Your Market Beliefs - page 9

 

Simba,

The analysis and explanation is mind numbingly brilliant... I'm still digesting some of the concepts from the fractal geometry link. Kenny seems to be picking up what you're dropping though. Unfortunately, math and statistics was not a focal point for me back during the school days. I'm more of a right side brain kind of guy.

And right now, right side is saying, "this Hurst exponent business is looking pretty tasty". The obvious question, as I'm sure you've probably pondered it before, is how can it be used in a trading scenario. Is it realistic to say that a program can take the data inputs (price) live, process it on the fly, and display real-time floating Hurst exponents. When the value begins to decrease we "assume" that the random qualities (noise) of the market is dying down, indicating a higher tendency toward non-randomness (trend).

I may have a woody forming in my shorts .

Let me guess... I'm way off here, aren't I?

 

Simba, did you try the software library from Ndustrix: Home ?

 

cointegration

Hello Simba and all

very impressive analysis Simba like always

I study the codebreaker thread from the beginning

very intersting stuff but of no use for an non programmer...

the only thing where i get cought was this idea of cointegration

here is a link of explanation

Cointegration is not the same as correlation

So i tried to make contact to codebreaker to help to find out a method how I can do a test to find two cointegrated pairs:

hi codebreaker

but its to complicate to figure this out for me.

i would need a step by step answer what to do to perform this cointegration fuller test.

i dont want to steal your time. so maybe you can say to me which pairs you testet with this fuller test and give me some advice which pairs i should try to trade?

i think it depends also on the time frame which you trade or?

some currency pairs cointegrate on 60 h timeframe but not on daily. Or Im wrong.

which excel sheet i need and how to do it

or let me know which pairs you know which are conintegrated ?

his answer ?

Hi

Cointegration is not easy to prove but not difficult either. Two time series are cointegrated if a LINEARLY WEIGHTED COMBINATION of them is STATIONARY. Individually, financial time series CANNOT be stationary due to the presence of TRENDS. (Actually, this brings us on to the most interesting part - how do you identify a trend?)

So start with your two currencies. You can do a Dickey-Fuller test on each (not hard in Excell - see previous posts) to prove to yourself that they are non-stationary. Then run various linear combinations. There are various methods but trial and error is easy on a computer. If you use 1% increments, e.g. 99%A + 1%B, 98%A +2%B, 97%A + 3%B etc you will only need 100 tests. At each combination, do a Dickey-Fuller test and test for the presence of a unit root.

IF you find a co-integrated pair you are very lucky and can exploit this for all it's worth! Wait until the pair wander apart then go long one and short the other. MAKE SURE THE HEDGE RATIO IS CORRECT!! If you do this right, you cannot really lose and long-term they will re-converge and you will gain on both the long and short leg.

This was how I bought my new car a couple of months ago.

Just test all pairs! There are only n.(n-1) of them so it won't take long. You can find the Dickey-Fuller test on the Web. I cannot spoon-feed you!! You must do the work yourself. Making money in trading is VERY HARD WORK!! You are right that the timeframe is important. That is why some charts work better than others in trading!

Cb

So this is great answer but not practicable for me, beause i have no idea of programming and math.

So if anayone has experience with this fuller test oder cointegration please inform us.

one other discussion is here

Cointegration of currency pairs - Page 2

thanks

LODOL2

 

^ Interesting. I will need to check if the statistics software I have access to has this test. I'm sure it does as it is being updated by an accomplished statistician in the industry. I'm still not understanding the presence of an unit root.

I wonder how long cointegration actually last, and what degrees of cointegration is exploitable.

 
Michel:
Simba, did you try the software library from Ndustrix: Home ?

Michel,

No,I didn`t,not yet at least..thanks for the link.

Regards

Simba

 
lodol2:
Hello Simba and all

very impressive analysis Simba like always

I study the codebreaker thread from the beginning

very intersting stuff but of no use for an non programmer...

the only thing where i get cought was this idea of cointegration

here is a link of explanation

Cointegration is not the same as correlation

So i tried to make contact to codebreaker to help to find out a method how I can do a test to find two cointegrated pairs:

hi codebreaker

but its to complicate to figure this out for me.

i would need a step by step answer what to do to perform this cointegration fuller test.

i dont want to steal your time. so maybe you can say to me which pairs you testet with this fuller test and give me some advice which pairs i should try to trade?

i think it depends also on the time frame which you trade or?

some currency pairs cointegrate on 60 h timeframe but not on daily. Or Im wrong.

which excel sheet i need and how to do it

or let me know which pairs you know which are conintegrated ?

his answer ?

Hi

Cointegration is not easy to prove but not difficult either. Two time series are cointegrated if a LINEARLY WEIGHTED COMBINATION of them is STATIONARY. Individually, financial time series CANNOT be stationary due to the presence of TRENDS. (Actually, this brings us on to the most interesting part - how do you identify a trend?)

So start with your two currencies. You can do a Dickey-Fuller test on each (not hard in Excell - see previous posts) to prove to yourself that they are non-stationary. Then run various linear combinations. There are various methods but trial and error is easy on a computer. If you use 1% increments, e.g. 99%A + 1%B, 98%A +2%B, 97%A + 3%B etc you will only need 100 tests. At each combination, do a Dickey-Fuller test and test for the presence of a unit root.

IF you find a co-integrated pair you are very lucky and can exploit this for all it's worth! Wait until the pair wander apart then go long one and short the other. MAKE SURE THE HEDGE RATIO IS CORRECT!! If you do this right, you cannot really lose and long-term they will re-converge and you will gain on both the long and short leg.

This was how I bought my new car a couple of months ago.

Just test all pairs! There are only n.(n-1) of them so it won't take long. You can find the Dickey-Fuller test on the Web. I cannot spoon-feed you!! You must do the work yourself. Making money in trading is VERY HARD WORK!! You are right that the timeframe is important. That is why some charts work better than others in trading!

Cb

So this is great answer but not practicable for me, beause i have no idea of programming and math.

So if anayone has experience with this fuller test oder cointegration please inform us.

one other discussion is here

Cointegration of currency pairs - Page 2

thanks

LODOL2

Lodol2,

Thanks for the post,yes ,cointegration is an interesting issue..

Regarding the Dickey-Fuller test ,I remember that both clahn and myself ended up being able to do it and posted our results on the thread,and the way we did,you should be able to repeat our steps and do so...and yes,CB sometimes borders on secret society methods when explaining trough hints...

BTW...Did you notice that what CB suggests..with 2 pairs..you can do with 3,4 or 5 pairs?...You can find a cointegrated basket(like Aleksey does) of 3,4,5 pairs,and you will probably reduce your risk level and the volatility of your returns.

Regards

Simba

 
Kenny Rogers:
Simba, very impressive analysis. BTW, I use 3rd party statistical software as well. Just finished a week training on the use of the software which is used in cutting edge medical applications. It is amazing the type of distribution curve in today's software. Most of which I have no idea what is being done to make the curve fit, but as long as the professor and designer of the software is up to date with the latest educational releases, it's no problem for me.

Hi Kenny(aka jbfx )

Thanks for the kind comments and the info about the software....3 questions...

1-Are the markets random or not?

2-In case they are not...are they predictable?

3-What is the name of the software you use?...;)

Regards

Simba

 

cointegration

thx kenny

here what i found on the net

unit root

Unit root - Wikipedia, the free encyclopedia

and fuller test excel sheet

Augmented Dickey Fuller Test in VBA [Excel VBA]

myebe its of use

regards

lodol

 

random numbers

I have taken the excel sheet posted by iGor on the first page of this thread,just a few hundreds of samples..I just used the "close" values,since the yale university soft just accepts 1 column data...I can repeat for open,or high,or close...no problem ..just check the attached .txt and check they are the same...

The results show a Hurst exponent of 0.4137..so,Hurst exponent clearly detects this is a randomly generated time series...one bin is way out of the "norm"..probably due to the scarcity of samples,only one bin,not several ones as before when I checked the 30 years of DM/USD data.......but Hurst exponent nails the randomness of this data series...while ,at the same time,Hurst exponent said that dmusd,yenusd,etc(I checked gbpjpy too H= 0.98XX ),Were not random..so,iGor,my friend,post as many series as you want,and i will detect if they are random or not...if possible,with more than 2k samples,if not possible,I will do an effort

If anybody can generate ,want to generate 2k or 3k random data..I will analyze them..and you will find the same result..if they are really random..H exponent will be near 0.5..if they are not,H will be near 0.9...you can`t escape Hurst,and,think about it...this is a name worth remembering .

Files:
 

Explanation

Mr.Marketz:
Simba,

The analysis and explanation is mind numbingly brilliant... I'm still digesting some of the concepts from the fractal geometry link. Kenny seems to be picking up what you're dropping though. Unfortunately, math and statistics was not a focal point for me back during the school days. I'm more of a right side brain kind of guy.

And right now, right side is saying, "this Hurst exponent business is looking pretty tasty". The obvious question, as I'm sure you've probably pondered it before, is how can it be used in a trading scenario. Is it realistic to say that a program can take the data inputs (price) live, process it on the fly, and display real-time floating Hurst exponents. When the value begins to decrease we "assume" that the random qualities (noise) of the market is dying down, indicating a higher tendency toward non-randomness (trend).

I may have a woody forming in my shorts .

Let me guess... I'm way off here, aren't I?

Mr Marketz,

There have been many intents to use the Hurst exponent(H) and its related cousin,the fractal dimension index(FDI) in trading...let me give several opinions on that issue and the related status of applying chaotic attractors in trading

1-Most if not all of the Hurst exponents and FDI that I have seen,do,not measure it right,the only one that seems to be rightly build is the wealth lab one,if any of the brilliant coders here can exactly reproduce that code for MT4,we may have a positive surprise....that is why I use the yale university tools.

2-The concept is simple..using H we determine,real time,the current status of the market...if persistent we use tools that work well for trading trends...if antipersistent we use oscillators ,etc and fade the moves...if random or quasi random,we either not trade or apply brilliant MM like the one iGor and you may be applying.

3-We have a related problem...which is...what must the sample size be?It may be solved by cycle analysis or we may be limited then to use above 2000 points of data,and consequently,the best strategy(mt4?)would be to use tick data trough an API with a ECN.

4-If you really want to have a hard on..check Aleksey Yudin`s work,he was published in this or the past month Stocks and commodities(just Google that)...and ,before that ,he had a thread at FF were he showed parts of his system and made brilliant predictions..I don`t remember the name of the thread,but his nickname at FF was/is Aleksey...I will summarize what he does,basically,he integrates several time series,composed of price,volume and open interest,..and then find 5 "strange attractors" where price must be in the near future..and price usually goes there ...he does this for an anticorrelated basket of currencies...and his results,from what I have been able to see,are impressive....Additionally you can check nickname Codebreaker at FF,his thread uses very sophisticated math to find cointegrated timeseries in order to find pockets of predictability,he is a scientist like Aleksey(actually a university professor) and is brilliant,but you need a very high math level to be able to learn from him,in any case,worth reading.

Regards

Simba