Trading Strategies Based On Digital Filters - page 51
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
Benefit of using such strategies based on digital filters is a possibility to make the most of profit by opening and closing positions more than a few times in case of continuance of flat. The basic shortcoming is that the breakout of lines of the channel can lead to substantial and baseless losses.
That sounds interesting. Where is this indiator located? cl
I have not published it yet, sorry :-)
I have not published it yet, sorry :-)
lol quite alright....i thought it was in some post that i missed....
cl
GOLD5 and noisy chirps
Here are my considerations about the tests signal given by Krzysztof.
GOLD1 - gaussian noise: MESA finds fake peaks. That's the nature of MESA. It always finds something to call spectrum peak. But if you add a sinusoidal signal of amplitude of the same order of the amplitude of the signal, then you have that all the fake poles fade away and only the true one remains. So you realize that that they where all fakes. It should not be difficult to program the analyzer to do that test.
GOLD15 - 2 sinusoids. It catches both in 200, 400 and 580 window length. If you put more than 580-585 it has problems. It must be some issue with the series being only 599 values. Always take care that you are not analyzing the end points.
GOLD5 - 2 sinusoids plus noise. Well, it is a lot of noise, but the analyzer correctly got the two frequencies as the main peaks of the spectrum, although there are other minor fake peaks (see first picture)
Things get slightly better when you apply noise reduction filters (especially kalman, median filter and FIR). Then other peaks lower a bit (but not that much)
Here is the spectrum with a pre-processing (digital filter with D1=18, P1=9) (see second pic)Hi
NOXA also made tests on GOLD5 so have a look for comparision
T2W Day Trading & Forex Forums
post 115
I attach 5 noisy chirps signal, underlying chirp is the same there. 3 i made myself they are with maks noise amplitude of 1,4 and 9 and two I extracted from NOXA charts. So you can make direct comparision with your MESA
Anyway, I think your MESA is working now so question is how to proceed.
Because if you want to choose P1 and D1 manually than this will work as long
as market conditions will not change so like CSSA. If automatically than there
is a problem how to sort and filter subcycles. For CSSA you are doing this manually using show eigenvector tool than Genetic Optimizer chooses the most profitable group of subcycles.
So what is your idea now ?? IS this adaptive FATL,SATL,FTLM,STLM,PCCI and RBCI way forward ??
Krzysztof
If you never trade with the trading strategies based on Digital filters then I have a good method for you..
just Replace your current indicators with the digital filters and see the difference, It will be more profitable...
Happy Trading
Here are my considerations about the tests signal given by Krzysztof.
GOLD1 - gaussian noise: MESA finds fake peaks. That's the nature of MESA. It always finds something to call spectrum peak. But if you add a sinusoidal signal of amplitude of the same order of the amplitude of the signal, then you have that all the fake poles fade away and only the true one remains. So you realize that that they where all fakes. It should not be difficult to program the analyzer to do that test.
GOLD15 - 2 sinusoids. It catches both in 200, 400 and 580 window length. If you put more than 580-585 it has problems. It must be some issue with the series being only 599 values. Always take care that you are not analyzing the end points.
GOLD5 - 2 sinusoids plus noise. Well, it is a lot of noise, but the analyzer correctly got the two frequencies as the main peaks of the spectrum, although there are other minor fake peaks (see first picture)
Things get slightly better when you apply noise reduction filters (especially kalman, median filter and FIR). Then other peaks lower a bit (but not that much)
Here is the spectrum with a pre-processing (digital filter with P1=18, D1=9) (see second pic)
Hi
NOXA also made tests on GOLD5 so have a look for comparision
T2W Day Trading & Forex Forums
post 115
100% of winners is OK
I like the approach of noxa guys. From what I understand they have a singular spectrum analisys + neural network to fill the missing data from the last unchanging bar of SSA to the current bar, so that it does not repaint.
I think it is a sound approach, I would like to peek into it one of these days
.
That said, I can see that CSSA-cycles is close to the real signal, but I could get it also with an FFT + frequency domain filtration + IFFT (see pictures), if I knew I had two dominant cycles plus noise (there are other issues with FFT, look at Meyer's EPFFT, but I just want to point out that its always easy do to the right thing... after you know)
I attach 5 noisy chirps signal, underlying chirp is the same there. 3 i made myself they are with maks noise amplitude of 1,4 and 9 and two I extracted from NOXA charts. So you can make direct comparision with your MESA
Anyway, I think your MESA is working now so question is how to proceed.
Because if you want to choose P1 and D1 manually than this will work as long
as market conditions will not change so like CSSA. If automatically than there
is a problem how to sort and filter subcycles. For CSSA you are doing this manually using show eigenvector tool than Genetic Optimizer chooses the most profitable group of subcycles.
So what is your idea now ?? IS this adaptive FATL,SATL,FTLM,STLM,PCCI and RBCI way forward ??
KrzysztofI will take a look at the chirp signal you posted, too.
Anyway, the idea was to pick P1 and D1 automatically (you can already do it with the MESA_cutoff_frequency indicator I posted) and use the ACTF methodology to make effective trend trading.
In that, I believe clahn can help (be patient clahn, I will post the indicators asap
) because he knows ACTF better than me.
The trading philosohy of CSSA (and, indipendently, Neuroshell), though, is very different from that of ACTF. I would like to get deeper into it, later.
Lastly, I have done a few tests with MESA and detrending.
Like for denoising, it depends mostly on what you define as "trend".
I have followed Codebreaker's suggestion to use a non-causal filter. As I don't have SSA at hand, I have used the same trick as in the previous post. I have filtered the signal in the frequency domain with the good old FFT of mine (see second and third pic, the yellow line is the "trendline" with different settings for the filter, all frequencies above a threshold are cut).
I have then subtracted the filtered signal from the signal (thus obtaining a non-causal hi-pass filter) .
It turns out that in the obseverved band the mesa spectrum is nearly invariant respect to the detrending.
CSSA illusion
100% of winners is OK
I like the approach of noxa guys. From what I understand they have a singular spectrum analisys + neural network to fill the missing data from the last unchanging bar of SSA to the current bar, so that it does not repaint.
I think it is a sound approach, I would like to peek into it one of these days
.
That said, I can see that CSSA-cycles is close to the real signal, but I could get it also with an FFT + frequency domain filtration + IFFT (see pictures), if I knew I had two dominant cycles plus noise (there are other issues with FFT, look at Meyer's EPFFT, but I just want to point out that its always easy do to the right thing... after you know)
I will take a look at the chirp signal you posted, too.
Anyway, the idea was to pick P1 and D1 automatically (you can already do it with the MESA_cutoff_frequency indicator I posted) and use the ACTF methodology to make effective trend trading.
In that, I believe clahn can help (be patient clahn, I will post the indicators asap
) because he knows ACTF better than me.
The trading philosohy of CSSA (and, indipendently, Neuroshell), though, is very different from that of ACTF. I would like to get deeper into it, later.Hi,
Regarding CSSA NOXA performance. Results discussed on TRAD2WIN were always out of sample. But even when it hit 100% on GOLD5 out of sample it's simple illusions. See this what's happen with performance after noise level change
T2W Day Trading & Forex Forums
post 175.
It collapsed !!! 443 crazy trades. Simple because 'curve fitter' lost the fit.
Regarding your automating method. That's interesting how you will sort it out because simple choosing tops and bottoms in the spectrum is not enough and to make combined dominant cycle will be very difficult.
The solution can be to have kind of filter banks or kind of running all the time optimizer which will choose those subcycles or maybe retraining NN indicators. I just wonder what will be the results of EA money wise.
Here you have some detrending methods, without test with EA difficult to say which one is the best.
a) Svd (ssa) This maybe could be the best for separabilities but it’s parametric and long process.
Book available (Analysis of Time Series Structure: SSA and Related Techniques)
b) Hodrick Prescott filter Parametric (MATLAB Central - File detail - Hodrick-Prescott Filter)
(MATLAB Central - File detail - Fast Hodrick Prescott Filter)
c) Wavelets
d) Differences of logarithms
e) Differences % (Rate of change)
f) MF-DFA (Detrended fluctuation analysis - Wikipedia, the free encyclopedia)
Krzysztof
Hi,
I was trying to create some cycle indicators using DFM, but sometime the cycles are completely wrong and I'm changing the parameters P1, D1, P2 and D2 only a few.
For example, if I use P1=90, D1=73, P2=100 and D2=138 (peak at 95 for EURUSD 4H) I'm getting very different results than if I'm using P1=87, D1=73, P2=102 and D2=138.
I'm only changing a bit P1 and P2, but the result is not the same at all.
About CSSA, what is it exactly? Is there some indicator or tool for MQ4?
I've downloaded the SSA.mq4 and SSA_normalize.mq4, but it does not work.