JamesH83
|
|
| Total Posts: 697 |
| Joined: Aug 2004 |
| |
|
| if one is interested in stat arb trading, what courses should you take? time series and financial econometrics? |
¦(X)=(Nh)-1K{h-1(X-x1) + ... + h-1(X-xT)}, |
|
|
 |
Anthis
|
| It's all Greek to me |
| Total Posts: 1180 |
| Joined: Jul 2004 |
| |
|
You should master cointegration techniques at least. Kalman filtering and Neural Nets also. |
Αίεν Υψικράτειν/Τύχη μη πίστευε/Άνδρα Αρχή Δείκνυσι/Νόησις Αρχή Επιστήμης //Σε ενα κλουβί γραφείο σαν αγρίμι παίζω ατέλειωτο βουβό ταξίμι
|
|
 |
FDAXHunter
|
| Founding Member |
| Total Posts: 7526 |
| Joined: Mar 2004 |
| |
|
| A signal processing course and the relevant coding experience is going to help you more than any "financial econometrics" course. |
On tue un homme, on est un assassin. On en tue des millions, on est un conquérant. On les tue tous, on est Dieu. |
|
|
|
Baltazar
|
|
| Total Posts: 1455 |
| Joined: Jul 2004 |
| |
|
time series are special cases of sig. proc.
it also encompass signal analysis (wavelets, filtering, adaptive filtering), signal detection (neural net, svm, bayesian theory, sequencial detection), parameter estimations....
i don't know however if sig proc is the only approach to stat arb.
|
Qui fait le malin tombe dans le ravin |
|
|
Patrik
|
| Founding Member |
| Total Posts: 1043 |
| Joined: Mar 2004 |
| |
|
Baltazar,
Sometime I'm going to dig into signal processing a bit more than the minimal basics I've done so far. Would you mind listing some decent material on the various topics you just mentioned? (and please, just in English so I have a slight chance of understanding anything  ) Sort of signal analysis/detection/etc essentials according to Baltazar. |
Paper Trading - Capital Structure Demolition, LLC  |
|
|
|
JamesH83
|
|
| Total Posts: 697 |
| Joined: Aug 2004 |
| |
|
Thanks for the info. Am i correct in believing that most MSc Fin Math programmes offer few if any courses related directly to stat arb? |
¦(X)=(Nh)-1K{h-1(X-x1) + ... + h-1(X-xT)}, |
|
|
Anthis
|
| It's all Greek to me |
| Total Posts: 1180 |
| Joined: Jul 2004 |
| |
|
| Signal processing is a physics/engineering field. You should be very happy if you find out a finmath program covering brief topics such as wavelets and filtering. |
Αίεν Υψικράτειν/Τύχη μη πίστευε/Άνδρα Αρχή Δείκνυσι/Νόησις Αρχή Επιστήμης //Σε ενα κλουβί γραφείο σαν αγρίμι παίζω ατέλειωτο βουβό ταξίμι
|
|
|
|
Baltazar
|
|
| Total Posts: 1455 |
| Joined: Jul 2004 |
| |
|
i'll do that next week from uni Patrick.
i'll try to focus on material available on the net. |
Qui fait le malin tombe dans le ravin |
|
|
Patrik
|
| Founding Member |
| Total Posts: 1043 |
| Joined: Mar 2004 |
| |
|
| great B - looking forward to that. |
Paper Trading - Capital Structure Demolition, LLC |
|
|
|
jungle
|
| Chief Rhythm OfficerCSD LLC |
| Total Posts: 2921 |
| Joined: Jul 2004 |
| |
|
| LSE has courses in time series / econometrics that can be done as optional courses for the econometrics msc. |
it's axiomatic, deal with it. |
|
|
JamesH83
|
|
| Total Posts: 697 |
| Joined: Aug 2004 |
| |
|
yes, i was looking in to the MSc in Applicable Mathematics there. Is anyone familar with it?
let me know what you think:
http://www.maths.lse.ac.uk/MSc_course_details.html |
¦(X)=(Nh)-1K{h-1(X-x1) + ... + h-1(X-xT)}, |
|
|
|
Baltazar
|
|
| Total Posts: 1455 |
| Joined: Jul 2004 |
| |
|
I'll start with signal (aka time serie) analysis. (wich is my main meat) (for process this book seem very nice:
books.pdox.net/Math/ An%20Introduction%20to%20Statistical%20Signal%20Processing.pdf)
Fourier Transform can be use to illustrate frequency contents of a signal, hidding in the process the time-related content.
what if the signal has a frequency behavior that changes with time?
One can look at the phase of the FT, not handy.
Or one can devise time dependant spectrums.
first try: short time FT and Gabor:
let's cut the signal in pieces using a sliding window, and compute the FT.
problem1: the window has an impact on the representation.
problem2: discrete version do not come easy (no mathematical base of decomposition)
corolary:
this stft can also be interpreted as the correlation of the signal with an atom (cf Gabor) localized around some time-frequency location.
second try: wavelet
let's use scale instead of frequency (they are related: for an admissible wavelet(oscilating at a frequency f_0 and energy conditions), scale a= f_0/f)
pro: discrete version has far better mathematical aspects.
pro: the resolution depend on the scale of analysis. That is long and low frequency components are rougthly caractized in time but well in frequency, the converse for high frequency events.
pro: can be fast
pro: many signals seem to be composed of the same patterns at different scales and that appears on such graphs.
con: graph depend on the chosen wavelet-> how do we choose such wavelet?
(refer to :http://www.bearcave.com/misl/misl_tech/wavelets/index.html, for a very deep introduction on wavelets (computer aspects, discrete and even application to financial series smoothing/analysis))
the two described methods are linear with respect to the signal. They can be interpreted as decomposition of a signal onto atoms parametrized by time and frequency for the first and time-scale for the second.
cons: use of window that impacts the result
cons: do not have marginals properties (integeration along time axis do not give the spectrum)
cons: cannot be interpreted as energy distribution in the time-frequency plane or time-scale one.
pro: simple
pro: wavelet allow inverse transform at some conditions
we turn to bilinear methods
they are synthetized differently depending on the required properties.
One well known example is the WignerVille distribution, straigth adaptation from quantum physics function aiming at a speed position graph.
it is way more precise than the linear methods in time of time-frequency/time-scale resolution.
but as it is quadratic, it has a lot of cross terms
some illustrations can be found here on the pages of my collegue
http://www.inrialpes.fr/is2/people/pgoncalv/pub/tf-esgco02.ppt
one can also find stuff on wavelets and fractals.
there is few introduction textbook on bilinear joint signal representation as it is an "advanced" topic but if you want more i can find more
|
Qui fait le malin tombe dans le ravin |
|
|
