backtesting the original turtles
- Thread starter mind
- Start date
Oh really ? You did so whereas I didn't provoke you.
Now after that he want to play the virgin that has been touched haha !
Now after that he want to play the virgin that has been touched haha !
It was modified statement of Marat.
Meaning was that you might be personally comfortable with big drawdown, but your boss ( customers ) will not . Therefore , if you want to keep managing money, you must do tweaking, like it or not .
It is very likely that customers will withdraw money before drawdown is over .
You, of course, have a right to do whatever you want.
Good luck with your project.
It is you and Mind that will go away into my ignore list so do the same if you are not masochist and don't thanks me for the suggestion since it's your interest not mine so I don't care that you do it or not 
!
!Correlation and dependance are two different things or rather correlation is only a particular case of dependance. A deterministic law is also a particular case of dependance in fact it is the most obvious case of dependance. Since nearly nobody supposed at the beginning that there can be any deterministic law they then suppose that the only kind of dependance would be correlation. And there have been largely shown by thousands of researchers in the world that the market except on very short term exhibits no evident statistical correlation. Then researchers turn towards detecting some non-linear dynamics detection tools - that is to say they finally came back to deterministic law hypothesis without having the ambition to find it but just have a clue if it could exist - like Lyapounov exponent but unhapilly it is still too weak to be really conclusive see also
http://www.elitetrader.com/vb/showthread.php?s=&postid=358735&highlight=lyapounov#post358735
"In econometric modelling - as in other fields - there are normally two class of modelisation :
- stochastic (from the simplest like ARMA, to more complex like Box & jenkins that are classically used in short term forecasting of products like champaign - I said that because it is on that that I learned them - and in financial modelling they have been extended with Garch and so like models - see the link I about Jean Philippe Bouchaud audio conference) and
- deterministic. The research on deterministic models is much recent in forecasting techniques and according to a statistical book "Methods for short term forecasting" (translated from french) it has only begun since the 1990 and at the moment nobody has never found something very convincing in financial modelling and one of the reason is that tools for detecting deterministic models are not very accurate. For example in another book I already quoted which is "modelisation of stock market returns" (sorry also translated from french) they show that the market must have some deterministic components but that the lyapounov exponents - which characterise such process - are too weak to be able to assess the hypothetical model. So by default they try to model long term memory effect with stochastics models that extend the classical ones above (farima, wavelet decomposition).
Is it astonishing ? No because to get a deterministic model you must know the real factors that is difficult since it would be like trying to read in the mind of God's market . And even using more sophisticated stochastic approach won't change the instrinsic wall of uncertainty attached to stochastic model not because of absolute impossibilty but because of the structural form of the class solution. Of course it will be possible to detect interesting things but it will stay shadows compared to a true deterministic model."
http://www.elitetrader.com/vb/showthread.php?s=&postid=358735&highlight=lyapounov#post358735
"In econometric modelling - as in other fields - there are normally two class of modelisation :
- stochastic (from the simplest like ARMA, to more complex like Box & jenkins that are classically used in short term forecasting of products like champaign - I said that because it is on that that I learned them - and in financial modelling they have been extended with Garch and so like models - see the link I about Jean Philippe Bouchaud audio conference) and
- deterministic. The research on deterministic models is much recent in forecasting techniques and according to a statistical book "Methods for short term forecasting" (translated from french) it has only begun since the 1990 and at the moment nobody has never found something very convincing in financial modelling and one of the reason is that tools for detecting deterministic models are not very accurate. For example in another book I already quoted which is "modelisation of stock market returns" (sorry also translated from french) they show that the market must have some deterministic components but that the lyapounov exponents - which characterise such process - are too weak to be able to assess the hypothetical model. So by default they try to model long term memory effect with stochastics models that extend the classical ones above (farima, wavelet decomposition).
Is it astonishing ? No because to get a deterministic model you must know the real factors that is difficult since it would be like trying to read in the mind of God's market . And even using more sophisticated stochastic approach won't change the instrinsic wall of uncertainty attached to stochastic model not because of absolute impossibilty but because of the structural form of the class solution. Of course it will be possible to detect interesting things but it will stay shadows compared to a true deterministic model."
Now I affirm to have this deterministic law at least one form - that is to say I have perhaps the Newton's form not the Einstein's form - but I have no intention to reveal it. But what I can say is that I can answer to the question: is there really dependancy in the market ? YES. But if you want to prove it by just using detection tools like Lyapounov good luck since researchers have tried and have not been very conclusive at least the mainstream researchers . And the fact that they can't prove means that the dependancy is not strong enough to exploit it in traditional way : the ARFIMA model which takes into account this dependancy gives poor surperformance.
. And the fact that they can't prove means that the dependancy is not strong enough to exploit it in traditional way : the ARFIMA model which takes into account this dependancy gives poor surperformance.agrau you are on my ignore list since one year: you can just piss off personal attack and never align one single technical argument.