|
PureBytes Links
Trading Reference Links
|
----- Original Message -----
From: "Bilo Selhi" <citadel@xxxxxxxxxxxx>
To: <systems-only@xxxxxxxxxxxxx>
Sent: Saturday, May 12, 2001 1:49 PM
Subject: Re: SO_Modeling The Market , study proposal... ( the unholy grail )
> ok, humor aside and whatever comes from this
> attempt at jump starting something good here
> i would like to at least go over the names of those
> market modeling and price forecasting techniques
> so that at least we know what's there. is that ok or
> should i forget bout it?
>
> anyway we can classify those as follows
> ( please, add subtract rearrange as you think is the proper
> because i can't catch them all ). first i will go over
> the classes and then we'll get into each one class at a time.
> for most financial engineers out there this is rudimentary
> info.
>
> 1. Stochastic ( probabilistic ) modeling same as
> Statistical approach or modeling the market based on
> stochastic process. aka the traditional approach.
> aka the main stream. aka the most common approach.
> aka the econometric approach. can be branched out
> into about hundreds of variation models and who knows
> maybe more.
> but can be generally divided into two main subclasses:
> Bayesian and non Bayesian.
> the concept is since we don't know the exact processes
> behind the price action all we can do is try to model
> price behavior based on prior statistical observations.
> there are a few models that i kinda like among the above
> but mostly only bits and pieces are worth taking.
> this is what i call the un holy grail approach.
> this approach is maybe 100 years old now. it's used
> very widely on just about every high end wall street firm,
> heavy options and just about every financial engineer
> is heavily trained in this field.
> FYI, the word stochastic has little to do with the stochastic
> indicator...
>
>
> 2. Deterministic modeling. aka the analytical approach.
> this class is non existent and has not been filled yet.
> aka the holy grail route. aka the non existing unifying
> market theory. that there we understand 100% of how
> the markets work. the laws are knows, the math is known
> and we can fly to the moon. to this day there are only
> vague attempts at modeling those processes. and there
> are no known solutions. ( compare that to the 100s of
> different stochastic techniques and think about all those
> financial engineers wasting time and research dollars ).
> this approach is touch and involves hard core dedication.
> i am not going to say that i got it figured out but what can
> and will say that i am and some of the buddies are trying
> hard at it. we found that getting to the truth helps in
> understanding and therefore helps in terms of the deep insights
> into the market action. the results are better, simpler and
> more elegant smart universal systems that kick ass of
> just about any other approach.
> there is one theory out there that attempts to be the holy
> grail ( it's not ) that i am sure everyone has heard of called
> "Efficient Market Theory". we will come back to those
> later.
>
> 3. Chaotic aka NLD ( non linear dynamics ) modeling.
> although i prefer to separate the two approaches. ie
> the chaos theory and NLD approaches, they though
> overlap. so we can break them down into two separate
> subclasses, namely Chaos based models and NLD
> modeling. the main argument for this approach is this:
> since a ). we don't know that deterministic process that
> controls the market behavior and b). the stochastic
> approaches don't really work well and c). price action
> kinda looks like chaos or d). price is highly non linear
> then a better model can be constructed by utilizing the
> laws of chaos and NLD. this approach is what i call
> the alternative approach. it requires some top knowledge
> and hours of research. the payoff might be a bit larger
> compared to the stochastic approaches.
>
> 4. this one is kind overlapping with the non linear approach
> but i would like to put it into separate class of it's own.
> The AI ( artificial intelligence ) approach. the subclasses
> are NN, NFL, FL, GA, adaptive rule based systems,
> and all possible combinations of the above. this approach
> gained momentum since about '80s when AI technology
> went mainstream . the argument is that a). again we
> don't know the processes behind the price movement
> b). we don't need to know them since no one knows them
> c). we have this wonderful new technology that models
> human "thinking" that should pick those tops and bottoms
> better than us so let's train the sucker to do the
> trading. therefore this approach is what i call "the endless
> tweaking" approach. since it gained so wide popularity
> there are now might be hundreds of different approaches
> within this class of market modeling. all kindsa stuff started
> popping out left and right. the best example now would
> be the models that "evolve" with the markets.
> however one has to admit that this approach is now dying
> off or at least tapering off quite a bit since to any sharp
> financial engineers the flaws are highly visible.
>
> 5. the ESOTERIC modeling approaches. those are interesting
> and sometimes funny to go through but there is quite a following
> mostly among not so math oriented traders. this class is therefore
> worth considering. mathematically usually there are no processes
> in there, just plain wooden theories. an example would be
> Elliott or Gan or Fib or Astro or Delta etc... sometimes those are
combined
> with stochastic modeling or with AI modeling. call it voodoo tech
> analysis but some traders swear by it.
>
> 6. HYBRIDS... take bits and pieces of 1 thru 5, mix them up and you
> got a hybrid model. there are hundreds... but by all means let's hear
> the names... at least and reference to some literature would help.
>
> 7. MANIPULATION
> i think special consideration should be given the following approach
> which is not really the modeling technique but the theory that
> the market are manipulated and therefore can not be modeled
> by the outsider
> ( you can't model manipulation without inside information )
> unless you are the one who controls it. this is a very narrow
> pessimistic approach that does not have a math description of the
> process and has very little following. but if you've been around
> enough you should know that it might well be the case the deserves
> consideration.
>
> 8. another class i would like to propose the NEW TECHNOLOGY
> class where we throw everything that's can not be classified as the
> above. if you got or heard of something that clearly not stochastic
> not chaos based not esoteric and not AI then we sure would like to
> hear about it.
>
> 9. next i would like to mention not really the class but a group
> of TECHNIQUES that we use many times without really modeling
> the process. there are hundreds of techniques borrowed from
> just about every field of mathematics such as all kinds of fit
techniques,
> smoothers, filters, detrenders, "predictors", LR, wavelets, FFTs you name
> it it's all there... but it is worth considering because some of the
those
> math techniques are like little helpers. so we can throw all those into
> a separate class...
>
> 10. now we come to the very interesting part. so far in look at the
> above classes there is one common thing that we can observe
> namely: stochastic, chaos, nld, ai, techniques are all field of
> mathematics that
> we utilized to do the modeling... they are "borrowed" ( except 2 )
> from vast math resources that we have available to solve the problem.
> therefore they are not market theory by themselves... and here come to
> the interesting part which is STAND ALONE MARKET THEORIES.
> and i would like to put those in the market theory class to avoid
> confusion
> with those modeling techniques that don't have no theory...
> there are just a few and you can count those on your fingers on one
hand
> probably, namely:
> The Efficient Market Theory
> The Random Walk Theory
> The Mean Reversion Theory ( although this one is more of a stochastic
> type deal )
> "The Elliott Wave Theory" :-)
>
> here is the summary:
> 1. Stochastic models class ( utilizing statistics )
> 2. Deterministic models class ( various )
> 3. Chaos / NLD models class ( chaos math and NLD )
> 4. AI models class ( NLD, neural mechanics )
> 5. Esoteric models class ( various )
> 6. Hybrid models ( various )
> 7. Manipulation theory ( unknown )
> 8. New Technology ( unknown )
> 9. Various Techniques ( various )
> 10. Stand alone market theories. ( various )
>
> now that i classified them for you, please add subtract or correct.
> if there is enough interest then we could go into individual models
> within the class, share what works, what is not, etc...
>
> bilo.
>
>
>
>
>
|