neural nets

[Ben-163]

For an artificial intelligence project a group of us wanted to write
something with neural networks.  Can anyone suggest a scenario where
Neural networks could be used for pricing? And what is the main idea in
that scenario.  I understand they can be used for pricing options, but
id like to work with something simpler as the lecturer and
the rest of the group have to understand it.  Id like to try and capture
bond price movements or something like that, but I cant see how we can
predict things like this based on past movements... Can anyone give me any
pointers in that right directions??


Regards
Ben


--
Ben Wootton
[hidden email]

[Ferdinando M. Ametrano-2]

Ben Wooton wrote:
>Id like to try and capture bond price movements or something like that,
>but I cant see how we can predict things like this based on past movements...
As far as I'm concerned I can't see how, but I would be happy if someone
can and is willing to share ... ;-)

I've always smiled at neural nets, but would stop if proven wrong

ciao -- Nando

[Ben-163]

I read something about capturing stock price movements.  This was
partially based on past movements, but could also factor in various news
events etc.

I think the idea is more that the neural network looks for long term
trends which might not be apparent.  Personally I think the only reason
technical analysis works anyway is because of the psychological effects -

for instance when a major resistance level is broken through- even if we
are spotting trends which most people havent found, then the herding
effect isnt likeley to be there.  That limits their use straight away IMHO.



Regards
Ben

--------------------
www.benwootton.co.uk
[hidden email]



On Sun, 20 Jan 2002, Ferdinando Ametrano wrote:

> Ben Wooton wrote:
> >Id like to try and capture bond price movements or something like that,
> >but I cant see how we can predict things like this based on past movements...
> As far as I'm concerned I can't see how, but I would be happy if someone
> can and is willing to share ... ;-)
>
> I've always smiled at neural nets, but would stop if proven wrong
>
> ciao -- Nando
>
>

--
Ben Wootton
[hidden email]

[Ferdinando M. Ametrano-2]

Ben wrote:
>I think the idea is more that the neural network looks for long term
>trends which might not be apparent.  Personally I think the only reason
>technical analysis works anyway is because of the psychological effects -
>for instance when a major resistance level is broken through- even if we
>are spotting trends which most people havent found, then the herding
>effect isnt likeley to be there.  That limits their use straight away IMHO.

I'm probably preaching to the choir, but there are 3 types of Quantitative
analysis: fundamental analysis, technical analysis and
derivative-pricing/financial-engineering.
QuantLib is about the latter.

ciao -- Nando

[Dirk Eddelbuettel]

  Ben Wooton wrote:
  Ben> Id like to try and capture bond price movements or something like that,
  Ben> but I cant see how we can predict things like this based on past
  Ben> movements...

Have you looked at the (fairly vast) existing literature on neural nets as
non-linear regression variants (and/or pattern matching algorithms) from both
the statistics and econometrics literature.

Pointers would be Brian Ripley "Pattern Recognition and Neural Networks",
Cambridge 1995, from the statistics corner, and papers from e.g. Hal White
from UCSD for the econometrics part. AFAIK neural nets mostly "survived" that
reasearch as valid non-paramatric tests for non-linearities (GNU R even has
two canned in its tseries package).

Hope this helps, Dirk

--
Good judgment comes from experience; experience comes from bad judgment.
                                                            -- F. Brooks

[Ben-163]

Indeed.... Id definetly be interested in learning more about how neural
nets can be used in derivative pricing too, its just this project has to
be kept simple.... thanks for the suggestions everyone--

Rgds
Ben



> I'm probably preaching to the choir, but there are 3 types of Quantitative
> analysis: fundamental analysis, technical analysis and
> derivative-pricing/financial-engineering.
> QuantLib is about the latter.
>
> ciao -- Nando
>
>

--
Ben Wootton
[hidden email]