Defining an AR(1) error term in FIML system

[vjayawic]

I was trying to specify a system to run Full Information Maximum Likelihood (FIML) estimator to replicate the research "Estimating New-Keynesian Phillips curves: A full information maximum likelihood approach" http://www.sciencedirect.com/science/article/pii/S0304393205000784

The system specified was

INF = C(1)*INF(+1) + C(2)*INF(-1) + C(3)*GAP
GAP = C(4)*GAP(+1) + C(5)*GAP(-1) - C(6)*(R - INF(+1))
R = (1-C(7))*(C(8)*INF + C(9)*GAP) + C(7)*R(-1)

The research suggests adding an AR(1) error term. According to the Eviews manual this requires an additional term [ar(1)=c(10)]

However, once the ar(1) error term is added, Eviews gives the error "ARMA terms not allowed in this procedure" under FIML. Is it possible to implement FIML estimation with ar(1) error terms?

Thankful if you can give some insights

[EViews Gareth]

No.

[vjayawic]

Thanks Gareth for your reply.

Is there any other convenient way to do Maximum Likelihood estimation with ar(1) error terms?

[startz]

You can probably write out the quasi-difference long-hand

Code: Select all

INF = C(1)*INF(+1) + C(2)*INF(-1) + C(3)*GAP


becomes

Code: Select all

INF = c(10)*INF(-1) + C(1)*(INF(+1)-c(10)*INF) + C(2)*(INF(-1)-c(10)*INF(-2)) + C(3)*(GAP-c(10)*GAP(-1))