Principles of working with an optimiser and basic ways of avoiding fitting in. - page 2
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
I disagree. Strongly disagree.
Which rows are you talking about?
We are talking about financial rows, specifically financial rows. Financial rows, due to their peculiarities, financial ones, have regularities on them.
Maybe there are no regularities on some other non-stationary series, but there are regularities on the financial ones.
This does not allow for the application of statistics, but does not contradict logic in any way. It is not necessary to look for patterns by statistical methods alone.
Contradictions to logic occur only when some botanists try to measure non-stationary data by statistical methods.
We diagnose: a person is sick, so we cannot cure them with medicine, we go to shamans, we poke fun at them.
But it is possible to clarify what the problem is, apart from the diagnosis - non-stationarity - and it breaks down as in all modelling into at least two parts: identification of BP and identification of the model. Once we have structured these two things we can raise the question of model fit to BP. To date, we cannot describe BP in all its diversity, but that does not mean that nothing can be done at all. By taking into account the error between the BP and the model, we can begin to talk about estimating the model itself. In doing so, we can put the problem of evaluation much wider than evaluation through a tester.
Maybe on some other non-stationary series there is no pattern, but specifically on the financial ones there is.
This is a verbal equivocation and nothing more. You have to define whether your series are non-stationary or with regularities. Because somehow the psyche cannot comprehend the phrase: "Regularities of nonstationary series" I understand that you have already found the regularities of nonstationary series?
Good. Then we need to define what nonstationary is. Do you have a definition of what non-stationarity/stationarity is?
A verbal contrivance and nothing more. You should define whether your series are non-stationary or with regularities. Because somehow the psyche cannot comprehend the phrase: "regularities of nonstationary series" I understand that you have already found regularities of nonstationary series?
This is a defeatist position.
Why not assume that a non-stationary series = the sum of several components. And the most interesting one is the deterministic component. If it does not exist or we admit it, then it is a random walk and forecasting is not possible by any means and methods (efficient market theory). If we acknowledge it, then our presence on the market and on this forum is justified.
When they talk about non-stationarity they usually mean the distribution of price increments. That the mo (trend) and dispersion (volatility) change over time. It's true, but non-stationary series may have stationary areas. If you can find them, you can trade on them with the corresponding MO and dispersion in deals. I.e., trading on fixed segments leads to the fact that equity increases stationary (or close to it quasi-stationary). That is, the mo and variance of the trade change slowly.
I.e. the trader's task is to find stationary plots on a non-stationary series of price increments.
When they talk about non-stationarity they usually mean the distribution of price increments. That the mo (trend) and dispersion (volatility) change over time. It's true, but non-stationary series may have stationary areas. If you can find them, you can trade on them with the corresponding MO and dispersion in deals. I.e., trading on fixed segments leads to the fact that equity increases stationary (or nearly "quasi-stationary"). That is, the mo and variance of the trade change slowly.
I.e. the trader's task is to find stationary areas on the non-stationary series of price increments.
Sort of a piecewise stationary series. It is a very strong assumption and practically impossible to identify in trading - forecast of the future, because identification requires some number of observations and there is no guarantee, that the next observation will not lead to non-stationary series.
It is much easier and more practical to consider that the series consists of a deterministic residual + noise.
A kind of piecewise stationary series. It is a very strong assumption and not practically feasible for identification in trading - forecasting the future, because identification requires some number of observations and there is no guarantee that the next observation will not start a non-stationary series.
It is much easier and more practical to consider that the series consists of deterministic leaving + noise.
You are confusing prediction model and what we should eventually get (purpose of prediction).
You should always enter a trade when the forecast of this deterministic component gives positive mo and fixed variance. I.e. the deterministic component forecast assumes stationarity of price increase in this area. Well, the problems are similar - if the model made a prediction before by singling out such segments, it may not do so from the next transaction. There will be a forecast but no positive mo.
You are confusing the prediction model and what we should end up with (the goal of the prediction).
I don't think I'm confusing anything, I'm always saying the same thing.
I believe there is a deterministic component, which I isolate by some method of smoothing. Then I look at the residual = cotier - deterministic component. Obviously, the residual is non-stationary (non-stationarity has nowhere else to go) and the whole problem is now buried in it.
When predicting, we compare the incremental mo and account for the prediction error by variance. We can only predict if these quantities are almost constants, and if not? That's the whole problem. It is because of the residual that testing cannot be trusted until our model at least partially accounts for non-stationarity. We should purposefully deal with non-stationarity, not turn a blind eye to it.
I don't think I'm confusing anything, and I always say the same thing.
I believe there is a deterministic component, which I isolate by some method of smoothing. Then I look at the residual = cotier - deterministic component. Obviously the residual is non-stationary and the whole problem is now buried in it.
When predicting, we compare the incremental mo and take into account the prediction error on the variance. We can only predict if these quantities are almost constants, but if not? That's the whole problem. It is because of the residual that testing cannot be trusted until our model at least partially accounts for non-stationarity. We should purposefully deal with non-stationarity, not turn a blind eye to it.