Stochastic resonance - page 32

New comment
 
No one has watched "Time Series Analysis and Forecasting" http://www.gistatgroup.com/gus/
 
Vinin:
Has anyone watched "Time Series Analysis and Forecasting" http://www.gistatgroup.com/gus/

I have made two solid runs at different times on SSA with Caterpillar. The amount of work done allows us to talk about the representativeness of the result obtained, and the result was negative both times. More precisely, prediction of BPs of currency type is possible, but the confidence prediction range (CFR) has an exponential divergence depending on the prediction horizon. Realistically, it appears in that DPD borders open almost symmetrically relative to the horizontal line originating on the last bar, and the absolute value of asymmetry does not exceed one transaction commission per DC.

Generally, I have the intuitive impression that for each instrument of the Forex market a priori there is an absolute arbitrage strategy, allowing to get the highest possible average yield (pips per transaction) and this yield (on average) does not exceed the brokerage company commission! This does not mean that the brokerage companies are so smart that they know this strategy and therefore can determine the level of commission from above, but it means that the human gnawing at the market with everything possible, adiabatically close their tail FR to this theoretically possible limit asymptotically defining it for brokerage companies ...

The problem can be solved in two ways - strictly and mathematically correct, and approximated by statistical (and/or iterative) methods. In the case of the market, there is a collective, approximate solution. Which, it seems to me, tends towards the exact one with great accuracy (because of the huge number of players). In other words, no matter what we invent, the statistical profitability of our strategy will not (cannot) exceed the average brokerage company commission for the given instrument!

All said above does not pretend to be true and is my personal opinion, which in turn correlates with my current mood;-)

 
Prival:

Found FFT_MA prototype here on the forum, and redid it according to figures posted earlier (FFT_MA_mod). The only thing is, it redraws, which makes analysis difficult. If anyone can fix this defect, please help.



This defect is easy to fix (look carefully at the source code I posted), but after that the muwings become lagging :)
 

to Prival

Sorry, I got confused in my haste, I should have askedYurixx. for the histogram. When the pictures appeared I realised my mistake. I continue to work on the idea of Resonance, based on my definition of "signal energy - moves the market. Noise energy - prevents us from seeing that movement". (Thanks for the tip about IIH or IIH, but about 12 years ago I read lectures on them to the cadets, and I even remember giving them bad marks :)).

Then all the more it is not clear why you offer one of the worst ways of filtering. Actually, it works fine for periodic signals, but it cannot work well for quotes. I know about this method and it is well described in MathCAD documentation, in "Signal Processing/Filtering vs Exponential Smoothing" section, but I strongly recommend not using it for this task.

I've been doing it for a long time, so I dug it out, the only input parameter controls the percentage of power that passes through (you can get twisted here, but by definition this filtering won't give any acceptable solutions anyway):

You can see that not only marginal effects are present, but also that the local extrema are noticeably shifted from the "true" ones (dumb or clever parameter enumeration won't help either). It would be better to deal with adaptive filters.

to eugenk

Sergey, about potential pits I deeply apologize and admit my own stupidity :) It's true, the level of support and resistance can only be compared with a potential barrier from which the price bounces. But concerning the fiction of the phenomenon I'm afraid I have to argue. Moreover, IMHO it is the only reality on the market, contrary to those fantasies about waves, fibos, pitchforks and alligators. At least, it is the only thing that can be easily explained without involving additional non-obvious postulates. Congratulations on discovering an interesting criterion X ! That's all. I'm off to read more, I haven't been here for two years and it's been 11 pages since then :)

I must admit that deep down I believe it myself... but I haven't found any proof yet...

 
grasn:

Been doing this for a while, so I dug it up, the only input parameter controls the percentage of power transmitted


So, grasn didn't look at my source either :). It's true that it's a cosine transform, but it's not crucial, it's easy to change.
 
lna01:
Prival:

I found here on forum FFT_MA prototype and reshot it according to pictures posted earlier (FFT_MA_mod). The only thing, it overdraws, that makes analysis difficult. If anyone can fix this drawback, please help.



This flaw is easily remedied (look closely at the source code I posted), but after that the muving becomes delayed :)

The only thing I found was 'Spectral Analysis', but there's some kind of error. Nothing appears on the screen.

If anyone has a chance, please help me make an indicator. It should be based on FFT_MA_mod_2.

It should reflect how the signal and noise energy has changed over time. I have to make changes in attached file - with appearance of new bar remember two variables energi_sign,energi_shum. And don't touch them again (don't redraw).

I'm not building an indicator that should smooth and predict the price. For this it is better to use Kalman filter. If interested, I am ready to discuss its use.

I am also looking for resonance here. I think that the resonance appearance should be indicated by the change of energy. I would like to see this curve. Then there will be material for further analysis.

I am grateful in advance.

Files:
fft_ma_mod_2.mq4  3 kb
 
grasn:

to Prival

I've been doing this for a long time, so I dug it up, the only input parameter controls the percentage of power that passes through (you can get twisted here, but by definition this filtering won't produce any acceptable solutions anyway)

I agree that percentage won't do it, and in general price smoothing (not sum of monochromatic signals) and its prediction with Fourier won't do anything. I want to see how energy behaves, because it doesn't go anywhere, it just goes from signal to noise with this kind of processing. Then I plot ACF of this curve and think about it further. Maybe I have made my point wrong :(. If someone wants to participate and help in research I am in Skype search for privalov-sv
 
Prival:
lna01:
Prival:

Found FFT_MA prototype here on the forum, and redid it according to pictures posted earlier (FFT_MA_mod). The only thing, it overdraws, that makes analysis difficult. If anyone can fix this defect, please help.



This flaw is easily remedied (look closely at the source code I posted) but after that the mouving becomes laggy :)

The only thing I found was 'Spectral Analysis', but there's some kind of error. Nothing appears on the screen.

I meant the one I posted in this thread https://c.mql5.com/mql4/forum/2007/10/oFFTMA_E.mq4. And the one https://forum.mql4.com/ru/6275 shows, but you have to look for the spectrum at the date, which is in the parameters. Actually it was just a blueprint. Simply by making a code to calculate the spectral density I decided to see how the result looks like visually. I used the code later, but not this indicator :)
 
Prival:

If anyone has an opportunity, please help me make an indicator. It should be based on FFT_MA_mod_2.

It should reflect how signal and noise energy has changed in time. I have to make changes in attached file - with appearance of new bar remember two variables energi_sign,energi_shum. And do not touch them again (do not redraw them).

To leave the variables untouched, we should put an indicator buffer for each of them and write them into the element with the same number (usually 0 or 1) on each bar.

But I don't understand the meaning of this fragment, can you explain in more details, what do you mean?

ArraySort(data1,WHOLE_ARRAY,0,MODE_DESCEND); // сортируем его
// теперь пороговая обработка
// удаляем все что ниже по амплитуде гармоники с номером hmax
for(i=hmax;i<N;i++)   if (data[i]<data1[hmax]) data[i]=0.0; 
for(i=hmax;i<N;i++)  energi_sign=energi_sign+data[i];   // сумма всех составляющих спектра (энергия сигнала)
// шум
// удаляем все что выше порога 
for(i=hmax;i<N;i++)   if (data[i]>data1[hmax]) data[i]=0.0;
for(i=hmax;i<N;i++)  energi_shum=energi_shum+data[i];   // сумма всех составляющих спектра (энергия шума)
 
lna01:
Prival:

If anyone has an opportunity, please help me make an indicator. It should be based on FFT_MA_mod_2.

It should reflect how signal and noise energy has changed in time. I have to make changes in attached file - with appearance of new bar remember two variables energi_sign,energi_shum. And do not touch them again (do not redraw them).

To leave the variables untouched, we should put an indicator buffer for each of them and write them into the element with the same number (usually 0 or 1) on each bar.

But I do not understand the meaning of this fragment, can you explain in more details what was meant?

ArraySort(data1,WHOLE_ARRAY,0,MODE_DESCEND); // сортируем его
// теперь пороговая обработка
// удаляем все что ниже по амплитуде гармоники с номером hmax
for(i=hmax;i<N;i++)   if (data[i]<data1[hmax]) data[i]=0.0; 
for(i=hmax;i<N;i++)  energi_sign=energi_sign+data[i];   // сумма всех составляющих спектра (энергия сигнала)
// шум
// удаляем все что выше порога 
for(i=hmax;i<N;i++)   if (data[i]>data1[hmax]) data[i]=0.0;
for(i=hmax;i<N;i++)  energi_shum=energi_shum+data[i];   // сумма всех составляющих спектра (энергия шума)

At this point the task of extracting N highest amplitude components from the spectrum is solved. The a priori frequency of the components is not known. Here is the figure.

If we go by the classic example, we should determine parameters of the Rayleigh-Rice distribution law and set the threshold with a given probability of second-order error. (In radar this is called signal detection with a given probability of false alarm).

But it can be simpler: we sort the spectrum in descending order and select the component with the number given in the hmax indicator.

The amplitude of this component determines the value of the threshold (see fig.1).

All that remains is to compare the original spectrum with this amplitude and select in 1 case

Signal, everything below is 0

for(i=hmax;i<N;i++) if (data[i]<data1[hmax]) data[i]=0.0;

or noise (everything above is 0)

for(i=hmax;i<N;i++) if (data[i]>data1[hmax]) data[i]=0.0;

In the first case we get signal energy that according to the hypothesis moves the market and in the second case we get noise. These are the charts we need. I seem to succeed, but the chart is only plotted in visual testing mode :(. I have to wait very long time

1...2526272829303132333435363738
New comment