Re: Best search algorithm?

Martin Sewell (
Mon, 26 Oct 1998 05:10:47 +0100 (MET)

Hello Lester (and everyone else)

>No, the goal is not solely to "predict" in the sense of trying to guess
>where the time series will (statistically or chaotically) end up in the
>future, as you precisely suggest. I think by now that fad has faded.
>Rather, good indicators and algorithms defining good indictors
>typically are used as input to decision making on current patterns of
>information. Many succesful trading strategies still make lots of
>money just discerning mispriced bids and asks.

Yes, I accept this, but I didn't use words like 'solely', I used 'boils down
to' - which implies I was reducing my statement to the essentials, i.e.
summarizing. Your point about arbitrage is very true - but it is not
something I was trying to model. The most important part of any trading
model which does not fall under the heading 'time series prediction' is
probably risk management.

>BTW, I think that Matt Kennel's reply to your posting in sci.nonlinear
>has much of the same criticism as I have presented.

Yes, his comments are appreciated.

>I saw your same posting in sci.nonlinear and sci.op-research. You
>should have cross-posted, as I am doing now.

I have this time! I posted separately before because I thought
cross-posting was not considered the 'done thing'. I'm learning . . .

>The relevant issue is
>that what you said was incorrect.

'Incorrect' is a little strong, 'appearing naive', 'misleading',
'misunderstood' or 'oversimplifying' I'll accept. Although I think my
earlier comments have adequately explained the reasoning behind it. In my
initial simple question to mathematicians about search algorithms it was in
context. In this later discussion with experts in the field of trading
systems it is now out of context.

>I happen to think it's a public
>service to correct some people's misconception of how they can make a
>quick buck by "simply" applying their raw confidence in their
>mathematical abilities to trading.

As a public statement that's fine. However, I am embarking on a three year
PhD, probably for the sum of money which one can expect for such studies -
so I'm most unlikely to make a quick buck!

>:Thanks for the ad........vice :)
>Strange reaction? I was offerring some insight based on experience. I
>guess you'd feel better if you didn't know that statements were made on
>the basis of detailed knowledge of the subject you were presenting? I
>think we have plenty of opportunities to make money without having to

Okay, sorry! I (publicly) apologise for such a cheap jibe. I hoped the
smiley would lessen it's impact!

>:>Yes, markets are ruthlessly random, as many suprised "experienced"
>:>traders have found recently.
>:I don't think we want to go around telling people the markets are random.
>:EMH and and technicals do not a happy couple make!
>I think most intelligent people already believe there is a strong
>"random" component in markets. Last year's Nobel prize recognized that
>explicitly. More recent work since 1973 has demonstrated even more
>stochastic influences on markets, e.g., some recent papers by lots of
>people, including myself, have made the case that even "volatility"
>(the term used for the scale of a stochastic process) in many markets
>should be considered a stochastic process itself. (This does not
>specify just how this stochasticity -- or "apparent" stochasticity
>might arise from information flow between many people, etc.)
>OF course the markets are anything but a "pure" random process. If you
>examine even these zeroeth order stochastic models, you will see lots
>of nonlinear multivariate "deterministic" influences that require more
>than just flipping a coin to decipher.

Okay, okay - it all depends on our definition of random. I was thinking of
random meaning 100% random.

>I'm sorry if some of this sounds "testy," but I see no reason not to
>respond in detail to public statements about my postings.

No problem. That is precisely what I was doing! (Note also the impressive
speed at which we've both responded!)

Now we can get down to business?

Do you have any opinions on the various nonlinear dynamic models used for
prediction? Such as neural networks, Markov chains, local polynomials, etc.
What else should I be looking at in this area?

Regarding search algorithms, you see it all boils down to . . .(only
kidding). I was considering using genetic programming with binary trees and
the logical operators to combine trading rules which had each been optimized
individully using GAs. Is this wise - or should the optimization be done in
one go? I guess the first method would appear more sense - but then ignores
interdependencies. Would ASA help solve this problem?

I often think of there being two main ways of tackling the problem. One
route has more emphasis on prediction and little emphasis on trading rules.
The other puts more emphasis on search algorithms and searches for optimum
trading rules. Any suggestions? The disadvantage of the first method is
that it is more of a black box approach. The disadvantage of the second is
that you still end up with trading rules - which everyone else has.

Finally, is there any point at all in using fuzzy logic? I've yet to find
it. Why not use probability instead?

If you're interested in some serious rocket scientist stuff, take a look at: (1-page abstract) (17-page paper)
It's by a guy from the University of Birmingham (nope - not Alabama!) and
involves Gauge Theory of Arbitrage.

Thanks for the constructive comments - and for giving me more of an insight!



This message was posted through the fuzzy mailing list.
(1) To subscribe to this mailing list, send a message body of
"SUB FUZZY-MAIL myFirstName mySurname" to
(2) To unsubscribe from this mailing list, send a message body of
(3) To reach the human who maintains the list, send mail to
(4) WWW access and other information on Fuzzy Sets and Logic see
(5) WWW archive: