Econometrics: one step ahead forecast - page 54
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
You have already given this many times, but it is only part of the prediction. In a previous post I already wrote about the rest.
It's a dark matter with volatility.
The aim of the modelling is a stable residual, i.e. mo and dispersion are practically constants. This has already been pointed out a few times above. This is the result of applying GARCH to the residual.
If you take the volatility of the original quotient, I take it into account as two bars.
Or anything like stochastic?
I don't see how you can't do it - and you can even google it: "martingale site:mql4.com". Have you even seen the Forum Navigator thread and Answers to Frequently Asked Questions. Highly recommended reading!?
Simply type one of the following words into the 'Search' (the field with the magnifying glass) at the top right: "martingale", "martin", "martini", "avalanche". Already this will be enough, there will be dozens of links on the search page.
It is a dark matter with volatility.
The aim of the modelling is a stable residual, i.e. mo and dispersion are practically constants. This has already been pointed out a few times above. This is the result of applying GARCH to the residue.
If I take the volatility of the initial quotient, I take it into account as two bars.
Or something like stochastic?
You can formally measure the return. Perhaps it will not be possible to apply Hearst's index or h-volatility in its pure form:
Plot the variation of the error against the forecast horizon. Now you are forecasting 1 day bar. How will the error change if you forecast 2 or more bars. If it grows less than the root of the forecast time, then there is a return. After all, the error is sko? I.e. if the error of forecast for 1 bar is 80 pips and for 2 bars it would be less than 80*SQRT(2)=113. Plot the change in the real and theoretical error for the case when there is no reversion.
it is possible to formally measure the return. It is probably not possible to apply Hearst index or h-volatility in its pure form, but you can do this:
plot the error from the forecast horizon. Right now you are forecasting 1 day bar. How will the error change if you forecast 2 or more bars. If it grows less than the root of the forecast time, then the return is present. After all, the error is sko? I.e. if the error of forecast for 1 bar is 80 pips and for 2 bars it would be less than 80*SQRT(2)=113. Plot the variation of real and theoretical error for the case where there is no reversion.
Apply the Hurst index or h-volatility
Hurst is more than a dark matter.
Plot the error from the forecast horizon. Right now you are forecasting 1 day bar. How will the error change if you forecast 2 or more bars
There are two forecast modes in EViews: static (one step ahead) and dynamic - for many steps ahead, when the previous value is taken as the previous forecast value, where the previous value is the last measured value. An error is two diverging lines around the forecast. How it relates to your value - I don't know.
I don't understand the very idea of a multi-step forecast. One step is quite enough. Not enough - enlarge the time frame.
Apply the Hearst figure or h-volatility
Hearst is more than a dark matter.
plot the error from the forecast horizon. Right now you are forecasting 1 bar of days. How will the error change if you forecast 2 or more bars
There are two forecast modes in EViews: static (one step ahead) and dynamic - for many steps ahead, when the previous value is taken as the previous forecast value, where the previous value is the last measured value. An error is two diverging lines in the forecast. How it relates to your value - I don't know.
I don't understand the very idea of a multi-step forecast. One step is quite enough. Not enough - enlarge the time frame.
It is not how many bars the forecast is, but how the error varies with the forecast horizon. This allows you to see whether or not there is a return to the forecast value
It is not how many bars the forecast is, but how the magnitude of the error varies with the forecast horizon. This allows you to understand whether there is a return to the predicted value or not
A +1 forecast uses the measured "true" value of the quotient and the error of that forecast is determined by the stationarity of the residual between the quotient and the model. If stationary the residual is a constant and no square roots. If not stationary, no square roots either, as it is not predictable and any measurement on the test sample is not meaningful.
A +1 prediction uses the measured "true" value of the quotient and the error of this prediction is determined by the stationarity of the residual between the quotient and the model. If the residual is stationary, it is a constant and no square roots. If not stationary, no square roots either, as it is not predictable and any measurement on the test sample is not meaningful.
Clearly that's not what I'm saying. Are you counting the prediction error as rms?
clearly that's not what I meant. Do you calculate the prediction error as a RMS?
made a suggestion for your model. Probably looked it up. Look above in the thread.
If he has an error as a constant, it will never cease to be a constant, no matter how you count it.
It cannot be a constant in each separate trade. And the slope of the price from the forecast (error) may converge to a constant, if the error distribution is stationary.