing filtering even on a very low number of bets within the same combination. For instance:

I have a track record of 10 bets in "esports LOL", "kills O" with "bookmaker X".
Let's say I have won 8 out of 10, I would then add a factor of 25% to the stake size.
Let's say I have lost 8 out of 10, I would then still bet, but reduze my stake size to half.

My approach is not very organized or scientific, and statisticians might moan over the lack of a significant sample size and claim this is purely variance and nothing else.

So I do not entirely trust what might be a lucky streak or a bad run, I just make small corrections to the stake and the bigger the sample size, the more certain I feel about these adjustments. Ultimately I would increase stakes 100% or I would start betting the sharp side, when samples are big enough.

I am not looking for a hard proof in the track record, just an indication and in case making adjustments is justified, I am looking for a more scientific and structured approach.

So what are your thoughts on this? Do you find this sound or nonsense? What are your own experiences? Please share.
Logged
Alfa1234
Totally Pro
*****

Karma: 38
Posts: 885


View Profile

Alfa1234

Re: Thoughts on dynamic filtering techniques
« Reply #1 on: October 02, 2019, 10:56:21 AM »

Reducing or increasing stakes that quickly is a horrible idea as it would take you longer to get back to the "mean" ROI after a bad run and after a good run you would revert much quicker to the lower "mean" ROI.  Don't change stakes unless you are asbolutely sure your edge has changed.
Logged
VidaBlue
Has experience
**

Karma: 35
Posts: 137


View Profile

VidaBlue

Re: Thoughts on dynamic filtering techniques
« Reply #2 on: October 02, 2019, 11:10:53 AM »

It is an interesting answer Alfa1234, but also kind of text book I think. When would you be absolutely sure then?

I have very satisfying ROI on esports in very specific setups and referencing softs vs softs, but it is based on common sense in the first place, subsequently trial and error and then gradually adjusting stakes.

I would never have gotten there, had I had to wait to be "absolutely sure".

What is your definition of "absolutely sure"?
Logged
Alfa1234
Totally Pro
*****

Karma: 38
Posts: 885


View Profile

Alfa1234

Re: Thoughts on dynamic filtering techniques
« Reply #3 on: October 02, 2019, 11:17:48 AM »

It would depend on what your normal ROI is.  If it's say 30%, I would consider reducing my stake if I had 100 bets with negative result as that's statistically significant.  If it's 3%, you'd need a much bigger sample size.
Logged
VidaBlue
Has experience
**

Karma: 35
Posts: 137


View Profile

VidaBlue

Re: Thoughts on dynamic filtering techniques
« Reply #4 on: October 02, 2019, 11:41:51 AM »

Ok. The statistical significance level is usually chosen, commonly to 5% and of course then 10 bets with 10% ROI is not enough. But is it so, that if you have a statistical significance level of 4% you're ok to bet, but 6% you're not ok? I don't believe so. It is not a hard limit in that sense. It is just a chosen number and stakes size should take this number into account.

So I in that regard I find it reasonable to increase or decrease stakes accordingly if there's a chance that the track record shows some significance, even if the level is far from the commonly used 5%. After all we are talking probability here as well, just as we are with the odds of all the games.

There's also the seasonal variation. For instance, some bookmakers may have misrated certain teams in the tournaments ahead, so maybe your next 50 bets would be influenced by this, and such short term tendency is not depicted in the larger "absolutely sure" sample size of a wider time span. Dynamic short term filtering could perhaps make slight weighted adjustments to the stake size in the short run? I think the value betting index posted in this forum, assumes seasonal variation.

If someone here thinks this makes sense, I would like your shares on a structured approach to dynamic filtering.
« Last Edit: October 02, 2019, 11:44:39 AM by VidaBlue » Logged
4lexy4
Guest


Email

4lexy4

Re: Thoughts on dynamic filtering techniques
« Reply #5 on: October 03, 2019, 06:26:44 AM »

if you bet premach, you can make faster conclusion who is sharper on closing odds
Logged
arbusers
Administrator
Totally Pro
*****

Karma: 370
Posts: 3808



View Profile WWW

arbusers

Re: Thoughts on dynamic filtering techniques
« Reply #6 on: October 03, 2019, 08:21:47 PM »

Thank you for the excellent post Vidablue.

IMHO, there are 4 dimensions in Value Betting.
1. Finding the Value
2. Filtering the Value
3. Have a staking plan
4. Placing bets.

These are 4 different dimensions complementing each other. A successful Value Bettor is the one that can satisfy all of them. If you lose one, then you are not that sharp. If you lose two you are not sharp at all. If you lose three you are not a Value Bettor. And if you lose all, you are not a bettor.

Deming once said ''Without data, you're just another person with an opinion.'' That means that in the end, you have to analyze these thousands of bets and see where you are doing mistakes. For example, you might notice that Bookmaker X is extremely successful in Championship A. Maybe there are some good reasons for that, and maybe you should avoid what is deemed as a value in this Championship from this bookmaker. Maybe bookmaker Y is extremely good with the odds range 3.00 to 3.50. There must be some good reasons for that.

It is my firm belief that many bookmakers often set value traps. They are willing to have an unbalanced event, simply because they themselves are doing Value Betting. They have some very good info on an event and they are willing to offer an arb against all others, just to get a good number of bets on their desired leg.

When it comes to staking, I believe it worths starting with a D' Alembert approach that would be later modified to cover your needs and according to your experience. Finding turning points is not an easy task, but you could find a supercilium of a turning point and start reducing or increasing stakes. If you know the general direction of your performance, then any big deviation should be an alert for you.

I remember in the old days, most arbers weren't easy to share their tricks. Now it is the same with Value Bettors and their staking and filtering plans. These are best-kept secrets these days.
« Last Edit: October 10, 2019, 09:16:03 AM by arbusers » Logged
SticyBandit
Has experience
**

Karma: 1
Posts: 141


View Profile

SticyBandit

Re: Thoughts on dynamic filtering techniques
« Reply #7 on: October 04, 2019, 04:40:44 PM »

If you are trying to beat bookies by varying the stakes then I suggest you can save your time and stop right away.
I will say only two things.
1) only thing that brings profit is value that bet has
2) going up and down with stakes with respect to available bank for betting off course vs flat stake that is also measured according to available bank is returning 0 advantage for you...having not-fixed stake just means that you are most of time under-staking your bets or over-staking them in respect to ideal stake. And if you are finding out time when things will change their way, so you can stake more,  you are introducing new variable, you are betting in fact on this to happen. And this has it's price too.
Interesting how many people are messing with stats and probability and doesn't know or understand simple fact that past events doesn't have any impact on those future ones if we are talking about independent events, which those we are talking about are.
This thread is witness of that.
Logged
VidaBlue
Has experience
**

Karma: 35
Posts: 137


View Profile

VidaBlue

Re: Thoughts on dynamic filtering techniques
« Reply #8 on: October 10, 2019, 12:40:23 AM »

Thank you for your contributions and thank you Arbusers for pinpointing that this thread is really about the 3rd dimension, the staking plan and less on filtering. I will dig into the D’Alembert approach and I appreciate your constructive answer. The beauty of having your own betting history, your own data, is that you can always model what would the result have been had a modified staking plan been used.

I find it a bit disturbing that when you try to introduce and discuss new and more elaborated chapters to the same small bible called “How to bet using pinnacle lean and Kelly criterion” you are often being targeted as someone ignorant and unaware. We (value bettors) are increasingly being challenged (by bookmaker methods, bots and people), so one needs to evolve in order to ensure future success (Pinnacle lean is not by far as effective as it has been). I think this is part of the reason why a forum like this exists, a smart betting community. Otherwise we could all just read the same book.

I have some real numbers in lieu of the hypothetical example in the original post and I will try to either round up the thread or invite more constructive comments in case anyone wants to. So, overall data from my shop betting during the last 18 months is the following:

ROI%(number of bets)
Overall ROI: 5.74% (5.501)
(value bets are made based upon various references, the pinnacle lean part (around 20%) is actually less effective than the overall 5.74% ROI)

Bets decided on:      Bets placed on:
  • Mondays: -4.97% (407)      -2.55% (662)
  • Tuesdays: 8.52% (636)      12.21% (707)
  • Wednesdays: 8.00% (672)   8.58% (753)
  • Thursdays: 8.35% (645)      7.59% (710)
  • Fridays: 5.18% (770)      3.57% (844)
  • Saturdays: 0.43% (1.246)   6.45% (986)
  • Sundays: 9.37% (1.125)      1.25% (839)
For instance, could the bad performance on bets decided on Mondays just be random? Sure, however, when modeling the outcome on these data an infinite number of times, the probability converges towards a 6 % chance that this is pure variance given the EV and number of bets (407). Would I try to identify the cause? Sure, but I can’t, maybe Mondays have more value traps, the soft is more sharp in certain sports on Mondays, etc… Would I stop betting on Monday’s matches? No, but I would add a correction factor to the stake, seems logical when considering the low probability for variance being the only cause.

Another example:
  • Basket 1Q O/U: 29.5% (100)
  • Soccer whole match O/U: -12.8% (103)
In these 2 examples, the probability for variance being the only cause calculates to less than 1% for basket and around 5 % for soccer. So it could still be variance of course, but I’ll decide to increase stakes on 1Q basket and decrease stakes on OU soccer.

Would I regularly review all this and model what the ROI had been using other staking plans? Absolutely!

So my question is still the same as in the beginning of this thread. If you agree, that adjustment of stakes in these circumstances above seems logical, how much would you adjust? I am after something calculable that takes my own gut feeling out of the equation.
« Last Edit: October 10, 2019, 08:18:46 AM by VidaBlue » Logged
arbusers
Administrator
Totally Pro
*****

Karma: 370
Posts: 3808



View Profile WWW

arbusers

Re: Thoughts on dynamic filtering techniques
« Reply #9 on: October 10, 2019, 03:51:19 PM »

Before anything else, I extend a big ''thank you'' for sharing your thoughts and your stats in such a way.
Then I would like to stress that we are talking about shop action. I m very much curious to know which companies you are using to place your bets, and what your operating procedure is. I know these are cornerstones of your business, I will still consider you as a friend if you don't answer.

Now let's go to your stats. I believe there is a credible interpretation of why Mondays are bad days. Most probably, it is because the results of the previous days (Saturdays and Sundays) are playing a role, influencing the results of the next day. This could happen for a number of reasons, I will quote some:
1. The teams that play on Mondays know much better what is the desired outcome of their own game because they already know what their competitors on the championships did the days before.
2. Bookmakers are getting more efficient because they get hints and more analysis from the days before.
3. It could be a matter of luck, but I doubt it is because your sample is adequate.

In your shoes, I would simply lower the stake on Mondays without touching the stake of the other days. One would say, why not to stop betting completely on Mondays, after all, you are losing money. That would be the wrong thing to do because Mondays also contribute to your data and you keep in touch with the market. Keeping in touch with the market is an important factor as it helps identify when something starts turning around.

And of course, through trial and error, you should adjust everything until you reach the optimal performance. You mentioned an overall ROI of 5.74%. Who knows, after lowering your Monday's stake, this could go to 5.80%. That is a significant profit.

Many thanks once again.
Logged
VidaBlue
Has experience
**

Karma: 35
Posts: 137


View Profile

VidaBlue

Re: Thoughts on dynamic filtering techniques
« Reply #10 on: October 11, 2019, 10:40:33 AM »

The way I see it, the greatest potential of a forum like this, is to improve personally sharing ideas and thoughts with others. I value equally approvals as well as disapprovals, as success implies boosting profitable patterns as well as learning from failures, so I thank all comments. They have been helpful. This is not a closed subject of course, so by all means continue this thread as required.

What I take from this thread up to now, is that I have some work ahead modelling a dynamic staking plan. Apart from what the simple staking plan already uses (odds and EV) this plan should also take the following into account: sport, bettype, day of the week, bookmaker, ROI, D'Lambert approach(?) and most importantly, the sample size. I am just thinking of a simple table that is recalculated and memorized now and then, giving bonuses or penalties to the stake size. During the actual betting action, for practicality, it will probably just be the nearest round number.

Some time ago I ran a simulation of my ROI had I been using a more aggressive staking plan (30-100% Kelly) and obviously profits would have been higher. However, I am not sure that the larger outswings would have been healthy for my psyche and there are stake limits even in shop betting. I will of course run this dynamic staking plan on my betting history to get an indication of how well it works had it been implemented earlier. I have expectations, but I am also prepared to cancel this idea in case it seems to have little or negative effect.

Arbusers, I am so glad for your thoughts on Monday’s games and I will take this into account right away, dynamic staking plan or not. I will share the other information with you, but I think this is too sensitive for public so expect a pm some time. One tends to get paranoid after some years of betting :-)
« Last Edit: October 11, 2019, 10:44:26 AM by VidaBlue » Logged
pythonic
Gaining experience
*

Karma: 6
Posts: 61


View Profile

pythonic

Re: Thoughts on dynamic filtering techniques
« Reply #11 on: October 12, 2019, 12:06:31 PM »

Two comments:
- Apart from just sampling your bets according to all sorts of possible criteria you should not forget to think about possible explanations. For example it might be a good hypothesis that a certain bookie is better at pricing certain sports or certain leagues, but it would be weird if you found an edge when betting on teams with certain names, say names starting with an A.
Not impossible, because perhaps certain names just happen to be much more common in certain leagues, but still it is not a good explanation and if you just randomly test against every possible criterion regardless of the explanatory power of your hypothesis then you are just doing data mining and will probably end up with lots of spurious patterns in the data which would not carry over into the future.
So that's the qualitative part and you can only rely on your judgement and careful handling of your data for that.

- About the quantitative part, it seems to me like bayesian statistics would be a natural tool for you to formally structure your approach because it is exactly about dynamically updating your beliefs about probability distributions in the evidence of new data. I should add that I have little experience with bayesian statistics myself and you might find it too complicated or not very helpful in practice, but in principle it should be exactly the formal approach that was developed for problems like this.
Logged
VidaBlue
Has experience
**

Karma: 35
Posts: 137


View Profile

VidaBlue

Re: Thoughts on dynamic filtering techniques
« Reply #12 on: October 12, 2019, 02:25:31 PM »

I agree with you photonic on seeking explanation on the performance of bookmakers at certain sports/leagues is important. I think this is the main factor to be studied and what everyone wants to know, as this information is very valuable. Day of week is nice and useless to know, at least I thought. But looking into my own bets I discovered significant difference between weekdays and I am giving this much more attention from now on. Another study is certain bookmakers sudden deviation when odds reach the high end. For instance odds my be within range of each other when they are around 2.00, but as soon as one team is given a large handicap, say -2.5 in a hockey or soccer match, one odd may be 4.0 while the other is 3.2. This indicates that something is not right. As for teams starting with a "A", I think it is obvious that the A-team must be better than the B-team  ;D

I usually do Monte Carlo simulations in order to visualize potential outcomes. I think it is viable, and that way I don't have to read up on math. But I must read up on bayesian statistics and try to implement this, in order to help with prognostics. Thank you for this hint.
« Last Edit: October 12, 2019, 02:35:39 PM by VidaBlue » Logged
Pages: [1]
Print