System estimation dlog spec - offset 1 period?

[ssperling]

I'm estimating a system of equations (essentially a pooled regression) with a series of equations of the following form:

dlog(y) = c(1)*dlog(x1) + c(2)*dlog(x2) + c(3)*dlog(x3)

The results i'm getting appear to be offset by one period. I.e. charting @pchy(y) vs @pchy(yhat) looks off by a period, but @pchy(y(-1)) vs @pchy(yhat) looks spot on. Do I have my econometrics off here or is it the output?

P.S. I'm getting similar results using the specification:
log(y) = c(1)*dlog(x1) + ... + log(y(-1))

[ssperling]

Let me rephrase:

For an equation of the specification: dlog(y) = c(1)*dlog(x1) + c(2)*dlog(x2) + ... + c(n)*dlog(xn),
with annual historical data for the endogenous and exogenous variables from 1990 -2012 (and exogenous variables with values to 2023),
I would expect the estimate/static forecast to yield results for 1991-2012.
Should one expect to get an estimate/forecast value for 2013 (which is what I am getting - note output below which is the result of a static forecast on a single equation solving for y [instead of dlog(y)] )?

1990 25.80000 NA
1991 25.00000 28.12299
.
.
.
2012 42.80000 40.80531
2013 NA 43.34411


For clarity, I am using EViews 6 (if that matters)

[EViews Gareth]

Code: Select all

create a 1991 2015 smpl 1991 2012 series y=rnd series x1=rnd series x2=rnd equation eq1.ls dlog(y) dlog(x1) dlog(x2) eq1.fit yf show yf

Seems to only produce results up to 2012...

[ssperling]

Got it. Appears I didn't truncate my sample when executing the fit command. But the question remains why would a value be calculated in 2013 with the code:

Code: Select all

create a 1991 2015 smpl 1991 2012 series y=rnd smpl 1991 2015 series x1=rnd series x2=rnd equation eq1.ls dlog(y) dlog(x1) dlog(x2) eq1.fit yf show yf

...if no value for y exists in 2013?

[EViews Gareth]

Ignoring the log part (since it is irrelevant), and assuming you just have d(y)...

If your dependent variable is D(y) [i.e. Y-Y(-1)], then when you perform a static forecast of Y in 2012, you need data for the X variables in 2012, and for Y in 2011. You don't need data for Y in 2012.

To forecast for 2013, you need data for X in 2013, Y in 2012, and you don't need Y in 2013.

To forecast for 2014, you need data for X in 2014, Y in 2013, and you don't need Y in 2014.