Econometrics: one step ahead forecast - page 72

 
Vizard:


the fuck it's supposed to be... but in practice it won't be...

It will. Take two samples of random data with the same expectation and variance. Combine results, i.e. increase sample size, calculate variance and expectation for the combined sample and get the same.

Even theoretically, it is not difficult to understand why this happens, for example, if the data volume in the samples is the same, then in both cases, i.e. for expectation and variance, both numerator and denominator will double. The twos in the numerator and denominator are mutually reduced and we get the same result. If the size of both samples is not the same, the numerator and denominator will still increase by the same amount relative to the numerator and denominator of the first sample: how many times the size of the first sample increased after combining.

 
Reshetov:

It will be. Take two samples of random data with the same expectation and variance. Combine the results, i.e. thereby increase the sample size, calculate variance and expectation for the combined sample and get the same.

That is, of course it will...but we are talking about reality...and in reality the model will be constantly fed with new (and possibly unsuitable for this model) data...
 
faa1947:


When forecasting inside the sample I have a fantastic profit factor, especially please pay attention to the profit factor in observations. But outside the sample ..... Why are such rosy results not extended one step further? I can't understand it.


Because there are only 40 observations. Even though you don't like classical statistics)), the root of the evaluation of test results is in them.
 
Reshetov:

At last the adherent of the cult, has revealed the main secret of the religious trick!

Elementary, Watson! Because they are non-stationary. Stationarity is when dispersion and expectation are constants and do not depend on the sample, on which they are measured. I.e. in any other independent sample, we should get approximately the same constants. If we don't, then the stationarity hypothesis is disproved.

The stationarity hypothesis can be tested in another way by increasing the sample dimension. In the case of stationarity both variance and expectation should also remain constants.

Astonishing deafness.

I've been arguing for several years - the kotier is non-stationary and cannot be predicted.

I've been arguing for years - kotir is non-stationary, but it can be predicted if the residual from the model is stationary. The residual is of interest because then you can add up the model (analytical) with a stationary residual. This sum is equal to the quotient, not a pip is lost. I have written a hundred times above. No same thing, adept chukchi who are writers but not readers.

 
Avals:
because there are only 40 observations. Even though you don't like classical statistics)), the root of the estimation of test results is in it.

Yes, 40 is a bit small. Did the test and wrote above. After 70 further increasing the sample does not affect the result. Here's the result on sample length. It is noteworthy. The model coefficients are estimated:

EURUSD = C(1)*HP1(-1) + C(2)*HP1(-2) + C(3)*HP1_D(-1) + C(4)*EQ1_HP2(-1) + C(5)*EQ1_HP2(-2) + C(6)*EQ1_HP2(-3) + C(7)*EQ1_HP2_D(-1) + C(8)*EQ1_HP2_D(-2) + C(9)*EQ1_HP2_D(-3) + C(10)*EQ1_HP2_D(-4)

There are 10 in total. All coefficients are random variables. Question: at what sample length they will become approximately a constant. I will show all the coefficients in one fig:

Here sample = 80 observations. You can see that after half of the sample all adjusts and especially the error of the evaluation of the coefficient. For the first coefficient I will give a larger one:

This is an estimate of the coefficient itself - we see that its value is not a constant.

And now the estimation error of the coefficient:

Hence I conclude that the sample should be somewhere over 60 observations.

We need stable coefficients with a small error - this is a measure of sample length!

 
Vizard:
Of course it will...but we're talking about reality...and in reality, the model will constantly be fed with new (and possibly unusable for this model) data...
This is indeed the case. We've got an almost stable residual. Shift the window by 1 bar and you have to change model parameters (number of lags). This is clearly visible in the table in the two outermost columns, where the number of lags is shown.
 
faa1947:

Astonishing deafness.

I've been saying for years - the cotier is non-stationary and cannot be predicted.

I've been saying for several years - kotir is non-stationary, but it can be predicted if the residual from the model is stationary. The residual is of interest because then you can add up the model (analytical) with a stationary residual. This sum is equal to the quotient, not a pip is lost. I have written a hundred times above. No same thing, adept chukchi who are writers but not readers.


And by the way, about the analysis of the residuals for a normal distribution: only 116 observations is very small for the results to be reliable. I.e. of course the test can be applied and it will assign the distribution to normal with some probability, but what is the confidence interval of this prediction? I.e. 25% is again very approximate value and may be the range 0...50 with 95% confidence for example, and may be 22...28. It depends on both number of observations and variance. It seems to me that with 116 observations the CI would be huge
 
gpwr:

Usually creators of such models quickly run them in the tester, make sure they fail, and move on to new models. But here the starter shows daily predictions in real time expecting a miracle - masochism of some kind.

Forumers who sit with an open beak where they should put the grail can disperse.

The problem I have outlined and for which I have no solution, is to predict the predictability of the model by the statistical characteristics of the model on history. I'm not interested in TA methods.

 
Vizard:

in any algorithm you can use any error...and r-Q in NS as well...
You can, but they don't. Give an example of an indicator whose text is accompanied by R-squared. Indicators are used and it is not known to what extent they reflect cotier and whether they reflect at all. Judged by eye, "surely a great indicator"
 
faa1947:

.....Don't know for several years - kotir is non-stationary and cannot be predicted.

I've been saying for years - kotir is non-stationary but it can be predicted.....

Make up your mind somehow.....