The RATS Software Forum

I have quartely data on GDP. Could anyone advise me how to change to monthly data?
Thank you very much

Use the @DISAGGREGATE procedure. See the story in http://www.estima.com/newslett/July2008RATSLetter.pdf.

Suppose I have 8month data for each year, I would like to estimate a model with monthly data.
And the timing of the data each year is different, e.g. 1,5,6,7,8,10,11,12for some years while 1,2,3,4,8,9,10,11 for some others.

Is that possible to disaggregate the data?

Kalman smoothing with a state-space model. That's what DISAGGREGATE is doing. Since it appears that you want the simpler "interpolation" (as opposed to "distribution", where sums across periods need to match), you can pick a relatively simple state space model, like a RW-AR(1), and Kalman smooth. You just want to set your data up as monthly with NA's where you have skips.

TomDoan wrote:Kalman smoothing with a state-space model. That's what DISAGGREGATE is doing. Since it appears that you want the simpler "interpolation" (as opposed to "distribution", where sums across periods need to match), you can pick a relatively simple state space model, like a RW-AR(1), and Kalman smooth. You just want to set your data up as monthly with NA's where you have skips.

What if I just want to fill in the missing data points without changing the current ones?

That's what will happen. You use a state space model with no measurement error, so if there's an actual observation at a time period, that's your smoothed value.

TomDoan wrote:That's what will happen. You use a state space model with no measurement error, so if there's an actual observation at a time period, that's your smoothed value.

Great. Rats is really versatile.

ivory4 wrote:Suppose I have 8month data for each year, I would like to estimate a model with monthly data.
And the timing of the data each year is different, e.g. 1,5,6,7,8,10,11,12for some years while 1,2,3,4,8,9,10,11 for some others.

Is that possible to disaggregate the data?

That's similar to http://www.estima.com/forum/viewtopic.php?f=3&t=545. If you set up your data file with date fields like

1990:1 xxx
1990:5 yyy
1990:6 zzz

etc.

and read into a monthly calendar, DATA will insert NA's for the missing months. In that case, you don't even need to patch NA's into the Y series on DLM---just use the data as it's read in.