A Fractal Market

[jonbig04 10]

Here at TL it seems to me that most of us rely on PA to make our trading decisions in one way or the other. Something I am curious about is whether you think the market is fractal in its patterns and general behavior.

 

I have always been taught that it was. However when someone asks me if the economy, or the S&P has more upside or more downside I always tell them I simply don't know. I tell them that I'm a speculator, and there are too many factors going into whether or not the S&P has bottomed for me to make a sound guess. Of course if the market is truly fractal, I would be able to perform the same analysis (right or wrong hehe) on the macro picture of the S&P as I during the day or week.

 

But I can't. DB mentioned in another thread that he didn't think the market was fractal. Is the market fractal? Can we perform the same analysis on 60 minutes of data as we can for 60 years of data?

 

Just looking for thoughts and opinions.

[HAL9000 10]

Jon,

 

I believe it is fractal,

 

but I don't think that it is right to assume that you can "predict" the macro derived from the micro picture.

If at all, you can "predict" the better choice for the micro picture from the macro picture.

 

Basically the fractal thing tells me, that you should be able to use systems in their own context.

 

Another thought of mine is about drift and noise. At some point, if you go in smaller and smaller time frames, drift should disappear while noise should stay. While noise might just end up only in commissions, drift should lead to results.

But what about playing drift using noise.

 

And this all in a fractal context.

 

I guess this was all academic.

Anyway, if you find a good system that uses drift and noise at the same time,

please tell me, I am searching it.

 

[HAL9000 10]

Am I confusing myself?

 

Reading my answer, I think I have to say, I don't believe that markets are fractal.

 

I believe that markets are like music.

 

Lets say you have a wave with 440 Hz (Drift).

Then you overlay this wave with a quicker wave (Noise).

And so on. (More noise)

 

(Sorry have to search for the correct English phrases tomorrow.

But I think you understand).

 

I believe in drift and noise.

 

[sep34 10]

johnbig04,

 

Just a suggestion...You might want to provide a clear definition of what your definition of Fractal is before this thread gets going. You might get better more accurate to the point response.

 

FWIW I used to think that there was market noise, however that has changed now. I no longer believe in market noise. I only believe that there is price action that I understand and price action that I dont understand. IMO this(market noise) is one of those things that gets shoved into our heads before we have a chance to make a independent determination if such exists.

Edited May 25, 2009 by sep34

[jonbig04 10]

By fractal I mean the patterns and behaviors of the market are equally relevant, and traded the same, regardless of the time frame...more or less.

[HAL9000 10]

However what is noise.

 

I once studied Browneen motion in a different context,

but its still in my mind.

 

So here is my problem.

If I assume that markets are fractal, and this should imply that I am able to use

one setup on every time frame (context), I think that there should be no drift (trend).

Or somehow something (the indicators) have to adjust for the drift or tolerate it or ...

I don't know.

 

But if there is a large drift on a higher time frame, you should get large errors in lower ones (if you go both sides, i.e. reversal strategy).

So I think, that it is possible, that the market just looks like it is fractal over time,

because they don't like to go to zero and they don't like to go to the sky.

 

 

Somehow its all academic, the practical problem is (at least for me), what stop will keep me out of noise while riding the drift. And, how do I recognize, when the drift changes direction. I don't have to recognize it in all frames, just within the one I like to trade,

based on time, volume, ticks, range, whatsoever.

 

Today I found out one mistake I made on my loosing days, too tight stops, simply placed within noise. Simple problem. Deadly result.

 

Hal

Edited May 25, 2009 by HAL9000

[sep34 10]

By fractal I mean the patterns and behaviors of the market are equally relevant, and traded the same, regardless of the time frame...more or less.

 

Generally speaking, I use the same methods that I use on the 1, 2, & 3min to trade intraday, on 15min,60min etc.., Daily, Weekly interval--It makes no difference for me.

[BlowFish 10]

There was a thread that discussed this not so long ago. People agreed to disagree on the whole.

 

As for a definition how about "An irregular geometric object that is self-similar to its substructure at any level of magnification", there are various definitions knocking around but they boil down to essentially the same thing.

 

For practical purposes I think they are close enough though current thinking favours multifractals if you are actually trying to model them (financial time series).

 

I have a couple of 'conceptual models' when I am thinking about time series data. One is kind of fractal. The other is more sampling (so kinda signal processing orientated).

 

If it helps you to think about things its useful if it doesn't its not. Of course if you are trying to model data your requirements might be more stringent.

[Hlm 10]

By fractal I mean the patterns and behaviors of the market are equally relevant, and traded the same, regardless of the time frame...more or less.

Using that definition...absolutely (imo).

 

But then again it's only a matter of time until the math freaks and definition police come in to explain the "true" definition of a fractal. Which is why I now use the phrase "fractal like" and usually with heavily/extremely influenced. Life is too short to waste time on semantics.

[HAL9000 10]

... the phrase "fractal like" and usually with heavily/extremely influenced. ...

 

Hi HLM,

 

nice to see you again (last time on CG ).

 

Now my question: Influenced by drift (trend)?

 

In that case I have found my syntheses.

 

Fractals, drift, noise ... everything in a nice co-existence.

Some might add s/r as well.

[atto 20]

But then again it's only a matter of time until the math freaks and definition police come in to explain the "true" definition of a fractal.

*ahem* :rofl:

 

.................

[Tams 20]

[dangermouseb 10]

However what is noise.

 

I once studied Browneen motion in a different context, but its still in my mind.

 

So here is my problem.

If I assume that markets are fractal, and this should imply that I am able to use

one setup on every time frame (context), I think that there should be no drift (trend).

...

 

if there is a large drift on a higher time frame, you should get large errors in lower ones ...

 

what stop will keep me out of noise while riding the drift. And, how do I recognize, when the drift changes direction...

 

one mistake I made on my loosing days, too tight stops, simply placed within noise. Simple problem. Deadly result....

Hal

 

 

As a newcomer to this forum forgive my late reply.

 

I’ve also studied Brownian motion – as a quant at a bank – and I share your questions.

 

First let’s not forget that Brownian motion (pure noise I think) is fractal but with no drift.

 

My assumption is that the challenge of modelling timeseries was to build in the correlation to the past – which is hard to do with Brownian processes, especially in a self similar way – take an OU process for example, there is only one mean to revert to and the process is not fractal. How important is 100% self similarities. For me the idea that there can be drifts at all timescales is important. I can catch a trend at the 1000 tick level, I can catch a trend at the monthly level and everywhere inbetween. A fractal approach is the only way I currently know that such a drift function can be constructed.

 

Fractal to me doesn’t mean you use exactly the same setups etc (i.e. similar parameters) it just means that we can see the tendency at any level for trends, or counter trends (can’t even get my head around that one) and nothing is smooth.

 

I the markets were noise it would be impossible to find an edge. So many people here know they have an edge and so I believe there is drift in the noise.

 

Fractals allow me to believe that there is still drift and that I can make my livelihood in the future because of that. Fractals don’t tell me why the trends are still there or how they can be traded (other than with an asymmetric bracket order with random entry). The various forums here are focussed on finding reasons and ways to trade the persistent drift.

 

How important to you is it to recognise when drift changes? A bracket order or a SAR will produce an edge in the presence of drift. How much effort is expended on the decision that drift has changed on a trade by trade basis? Could that effort be better spent elsewhere?

 

And having said all this – I’m not making my livelihood yet ;o)

 

-- DM

[dangermouseb 10]

Ok here's a thought experiment... would Larry Conners (I pick that because I'm having a discussion about him with a friend) systems work on 5 minute charts?

 

My hunch is that although we might see trends (drifts) at different timescales they will have differing personalities and although there are strategies that might be reasonable across drift personalities there could well be some drifts that are best traded in a particular style.

 

Just a thought.

 

-- DM

[BlowFish 10]

One thing that seems plausible to me is that if markets are in fact multi fractals (do not have a single fractal dimension) that trending markets and stationary markets (cycling or range bound if you like) possibly have different fractal dimensions. I wonder if separating the cyclical components and the non-stationary components (a la Ehlers) and then modelling those separately might bear fruitful results. To be honest this was just a gut feel from someone (me) that is not at all well schooled in the mathematics. Intuitively it makes some sense.

[zdo 20]

With that definition "By fractal I mean the patterns and behaviors of the market are equally relevant, and traded the same, regardless of the time frame...more or less."

My short answer would have to be NO!

 

Long ago I bought into fractals, but my own direct experiences lead me to this ---

They lied to us! As above, so below - NOT!

[Tams 20]

in the market... as in life...

 

people tend to see things they want to see... even if it is not there,

 

and

 

people tend to not see things they don't want to see... even if it is there,

[zdo 20]

in the market... as in life...

 

people tend to see things they want to see... even if it is not there,

 

and

 

people tend to not see things they don't want to see... even if it is there,

 

Tams, .. but I wanted to see them - still do! How come I can't ? :rofl:

[Tams 20]

Tams, .. but I wanted to see them - still do! How come I can't ? :rofl:

 

 

because you have already made up your mind... LOL

[zdo 20]

You're missing something or you're jerking my chain

 

I "made up my mind" in favor of fractals long ago... conceptually, market fractals are so 'wired'... I liked what I 'saw', still do ... but extended, actual experiences have forced me to unmake up my mind... there are neighborhoods of tf's that are 'fractal' but way above and way below and etc NOT!... false fractals - a result of culturally accepted, but inadequate representations.

 

I love Mandelbrot, etc.! I use fractal dimensions in some systems. But that said - I’ve concluded the only thing fractal across tf’s is up is up, and down is down. DBP has written so much better than I could about the pitfalls in the representations of the auctions – especially in bars. The map is not the thing mapped etc. If a three tick bar is already a rapidly degrading misrepresentation, then a weekly is not going to be an equivalent representation - it's just a worse representation, not a fractal in scale or behaviour.

 

Bottom line for me – the real auction has NO fractals above or below it…ie the actual auction is not a ‘fractal’ of a larger or smaller ‘economy’ - so I can’t figure why to be looking for fractals in the representations of the auction – especially since what I found when looking for them (for freakin years) didn’t hold up in practical trading and systems development…

Edited November 12, 2009 by zdo
don't know

[dangermouseb 10]

I guess I see the idea of fractals like the idea of any stats - a high level summary. The auction process I see as a detail. The fact the auction process can account for trends would explain a fractal timeseries with trends. The auction process (although concretely in the moment by moment order book) happens at different timescales etc and a larger order at a different timescale my not manifest quickly or continuously in the order book. E.g. large orders get worked over hours, days and weeks not just at the smallest level and a volume by price summary is fractal itself depending on the period viewed.

 

I'm still at the beginning of fractals - I don't expect to use them as a trading strategy other than maybe getting good at Hurst Exponent estimation. The higher the exponent the more trending and I understand different stocks exhibit different exponents (and probably at different times - heresy!!!) so I can use that as a sonar to find fish as it were. Other than that fractals at least explain that there should be money on the table until all the price series have a Hurst of 0.5 or Brownian noise - in which case it'll be time to find another livelihood.

 

--DM

[motorway 10]

I would refer people interested in Fractals To the link I posted on intrinsic time

 

You do not measure "coastlines" with a clock BUT a ruler

and when you use that ruler . It is end to end..

If you to do use a clock ( a wrong type of ruler esp because it allows overlap and noise ) you so deform that coastline as to destroy any fractal relationships..

 

conventional time series analysis, focusing exclusively on a time series of regularly

spaced observations, is far removed from both the fractal viewpoint and the real nature of

the raw data: a regularly spaced economic time series is not original data but a preprocessed

artifact. R B Olsen

 

So all this talk of 1 min 5 min 1 hour ,daily ,weekly etc

Is of no use ... Those increments are too static and coarse..

BIG problem is the incorporation of various volatility lags

 

The clock ticks

But the Markets Rhythms are broken when seen in terms of the clock

and cycles are APERIODIC

 

eg OLSENS ( FRACTAL ) 17, and counting, scaling laws

are revealed only when a ruler is used to measure the price "COAST LINE "

 

and tIME becomes simply what WYCKOFF tells us it was

 

a measure of DURATION...

 

I do not believe you will find true fractal relationships

on fixed time scale charts

or ones that incorporate overlap ( The ruler MUST be used END TO END )

 

 

Motorway

[dangermouseb 10]

Good thoughts.

 

Olsen seems to be presenting a stochastic time view of the world - which is fine, it certainly helps with the fat tails seen in the market, but I'm not sure it helps to understand or determine the existence of trends in the price series or not. Also I'm not certain how much would be lost because of the one-way transformation from intrinsic time into clock time. My gut feel is that given the roughness of tools like Hurst Exponent estimation in the first place it wouldn't matter too much.

 

Not sure what you mean by "BIG problem is the incorporation of various volatility lags"?

 

--DM

[zdo 20]

motorway,

 

Thanks for your post.

Unfortunately after an as yet undefined number of wash and rinse cycles in the real world, we'll find that “end to end” is ultimately just as artificial and unworkable as “timeframes”

[motorway 10]

I also think we can expect to much from "FRACTALS"

 

END to END

 

Point and Figure is

 

RENKO ( Interesting exploded form of P&F ) is

 

RANGE BARS ? Ain't ! They overlap and you can not measure anything very well if your ruler over laps

 

OK

 

From Olsen's research

 

17 scaling laws discovered so far, e.g.

– When a directional change of r% occurs, it is

followed by an overshoot of r%

 

– The time for the overshoot to happen is also

highly correlated to the time taken for the

change of direction to happen!

 

OK this is Fractal because it does not matter what the size of r is

 

Directional Change Threshold can be .01% or it can be 5% or nearly ( there are boundary effects ) anything

 

BUT these relationships are ON AVERAGE..

 

They are useful as measures of divergence , trend , changing of behaviour etc

 

Why are there trends when at any time there is 50/50 chance of movement up or down ?

Because the size of moves are not a 50/50 proposition up or down..

 

My interest ?

 

P&F chartists were doing this over a 100 years ago

They called r ---> BOX SIZE

 

and knew that the P&F chart ( call it an intrinsic time chart . or unit chart as some did )

lent itself to such statistical analysis...

 

Olsen's heterogeneous markets hypothesis is important

 

based on the theory of heterogeneous markets. The models follow a top down approach based on this theory.

 

There are four primary trend components ( at least )

causing fractal behaviours

They have different trading time horizons

 

But all have to trade AT the same time ( zero lag There are NO TRADING TIME FRAMES )

 

There is an immediate trend

a short term trend

an intermediate trend

and a long (pull) trend

 

This goes back to DOW and WYCKOFF

 

 

The activity of the larger scale constrains and has large impact on the smaller scale..

 

Olsen in his olsenscale uses 100 trend increments

But thinking in terms of these four main trends is a good starting point

 

intra day

day

short term

long term

 

 

Motorway