Market prediction based on macroeconomic indicators - page 10
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
It wouldn't hurt to give a clear definition, or at least clarify what is meant by "prediction", "prediction", etc. What is the horizon of "prediction"? Without that, "predictions" are meaningless. Because, depending on the horizon, the same "forecast" may be correct for one horizon and incorrect for another horizon. Moreover, such plots may alternate many times.
I don't take on universal generalisations, but I will give the following definition of prediction and prediction. These definitions are based on algorithms available in R (and not only). I could probably give other definitions, but my definitions are constructive in the sense that they have program code.
So.
1. Prediction. The principle is that the previous value is always used for prediction. I.e. real history bars are used for making 1-step forecasts. And then there are the following nuances. For a 2-step forecast the value for the first step forward is always used. For a 3-step forecast the values from the previous 2 steps are used, etc. This is fundamental because it adds up the errors of all previous steps of the forecast. So you get an expanding funnel-type error graph. Error on n steps ahead is always greater than the error on n-1 steps ahead. The most widely known representative of this direction is the forecast package.
2. Prediction. Here you specify the set of values from which a forecast is made which is called prediction to distinguish it, because you may use one set of bars to make a one-step, two-step or n-step prediction. Whether the prediction values are used or not is not determined by the algorithm and is determined by a developer. The behavior of the prediction error is unknown. One can make up one's mind. If there are predictors in a set of predictors from H1 that inherently have predictive ability at H4, there may be less error at every fourth hour than at the third hour. Predictors are all classification-type models.
I will not make universal generalisations, but I will give the following definition of prediction and prediction.
For regressions - the non-stationarity of financial series is a basic problem. So when choosing a tool one should see how the chosen tool solves the non-stationarity problem. My mentioned ARIMA solves the non-stationarity problem to some extent, but I've never heard of Taylor series solving the non-stationarity problem.
Non-stationarity is seen as a problem when applying stationary models. If a non-stationary model is used, then non-stationarity is not a problem, but a problem to be solved.
ARIMA does not solve the so-called "non-stationarity problem" - it is not designed for it.
Taylor series are in some sense universal -- if coefficients are constant, we have stationary model (and ARIMA is here too), but if coefficients are functions of time and\or state, we get non-stationary model. That is, in a nutshell, for a quick reference.
In my case, the forecast 1 step (quarter) ahead with date E uses all available values of inputs on dates e=0 to e=D-1. Forecast 2 steps ahead with date E uses all available values of inputs on dates e=0 to e=D-2. And so on. In other words, the two-step forecast at date E does not use the forecast at date D-1 because it implies that if the forecast at date D-1 used a set of inputs on dates 0...D-2, then the same inputs can be directly used for the two-step forecast at date D without intermediate forecast at date D-1.
Non-stationarity is seen as a problem when applying stationary models. If a non-stationary model is used, then non-stationarity is not a problem, but a problem to be solved.
ARIMA does not solve the so-called "non-stationarity problem" - it is not designed for it.
.
Either we admit the presence of non-stationarity or we do not.
If we do, then either our model should immediately eat non-stationary data, or most likely we need a number of preliminary actions that will prepare the raw data so that they are suitable for the model.
And here ARIMA is a classic example. It is precisely a model for non-stationary data. In the first step, the original non-stationary series is converted to stationary, and then the resulting series is modelled.
Specifically.
For stationary data, it is a model without the letter I, which means how many times to differentiate (take differences) the original data so that they become stationary and the ARMA model can be applied. Another thing is that the criteria that are used to determine the stationarity in ARIMA models, weak as a result of which the ARMA model applied to the results of differentiation, not applicable to these results and require additional research, usually by modeling variance - ARCH, but there too nuances..... As a result, it turns out that you have a quote at the input, you model something dissected, but it is impossible to understand where to put the result.
Either we acknowledge the presence of non-stationarity or we do not.
If we acknowledge, then either our model must immediately gobble up non-stationary data, or it is likely that a number of preliminary actions are needed to prepare the raw data so that it is suitable for the model.
And here ARIMA is a classic example. It is precisely a model for non-stationary data. In the first step, the original non-stationary series is converted to stationary, and then the resulting series is modelled.
Specifically.
For stationary data, it is a model without the letter I, which means how many times to differentiate (take differences) the original data so that they become stationary and the ARMA model can be applied. Another thing is that the criteria that are used to determine the stationarity in ARIMA models, weak as a result of which the ARMA model applied to the results of differentiation, not applicable to these results and require additional research, usually on the modeling of variance - ARCH, and there are also nuances..... As a result, it turns out that you have a quote at the input, you model something dissected, but it is impossible to understand where to put the result.
You're repeating the old mistake again, about which much has already been said before...
It is impossible to make a transformation of the original non-stationary series into an equivalent stationary series. It is possible to do all sorts of manipulations with the initial series, but it is necessary to understand that the obtained result may not be equivalent to the initial one. This is exactly what happens when one performs "transformation of nonstationary series into stationary ones".
A lot has already been said about this. But I see that you do not notice the principle points. Figuratively speaking, converting a cat into a dog by driving it on a leash will not work.
You are again repeating the old mistake, which has been said a lot before...
It is impossible to transform an initial non-stationary series into an equivalent stationary series. It is possible to do all sorts of manipulations with the initial series, but it is necessary to understand that the obtained result may not be equivalent to the initial one. This is exactly what happens when one performs "transformation of nonstationary series into stationary ones".
A lot has already been said about this. But I see that you do not notice the principle points. Figuratively speaking, converting a cat into a dog by driving it on a leash will not work.
Dear participants of the discussion of this topic! I dare to assure you that my research showed that none of the known methods of regression analysis, including Fourier transforms, neural networks, linear and non-linear regression models and other models, methods and techniques used to describe and/or predict the behavior of a numerical series, including the market price flow, can compete with the universal regression model I proposed and known to all by any evaluation parameter. I am ready to challenge any objection with specific, comparative, examples of any series analysis. I would be happy to find, with your help, the drawbacks of my model, if any. I am sure that as soon as you seriously study and understand the proposed model, you will discover its power and omnivorousness. To explain primitively, the model is an extension of Gaussian MOC to the nonlinear domain, and as a special case, it covers Gaussian MOC as well. Consequently, if in the linear domain Gaussian MNA is the recognised favourite, then in the general case the proposed method may prove to be so. I am ready to parry objections, if any. Respectfully, Yusufhoja.