Dynamic Micture Models

Quote from Terrylee:

limitdown
A mixture model is an ensemble of functions which number, location and parameters have been chosen to represent the density in a data set. The functions (also called kernels) can be any kind of distribution. This applet shows an example where you can choose a Gaussian or a line as kernel. http://www.neurosci.aist.go.jp/~akaho/MixtureEM.html

Dtrader98
The paper that first caught my attention describes a “dynamic” version of mixture models for which the out-of-sample extension is done through signal reconstruction.
http://www.aaai.org/Papers/IJCAI/2007/IJCAI07-468.pdf


In my case the data points represent the shifts of orders from one timeframe to the next so a natural out-of sample extension for it is the cycle in the larger time frame itself. Now that I know more about it, I think that Dynamic Mixture Models are able to learn the “quasi periodic” occurrence of these shifts. I might go ahead with some experiments…

Stoxtrader
I make dlls in c++. This way I know what I do. But you are right I could also ask the MatLab community about it although I am mostly interested in getting some impressions from traders on the cycle shift phenomenon.

Hi Terry,

The paper looks interesting. I haven't read too much yet, but-- there is a thesis I think will interest you.
When I track it down, I'll PM you. It has to do with using hidden markov models, whereby each transition and emission probability is based on a mixture model that gets updated periodically (in this way it is constantly adapting). The results were pretty good, although the sample was pretty small, unfortunately.

Regarding regime switching using markov models, there is a well referenced econometrics paper under the author, Hamilton, if you want to look it up.
 
FE MS (financial engineering masters science) degrees deal with these things as class projects...

just because one can emulate and simulate does not mean advantage over other traders or market participants...

however, it does sell well to employers as well as other trading departments so that others will be employed at very high salaries, starting over $400,000 with bonuses ranging in the 2.5x factor (some guaranteed in writing, some only promised)...

what you are showcasing here are generally factored into black box systems and algorythmic trading systems....

one factor is or remains is even the so called competition has employed these methods and hence the advantage just about gets cancelled out in application....

so, will this be used for your personal trading or for an employer?
 
limitdown

I actually don't understand why quants are paid so much. Anyway, I do run my own investments. Sorry if I offended you and the readers of this thread by asking silly questions. I won't importune you with them anymore.

Wish you well.

Terry
 
Quote from Terrylee:

limitdown

I actually don't understand why quants are paid so much. Anyway, I do run my own investments. Sorry if I offended you and the readers of this thread by asking silly questions. I won't importune you with them anymore.

Wish you well.

Terry

intelligent questions don't in them selves offend, however, what you might have tapped into were the bruised feelings that so many of us maintain by all the carefully worded advertisings and research for free questions that are used to enrich the salesmen or overpaid employed quants seeking "alpha", as they like to say.

feel free to contribute, participate and laugh with the rest of us traders, but again, from the traders perspectives

hint,

it should be evident in one's comments that there is personal benefit from the thread, from the perspective of the trade and trades aspects....

sometimes we actually get huge hedge fund managers under guise, and someone later comes along and says that key wall street figures use those handles just to tap into the vibe, as it were....

mind you those are highly compentated, highly leveraged owners, who long past earned their keep on wall street, which collectively makes most of us traders, struggling new traders and soon to be traders wonder why they would associate in these grades instead of at their vaunted walled cathedrals?

ok, look forward to your next post...ok?, ok!
 
Quote from Terrylee:

limitdown
A mixture model is an ensemble of functions which number, location and parameters have been chosen to represent the density in a data set. The functions (also called kernels) can be any kind of distribution. This applet shows an example where you can choose a Gaussian or a line as kernel. http://www.neurosci.aist.go.jp/~akaho/MixtureEM.html

Dtrader98
The paper that first caught my attention describes a “dynamic” version of mixture models for which the out-of-sample extension is done through signal reconstruction.
http://www.aaai.org/Papers/IJCAI/2007/IJCAI07-468.pdf

In my case the data points represent the shifts of orders from one timeframe to the next so a natural out-of sample extension for it is the cycle in the larger time frame itself. Now that I know more about it, I think that Dynamic Mixture Models are able to learn the “quasi periodic” occurrence of these shifts. I might go ahead with some experiments…

Stoxtrader
I make dlls in c++. This way I know what I do. But you are right I could also ask the MatLab community about it although I am mostly interested in getting some impressions from traders on the cycle shift phenomenon.
Interesting paper, terry, thanks...

I have started reading it, but I notice some mention of SVD, as well as PCA and ICA. It also appears that there is some similarities between the problems these methods attempt to address.

Just out of curiosity, have you looked at/used simpler SVD-like techniques?
 
Quote from Martinghoul:


Just out of curiosity, have you looked at/used simpler SVD-like techniques?

How can you have a "simpler" singular value decomposition 'like' technique? It's just a matrix decomposition...
 
Quote from Corey:

How can you have a "simpler" singular value decomposition 'like' technique? It's just a matrix decomposition...
Simpler than DMM... I am wondering if the OP experimented with SVD-based methods before progressing to what seem to be more complex ideas.
 
Back
Top