IPSHIN—Im-Pesaran-Shin Panel Unit Root

[TomDoan]

@IPShin implements tests from Im, Pesaran and Shin, "Testing for Unit Roots in Heterogeneous Panels", J. of Econometrics, 2003, vol 115, pp 53-74. This aggregates the ADF test statistics computed separately across individuals. Under the null, the average ADF statistic has a Normal limiting distribution as N->infinity.

Detailed description

[tflavin71]

I am trying to run this procedure in WinRATS 8.0 and keep getting the following error message

## OP10. Instruction PSTATS Requires Option BLOCK
The Error Occurred At Location 489, Line 190 of IPSHIN

I can't see where I am going wrong - any help appreciated.

Thanks

[TomDoan]

Do you have a panel data CALENDAR? If you don't, it has no idea how to separate the data into individuals.

[tflavin71]

Thanks for pointing that out.. the proper Calendar instruction works but I have one further problem. Now the procedure runs the test on one series, even though I want it to run on the whole panel. The data reads in properly

Here is my instruction and output

OPEN DATA "G:\PPP\CIPS.csv"
CALENDAR(PANELOBS=588,M) 1957:1
ALL 10//588
DATA(FORMAT=PRN,nolabels,ORG=COLUMNS) 1//1957:1 1//2005:12 AUSTRIA BELGIUM FINLAND FRANCE GREECE IRELAND $
ITALY NETHERLANDS PORTUGAL SPAIN

@ipshin AUSTRIA BELGIUM FINLAND FRANCE GREECE IRELAND $
ITALY NETHERLANDS PORTUGAL SPAIN

------------------------------------------------------

Output:

Im-Pesaran-Shin Unit Root Test: Series AUSTRIA
Test has large N on large T asymptotics
Null is Unit Root. Alternative is rho(i)<>1 for some i
Individual Specific Components: Constant

N 1
T 1
P 0
Statistic Signif Level
Z tbar 1.150657 0.875063
Z ttildebar NA NA
TBar 0.000000
T-tilde-bar NA
1% CV -3.790000
5% CV -2.760000
10% CV -2.380000



Thanks in advance..

[TomDoan]

You have to PFORM the separate time series into a panel data set first.

Code: Select all

OPEN DATA "G:\PPP\CIPS.csv"
CALENDAR(M) 1957:1
DATA(FORMAT=PRN,nolabels,ORG=COLUMNS) 1957:1 2005:12 AUSTRIA BELGIUM FINLAND FRANCE GREECE IRELAND $
ITALY NETHERLANDS PORTUGAL SPAIN
PFORM cips
# AUSTRIA BELGIUM FINLAND FRANCE GREECE IRELAND $
   ITALY NETHERLANDS PORTUGAL SPAIN
*
* Reset the CALENDAR to a panel date schem
*
cal(panelobs=2005:12,m) 1957:1
all 10//2005:12
@ipshin cips

[celineboulenger13]

Hi, I am working on testing for unit roots in a multivariate panel but when I use @ipshin for gdp, the results look different than when I do it for the variable re.
Here are the 2 outputs/inputs

@ipshin gdp

Im-Pesaran-Shin Unit Root Test: Series GDP
Test has large N on large T asymptotics
Null is Unit Root. Alternative is rho(i)<>1 for some i
Individual Specific Components: Constant

N 23
T 22
P 0
Statistic Signif Level
Z tbar 4.317106 0.999992
Z ttildebar 4.475388 0.999996
TBar -0.710339
T-tilde-bar -0.694758
1% CV -1.940000
5% CV -1.820000
10% CV -1.750000

@ipshin re

Im-Pesaran-Shin Unit Root Test: Series RE
Test has large N on large T asymptotics
Null is Unit Root. Alternative is rho(i)<>1 for some i
Individual Specific Components: Constant

N 23
T 21 to 22
P 0
Statistic Signif Level
Z tbar 6.731118 1.000000
Z ttildebar 6.318091 1.000000

This is a problem since I need to report the tstats. I don't know why it doesn't show for RE.

Thank you for your help!

[TomDoan]

The t-statistic is only available when the data set is balanced. Yours isn't---at least one of the individuals is missing a data point for RE.

It looks like you're testing your raw data. Shouldn't you do using the logs instead?

[celineboulenger13]

I did have one missing value but I fixed that but I still cannot get the t-stats reported when using the series re (these are in logs).
@ipshin(lags=5) re

Im-Pesaran-Shin Unit Root Test: Series RE
Test has large N on large T asymptotics
Null is Unit Root. Alternative is rho(i)<>1 for some i
Individual Specific Components: Constant

N 23
T 17
P 5
Statistic Signif Level
Z tbar 5.653283 1.000000
Z ttildebar 5.647141 1.000000

Any other reason why it could not give me tstats? thanks!

[celineboulenger13]

Is there any way to include the max amount of lags and also get tstats? I am trying to report both but it seems that when lags are included, the tstats are not reported. THanks !!!

@ipshin(det=trend,crit=aic) re

Im-Pesaran-Shin Unit Root Test: Series RE
Test has large N on large T asymptotics
Null is Unit Root. Alternative is rho(i)<>1 for some i
Individual Specific Components: Constant + Trend

N 23
T 22
Avg P 0.000000 chosen from 0 by AIC
Statistic Signif Level
Z tbar -0.125701 0.449984
Z ttildebar 2.259853 0.988085
TBar -2.189132
T-tilde-bar -1.769112
1% CV -2.580000
5% CV -2.460000
10% CV -2.400000

@ipshin(crit=aic,lags=5) re

Im-Pesaran-Shin Unit Root Test: Series RE
Test has large N on large T asymptotics
Null is Unit Root. Alternative is rho(i)<>1 for some i
Individual Specific Components: Constant

N 23
T 17 to 22
Avg P 1.347826 chosen from 5 by AIC
Statistic Signif Level
Z tbar 6.585091 1.000000
Z ttildebar 6.217889 1.000000

[TomDoan]

The t-stats are only available in very, very, controlled situations (balanced sample, no augmenting lags). The main test statistic is the average Z.

Yes, @IPSHIN has options for choosing the number of lags, which, as is typical of panel UR tests, does a separate analysis for each individual. See the detailed description link at the first page.

However, from the look of it, this series isn't a candidate for a unit root test. It has to be really explosive to generate a test statistic like that (probably not surprising given what it is).