Determining the Number of Factors

[Frauke]

Hello,

I want to analyse a money demand function for a panel of countries and would like to determine the number of factors to apply the PANIC methodology proposed by Bai and Ng (2004), A PANIC attack on unit roots and cointegration. Econometrica, 72, 1127–1177. Thus, I use the RATS procedure baing.scr (with fewer features since I work with RATS 6.3) to estimate the required number of factors of the set of countries for each variable (using the formulas in Bai and Ng, "Determining the Number of Factors in Approximate Factor Models", Econometrica, 2002, vol 70, pp 191-222).

First of all, I am not sure whether I have to define the original (differenced)* variables as matrix x or already estimated factors (using principal component analysis) or even both (I attach both as xls-file).
(Parameters: x = nper x nvar matrix of data. The factors are nper x 1 vectors)

Further, even though there are no error messages I get no output as I would expect from the code:

report(action=define)
report(atrow=1,atcol=1,align=center) "Factors" "PCP1" "PCP2" "ICP1" "ICP2"
report(atrow=2,atcol=1,fillby=cols) %seq(1,max)
report(atrow=2,atcol=2) stats
report(action=format,atcol=2,tocol=2,tag=min,special=onestar)
report(action=format,atcol=3,tocol=3,tag=min,special=onestar)
report(action=format,atcol=4,tocol=4,tag=min,special=onestar)
report(action=format,atcol=5,tocol=5,tag=min,special=onestar)
report(action=format,atrow=2,atcol=2,picture="*.#####",align=decimal)
report(action=show)

What might be the problem(s)?

Thanks in advance!
Frauke

*Following the PANIC methodology proposed by Bai and Ng (2004) I want to apply the method of principal components to the first differenced data.

[Frauke]

Hello again,

I just run the more recent version of baing.scr again with RATS 7.
Unfortunately I get a lot of error messages:

## SX9. REPORT is not a Basic Data Type
>>>>local report <<<<
## CP3. Instruction is Illegal Outside Compiled Section
>>>>local <<<<
## SX11. Identifier WX is Not Recognizable. Incorrect Option Field or Parameter Order?
>>>>dim wx(<<<<
## SX11. Identifier CENTER is Not Recognizable. Incorrect Option Field or Parameter Order?
>>>>if center.or.<<<<

and so forth.

These four messages appear when running the program lines:

local report freport
local string ltitle

dim wx(nper,nvar)
if center.or.standardize {

Have I to define more than a x matrix with the data which I want to reduce to factors?
Are there any bugs in the program code baing.scr?

Regards,
Frauke

[TomDoan]

Hello again,

I just run the more recent version of baing.scr again with RATS 7.
Unfortunately I get a lot of error messages:

## SX9. REPORT is not a Basic Data Type
>>>>local report <<<<
## CP3. Instruction is Illegal Outside Compiled Section
>>>>local <<<<
## SX11. Identifier WX is Not Recognizable. Incorrect Option Field or Parameter Order?
>>>>dim wx(<<<<
## SX11. Identifier CENTER is Not Recognizable. Incorrect Option Field or Parameter Order?
>>>>if center.or.<<<<

and so forth.

These four messages appear when running the program lines:

local report freport
local string ltitle

dim wx(nper,nvar)
if center.or.standardize {

Have I to define more than a x matrix with the data which I want to reduce to factors?
Are there any bugs in the program code baing.scr?

Regards,
Frauke

The ability to define a LOCAL REPORT was added with version 7.1. The number-crunching didn't change, just the output, so there's no reason to use an incompatible version of BaiNg.

[TomDoan]

Hello,

I want to analyse a money demand function for a panel of countries and would like to determine the number of factors to apply the PANIC methodology proposed by Bai and Ng (2004), A PANIC attack on unit roots and cointegration. Econometrica, 72, 1127–1177. Thus, I use the RATS procedure baing.scr (with fewer features since I work with RATS 6.3) to estimate the required number of factors of the set of countries for each variable (using the formulas in Bai and Ng, "Determining the Number of Factors in Approximate Factor Models", Econometrica, 2002, vol 70, pp 191-222).

First of all, I am not sure whether I have to define the original (differenced)* variables as matrix x or already estimated factors (using principal component analysis) or even both (I attach both as xls-file).
(Parameters: x = nper x nvar matrix of data. The factors are nper x 1 vectors)

Further, even though there are no error messages I get no output as I would expect from the code:

report(action=define)
report(atrow=1,atcol=1,align=center) "Factors" "PCP1" "PCP2" "ICP1" "ICP2"
report(atrow=2,atcol=1,fillby=cols) %seq(1,max)
report(atrow=2,atcol=2) stats
report(action=format,atcol=2,tocol=2,tag=min,special=onestar)
report(action=format,atcol=3,tocol=3,tag=min,special=onestar)
report(action=format,atcol=4,tocol=4,tag=min,special=onestar)
report(action=format,atcol=5,tocol=5,tag=min,special=onestar)
report(action=format,atrow=2,atcol=2,picture="*.#####",align=decimal)
report(action=show)

What might be the problem(s)?

Thanks in advance!
Frauke

*Following the PANIC methodology proposed by Bai and Ng (2004) I want to apply the method of principal components to the first differenced data.

You indeed want to apply BaiNg to the first differenced data. This seems to work. I tested it with an older version of RATS.

Code: Select all

open data moneydemand.xls
data(format=xls,org=columns,sheet="dm1") 1 98 m1_au m1_cn m1_dk m1_fr m1_bd $
  m1_it m1_jp m1_nl m1_sd m1_sw m1_us
make x
# m1_au to m1_us
@baing(max=4,standardize) x

[Frauke]

Thanks a lot!
Now it works without any problems.

However, can you please confirm that it is correct to include the original, i.e. not decomposed, data in x but not the already estimated factors?
Or to put it in another way: Implies baing.scr a principal component analysis?

Regards,
Frauke

[TomDoan]

Thanks a lot!
Now it works without any problems.

However, can you please confirm that it is correct to include the original, i.e. not decomposed, data in x but not the already estimated factors?
Or to put it in another way: Implies baing.scr a principal component analysis?

Regards,
Frauke

That's correct. You input the original set of data (in this case in differences). Once you've decided upon the number of components, you use either PRINCOMP or PRINFACTORS (both standard RATS procedures) to extract the desired principal components for the next step.

[Frauke]

Thanks for the quick and informative reply!