Indicators: Fourier extrapolation of price - page 5
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
Good day, forum members.
I can't find what is the Quinn-Fernandes frequency calculation algorithm, or maybe I don't understand something. :)
Let me explain. We take the graph and average it so that it does not ripple in the eyes. We take the last 22 points, between them 21 intervals.
We select sinusoids with periods 21, 10.5, 7, 5.25, etc. That is, 21 divided by 1, 2, 3, 4, 5, etc. on this interval to a maximum of 10.
Extrapolate by the sum of these sinusoids and we get what the theorem suggests. What is the trick?
It's probably more obvious that way.
The extrapolation in the form of candles rather than a line would look interesting (not a criticism but a suggestion. I'm weak as it is).
This is how I do it. There is an indicator in the market. Bar Predictor. It separately predicts ohlc and ratios between them, bars are built on the basis of ratios. Sometimes the predicted bar with accuracy to a point coincides with the real one.
How often is "sometimes"?
I wonder where the "pi" went in that indicator?
There is a declaration #define pi 3.14.....
But it is not used in any formula. Error.
yes the idea is good but wrong - any approximation has the character of the curve that was calculated on the basis of the historical series, even the extrapolator confirms it.
to understand this, take a simple oscillator - stochastic, which will show what the frequency of the series is at the moment - it will be total and cannot be the basis for making a decision on a trade operation....
If we make a simple check - take several frequencies as a basis and build a summarised curve, which in some way will resemble a historical series, but the continuation of this series in comparison with the continuation of the summarised curve will differ by a number of parameters with the fact that the curve has a certain base frequency, and the series has a dynamic base frequency.....
In such cases one resorts to normalisation of the series, i.e. an attempt to bring the series to some "constant" amplitude value - for this purpose MA is used - and here arises the next moment of error - the presence of equilibration of calculated data of MA relative to the historical series, even say oscillators with large periods do not allow to get an accurate picture for the subsequent approximation - any oscillator strives for balance - averaging of data. so we need a method that can reduce the error of normalisation to "0"....
I, as a former electronic engineer, at the beginning of my acquaintance with Forex, also thought that now I will run Fourier on Matlab, isolate the main harmonics and beat everyone up )) It didn't work, Fourier doesn't work on Forex, because the signal is non-periodic and non-deterministic. I tried it for a long time, I tried it on bars and tick data, all to no avail.
The author has done a lot of work, so I have a lot of respect for it, but the result was not there (I mean the Fourier approach).
To be sure, just run the indicator in the tester at maximum speed and see how the prediction curve behaves. It twists its tail up and down, which is to be expected.
I ran it with default parameters, is everything correct?
I, as a former electronic engineer, at the beginning of my acquaintance with Forex, also thought that now I will run Fourier on Matlab, isolate the main harmonics and beat everyone up )) It didn't work, Fourier doesn't work on Forex, because the signal is non-periodic and non-deterministic. I tried it for a long time, I tried it on bars and tick data, all to no avail.
The author has done a lot of work, so I have a lot of respect for it, but the result was not there (I mean the Fourier approach).
To be sure, just run the indicator in the tester at maximum speed and see how the prediction curve behaves. It twists its tail up and down, which is to be expected.
I ran it with default parameters, is everything correct?
For such posts about 10 years ago I was personally criticised on all corners here.
DSP is a fundamental theory that has a wide practice and a corresponding number of very specific and profound specialists on the forum. Some time ago Vadim Junko disappeared, who, as I think, with the help of Fourier managed to achieve results on the real. But he had much more than an indicator....
The whole problem is that DSP together with Fourier has nothing to do with the market, none at all.
This is based on the fact that on the market only NONSTATIONARY random processes, moreover non-stationary and uncertain processes, i.e. non-stationary processes, in the control loops of which there is a person who at some moments behaves like a molecule in Brownian motion, and then, suddenly, all in a row and in step.
If we want to make useful models, we should always keep in mind non-stationarity and the tools used should either directly solve non-stationarity problems (ARIMA, ARCH) or indirectly in the form of classification models that look for patterns during training.
I have only one question: will I live to see the time when there will be a critical number of forumites who will start trying on tools from the caret shell to trade ideas day after day?
It is clear that such people will never turn to matlabs, matcads and so on, which are alien tools for the market.