The ideal timeseries to trade

To add comments, please log in or register
MrM
271
MrM  

I would like to start a thread where we could dicuss econometrics, non-linear dynamics and chaos theory. There's a lot of info on wikipedia for the uninitiated, but I would appreciate feedback from people with a good background in these fields. I have a Master in economics myself and I only do statistical program trading and I've started playing with Matlab recently (trying to be Jim Simmons :-).

So here's my way of looking at the trading game: instead of trying to find indicators or constructing custom indicators that try to predict the future as accurately as possible on the given timeseries (like EUR/USD, 30min or so), why not try to find the ideal timeseries to trade for a simple trading system?

I guess the perfect one would be:

-A stationary process?

-with a very high Hurst exponent?

-presence of a strange attractor?

-autocorrelation in the log-returns?

And then we could just apply a very simple trendfollowing system. But because all instruments are different on all timeframes, we would need to construct a statistics application for MQL4 (or matlab) that automatically scans all timeframes of all pairs and only selects the best ones to trade for our trendfollowing system.

what do you think?

mezarashii
189
mezarashii  

just a suggestion,

try to trade on the day before on and after a moon phase, markets tend to have higher velocity in their direction on these days.

Sergey Golubev
Moderator
113448
Sergey Golubev  
PipTrip
192
PipTrip  

Hey - you're thinking way too much now. Currency trading is far too simple than this. People don't become casualties in this biz b/c they don't and/or can't get the entries/indicators/trends/etc. right.

It's ALL discipline. (Oh, how many threads are on this subject in any given forum? Next to none!)

I like indicators. Most good ones can get you good entries/exits w/o even looking at the price. However, it's the mind stuff that gets me when I do get into trouble. Has NEVER been anything else. We spend WAY too much time feeding our addictions looking at charts, indicators, and past data. We pinpoint where we would've, could've, and should've gotten into trades when really it's only about trading live and from the right side of the chart.

My piece of advice to you would be to throw a MACD and perhaps 2 MA's onto a chart, be disciplined, use stop losses that have a purpose, use take profit areas that have a purpose, and do the same thing every single time you place a trade.

I say this not b/c I found this out on some silly website like Newbie Forex Factory but out of trading for a living, making lots of mistakes, and being successful.

MrM:
I would like to start a thread where we could dicuss econometrics, non-linear dynamics and chaos theory. There's a lot of info on wikipedia for the uninitiated, but I would appreciate feedback from people with a good background in these fields. I have a Master in economics myself and I only do statistical program trading and I've started playing with Matlab recently (trying to be Jim Simmons :-).

So here's my way of looking at the trading game: instead of trying to find indicators or constructing custom indicators that try to predict the future as accurately as possible on the given timeseries (like EUR/USD, 30min or so), why not try to find the ideal timeseries to trade for a simple trading system?

I guess the perfect one would be:

-A stationary process?

-with a very high Hurst exponent?

-presence of a strange attractor?

-autocorrelation in the log-returns?

And then we could just apply a very simple trendfollowing system. But because all instruments are different on all timeframes, we would need to construct a statistics application for MQL4 (or matlab) that automatically scans all timeframes of all pairs and only selects the best ones to trade for our trendfollowing system.

what do you think?
zupcon
313
zupcon  
MrM:
I would like to start a thread where we could dicuss econometrics, non-linear dynamics and chaos theory. There's a lot of info on wikipedia for the uninitiated, but I would appreciate feedback from people with a good background in these fields. I have a Master in economics myself and I only do statistical program trading and I've started playing with Matlab recently (trying to be Jim Simmons :-).

So here's my way of looking at the trading game: instead of trying to find indicators or constructing custom indicators that try to predict the future as accurately as possible on the given timeseries (like EUR/USD, 30min or so), why not try to find the ideal timeseries to trade for a simple trading system?

I guess the perfect one would be:

-A stationary process?

-with a very high Hurst exponent?

-presence of a strange attractor?

-autocorrelation in the log-returns?

And then we could just apply a very simple trendfollowing system. But because all instruments are different on all timeframes, we would need to construct a statistics application for MQL4 (or matlab) that automatically scans all timeframes of all pairs and only selects the best ones to trade for our trendfollowing system.

what do you think?

Looking at this simplistically, I think that in general, diversification across different time frames might just be as least as good a solution, as trying to identify a single optimum time frame.

MrM
271
MrM  

This thread is so not going in the right direction..

jimven
233
jimven  
MrM:
This thread is so not going in the right direction..

I agree and sympathize with you. But I suspect that very few traders have the background and expertise to contribute much to nonlinear dynamics and chaos theory.

I'm an engineer, so I have some math background. Still, I'm not too sure such studies would be helpful.

Frederik Gert Johannes Breytenbach
159
Frederik Gert Johannes Breytenbach  

Hi

Hi MrM,

I think this econometrics idea of yours is great. I've always been interested in it but I can't really contribute in terms of econometrics. However I'm more than willing to help with the programming. I know matlab as well but can program in mql as well.

PM me if I can help you in some way. I think this is the way to go. I don't believe in all these indicators because most of them stop working and need reoptimizing etc.... This sounds like the best idea on this forum - even better than NN.

Best regards

Derik

SIMBA
2078
SIMBA  
MrM:
This thread is so not going in the right direction..

MrM,

Yes,this thread is not going in the right direction,so,we are going to steer it.

I too have an MBA(1984),Though I don`t believe it gives me any edge in trading,nor in much things in life,besides having a head start ,years ago,when hiring for certain positions..since I became self employed several years ago ,I don`t mind about it.....and I apply several statistical methodologies to filter out trades,I have found that datamining programs and genetic Algorithms can help you in finding non overfitted rules for trading decisions..I am not confident enough about NNs,due to their obvious overfitting,but it sems that there are people out there that know BETTER

Back to basics..1-Yes,Hurst exponent is a must,since it gives us the autocorrelation exponent of sequential points in a time series..basically we need to devise an indicator that gives us the right Hurst exponent,so,that we can trade either the trendfollowing startegies or the countertrend..in theory this is the perfect trendfilter...If I give you a template ,can you reprogram it and do it right??

2-Strange attractors,yes they sound well..but,how do you find them?I have done 2 courses on Chaos theory applied to trading and ,besides having found some fractal rules with an edge..I have never found any way to estimate a fractal attractor,though I know there are ways to do it..Do you know how to do it?

We can steer this thread,just say if you can do what I asked for,and do not have any doubt that ,when we start here posting our joint efforts,if they show a minimum promise of success..we are going to get the help of several people that know how to kick ass..trading related

Waiting for your reply and regards

Simba

zupcon
313
zupcon  
SIMBA:

Back to basics..1-Yes,Hurst exponent is a must,since it gives us the autocorrelation exponent of sequential points in a time series..basically we need to devise an indicator that gives us the right Hurst exponent,so,that we can trade either the trendfollowing startegies or the countertrend..in theory this is the perfect trendfilter...If I give you a template ,can you reprogram it and do it right??

Hi Simba

I'm not sure if this helps anyone but there's a spreadsheet which can be used to estimate Hurst exponents down loadable from:

Index of /staff/M.Sewell/hurst/

regards

zu

MrM
271
MrM  
To add comments, please log in or register