Trading Strategies Based On Digital Filters - page 38
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Reply
Hi,
"Question to Simba. You tried to make a composite oscillator I believe. What was the idea about the weigth of different TF in this oscillator ??"
The first stage only included d1 cycles,at the end of the thread when the mtf composite indicators were posted at the thread I tried to develop a composite of the DOMINANT Cycles for each timeframeW1 to m1...if you try to visualize it,the composite should be attached at the lowest tf,so,at m1.
It was not useful for practical trading,I don`t know if due to mt4 platform or because really we were adding pears with apples,but the composite cycles were not tradable...so,I just ended up trying h4+h1...also to no avail...so,I didn`t post anything about it.
The solution was to use the mtf indicator in order to have,in a h1 chart the dominant h4 cycle,as mtf,and the dominant h1 cycle as a standalone.
I believe that each timeframe has different dominant cycles,and,even if some of them are harmonic,not all of them are..so,you hav eto analyze each timeframe,IMHO,Goertzel is best for fast ,dirty and practical results in real time.
Your Hurst exponent,I will check it,I have tried and tested many modifications of Hurst and FDI(FDI=2-H),and frankly,never found one that worked as a good filter with my cycles,what I found useful was volatility indicators,specially the synthetic vix ,since vix tops usually coincide with cycle bottoms and viceversa for vix bottoms and cycle tops,using Goertzel you can nail a robust period(usually 20 to 30 periods work best for most timeframes) for the vix by fitting it to the cycle or half cycle.
I am still interested in the idea of a multiple timeframe composite,if we can solve its "chart visualization apparent problem",so,if you have any ideas please say so.
regards
Simba
Hi Simba,
You wrote
1-MESA is not very good for noisy data,so,either we use it with a S/N filter,like Damiani`s volatimeter or we use it on smoothed data or we expose ourselves to nasty surprises.
Than I made a test of Damiani Volatmeter. I applied to it gauss noise so it should show no signal. See below. It shows total b.s. a lot of green signal
above grey.
I checked the code and this what it shows is
ATR(1) STD(1)
------- - -------
ATR(2) STD(2)
So kind of change of range or volatility but you dont know if it is because
of chage of signal amplitude or noise amplitude....so it has nothing with S/N ratio.
If you still have Dickey-Fuller document on your PC can you post it here. It disappeared from the link in FF (and the excel sheet)
KrzysztofKrzysztof,
I wrote that I prefer Goertzel to work with noisy data,this is what I do,and I do other things too like smoothing the data first before passing it trough the dfg...I don`t use Damiani`s indicator,I just said you needed to use a S/N filter,if that was your choice,or smoothe the data,or,etc,etc...I mentioned Damiani because I know it has been studied ,as a S/N filter,at several threads at this forum,and he is a trader whose work I respect..
I find your posts extremely perceptive and interesting,so,please,do not take my words out of context,I don`t have neither the time nor the wilingness to discuss minor issues,my main point was that you shouldn`t work with raw data unless you use Goertzel,and that if you wanted to use Mesa you should at least smoothe the data or /and use a S/N filter...I wasnot peddiling Damiani`s indicator,so,why did you focus on it is interesting to say the least.
If you want to expand your already notable contributions to the thread,please be so kind to do the same exercise you did with Damiani`s,with several other options we could use to filter noise..I am specifically interested in the synthetic vix,your Hurst indicator or any other you may find interesting,I don`t know,probably a homodyne discriminator too?
Sorry,I don`t have the excel sheet at my pc,nor did I delete it at FF thread...you can check at the "attachments button"-top right part of the thread,at FF,if it is not there I am lost,you could ask clahn,since he too posted an excel sheet,or you could ask CB or FF administration why did they delete the attachments.
The Cause of Cycles:What causes cyclic behaviour in moneyflow?what causes cyclic behaviour in investors ,traders and speculators?why is that so at ALL timeframes?I know many people interested in cycles,but few of them even wonder about what causes cycles..what causes the 56 bars cycle present at GBPUSD and GBPJPY both at h1 and h4 timeframes?...It appears,then disappears,then appears again..but it is there.oh,yes you can see it as 55 or 57 or whatever similar period to 56...but it is the same cycle and it is 4x(good choice of words
)harmonic
AS a sidenote,you may have proven that Damiani indicator is bs..but he was a very good trader and obtained exceptional results with it,you know,most of the time it is not the tool,it is the user.
"In early times in Japan, bamboo-and-paper lanterns were used with candles inside. A blind man, visiting a friend one night, was offered a lantern to carry home with him.
"I do not need a lantern," he said. "Darkness or light is all the same to me."
"I know you do not need a lantern to find your way," his friend replied, "but if you don't have one, someone else may run into you. So you must take it."
The blind man started off with the lantern and before he had walked very far someone ran squarely into him.
"Look out where you are going!" he exclaimed to the stranger. "Can't you see this lantern?"
"Your candle has burned out, brother," replied the stranger."
Best Regards
Simba
gaussian noise
Could you run the HUrst exponent on you so called gaussian noise?...just from a visual inspection it looks apparent that it exhibits cyclic behaviour...antipersistence loos present there?And Damiani bs nailed 7 out of 8 signals it gave for your so called gaussian noise..surprising.
different spectrum analysis
Hi,
I believe MESA results were your inputs. First I don't know if it is possible to use MESA for non stationary series (changing frequency). I didnt make any test except which I published here when in the signal was 300bars of gauss noise and 300 bars of stationary signal with noise. So it was a noise + stationary serie.In this case it shown wrong results already.
For sure it is not possible also to use FFT for this, only wavelet transform works well for non stationary. In order to use FFT it is necessary to use window function and it is assumed that inside window data is stationary and we will not have an information describing how frequency is changing in time.
Here is the example of attempt of using DFT for stock analysis
DFT of a non-stationary time series
this site describes in detail application of wavelets to financial series. It has also chapter about Hurst component.
here is another very good tutorial where DFT and Wavelet are described
and compared step by step.
INDEX TO SERIES OF TUTORIALS TO WAVELET TRANSFORM BY ROBI POLIKAR
So we are dealing with non stationary, changing frequency series which are most likely non continous with random data in between. In this case taking measurement from High TF first introduces errors because of low number of samples. See this.
https://www.mql5.com/en/forum/178842/page6
than we have a big chance to have a random data in between which influence for this measurement in unknown for me.
So I believe it is necessary to approach this problem step by step and find the method first which will signalize randomness of the market (Hurst exponent or noise measurement and filter) than analyze data which we are sure is not random e.g. by inverse wavelet trasform.
The other aproach is Ehler approach of using FFT with window for I believe
short term trends. See attachement.
Krzysztof
gaussian noise
For generating the signals I used Sigview, than I smooth it with 15sma
https://www.mql5.com/en/forum/178842/page7
I already run Hurst tool against gaussian noise and got similar results. Here is a panel view for generating signals.
Krzysztof
noise and noise + signal FFTs
and here comparison of FFTs of noise and signal with noise. In second case 2 clear peaks, the same like in DFs MESA with amplitude 0.4 and 0.54. Pure
noise has one peak 0.22. perhaps this is the reason that Hurst tool shows something.
Krzysztof
Hi,
I believe MESA results were your inputs. First I don't know if it is possible to use MESA for non stationary series (changing frequency). I didnt make any test except which I published here when in the signal was 300bars of gauss noise and 300 bars of stationary signal with noise. So it was a noise + stationary serie.In this case it shown wrong results already.
For sure it is not possible also to use FFT for this, only wavelet transform works well for non stationary. In order to use FFT it is necessary to use window function and it is assumed that inside window data is stationary and we will not have an information describing how frequency is changing in time.
Here is the example of attempt of using DFT for stock analysis
DFT of a non-stationary time series
this site describes in detail application of wavelets to financial series. It has also chapter about Hurst component.
here is another very good tutorial where DFT and Wavelet are described
and compared step by step.
INDEX TO SERIES OF TUTORIALS TO WAVELET TRANSFORM BY ROBI POLIKAR
So we are dealing with non stationary, changing frequency series which are most likely non continous with random data in between. In this case taking measurement from High TF first introduces errors because of low number of samples. See this.
https://www.mql5.com/en/forum/178842/page6
than we have a big chance to have a random data in between which influence for this measurement in unknown for me.
So I believe it is necessary to approach this problem step by step and find the method first which will signalize randomness of the market (Hurst exponent or noise measurement and filter) than analyze data which we are sure is not random e.g. by inverse wavelet trasform.
The other aproach is Ehler approach of using FFT with window for I believe
short term trends. See attachement.
Krzysztof"So I believe it is necessary to approach this problem step by step and find the method first which will signalize randomness of the market (Hurst exponent or noise measurement and filter) than analyze data which we are sure is not random e.g. by inverse wavelet trasform."
Yes,I agree with you re noise that this is the optimal road...regarding wavelets,they repaint,I used the foretrade wavelet for more than a year testing data in xls and they repaint,same as for SSA though they are much better than DFT which is not worth a dime for noisy non stationary data..IMO
I really would like to know more about how you use your Hurst exponent.
Regards
Simba
For generating the signals I used Sigview, than I smooth it with 15sma
https://www.mql5.com/en/forum/178842/page7
I already run Hurst tool against gaussian noise and got similar results. Here is a panel view for generating signals.
KrzysztofCould you please post the Hurst tool similar results,exactly what they were?What did it tell you about gaussian noise?Did the sma15 smooth affect the results?
Regards
Simba
and here comparison of FFTs of noise and signal with noise. In second case 2 clear peaks, the same like in DFs MESA with amplitude 0.4 and 0.54. Pure
noise has one peak 0.22. perhaps this is the reason that Hurst tool shows something.
KrzysztofThe 0.22 noise peak had a frequency of about 0.025 hz...so,about 40 periods...why?Smoothing method?
Hurst results for gauss noise
Don't ask me why it shows like this. Obviously smooth of gauss noise changed
results. For clear gauss noise only whittle estimator shown exact 0.5 however others are close.
I will try those wavelet indicators from this site but I think you agree that the most important is to know when we play roullette and when we really trading
/Krzysztof