Obtaining a stationary BP from a price BP - page 8
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
Transformations of BP into something more decent are plentiful - all (or almost) indicators, but no profit is visible. Whenever an indicator is developed, it is always the idea first and then the implementation. Here they say "it's a good thing if VR is stationary in place of non-stationary". What is good? The development of all indicators is aimed at the fact that they reflect some characteristic of the initial BP. Here we don't set such a task at all, we set the task of statistic characteristic of the result and we don't know what this result will reflect of the initial BP.
What is the advantage of stationary processes over non-stationary ones? A stationary process is predictable. It has some important characteristic (in its most general form it is not necessarily the mean), and it is known how this characteristic can change (in its most general form it is not necessarily the variance). It is also known that this characteristic and its variability are independent of the sample. And in a particularly strict form, it is also known that the distributions of variability are constant. All this, with a degree of rigour, makes it possible to predict. This is not the case with non-stationary processes. In this case, strictly speaking, it is impossible to predict (in practice, of course, it is not so bad). It is the non-stationarity that leads to some kinds of plummet. However, stationary pluming is also possible, and there are a lot of fans of such type of pluming. But this is another topic.
For this reason I write: profit = f(price), which means - getting a stationary profit from non-stationary price.
This is not a guarantee of profit, but it is a very desirable property.
By the way, I've seen graphs on the forum that show that the length of candlesticks depends on the time of day.
What is the advantage of stationary processes over non-stationary ones? A stationary process is predictable. It has some important characteristic (in its most general form it is not necessarily the mean), and it is known how this characteristic can change (in its most general form it is not necessarily the variance). It is also known that this characteristic and its variability are independent of the sample. And in a particularly strict form, it is also known that the distributions of variability are constant. All this, with a degree of rigour, makes it possible to predict. This is not the case with non-stationary processes. In this case, strictly speaking, it is impossible to predict (in practice, of course, it is not so bad). It is the non-stationarity that leads to some kinds of plummet. However, stationary pluming is also possible, and there are a lot of fans of such type of pluming. But this is another topic.
That's why I write: profit = f(price), which means - getting a stationary profit from non-stationary price.
This is not a guarantee of making a profit, but it is a very desirable feature.
This is an inherent property of some markets. Why are you interested in it? In the context of this discussion.Stationarity is desirable with positive mo. Then yes, it is a source of profit. Are there any cases of stationary mo=0 on which one can systematically earn? It is possible to predict quite adequately, but earning? Positive mo can be played most efficiently if you know the other features/parameters of the distribution.
The stationary series is cleared of dependencies. It has no "memory" because its distribution does not depend on the time shift of the reference t-score. Looking for dependencies in it, building TA is useless. Only play out the positive mo in purely statistical ways if it is present.
stationarity is desirable with positive mo. Then yes, it is a source of income. Are there possible cases of stationary mo=0 on which one can systematically earn? Predicting is possible and quite adequate, but earning? Positive mo can be played most efficiently if you know the other features/parameters of the distribution.
The stationary series is cleared of dependencies. It has no "memory" because its distribution does not depend on the time shift of the reference t-score. Looking for dependencies in it, building TA is useless. Only to play out the positive mo in purely statistical ways if it is present.
I hope this is a clarification and not a refutation of what I wrote.
Are we talking about the same thing?
Probably yes, it just wasn't there yet when I read it.
"stationarity is not a guarantee of profitability, but it's a desirable attribute."
and that's a point I wanted to clarify :)
stationarity is desirable with positive mo. Then yes, it is a source of income. Are there possible cases of stationary mo=0 on which one can systematically earn? Predicting is possible and quite adequate, but earning? Positive mo can be played most efficiently if you know the other features/parameters of the distribution.
The stationary series is cleared of dependencies. It has no "memory" because its distribution does not depend on the time shift of the reference t-key. Looking for dependencies in it, building TA is useless. Only to play out the positive mo in purely statistical ways if it is present.
A few posts earlier I wondered about the consistency of the model with the original. Does the mashup match the original unsteady signal? A whole bag of answers both theoretical and in the form of mashup modifications. You can use regular models but you must always answer the question of model correspondence (adequacy) to the original. We are trying to take a trend. What do we take by increments? If we assume that BP is a sum of strong trends, weak trends, cycles (oscillations in the corridor and white noise), then what do we take as increments?
This is an inherent property of some markets. Why are you interested in it? >> In the context of this discussion.
The dependence of candle length on the time of day was interpreted as non-stationarity of increments.
The dependence of candle length on the time of day was interpreted as non-stationarity of increments.
is there a link?
is there a link?
Unfortunately, no. It was on the spider, then moved here. The most unpleasant sign of BP's non-stationarity is the varying periodicity of the trends. The length of the candle also varies with time. This seems obvious to me as volatility is not stationary.
That's it, my model is working! Today I identified errors in the neural network extrapolation algorithm by weighting coefficients.
In short, we divide the story into three parts.
In the first part we feed the NN inputs with residues of price difference and (I won't say what). The neural network is lagging. We obtain weight coefficients. We extrapolate.
In the second part we check for extrapolation. We see that there is a fit. Subtract first NN's extrapolation from the remainders, i.e. we get one more BP of the remainders (errors of the first NN). Feed to inputs of second NN.
On the third part the second NN corrects errors of the first NN. Forward is successful.
The branch can be considered closed. My hypothesis from the first post turned out to be correct. At least in the tester the result is stable.