The RATS Software Forum

@DISAGGREGATE is an "all-purpose" temporal disaggregation procedure, which can handle linear and log-linear versions of "Chow-Lin" type distributions, as well as proportional Denton. If you provide an empty list of regressors, it will do the same calculations as was done by the @INTERPOL and @DISTRIB procedures (though with a slightly different set of parameters and options).

Detailed description

Algorithm

The calculations are done using Kalman smoothing (for log-linear, with iterated linearizations). All of the above references use other constructive calculations, but they, in fact, are simply derived special cases of Kalman smoothing. This is the chapter from the State Space/DSGE course which provides detailed information on how this is done.
Example

What is the parameter setting to correspond to DISTRIBUTE procedure where no related series are used?

Also in UG, there is an example using DISTRIB It says we need to multiply the result by 3 to maintain the original level.

In this case, M1+M2++M3 is not equal to the original Q1 GDP. I think if for unemployment rate, we should do this?

ivory4 wrote:What is the parameter setting to correspond to DISTRIBUTE procedure where no related series are used?

MODEL=LINEAR,MAINTAIN=SUM plus the choice for the time series model.

ivory4 wrote:Also in UG, there is an example using DISTRIB

Code: Select all

cal(m) 1947
open data haverexample.rat
data(for=rats, verbose) 1947:1 2006:4 gdp
@distrib(factor=3) gdp gdpm
set gdpm = 3*gdpm

It says we need to multiply the result by 3 to maintain the original level.

In this case, M1+M2++M3 is not equal to the original Q1 GDP. I think if for unemployment rate, we should do this?

GDP is generally quoted in annual rates. Multiplying by 3 maintains that so M1+M2+M3 has the same average as Q1. Yes. You would want to do the same for unemployment rate.

TomDoan wrote:

ivory4 wrote:Also in UG, there is an example using DISTRIB

Code: Select all

cal(m) 1947
open data haverexample.rat
data(for=rats, verbose) 1947:1 2006:4 gdp
@distrib(factor=3) gdp gdpm
set gdpm = 3*gdpm

It says we need to multiply the result by 3 to maintain the original level.

In this case, M1+M2++M3 is not equal to the original Q1 GDP. I think if for unemployment rate, we should do this?

GDP is generally quoted in annual rates. Multiplying by 3 maintains that so M1+M2+M3 has the same average as Q1. Yes. You would want to do the same for unemployment rate.

But the data is for quarterly GDP level not for the growth rate?

ivory4 wrote:But the data is for quarterly GDP level not for the growth rate?

That's correct. But quarterly GDP is quoted on the basis of what the annual GDP number would be at the current rate of production.

There is another example using denton method to distribute data with a related series.
Do we need to multiply the distrib by 4?

TomDoan wrote:

ivory4 wrote:What is the parameter setting to correspond to DISTRIBUTE procedure where no related series are used?

MODEL=LINEAR,MAINTAIN=SUM plus the choice for the time series model.

What to put under "#"

Another question is about option LOGLINEAR.

My understanding is that this "loglinear" is for equation connecting interpolated series with observed series (L_t = exp(logI_t) + .....exp(logI_t-q) ) ; NOT the relation between interpolated series with high frequency information is still linear? (I_t = H_t+u_t ?); But when I check July2008 RATSLETTER it says (3) logI_t = H + Z (LOGLIN) (referring to the relation between interpolated series with high frequency information).

Could you clarify this?

ivory4 wrote:There is another example using denton method to distribute data with a related series.
Do we need to multiply the distrib by 4?

Code: Select all

set distrib =distrib*4

For a series quoted like GDP, yes.

ivory4 wrote:

TomDoan wrote:

ivory4 wrote:What is the parameter setting to correspond to DISTRIBUTE procedure where no related series are used?

MODEL=LINEAR,MAINTAIN=SUM plus the choice for the time series model.

What to put under "#"

Leave it blank.

ivory4 wrote: Another question is about option LOGLINEAR.

My understanding is that this "loglinear" is for equation connecting interpolated series with observed series (L_t = exp(logI_t) + .....exp(logI_t-q) ) ; NOT the relation between interpolated series with high frequency information is still linear? (I_t = H_t+u_t ?); But when I check July2008 RATSLETTER it says (3) logI_t = H + Z (LOGLIN) (referring to the relation between interpolated series with high frequency information).

Could you clarify this?

Aren't they the same thing? One's just an expansion of the other.

I see. I thought (L_t = exp(logI_t) + .....exp(logI_t-q) or L=I_t+...I_t-q; can be combined with any one of (7.1, 7.2, 7.2), which are I=H+Z; I=Hz; LogI=H+z
As you pointed out, DISAGGREGATE 's option LOGLIN actually indicates a combination of L_t = exp(logI_t) + .....exp(logI_t-q) and LogI=H+z;
while option LIN indicates L=I_t+...I_t-q (sum or average) and I=H+Z;
while option PROPORTION indicates L=I_t+...I_t-q (sum or average) and I=Hz

As for Stock and Watson's research memorandum "Distribution of quarterly values of GDP/GDI across months within the quarter
Background", is there a replication here?

http://www.princeton.edu/~mwatson/mgdp_gdi.html

Dear Tom,

i have a monthly time series and i want to make it weekly to match the database i am analyzing but i have two problems,

First is, the observed weekly data with is not uniformly distributed, so my question here is how to fill in exactly the empty weeks? is it correct that factor=3 increase the frequency from months to weeks?

Second is, when i use the code below, i constructed the variable [rapst] (monthly data of inventories that i want to disaggregate to weekly) and made and fixed the observed value of each month for its corresponding weeks to be a related series, is it correct to use it as related series? can i use a constant here?
when i remove the related series i get this error
thank you so much for your help

Fadi

You can't use @DISAGGREGATE to do monthly to weekly. That procedure is designed to handle only frequencies which are a fixed number of subperiods. The same underlying idea can be used from monthly to weekly, but you have to figure out how you want to handle the weeks that split across months. There are many ways that one could do that, since for some data sets, there are day-of-the-week effects, so Saturday and Sunday might need different weights than the weekdays. If there are day-of-the-week effects, the input monthly data may or may not have already been adjusted for those, which makes the weight handling in the disaggregation more complicated.