EViews.com

Hi,

I was trying to estimate a dynamic factor model in EViews, but I seem to get stuck every time.

Here's the idea: to keep things simple, I took inflation data from Germany, France and Denmark. The model I would like to estimate is:

Inflation (i,t) = B(i) * F(t) + e(i,t)
F (t) = c1 + c2 * F(t-1) + c3 * F(t-2) + u(t)

with F(t) being the dynamic factor.

I thought that the following state-space model would do the trick:


@signal france = c(11)*sv1 + [var = exp(c(1))]
@signal germany= c(12)*sv1 + [var=exp(c(2))]
@signal denmark = c(13)*sv1 + [var=exp(c(3))]

@state sv1 = c(21) + c(22)*sv1(-1) + c(23)*sv2 + [var=exp(c(24))]
@state sv2 = sv1(-1)

However, the error I get is:
"Invalid lags or leads in state variables in state equation '@state sv1 = c(21) + c(22)*sv1(-1) + c(23)*sv2 + [var=exp(c(24))]'".

I've double-checked the manual, and I don't see why the lagged sv1 in the state equation should give me an error.... I'd be grateful if anyone could point me into the right direction.

Many thanks, Philipp

@signal france = c(11)*sv1 + [var = exp(c(1))]
@signal germany= c(12)*sv1 + [var=exp(c(2))]
@signal denmark = c(13)*sv1 + [var=exp(c(3))]

@state sv1 = c(21) + c(22)*sv1(-1) + c(23)*sv2 + [var=exp(c(24))]
@state sv2 = sv1(-1)

However, the error I get is:
"Invalid lags or leads in state variables in state equation '@state sv1 = c(21) + c(22)*sv1(-1) + c(23)*sv2 + [var=exp(c(24))]'".

I've double-checked the manual, and I don't see why the lagged sv1 in the state equation should give me an error.... I'd be grateful if anyone could point me into the right direction.

Many thanks, Philipp

Is it possible you wanted "c(23)*sv2(-1)"?

.... sigh. You are absolutely correct.

After you pointed me in the right direction, I changed the coefficient and estimated the model (see attached workfile). When I examine the series for the dynamic factor, it basically looks like the model works. However, I noticed a lot of missing standard errors/probabilities and the warning "Singular covariance - coefficients are not unique".

Is this a sign that the data is not very informative (a lot of parameters to estimate with just 3 series) or am I (still) doing something wrong when specifying the model?

Thanks again,

Philipp

I don't see anything wrong. You might want to try allowing separate constants in each signal equation.

Excellent! Thanks, startz, and I'll try your suggestions using separate constants.

Philipp