EViews.com

Hi,

I am having trouble estimating the following nonlinear model:

∆y_t=φ'w_t+φ'w ̃_t*z_(t-d)+φ'w ̃_t* z_(t-d)^2+φ'w ̃_t*z_(t-d)^3+η_t

where _t is the time subscript and d is the delay parameter, η_t is the error term which is niid (0, σ^2) . w is the original regressors in the linear EC model eq w = { c, x1, x2, x3 } and w ̃ = { x1, x2, x3 }

my problem is how to construct w. I assume the way i have tried is completely erroneous : construct a matrix with each column being equal to that of w or w ̃ as defined above since it when i click "ok" i get the error message "W is not a series"

maybe i am trying to over complicate things and there is an easier way?

help would be greatly appreciated

M

To be honest, I'm not sure what you are trying to do.

Apologies for not making myself clear.

having estimated a linear ECM of the form:

∆pt=β0+β1∆p(-1)+β2x(-1)+β3∆m(-1)+β4EC(-4)+u

where EC is the error correction term from a long run relationship. i wish to test against a smooth transition EC model of the form:

∆pt= π1'wt+π2'wt F(z(t-d))+ηt

in order to do so F(z(t-d)) is replaced by a thrid order Taylor approximation so defining the nonlinear model as:

∆pt=φ'wt+φ1'w ̃t*z(t-d)+φ2'w ̃t*(z(t-d))^2+φ3'w ̃t*(z(t-d))^3+ηt

where wt= {β0, β1∆p(-1), β2x(-1), β3∆m(-1), β4EC(-4)}, w ̃t = {β1∆p(-1), β2x(-1), β3∆m(-1), β4EC(-4)} and z(t-d) = EC at time t minus d (the delay parameter which i will be using d = 1, 2, 3, 4 when it comes to testing hypotheses). This is estimated by NLS, however i am wondering how i define wt and w ̃t. My thought was to construct a matrix and place each series in the columns however, when i try this i get an error message. I hope this is enough information.

You can compute them using the matrix language and place results in a matrix, but to estimate using EViews regression tools, they'll have to put into series. The mtos function may be used to convert a matrix into series. You just have to make sure you are matching rows of the matrix to the rows of the workfile. But looking at your expressions, I'd probably just generate series for each of the elements of the w in the usual fashion.