DOLS - dynamic vs static forecasts

[maragloria]

Hi there,

I'm running the following code to compare dynamic with static forecasts by running cointreg(dols):

Code: Select all

equation dols1_br.cointreg(method=dols, trend=const, lag=1, lead=1, cov=hac) reerln_br trdbal_br eff_totln_br @determ eff_rint_br forecast reer_dyn forecast(s) reer_lt plot reerln_br reer_dyn reer_lt dols1_br.resids(g)

"forecast(s) reer_lt" is giving me the same values as "forecast reer_dyn" (see [img]

graph1.jpg (21.57 KiB) Viewed 8608 times

[/img].
What I would like to get is the "fitted" in the graph below [img]

graph2.jpg (27.67 KiB) Viewed 8608 times

[/img]. How can I get that?

Many thanks,

Mara

[trubador]

Use "fit" instead of forecast(s) to get static forecast. The latter is used in models containing ARMA terms and instructs EViews to ignore those terms and produce forecasts only for the structural part.

[maragloria]

Hello Trubador,

Thanks for your reply.

The problem is that "forecast reer_dyn" is not taking into account the dynamic term of the DOLS, as I thought it should. Is that so?

"forecast reer_dyn" and "fit reer_fit" gives the same forecasts.
See [img]

graph3.jpg (21.93 KiB) Viewed 8592 times

[/img]

Warm regads,

Mara

[trubador]

That is because you are estimating a DOLS model, which produces forecasts only for the long term relationship that does not take into account the short term dynamics.

[maragloria]

I got the impression that the "fitted" in the graph with the residuals made using resids(g) do take into account the dynamic short-term term.
The "fitted" in graph 2 is quite different from the ones in graph 3 or graph 1.

[maragloria]

Indeed, I did an OLS regression specifying the dynamic term explicitly and its fitted value is quite similar to the one in the residuals graph of a DOLS regression.
See graph4 and graph 5 below.

[img]

graph4.jpg (24.27 KiB) Viewed 8568 times

[/img]
[img]

graph5.jpg (27.38 KiB) Viewed 8568 times

[/img]

My question is: is that a way for me to program a DOLS regression to get these fitted values (taking into account the dynamic term) given by the command resids(g)?

Many thanks,

Mara

[trubador]

I also cannot think of any other way to obtain dynamic forecasts other than explicitly specifying and estimating the model with OLS.