Trading Strategies Based On Digital Filters - page 40

 

.hst files

by work offline i mean without internet connected. It is just enough replace

existing gold files with my files and start MT than open new chart with gold. To this PC internet connection must be disabled/disconnnected otherwise golde files will be updated from the broker server.

OK, Christmas dinner time, I'm in Poland and we are celebrating Christmas

on 24....so MERRY CHRISTMAS to everybody !!!!

Krzysztof

 

trading Gaussian noise

Hi,

Back from Christmas dinner. I hope you figured out dealing with .hst files.

To be honest your tries of trading gaussian noise confused me a bit.

I'm not an expert of DSP but it was clear for me as a basic thing that you can not trade it.

There are two simple explanation of this.

1) We are doing technical analysis in order to extract information about

signal which for us in trend, patterns or oscillations. Noise is random

and does not have any information, hope you agree with this. How you can extract information if you dont have it !!!! That's the problem here,

random noise or signal with low S/N ration and we have a fake outs

2) From definition of DSP measurements. Please read below. Simply

noise only === 0 measurement precision.

That it shows that it works...curve fitting it's called I think. Meyers have very

good publication about it. You are adjusting your strateggy to amplitude of noise but you can not adjust in this way because we are at random conditions and it can change anytime... imagine trade inside this amplitude range of noise on small subrange like 1/100 than you dont' know if it will up or down because is random.

Krzysztof

Files:
1_3.jpg  114 kb
2_1.jpg  23 kb
ch2.pdf  470 kb
 

trading Gaussian noise continuation

I made a small test to prove that the curve fitting to amplitude happens

in attempt of trading gaussian noise. In order to obtain more random amplitude i simply multiplied noise signal by itself 6 times x^6.

See results below. Average Hurst component for raw data changed from 0.41 (previous signal GOLD1) to 0.52 so pure random. From smothed data didn't change at all (0.96-0.97).

Than somebody can transfer those files to MT and try to trade them offline or with trading simulator to see if it is really possible to make money trading noise.

I believe it is still not perfect random signal as it has still flat shape. The perfect one would be with random shape and contents of this signal but so far I don't know how to obtain this. Anybody can help ??

Krzysztof

Files:
 
fajst_k:
Hi,

Back from Christmas dinner. I hope you figured out dealing with .hst files.

To be honest your tries of trading gaussian noise confused me a bit.

I'm not an expert of DSP but it was clear for me as a basic thing that you can not trade it.

There are two simple explanation of this.

1) We are doing technical analysis in order to extract information about

signal which for us in trend, patterns or oscillations. Noise is random

and does not have any information, hope you agree with this. How you can extract information if you dont have it !!!! That's the problem here,

random noise or signal with low S/N ration and we have a fake outs

2) From definition of DSP measurements. Please read below. Simply

noise only === 0 measurement precision.

That it shows that it works...curve fitting it's called I think. Meyers have very

good publication about it. You are adjusting your strateggy to amplitude of noise but you can not adjust in this way because we are at random conditions and it can change anytime... imagine trade inside this amplitude range of noise on small subrange like 1/100 than you dont' know if it will up or down because is random.

Krzysztof

That is exactly my point,if I can trade your gaussian noise it is because it was not random(not because I have any specific freudian fixation with noise ),that is why I insisted you rechecked your gold1 file...it has an embedded cyclic component equivalent to 20 periods...from Wikipedia

"Gaussian noise is properly defined as the noise with a Gaussian amplitude distribution. This says nothing of the correlation of the noise in time or of the spectral density of the noise. Labeling Gaussian noise as 'white' describes the correlation of the noise. It is necessary to use the term "white Gaussian noise" to be correct. Gaussian noise is sometimes misunderstood to be white Gaussian noise, but this is not the case."

White noise`s sequence of values is statistically uncorrelated..Gaussian noise not necessarily so.

I doubt I could find any hidden structure in white noise,...But I found,easily,a tradable structure(20 periods cycle) in your gaussian noise,so,it is not white(random)noise...The fact that Gaussian noise has a smaller range of frequencies than white noise,all of them at an average power level...and that BOTH are mean reverting and have a gaussian pdf ,means that anybody could trade Gaussian noise by just using a volatility filter or std deviation bands...mean reversion +distribution among a smaller range of frequencies=tradable with "low" risk.

I would like to focus back on the task we both agreed was of fundamental interest:Elimination/reduction of noise

Now,the question is:What kind of noise do we find among time series of forex quotes?Pink,White,Brownian,All?Because we need to know (I don`t know,yet)WHAT we want to filter out in order to design the right filter or detector.

Regards

Simba

 

std

fajst_k:
I made a small test to prove that the curve fitting to amplitude happens

in attempt of trading gaussian noise. In order to obtain more random amplitude i simply multiplied noise signal by itself 6 times x^6.

See results below. Average Hurst component for raw data changed from 0.41 (previous signal GOLD1) to 0.52 so pure random. From smothed data didn't change at all (0.96-0.97).

Than somebody can transfer those files to MT and try to trade them offline or with trading simulator to see if it is really possible to make money trading noise.

I believe it is still not perfect random signal as it has still flat shape. The perfect one would be with random shape and contents of this signal but so far I don't know how to obtain this. Anybody can help ??

Krzysztof

Krzysztof,

Forget about wasting time with gaussian noise,it has been an interesting exercise ,for both of us I presume,until now...you can multiply by whatever factor you want *6,*9,*33...The fact that Gaussian noise is MEAN REVERTING and has a GAUSSIAN pdf,means that anybody could trade it with standard deviation bands and split entries(1 lot at 2std,2 more lots (if necessary) at 3 std...then close ALL when recross back trough the mean)...Please see attachment,your gold1 file(Thanks to you and to Daniil,I was able to open in mt4 as specified)with 2 and 3 std bands...self explanatory.

I suggest that we focus on discovering what kind of noise(s) we find among financial time series,their properties and the potential ways to reduce them.

My model of the market is that this is a beast that can be in any of 3 states,and switches among them with unknown periodicity:

1-Random:aka don`t trade

2-Antipersistent:Find the mean(dynamic mean)and trade deviations from it.

3-Persistent:Find the direction and jump on the bandwagon as long as the tickets are cheap

In theory,Hurst exponent should be able to determine the state of the market,then cyclic or trend analysis could easily take care of the rest...the problem is that Hurst exponent applications to Financial time series have produced dismal results until now...so,we need to focus on noise in order to erase it and apply cyclic analysis with digital filters to denoised price series.

JM Hurst`s(no relation with Harold Edwin Hurst,the hydrologist whose research led to the Hurst exponent) definition of trend was:Trend is the longest period cycle at work...so,by using cyclic analysis we can trade the market,be it either trending or ranging...but ,first,we have to eliminate noise.

Best Regards

Simba

Files:
gold1_1.gif  121 kb
 

Mean reverting

Yes that's true it is mean reverting than it is tradable because this is the information which we can extract to make trade decisions. It is even visible

in last amplified gaussain noise that it wants to return to the mean. Exercise was good. I was concentrating on signal/information extraction but it seems

that we have to consider pdf sometimes also.

WIth your statements I agree completelly. We must build test data series

which are non stationary and mixed with noise than try to filter it out.

From my site I can find the way to generate brownian motions as a next step,

I can also investigate some ways to filter the noise but I believe pararell actions can be more efficient.

Simba, are you using mathlab ??? It has 4GB + 1GB mathlab books but it is really worth, I believe without this it will be difficult to proceed. You can download it from the web (azureus) takes a few days but....

Another good tool is findgraph. it has 200 algoritms for interpolation, extrapolation, wavelet, FFT, NN, etc everything in one tool. Meabe some outputs from filtering must be sent to NN ??? I think Noxia did this way to make SSA casual..This tool is really worth to see. Of course initially it is a pain to learn to use new tool but at least you don't have to compile it like a GRACE from MTM toolkit. It took me 1.5 day to compile it and 0.5 day to learn

how to use grace even for several years I was using openwin under UNIX.

Krzysztof

 
SIMBA:
Thanks,I have been trying to do so,no appearance of gold1 hst...then I pasted and copied gold1 .hst on History files..when I try to access it trough File>Open Offline...I can`t.

Can you explain in detail how to do it?

Regards

Simba

1. Put gold hst files in your current data provider (e.g., I put them MetaTrader 4\history\Alpari-Demo, because I have demo account in Alpari)

2. Open File -> Open Offline

3. Gold files should appear here

4. Offline charts have "Offline" word in window title

Files:
openoffline.jpg  104 kb
 
Daniil:
1. Put gold hst files in your current data provider (e.g., I put them MetaTrader 4\history\Alpari-Demo, because I have demo account in Alpari)

2. Open File -> Open Offline

3. Gold files should appear here

4. Offline charts have "Offline" word in window title

Daniil,

Thanks, already solved this problem ,I even thanked both of you a few posts back

BTW:this was not the solution,this was what I already did,problem was to shut off terminal from internet,you can`t put the gold hst files in metatrader,unless using what is there,so,even if you open offline,while online,the file gets updated so you end up getting mixed data,try it and you will see ..the actual process MUST be done while physically offline..anyway,water under the bridge ,so,let`s focus on next steps,and thanks again for your kindness.

Regards

Simba

 

Mathlab

fajst_k:
Yes that's true it is mean reverting than it is tradable because this is the information which we can extract to make trade decisions. It is even visible

in last amplified gaussain noise that it wants to return to the mean. Exercise was good. I was concentrating on signal/information extraction but it seems

that we have to consider pdf sometimes also.

WIth your statements I agree completelly. We must build test data series

which are non stationary and mixed with noise than try to filter it out.

From my site I can find the way to generate brownian motions as a next step,

I can also investigate some ways to filter the noise but I believe pararell actions can be more efficient.

Simba, are you using mathlab ??? It has 4GB + 1GB mathlab books but it is really worth, I believe without this it will be difficult to proceed. You can download it from the web (azureus) takes a few days but....

Another good tool is findgraph. it has 200 algoritms for interpolation, extrapolation, wavelet, FFT, NN, etc everything in one tool. Meabe some outputs from filtering must be sent to NN ??? I think Noxia did this way to make SSA casual..This tool is really worth to see. Of course initially it is a pain to learn to use new tool but at least you don't have to compile it like a GRACE from MTM toolkit. It took me 1.5 day to compile it and 0.5 day to learn

how to use grace even for several years I was using openwin under UNIX.

Krzysztof

Krzysztof,

I don`t use MATHLAB...I am currently testing SYNAPSE by PELTARION,I suggest you check the 30 days free functional demo.

I will check MATHLAB and Findgraph too.

Yes,parallel actions are best at this stage.

Regards

Simba

 

Brownian motion

I posted info and time series here

https://www.mql5.com/en/forum/178285

Krzysztof