How do you measure noise? - page 4
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
And this is what the difference between the smoothing and the closing prices looks like - the proverbial noise.
You can even see by eye that its character is changing all the time.
It is clear that instead of a muving, you can use any smoothing, based on Fourier, wavelets, or some kind of digital filters. The problem is the changing variance - or in simple econometric terms, heteroscedasticity. I once read an interesting article by chemists where they used a Savitsky-Haley filter to filter out noise in spectrograms. This filter builds a polynomial of a given period and returns the midpoint of the polynomial to the smoothing one, then shifts one point to the right and repeats the process until the data runs out. Generally speaking, it is a kind of analogue of a muving but using polynomials. So, they came up with the following trick - as long as the difference between the signal and filtered value is small - for plotting a large number of data and linear polynomial, if deviation starts to grow - the number of data to filter decreases and the degree of polynomial increases. Well it is like if if the deviations from the muving increase - decrease its period.
In general, you should first define the signal model and then use the signal model to sift out noise, i.e. movements that do not fit your signal model.
Then the topic will become more interesting and constructive, otherwise there is no point in looking for noise where it does not exist, there is simply no noise in the market, because you think that what I use for work is noise.
I used to work on the problem too, how to get rid of noise, to separate trend from flat and all that. Then getting rid of the time sampling of price I came to the conclusion that there are no trends and no flots.
On top of that, the noise must have a normal distribution and be really random, I found no evidence that any movements in the market are random.
do the same analysis with smoothing 1, because you need to measure the noise in one candle and not in 51
Well it will just get the difference between H4 opening prices in pips. Here you go.
In general, you should first define the signal model and then use the signal model to sift out noise, i.e. movements that do not fit your signal model.
Then the topic will become more interesting and constructive, otherwise there is no point in looking for noise where it does not exist, there is simply no noise in the market, because you think that what I use for work is noise.
I used to work on the problem too, how to get rid of noise, to separate trend from flat and all that. Then getting rid of the time sampling of price I came to the conclusion that there are no trends and no flots.
On top of that, the noise should have a normal distribution and be really random, I found no evidence that any movements in the market are random.
I agree, trying to look at noise I should clearly understand what I need it for - it will hardly be good for trading directly, in order to trade noise I need to know value of smoothing on the far right bar, then seeing that value I open towards the smoothing and make a lot of money. But the problem is that the smoothing e.g. a muving is always moved half a period backward in the picture above by 25 bars and if you know 25 muving values up to the far right it will automatically mean you know the future price 25 bars - why trade noise then you can just trade the future price. Even an accurate prediction of just one future muvinng value gives the exact value of the future bar.
Well that would just be the difference between the H4 opening prices in pips. Here you go.
In general I agree, trying to watch noise one should clearly understand what I need it for - it is unlikely to trade directly, in order to trade noise one needs to know the value of the smoothing at the extreme right bar, then seeing this value one opens towards the smoothing and rakes the money with a shovel. But the problem is that the smoothing for example muwing is always shifted half a period backward in the picture above by 25 bars and if you know 25 muwing values up to the rightmost one it will automatically mean you know the future 25 bars of price. Even an accurate prediction of just one future moving average gives the exact value of the future bar.
I've been doing this by eye, the signal is in the range of ~20 pips from the average price and the rest is noise.
The signal is in the range of ~20 points from the average price, the rest is noise.
Matlab says that the variance is 32.02 points, but practical statistics in cases with large outliers recommends counting the sum of moduli divided by the number of samples - if so, it really is 20.48 points.
Are you sure that the axis in the figure is the average price?