ARDL Bounds Test & ECM
Posted: Wed Apr 16, 2014 9:47 am
Hi all,
I wasn't sure which section to post this in as I have questions regarding eviews specifically and the theory of the model in question. I've post it here as the theory is more pressing for me. Apologies if this is the wrong place.
I have been investigating the relationships between 5 variables (call them Y, A, B, C & E).
I have found the F- and t-statistics for F(Y|A,B,C,E) to be significant after estimating the ARDL(p,q[1],q[2],q[3].q[4]) model:
(1) D(Y[t]) = a[0] + Σb[j]D(Y[t-j]) + Σc[j]D(A[t-j]) + Σd[j]D(B[t-j]) + Σe[j]D(C[t-j]) + Σf[j]D(E[t-j]) + g[0]Y[t-1] + g[1]A[t-1] + g[2]B[t-1] + g[3]C[t-1] + g[4]E[t-1] + v[t]
So, having found evidence of a long-run relationship, I have estimated the levels relationship (ARDL(p,q[1],q[2],q[3],q[4]), where p,q[1],etc are not necessarily equal to the p,q[1], etc used in the above ARDL model.:
(2) Y[t] = a[0] + Σb[j]Y[t-j] + Σc[j]A[t-j] + Σd[j]B[t-j] + Σe[j]C[t-j] + Σf[j]E[t-j] + u[t]
I then normalised on Y to find my long-run coefficients. This leaves me with something like:
Y = 0.2104 + 8.1531A - 2.9184B + 0.0070C + 0.0014E + u[t].
As far I can understand, so far so good.
1) My first question is how do I get t-statistics and p-values for the above coefficients?
Now is the part where I get confused.
I now construct the Error Correction Model by writing (2) in terms of the first differences of Y,A,B,C,E leaving me with an ARDL(p-1,q[1]-1,q[2]-1,q[3]-1,q[4]-1) as follows:
(3) D(Y[t]) = a[0] + Σb[j]D(Y[t-j]) + Σc[j]D(A[t-j]) + Σd[j]D(B[t-j]) + Σe[j]D(C[t-j]) + Σf[j]D(E[t-j]) + gEC[t-1] + u[t]
where EC[t-1] = Y[t-1] - 0.2104 - 8.1531A[t-1] + 2.9184B[t-1] - 0.0070C[t-1] - 0.0014E[t-1]
And
a[0] = a[0]
b[j] = b[j], c[j] = c[j], d[j] = d[j], e[j] = e[j]
g = 1 - Σb[j]
2) Is this correct? That the coefficients of (2) are in (3)? Or are the coefficients in (3) re-estimated? In other words, are the coefficients in (3) set to be equal to those of (2)?
3) If that is the case, how do I set coefficients in Eviews?
I have become extremely confused by this and the more I read the more confused I get. I'd really appreciate if anyone could answer any of the questions above.
Thank you!
I wasn't sure which section to post this in as I have questions regarding eviews specifically and the theory of the model in question. I've post it here as the theory is more pressing for me. Apologies if this is the wrong place.
I have been investigating the relationships between 5 variables (call them Y, A, B, C & E).
I have found the F- and t-statistics for F(Y|A,B,C,E) to be significant after estimating the ARDL(p,q[1],q[2],q[3].q[4]) model:
(1) D(Y[t]) = a[0] + Σb[j]D(Y[t-j]) + Σc[j]D(A[t-j]) + Σd[j]D(B[t-j]) + Σe[j]D(C[t-j]) + Σf[j]D(E[t-j]) + g[0]Y[t-1] + g[1]A[t-1] + g[2]B[t-1] + g[3]C[t-1] + g[4]E[t-1] + v[t]
So, having found evidence of a long-run relationship, I have estimated the levels relationship (ARDL(p,q[1],q[2],q[3],q[4]), where p,q[1],etc are not necessarily equal to the p,q[1], etc used in the above ARDL model.:
(2) Y[t] = a[0] + Σb[j]Y[t-j] + Σc[j]A[t-j] + Σd[j]B[t-j] + Σe[j]C[t-j] + Σf[j]E[t-j] + u[t]
I then normalised on Y to find my long-run coefficients. This leaves me with something like:
Y = 0.2104 + 8.1531A - 2.9184B + 0.0070C + 0.0014E + u[t].
As far I can understand, so far so good.
1) My first question is how do I get t-statistics and p-values for the above coefficients?
Now is the part where I get confused.
I now construct the Error Correction Model by writing (2) in terms of the first differences of Y,A,B,C,E leaving me with an ARDL(p-1,q[1]-1,q[2]-1,q[3]-1,q[4]-1) as follows:
(3) D(Y[t]) = a[0] + Σb[j]D(Y[t-j]) + Σc[j]D(A[t-j]) + Σd[j]D(B[t-j]) + Σe[j]D(C[t-j]) + Σf[j]D(E[t-j]) + gEC[t-1] + u[t]
where EC[t-1] = Y[t-1] - 0.2104 - 8.1531A[t-1] + 2.9184B[t-1] - 0.0070C[t-1] - 0.0014E[t-1]
And
a[0] = a[0]
b[j] = b[j], c[j] = c[j], d[j] = d[j], e[j] = e[j]
g = 1 - Σb[j]
2) Is this correct? That the coefficients of (2) are in (3)? Or are the coefficients in (3) re-estimated? In other words, are the coefficients in (3) set to be equal to those of (2)?
3) If that is the case, how do I set coefficients in Eviews?
I have become extremely confused by this and the more I read the more confused I get. I'd really appreciate if anyone could answer any of the questions above.
Thank you!