|
PureBytes Links
Trading Reference Links
|
At 07:43 PM 9/24/98 +0200, hans esser wrote:
>
>> Usual disclaimers. I have no connection, just a satisfied customer, blah
>> blah
>
>usual question - do you make money with it ? trade S&P with it ? :-))
>
>thanks
>rgds hans
Hans:
Aha, the $64K question.
The short answer:
Yes.
The long answer:
Probably because I'm doing something different than what most people try to
do and fail with. The most common approach to NN's is to think that they're
magic and that all you have to do is throw huge quantities of data at one
and it will make zillions of dollars. This, of course, doesn't work.
If you want to predict a time series with a neural network, you have to
start with data inputs that have predictive value. IMHO, this does not
consist of combinations of the traditional TA indicators -- what Bob
Brickey has so aptly labeled "gambler functions."
Thus, it still comes down to economics, market theory, etc. to determine
what (if anything) is helpful in this task. If you do come up with such
data, a NN can be extremely useful in finding the best combinations of such
data. This can be used for intermediate prediction, although I personally
believe that using neural networks as classifiers is better suited to a
trading system.
I'll also add that Fuzzy Logic is really useful in this regard. Fuzzy Logic
tends to mimic the way humans partition situations. As an example, crisp
logic would require the adjectives tall, medium and short as applied to the
height of a person to have distinct hard boundary values; e.g., tall is
over 5'11', short is under 5'3", medium is everything else. Fuzzy logic
would allow soft boundaries so that someone could be both "kind of tall"
and "kind of medium". This extends well to a trading situation (is the
market trending?), although at the end of the chain, a crisp decision to
buy / sell / hold obviously must be made.
Allan
"Common sense is the collection of prejudices acquired by age eighteen." -
Albert Einstein
|