Can a System be reverse engineered just from viewing daily trades/account statements?
- Thread starter CPTrader
- Start date
i have a system in place that will take an experienced person not more than two weeks to figure it out and i have one where several levels of complicated and non-intuitive steps make it literally impossible to reconstruct. as many other threads on ET an interesting subject is discussed so broadly that every statement is valid and nonValid at the same time.
peace
peace
Nonsense by definition something that is valid cannot be invalid because the term valid/invalid refers to binary logic. It exists ternary, quaternary logic etc. and many kind of other logics but then use another expression than valid/invalid. Even in the case of binary logic, a statement is CONDITIONAL to some PREMISCES that are different from one guy to another. If the guy doesn't even know the premisces then he is making a fuzzy statement which has no validity state yet because it's just an hypothesis.
It is in fact the premisces that constitute the real part of knowledge but some people prefer not to consider them as they just want to emit opinions.
It is in fact the premisces that constitute the real part of knowledge but some people prefer not to consider them as they just want to emit opinions.
If there is a large enough number of trades that have been done by following the same mechnical system then a NN can reverse engineer it.
Basically this is discovering the rules.
NN can be very powerful if used in competent hands. I was in one particular country where they were very concerned about the rigging and doping of horse races. Used a NN to detect these fraudulent activities with over a 90% succes rate.
I started out with NN in the futures market and made some very good money with it. (doubling in a month). But that was a system that had to detect its "own" rules, did not have a mechanical system as example. Later I gave most of it back again when it went out of whack.
If you have a succesful mechanical system then a NN can increase the results by finding additional rules. Or you can reverse engineer a mechanical system. How many trades are required all depends on the complexity of the system and what indicators have been used.
Personally I am not going back to NN any more as I have developped a few indicators that make for easy trading and easy adjustments in a shorter timeframe (read less trades required) than that a NN needs for succesfull deployment. No longer do I have a taste for "black boxes" as you do not know when they stop working.
marinus
