Econometrics: one step ahead forecast - page 126
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
The results that are posted in this thread above are obtained that way. The profit factor is just above 1. I came to the conclusion that the model has no predictability, and I'm stuck with that. For smoothing, HP was applied with ladd = 1. Could be here. But it's not clear what "predictability" is. If you look at what you get in the tester, the model doesn't hold the trend and it's not about false reversals.
(1) The question is not about the results of the forecast. I am not interested in that at all. How the model coefficients behave over time. Can you at least show graphs of their dynamics.
(2) Is HP a filter (Hodrick-Prescott)?, then it's even worse.
(2) HP is a filter (Hodrick-Prescott), then it's even worse.
Well yes, at first it didn't seem to matter. Need to solve the residual problem. Solved it. Now I have doubts. Do you have a valid complaint against HP?
1) The issue is not the results of the forecast. I'm not interested in it at all. How the model coefficients behave over time. You can at least show graphs of their dynamics.
Finally a real question. I did. Very interesting. I'll dig around now and try to post them again.
to faa
У Вас имеются обоснованные претензии к НР?
no way! I have no complaints about Prescott. You know how much I respect Prescott, Prescott is a head, you can't put your finger in his mouth...
Finally a real question.
Fucking hell, like before I was distracting you and asking you all this bullshit.
I'll do some digging and try to post it again.
Don't bother, you're wasting valuable kilocalories...
Model:
kotir hp1(-1 to -2) hp1_d(-1 to -1) eq1_hp2(-1 to -3) eq1_hp2_d(-1 to -4)
In brackets is lag. At each new bar I adjust the number of lags
HP_d - difference between kotir and HP.
eq1_HP2 - smoothing HP difference between kotir and HP1(-1 to -2) hp1_d(-1 to -1)
eq1_hp2_d( -1 to -4) 'this is the last residual
If it has heteroscedasticity, then I model GARCH
Without GARCH estimation we get the equation
KOTIR = 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)
Lots of coefficients.
Chalky, but a lot. Almost stable.
But there is a large error in estimating coefficients for some of them. We need to divide 100% by the value in the t-statistic
to faa
no way! I have no complaints about Prescott. You know how much I respect Prescott, Prescott is a head, you can't put your finger in his mouth...
like before I was distracting you and asking you all kinds of bullshit.
Don't bother, you're wasting valuable kilocalories...
So touchy!
Of course, the cof is extremely valuable information. And your opinion is very interesting. You are the first to ask and in that sense "finally"
So touchy!
Of course, coefficients are extremely valuable information. And your opinion is very interesting. You're the first to ask and in that sense "finally"
You can't see anything :o( At least give me an excel with data, I'll make the graphs myself, maybe I'll analyze something
Almost stable.
They're all crooked. What do you mean they are "almost stable"?
You can't see anything :o( At least give me an Excel file with data, and I'll draw the graphs myself, maybe I'll analyze them
They are all crooked, aren't they? What do you mean they are "almost stable"?
I am attaching. Please note that kotir is EURUSD.
For each coefficient, the value of the coefficient and the error of the coefficient
Attached. Please note that kotir is EURUSD.
For each coefficient, the value of the coefficient and the error of the coefficient
I took a closer look at the software evaluations, but I don't see any reason to be happy. If I understand the result correctly, EW shows you that the model is, generally speaking, fake:
(1) the coefficient is -0.48, with error standard deviation of 0.12, for example -4.89 with error standard deviation of 0.9 -2.9 with error standard deviation of 1.0 etc. these are very big errors, very big, i.e. they are almost on the verge of invalidating the estimate.
(2) t statistics for the first coefficient is very large, (if I remember correctly, it's been a long time since I worked with it, I need to refresh my knowledge), in other words, the very first coefficient doesn't describe the model in any way, in a sense - it's just lefty. By the way, what "trend" did you take for the HP model?
(3) yes there is no need to estimate the probability that the parameter is not zero. yes it is clear that it is not zero
(4) R-squared, not a correct estimate, I explained why, it should not be looked at at all in this case. Literally speaking, the scale of price bias is not normalised, it's as if you've moved miles away from the quote and say woohooo over there will be the price. Yeah, within the bounds of the deviation statistics yes, but you won't make any profit on that, you'll only lose
OK, if I don't understand something, I'll figure it out later.
I took a closer look at the software evaluations, but I don't see any reason to rejoice. If I understand the result correctly, EW shows you that the model is, generally speaking, fake:
(1) the coefficient is -0.48, with error standard deviation of 0.12, for example -4.89 with error standard deviation of 0.9 -2.9 with error standard deviation of 1.0 etc. these are very big errors, very big, i.e. they are almost on the verge of invalidating the estimate.
(2) t statistics for the first coefficient is very large, (if I remember correctly, it's been a long time since I worked with it, I need to refresh my knowledge), in other words, the very first coefficient doesn't describe the model in any way, in a sense - it's just lefty. By the way, what "trend" did you take for the HP model?
(3) yes there is no need to estimate the probability that the parameter is not zero. yes it is clear that it is not zero
(4) R-squared, not a correct estimate, I explained why, it should not be looked at at all in this case. Literally speaking, the scale of price bias is not normalised, it's as if you've moved miles away from the quote and say, woohooh over there will be the price. Yeah, within the bounds of the deviation statistics yes, but you won't make any profit on that, you'll only lose
OK, if I don't understand something, I'll figure it out later.
(1) .... these are very big mistakes, very big.
Yes. By individual coefficients.
(2)t statistics for the first coefficient are very large,
Wrong. T-statistic = coefficient/SCO
the very first coefficient does not describe the model
It is the first coefficient that does. We need 100 / t-statistics and get the error in %. But this does not solve the problem with other coefficients.
And what "trend" did you take for the HP model?
There is no trend. HP is smoothing to get noise in the residual.
(4) R-squared, not the correct estimate,
It's supposed to be correct. DW is about two, which means that the residual is normally distributed. There is still regression error = 11 pips, but the error of the dependent variable = 212 pips
But here is the prediction result
Please note that average error % = 5.7%!!!!