[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[amibroker] Re: AB and Neural Networks [was: The Transcendental use of Data in Optimizations]



PureBytes Links

Trading Reference Links

> 
> I'm quite familiar with NN's.  In the past I've tested every 
product 
> there was and/or followed postings from places that touted the 
> results all of which fell on their face pretty quickly. I'd like 
> nothing better then to see for example the results of using NN's 
to 
> trade NDX/QQQ.  Set it up to use whatever input data you want for 
as 
> long as you want prior to 9/11/01 as train / test etc. data and 
when 
> you are satisified with the results run it on out of sample for 
> 9/11/01 to current and post the results.
> 
> 

snip
 
> Fred
> 

snip

   A couple of comments on NN(Neural Networks): They try to train
themselves to find predictablilty in the data, the only problem
is that the price action is not a predictable function, this is
why they fail. The same may be said for optimization, it will
seem to work on past data, but if the basic trend is broken
then the optimization fails. It boils down to, how to predict the
future.

   The only time you have predictablility is when you find
the parameters of a trading system that is being used to
influence the price action. The program traders are very
careful not to tip their hand, so others can see what they
are doing. NN's would sniff out their actions because they
could be predictable. Some writers claim this is institutional
or fund traders, but even these folks don't like to tip their
hand.

   I have trading for a very long time now and have seen all
sorts of markets. I have learned the hard way about predictability.
Everbody would like the Holy Grail to be some sort of simple
function. 

   Do NN's have value, sure they do, if they are used to discover
predictibility. The time frame you chose in your comment is a 
test of this predicablilty because of the influence of external
functions. The basic question is should a system be reactive
or predictive. We can progam a reactive system, but I have never
had much success in programming a predictive system.

                          Cliff s



Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx 
(Web page: http://groups.yahoo.com/group/amiquote/messages/)

Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/