Yes your are true.
I know i made my formula just by feelings, from the AR(1) formula. I
just hoped it was near the true CI : )
Your formula are right and to expand it i will give the recursion
form i found :
for n > 1 (n is the number of step forward, fe the forecast error),
fe(n) = e(n) + Sum_i_from_1_to_n-1[a(i)*fe(n-i)],
with fe(1) = e(1)
>From this recursion we can assume like the author that error are
independant each other (because white noise from the same variance),
so COV are zero like you mention it.
And var{e(n)} = var; (it is the noise variance estimator from the AR
model).
We can find (varfe is variance from forecast error) :
varfe{n} = var^2 + 0 (because COV are zero) + var{Sum_i_from_1_to_n-1
[a(i)*fe(n-i)]}
var{n} = var^2 + .?.
Difficult here... i try for n=1, 2 or 3... we arrive maybe something
like this :
varfe{n} = var*( 1 + a(1)^2 + [a(1)^2+a(2)^2]^2 + ... + [a(1)^2+a(2)
^2+...+a(p)^2]^2*(n-1) )
This above formula is good for AR(1)... so maybe for AR(p) ? But i
think not. Final varfe should maybe appear some combination coef,
because there is polynomial dev. here because of the ^2.
In fact, that is why i simply use sum(a(i)) because there is such sum
but not exactly the same... and should be not far from real
formulation.
Any idea to compute final varfe from above formula ? We are near to
find good equation maybe.
Cheers,
Mich
It works for the
--- In [email protected], "Paul Ho" <[EMAIL PROTECTED]> wrote:
>
> Hi Mich
> thanks for the link
> I dont know if i can agree that for AR(p) is the sum all the coeff.
> the author of the paper made an assumption that error is
independent of
> variance, so sum of co-variances among terms are zero, and hence
the formula
> 5.71
>
> Why is AR(p) the sum of all coeff?
>
> From my own derivation
> fe(n+1) = e(n+1)
> fe(n+2) = a(1)e(n+1) + e(n+2);
> fe(n+3) = a(1)^2.e(n+1) + a(1).e(n+2) + a(2)e(n+1) + e(n+3)
> fe(n+4) = a(1)^3.e(n+1) + a(1)^2.e(n+2) + a(1)a(2)e(n+1) + a(1)e
(n+3)+
> a(2)a(1)e(n+1) + a(2)e(n+2)
> a(3)e(n+1) + e(n+4)
>
> A bit too complicated after 4 steps forecast error.
> But the variances for the 1st 4 steps dont seem to the same as
what you
> have suggested.
> Any thoughts
> /Paul
> ________________________________
>
> From: [email protected] [mailto:amibroker-
[EMAIL PROTECTED] On
> Behalf Of Tom Tom
> Sent: Thursday, 7 December 2006 12:29 PM
> To: [email protected]
> Subject: RE: [amibroker-ts] confidence level asymptoe of Predicted
AR
>
>
>
> Hi !
>
> Just finish to code confidence intervals for AR prediction.
>
> I posted code as a remarks in the AFL library :
> http://www.amibroker.com/library/detail.php?id=764
> <http://www.amibroker.com/library/detail.php?id=764>
>
> I find this paper and my work is based on it :
> http://www.tnet.teagasc.ie/fapri/downloads/rochemanual/mrochech5.pdf
>
<http://www.tnet.teagasc.ie/fapri/downloads/rochemanual/mrochech5.pdf>
> Page 176 and 177
> Maths explaination is given for AR(1).
> I think for AR(p), there is to sum all the coeffs, so that is what
i did in
> my code.
>
> sum(coeff) < 1 so the confidence intervals converge to the
unconditionned
> variance. should be ok.
> if you take just the first coef (like AR(1)) convergence is not
assume
> because first coeff can be 2... but the sum will always be <1.
>
> I take the student t-distribution for infinite number of freedom
degree :
> 1.96.
> Maybe 2 will be better... users can modify this in the code or code
a
> student t-distribution table so it adapts to the degree of freedom
> (deg_freedom = nb_sample - nb_coeff).
>
> i join a picture to illustrate CI
>
> Cheers,
> Mich
>
> __________________________________________________________
> Windows Live Spaces : crىez votre Space ـ votre image !
> http://www.windowslivespaces.fr/ <http://www.windowslivespaces.fr/>
>
> [Non-text portions of this message have been removed]
>
>
>
>
>
> Send instant messages to your online friends
http://au.messenger.yahoo.com
>
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/amibroker-ts/
<*> Your email settings:
Individual Email | Traditional
<*> To change settings online go to:
http://groups.yahoo.com/group/amibroker-ts/join
(Yahoo! ID required)
<*> To change settings via email:
mailto:[EMAIL PROTECTED]
mailto:[EMAIL PROTECTED]
<*> To unsubscribe from this group, send an email to:
[EMAIL PROTECTED]
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/