Blanchard-Quah AER 1989 paper results

[TomDoan]

The set of attachments below do pretty much the full analysis from Blanchard-Quah(1989), "The Dynamic Effects of Aggregate Demand and Supply Disturbances", AER, vol 79, no. 4. The BlanchardQuahAER1989.zip has the files that are specific to this application: the data and the two main program files.

BQExtras.zip has a number of procedures which can be used more generally. One of these is MCGraphIRF, which organizes graphs for the impulse response functions after Monte Carlo integration, and MCFEVDTable, which organizes a table with error band information for a decomposition of variance. BQDODRAWS.SRC does the simulations for a particular model.

The program files and procedure file in the main zip are:

• BQEXAMPLE.PRG. Program file which does everything that doesn't involve simulations. This includes the historical decomposition of the results of demand shocks and the calculation of the supply and demand shocks.
• BQMONTE.PRG. Program file which controls all the simulation analysis.

17 July 2009. Updated to use newer versions of procedures

[le_chien_de_feu]

Hello,

i am new with RATS. I tried to replicate this paper using the codes previoisly posted using RATS 7.0.

The results for the impulse-response functions of unemplyment are exactly the same of the ones in the paper (the demand shock response is exactly the same, while the supply shock response is slighlty wider).

Unfortunately the GDP responses to the two shocks are very different from the paper. Does anybody else tried to do it? Do you have the same different results?

Thanks, waiting for a reply.

[TomDoan]

If you're talking about the error bands, read the comments at the top of bqmonte.prg.

[asdavies6032]

What kind of modifications do I need to make to Blanchard and Quah for more than 2 variables
Albert

[TomDoan]

What kind of modifications do I need to make to Blanchard and Quah for more than 2 variables
Albert

The direct extension of BQ is fairly simple, since %BQFACTOR will work with any size VAR. However, it tends to be uninteresting, since it requires that you have one shock which has no permanent effects on y2 and another have no permanent effects on y1 and y2 which isn't a very likely pattern. Once you get above two variables, you usually need to go with a combination of short and long run restrictions. See 7.5.3 in the Version 8 User's Guide (10.5.3 for the Version 7 UG). Also the http://www.estima.com/procs_perl/galiqje1992.zip examples.

[ivory4]

Code: Select all

set loggnp = log(gnp)
set loggd  = log(gd87)
set logrgnp = loggnp-loggd+log(100)
set dlogrgnp = 100.0*(logrgnp-logrgnp{1})


What is gd87?

[TomDoan]

Code: Select all

set loggnp = log(gnp)
set loggd  = log(gd87)
set logrgnp = loggnp-loggd+log(100)
set dlogrgnp = 100.0*(logrgnp-logrgnp{1})


What is gd87?

GD87 is the GNP deflator. GNP is nominal GNP.

[asdavies6032]

Tom,
can you please help me solve this error:

## MAT2. Matrices with Dimensions 37 x 9 and 36 x 9 Involved in + Operation
The Error Occurred At Location 0425 of MCVARDODRAWS
Line 50 of MCVARDODRAWS

i am trying to draw impulse graphs for a 9 variable VAR
Albert

[TomDoan]

Tom,
can you please help me solve this error:

## MAT2. Matrices with Dimensions 37 x 9 and 36 x 9 Involved in + Operation
The Error Occurred At Location 0425 of MCVARDODRAWS
Line 50 of MCVARDODRAWS

i am trying to draw impulse graphs for a 9 variable VAR
Albert

The last thing you do before @MCVARDODRAWS should be an ESTIMATE on the model that you're inputting to the procedure. Offhand, it looks as if you're doing some other type of regression that doesn't match the VAR.

[Sonica Singhi]

Hello
I am new to Rats. I tried to replicate the Blanchard Quah paper and could successfully do it. However, i have got few questions pertaining that -
1.The data involves LHMUR series. what is the full form of LHMUR? Also in the codes we use URADJUST. Does URADJUST means unempolyment detrended?
2. I want to replicate the article for UK economy.what data should i procure - Is it Level of GNP and rate of unemployment? However,the unemployment rate data of UK is nonstationary in levels and stationary in the first difference.What do you think i can do now?
3 Can i directly use the attachment named BQEXTRAS-ZIP that is uploaded in the forum?
4. I want to conduct a granger causality test for the two variables in Blanchard and Quah. Could you please hep me sending the codes for the same?

Many Thanks
Sonica

[TomDoan]

Hello
I am new to Rats. I tried to replicate the Blanchard Quah paper and could successfully do it. However, i have got few questions pertaining that -
1.The data involves LHMUR series. what is the full form of LHMUR? Also in the codes we use URADJUST. Does URADJUST means unempolyment detrended?
2. I want to replicate the article for UK economy.what data should i procure - Is it Level of GNP and rate of unemployment? However,the unemployment rate data of UK is nonstationary in levels and stationary in the first difference.What do you think i can do now?
3 Can i directly use the attachment named BQEXTRAS-ZIP that is uploaded in the forum?
4. I want to conduct a granger causality test for the two variables in Blanchard and Quah. Could you please hep me sending the codes for the same?

Many Thanks
Sonica

The data actually used in the model are adjusted quite a bit from the observed values. You'll have to check the original paper to get the sources. However, yes, URADJUST is the unemployment rate detrended and the GDP series is in growth rates, which have had separate means extracted from different ranges. The model assumes the data are stationary, so these transformations were done to get rid of what were seen as obvious problems. However, I'm not sure anyone would detrend the unemployment rate any longer.

If you have a current version of RATS, use the Blanchard Quah programs on that.

I wouldn't be concerned about whether U has a unit root. The long-run zero restriction is on the growth of GDP, not on unemployment, so the identification of the demand shock isn't affected by the long run response of unemployment.

There's nothing special about doing a Granger causality test for this pair of variables. However, the whole B-Q model isn't very interesting if U doesn't Granger cause Y---the supply shock will then have to be the innovation to Y.

[FRegensburg]

Dear Mr Doan,

I was hoping to get some advice from you on how to adapt the BQ procedure to a country other than the US, and I apologise in advance for the pedestrian nature of my questions. (By the way, all I’m after really is the measure of the output gap predicted by the model, I’m aware that you’ve written a code specifically for that, but I think it would be easier for me to just use the BQ code as I’m just starting to come to grips with the software). There are three things that I’m concerned about: the NBER classification of peaks and troughs, the dummy variables, again, specific to the US, and how to obtain figures from two graphs.


How shall I extract the means without regressing the series against two dummies, i.e.:
set dummy1 = t<=1973:4
set dummy2 = t>1973:4
linreg dlogrgnp
# dummy1 dummy2
set gdpadjust = %resids
prj means_from_gnp

I thought the following lines referred to the US business cycle, obviously I’m missing something here, because if I remove them, I somehow get rid of the graph to which they apply as well.

@NBERCycles(peaks=peaks,trough=troughs)
*
* This gets the scale correct for the spikes showing the peaks and troughs
*
set peaks = .10*peaks
set troughs = -.10*troughs
graph(footer="Figure 8. Output Fluctuations Due to Demand",$
ovcount=2,overlay=spike,ovsame) 3
# histdecomp(3,1)
# peaks 1950:2 * 2
# troughs 1950:2 * 2

what shall I add to the following lines to extract the output gap numbers?

graph(footer="Figure 8. Output Fluctuations Due to Demand",$
ovcount=2,overlay=spike,ovsame) 3
# histdecomp(3,1)
# peaks 1950:2 * 2
# troughs 1950:2 * 2

…as well as to these lines to get the figures for trend GDP?

set lessdemand = histdecomp(1,1)+histdecomp(2,1)
graph(footer="Figure 7. Output Fluctuations Absent Demand")
# lessdemand


Many thanks in advance!

[TomDoan]

Dear Mr Doan,

I was hoping to get some advice from you on how to adapt the BQ procedure to a country other than the US, and I apologise in advance for the pedestrian nature of my questions. (By the way, all I’m after really is the measure of the output gap predicted by the model, I’m aware that you’ve written a code specifically for that, but I think it would be easier for me to just use the BQ code as I’m just starting to come to grips with the software). There are three things that I’m concerned about: the NBER classification of peaks and troughs, the dummy variables, again, specific to the US, and how to obtain figures from two graphs.


How shall I extract the means without regressing the series against two dummies, i.e.:
set dummy1 = t<=1973:4
set dummy2 = t>1973:4
linreg dlogrgnp
# dummy1 dummy2
set gdpadjust = %resids
prj means_from_gnp

diff(center) dlogrgnp / gdpadjust

would pull a single mean out of the data.

I thought the following lines referred to the US business cycle, obviously I’m missing something here, because if I remove them, I somehow get rid of the graph to which they apply as well.

@NBERCycles(peaks=peaks,trough=troughs)
*
* This gets the scale correct for the spikes showing the peaks and troughs
*
set peaks = .10*peaks
set troughs = -.10*troughs
graph(footer="Figure 8. Output Fluctuations Due to Demand",$
ovcount=2,overlay=spike,ovsame) 3
# histdecomp(3,1)
# peaks 1950:2 * 2
# troughs 1950:2 * 2

what shall I add to the following lines to extract the output gap numbers?

graph(footer="Figure 8. Output Fluctuations Due to Demand",$
ovcount=2,overlay=spike,ovsame) 3
# histdecomp(3,1)
# peaks 1950:2 * 2
# troughs 1950:2 * 2

…as well as to these lines to get the figures for trend GDP?

set lessdemand = histdecomp(1,1)+histdecomp(2,1)
graph(footer="Figure 7. Output Fluctuations Absent Demand")
# less demand


Many thanks in advance!

If you want to get rid of the peaks and troughs part, just do:

graph(footer="Figure 8. Output Fluctuations Due to Demand") 1
# histdecomp(3,1)

This is correct if you get rid of the extra space in "less demand" on the # line.

set lessdemand = histdecomp(1,1)+histdecomp(2,1)
graph(footer="Figure 7. Output Fluctuations Absent Demand")
# less demand

[FRegensburg]

Thanks a lot for that. To a novice, there is something slightly god-like about things materialising after the right words, as it were.
In your original code, you added the means previously removed back into GDP to get the BQ equivalent of potential output, via the line:

set histdecomp(1,1) = histdecomp(1,1)+means_from_gnp

When, however, one mean is taken out, that is:

diff(center) dlogrgnp / gdpadjust

Like you indicated, how do I go about storing it so that I can add it back in, as above?

Many thanks

[TomDoan]

If you do it right away, before something else redefines %mean:

diff(center) dlogrgnp / gdpadjust
set means_from_gnp = %mean