Kelly betting live test - is it a madman's goal as some quants around here suggest?

Remove this Banner Ad

brett128 said:
No it doesn't.



That's too far dude. Lines are only very rarely that wrong. A 15-16 point difference vs the spread implies a win % going into mid 60's% in AFL. Either increase your Kelly fraction or re-work your model.
Click to expand...
This is what sent you broke first time.

Lines are off by a fair margin alot more than 'rarely'.
 
RE the accuracy of the lines, here's the results so far this week:

Ess by 7, line 4.5, error 2.5
Syd by 22, line 18.5, error 3.5
Adel by 28, line 7.5, error 20.5
Port by 35, line 15.5, error 50.5
Melb by 3, line 27.5, error 30.5
Geel by 41, line 16.5, error 24.5
Hawks by 58, line 42.5, error 15.5

How can we then say that the lines are rarely 16 or more points out, when 4 of the 7 matches completed this weekend have been 16 or more out, and 5/7 have been 15 or more out??
 

Log in to remove this ad.

Duritz said:
It does though, doesn't it? Given that I am using $10K as a BR to work out the bets, no matter what the actual BR is, then simultaneous plays are not an issue. They're only an issue if you're overbetting as a result of having many plays in play at once, but that's only the case if you have a reducing or increasing BR. I am sure I am right that simultaneous plays don't matter if you're not adjusting your BR.
Click to expand...


It doesn't matter if you have a fixed or progressively changing BR. If you betting on different events in a simultaneous fashion, you MUST account for the fact that you allocating capital to those opportunities simultaneously.

You can't go commit to this line of logic:

EG:

1. I have 10 x 55% simultaneous ATS plays

2. 1 x 55% play normally allocates 4.55% of BR if bet in isolation.

3. Therefore I should allocate 45.5% of BR to those 10 x 55% plays.


If you are doing this, you are massively over-betting.
 
Duritz said:
But how can this be proven?

Edit: Wouldn't the way to prove be to look at the final lines for a couple of years, and how far off the final score they were? How often is a line 16 points off the final margin?
Click to expand...



A 200 game moving window of Mean Absolute Error (MAE) of the AFL line is 29.5 points. It has oscillated around 29.5 points for around a decade. This trend hasn't seemed to have changed due to the market becoming more efficient just yet. So you could define error in this way.

But I define how far a market is off to one's own model in terms of how the cream of your own plays covers the spread. Why?

Because your estimate of the game and the market's estimate of the game will meet somewhere in between, sometimes closer to your estimate, and sometimes closer (much closer) to the market estimate.

So I gauge how our model is 'functionally off' the market by evaluating the win % of my very best plays. This year since I've went full time we've gone 47-31 (60.3% ATS) on our very best plays. Speaking to the 'old bloke' in our group, speaking to other full time handicappers, and then examining the public records of touts like Dr Bob and RAS, everyone seems to have the same win % ceiling of between 58% and 62% ATS on their very best plays.

Given a win % ceiling of 62% ATS and the fact the value of a point to the spread is roughly 1.2% in the probability of covering, this 'meeting of model to market' implies that functionally, any AFL model output expressed in points should be on average not more than 10 points away from what the market spread is showing.

So yes it is possible for a model to be 20-30 points off the market in some games, but if this is not the rare exception, the model is out of wack.
 
Here's a tale of Kelly woe. Let this be a warning to prospective Kelly bettors:

6-15 ATS this week for us, 44% of BR and $40K down the hole. Every single 60% play lost and only the smaller bets won. BR went from $10K 5 months ago to within striking distance of $100K before the start of this weekend. Our record reverts to 57% ATS over 318 plays this year.
 
brett128 said:
It doesn't matter if you have a fixed or progressively changing BR. If you betting on different events in a simultaneous fashion, you MUST account for the fact that you allocating capital to those opportunities simultaneously.

You can't go commit to this line of logic:

EG:

1. I have 10 x 55% simultaneous ATS plays

2. 1 x 55% play normally allocates 4.55% of BR if bet in isolation.

3. Therefore I should allocate 45.5% of BR to those 10 x 55% plays.


If you are doing this, you are massively over-betting.
Click to expand...

How? I don't follow this at all. How can you be over betting when the BR would not change regardless of whether the 10 plays won or lost?
 
brett128 in your example, 10 x 4.55% bets. The first one is in isolation. I have 4.55%, or $455 on it on a 10k BR. It gets beat. The next one is in isolation. I have 4.55% or $455 on it. It gets beat. The next 7 are all together, then the final one is in isolation. REGARDLESS of what happens in those 7, I am having 4.55%, or $455, on the final one, on a static 10k BR. Therefore, to have anything other than $455 on the other 7, just because they happen at the same time, is absurd. Given their winning or losing wouldn't change the outlay of any subsequent bet, if they were in isolation, why should it change it if they co-incide?
 
Duritz you are correct. If you are using a static BR then simultaneity is irrelevant. Brett would argue that he is still correct because you are still overbetting, but that is an unavoidable possibility of using the static BR
 
Duritz said:
brett128 in your example, 10 x 4.55% bets. The first one is in isolation. I have 4.55%, or $455 on it on a 10k BR. It gets beat. The next one is in isolation. I have 4.55% or $455 on it. It gets beat. The next 7 are all together, then the final one is in isolation. REGARDLESS of what happens in those 7, I am having 4.55%, or $455, on the final one, on a static 10k BR. Therefore, to have anything other than $455 on the other 7, just because they happen at the same time, is absurd. Given their winning or losing wouldn't change the outlay of any subsequent bet, if they were in isolation, why should it change it if they co-incide?
Click to expand...


That's so wrong it's not funny. Thinking about this more, there's also no such thing as a $10K 'static' BR when betting Kelly. You're just using a bastardized version of Kelly to bet to edge. If you're not adjusting your BR size with the available (total) capital at hand, you're not optimizing expected BR growth.


If you choose to bet on 10 investments at once, that is defined as simultaneity. You have no say in the capital allocation of the last bet, even if your first 9 bets lost, as your intent was to bet them all at once (if that is what you actually wanted to do). It doesn't matter about their chronological order, you have a chosen a time prior to the first event going off to invest in all 10 events. Functionally, they are linked.


Let's take a look at 10 x 55% ATS plays with the SBR Kelly calculator:

1 x 55% ATS play = 5.55% (sorry it should have been 5.55%, not 4.55%)



10 x simultaneous 55% ATS plays = 3.30% per play


So 10 x 55% ATS plays is not 55.5% of BR, but 33% of BR. So if you choose not allowing for the fact you have simultaneous events, you are betting 68% more than what Kelly would allow.
 

(Log in to remove this ad.)

brett128 said:
That's so wrong it's not funny. Thinking about this more, there's also no such thing as a $10K 'static' BR when betting Kelly. You're just using a bastardized version of Kelly to bet to edge. If you're not adjusting your BR size with the available (total) capital at hand, you're not optimizing expected BR growth.
Click to expand...

Yes, but this is what I have stated. I am not betting Kelly in it's pure sense, I am using it to work out bet sizes. I am utilising Kelly's ability to maximise the best overlays, while not going with the up and down scaling nature of it, which has problems.

You say, "That's so wrong it's not funny." But it's only wrong from a pure Kelly point of view. Only wrong if you're dogmatic about betting Kelly EXACTLY as intended. I am not doing that, because IMO the scaling up/down of Kelly is flawed. Great in simulations, where betting doesn't run in cycles, shit in the real world, where it does.

brett128 said:
If you choose to bet on 10 investments at once, that is defined as simultaneity. You have no say in the capital allocation of the last bet, even if your first 9 bets lost, as your intent was to bet them all at once (if that is what you actually wanted to do). It doesn't matter about their chronological order, you have a chosen a time prior to the first event going off to invest in all 10 events. Functionally, they are linked.


Let's take a look at 10 x 55% ATS plays with the SBR Kelly calculator:

1 x 55% ATS play = 5.55% (sorry it should have been 5.55%, not 4.55%)



10 x simultaneous 55% ATS plays = 3.30% per play


So 10 x 55% ATS plays is not 55.5% of BR, but 33% of BR. So if you choose not allowing for the fact you have simultaneous events, you are betting 68% more than what Kelly would allow.
Click to expand...

I think you're missing my point here... I am not using the Kelly calculator, because it is only suited to a pure Kelly standpoint. I am using PARTS of the Kelly method. I am using the Kelly method to work out wager size, based on a static bankroll of $10,000. I am choosing the bet that has the best EG of the bets available. Simultaneity is only relevant if your bankroll is varying.

You need to remove your pure Kelly glasses and look at this for what it is: an adaptation of the Kelly method, NOT pure Kelly. That's why the simultaneous betting calculator has no bearing on what I should outlay, and it is why I am correct that I should not adjust my outlay for simultaneous events, because I am not adjusting my BR figure.

I fail to understand why this is difficult to comprehend.

If you like, call it the Duritz Kelly method, so that you can remove yourself from thinking in the limited scope of pure Kelly, then rethink why on earth simultaneity has any bearing.
 
Ok call it the Duritz method.

If you want to base your capital allocation on capital you don't actually have (if your BR becomes less than $10K), then go for it. Your exercise in 'Kelly betting' doesn't quite reach an experimental gold standard for a 'Kelly betting live test' that was the premise of this thread.

So don't blame Kelly if it all goes wrong.
 
brett128 said:
Ok call it the Duritz method.

If you want to base your capital allocation on capital you don't actually have (if your BR becomes less than $10K), then go for it. Your exercise in 'Kelly betting' doesn't quite reach an experimental gold standard for a 'Kelly betting live test' that was the premise of this thread.

So don't blame Kelly if it all goes wrong.
Click to expand...

Well, it is a Kelly betting live test, just with my twist on it. You might say, my improvements. I see what Kelly and the other maths guys were doing, but I recognise that all that time in the "lab" with their "computers" and their "lab coats" has dulled them to the experiences of the real world. I've got zero doubt that I've bet in more events (races mainly of course) than anyone on this forum, and that by quite a long way, too... I have a little experience therefore in the ups and downs of punting, that the "scientists" with their "logic" in their "labs" wouldn't recognise.

Your old mate in your group would get something of it. Punting is no mathematical simulation.
 
I've never used Kelly, but it seems entirely premised upon calculation of edge for each bet, which would appear to be (a) incalculable, and (b) habitually overestimated.

I quickly plugged some numbers into an online Kelly tool, and with 57.5% expectation on an even money bet, I would need to wager 15% of bankroll. LOL! In the early stages while you are learning and trying to build a bankroll this will lead to insane swings (see Brett) especially when you are still learning and vastly overestimate your edge, and then once you have some success and built a real roll you won't be able to get enough $ down as per Kelly's suggestions.

So yeah, Kelly seems like a waste of time.
 
"a)" is not the problem at all, though it is often cited as the flaw in Kelly betting. "b)" is the real flaw.

Kelly works on the logical premise that your bet size will be directly proportional to your current BR. This 'feels wrong' to a lot of people because it is easier to lose from a good position (because your bets will be bigger, ie going 5-0 then 0-4 could leave you worse than when u started betting at even money) and harder to win out of a bad position. Following from this, there seems to be a perception that when betting a % based amount, order of wins and losses matters. This is false; when all bets are a percentage the wins or losses are equivalent to applying multiplication to your bankroll, and order of multiplication doesn't matter.

Betting set units on the other hand, you are making bets based on a static 'starting bankroll' which is illogical. If I have 5k now it doesn't (or shouldnt) matter if I started with 1k or 10k. The correct bet should be the same.

Now, given you are betting a % of current bankroll, the Kelly bet is the best possible strategy if you run exactly to expectation, which is why it is referred to as the best strategy "long term". Re: your example, if you make 40 even money bets, betting 15% on each one is the best possible strat if you hit 23 of them.

/end rant

(Most of you probably know all this, but I see enough misconceptions that I thought I'd try to clear up a few things on the theory side. In practice I ignore logic and just bet set units like most of the test of you)
 
Dominion said:
"a)" is not the problem at all, though it is often cited as the flaw in Kelly betting. "b)" is the real flaw.

Kelly works on the logical premise that your bet size will be directly proportional to your current BR. This 'feels wrong' to a lot of people because it is easier to lose from a good position (because your bets will be bigger, ie going 5-0 then 0-4 could leave you worse than when u started betting at even money) and harder to win out of a bad position. Following from this, there seems to be a perception that when betting a % based amount, order of wins and losses matters. This is false; when all bets are a percentage the wins or losses are equivalent to applying multiplication to your bankroll, and order of multiplication doesn't matter.

Betting set units on the other hand, you are making bets based on a static 'starting bankroll' which is illogical. If I have 5k now it doesn't (or shouldnt) matter if I started with 1k or 10k. The correct bet should be the same.

Now, given you are betting a % of current bankroll, the Kelly bet is the best possible strategy if you run exactly to expectation, which is why it is referred to as the best strategy "long term". Re: your example, if you make 40 even money bets, betting 15% on each one is the best possible strat if you hit 23 of them.

/end rant

(Most of you probably know all this, but I see enough misconceptions that I thought I'd try to clear up a few things on the theory side. In practice I ignore logic and just bet set units like most of the test of you)
Click to expand...


This post is not pretentious enough for a kelly explanation
 
Dominion said:
"a)" is not the problem at all, though it is often cited as the flaw in Kelly betting. "b)" is the real flaw.

Kelly works on the logical premise that your bet size will be directly proportional to your current BR. This 'feels wrong' to a lot of people because it is easier to lose from a good position (because your bets will be bigger, ie going 5-0 then 0-4 could leave you worse than when u started betting at even money) and harder to win out of a bad position. Following from this, there seems to be a perception that when betting a % based amount, order of wins and losses matters. This is false; when all bets are a percentage the wins or losses are equivalent to applying multiplication to your bankroll, and order of multiplication doesn't matter.

Betting set units on the other hand, you are making bets based on a static 'starting bankroll' which is illogical. If I have 5k now it doesn't (or shouldnt) matter if I started with 1k or 10k. The correct bet should be the same.

Now, given you are betting a % of current bankroll, the Kelly bet is the best possible strategy if you run exactly to expectation, which is why it is referred to as the best strategy "long term". Re: your example, if you make 40 even money bets, betting 15% on each one is the best possible strat if you hit 23 of them.

/end rant

(Most of you probably know all this, but I see enough misconceptions that I thought I'd try to clear up a few things on the theory side. In practice I ignore logic and just bet set units like most of the test of you)
Click to expand...

Except that the ups and downs in punting in real life (not simulations) do tend to be cyclical. When this happens, Kelly is disastrous (without a static BR figure).
 
Duritz said:
Except that the ups and downs in punting in real life (not simulations) do tend to be cyclical. When this happens, Kelly is disastrous (without a static BR figure).
Click to expand...


I cannot fathom how you think capital allocation based on a static BR when say, you have 50% left of that static figure, is any less dangerous that normal Kelly adjusting for the actual capital at hand. With 50% of the static BR left, you are just betting 2X Kelly.
 
brett128 said:
I cannot fathom how you think capital allocation based on a static BR when say, you have 50% left of that static figure, is any less dangerous that normal Kelly adjusting for the actual capital at hand. With 50% of the static BR left, you are just betting 2X Kelly.
Click to expand...

Well for starters I'm betting half Kelly, as I've repeatedly stated, so a 5k br would mean I was still not over betting.
 

Remove this Banner Ad

Kelly betting live test - is it a madman's goal as some quants around here suggest?

Similar threads

H
HTP Purton Bets
Replies
11
Views
1K
HTPunter
H

Remove this Banner Ad

Back
Top