PANCOINT Procedure |
@PANCOINT does cointegration tests in heterogeneous panels with multiple regressors ("Pedroni tests"), from Pedroni (1999) and Pedroni(2004). Heterogeneous panels allow the cointegrating vector (and short-term dynamics) to be different from individual to individual. When applied to just a single series, it does a unit root test.
@PANCOINT( options ) start end
# list of variables (list dependent variable first)
Parameters
|
start, end |
range for regression. By default, the maximum range permitted by all variables involved in the regression allowing for lags. |
Options
DET=NONE/[CONSTANT]/TREND
TDUM/[NOTDUM]
Use TDUM to subtract out common time effects
SMPL=standard SMPL option[not used]
UNWEIGHTED/[NOUNWEIGHTED]
Use UNWEIGHTED for unweighted statistics. By default, they are weighted by the long-run variances.
LAGS=(maximum number of) augmenting lags in the ADF regressions [Schwert's]
CRIT=[FIXED]/GTOS/AIC/BIC/HQ
SLSTAY=significance level to keep lag in model with METHOD=GTOS [.10]
Criterion to use to select the number of lags. This is done on an individual by individual basis. CRIT=FIXED uses the input LAGS on each. CRIT=GTOS starts with LAGS and drops lags until it hits one which has the marginal t with significance level set by the SLSTAY option. CRIT=AIC/BIC/HQ select the lag length using AIC, BIC (or SBC) or HQ.
[PRINT]/NOPRINT
TITLE="title of report" ["Panel Cointegration Tests"]
Example
This does unit root tests on two variables (for each, with and without time dummies), then a cointegration test on a pair. All use general-to-specific lag pruning with a maximum of 3 lags.
*
* Do unit root tests
*
@pancoint(det=constant,lags=3,crit=gtos,title="Log Investment Shares-No Time Dummies")
# logiy
@pancoint(det=constant,lags=3,crit=gtos,tdum,title="Log Investment Shares-With Time Dummies")
# logiy
@pancoint(det=trend,lags=3,crit=gtos,title="Log PC Income-No Time Dummies")
# logypc
@pancoint(det=trend,lags=3,crit=gtos,tdum,title="Log PC Income-With Time Dummies")
# logypc
*
* Do cointegration test
*
@pancoint(det=trend,lags=3,crit=gtos,tdum,title="Cointegration Test-With Time Dummies")
# logypc logiy
Sample Output
The output looks quite different for the one-variable (unit root) and multi-variable (cointegration) tests. This is from one of the unit root tests:
Log Investment Shares-No Time Dummies
raw panel unit root test results
Levin-Lin rho-stat -4.60
Levin-Lin t-rho-stat -1.15
Levin-Lin ADF-stat -0.10
IPS ADF-stat -1.02
(using large sample adjustment values)
Individuals 29
Time Periods 43
Regressors 0
All reported values are distributed N(0,1)
under null of unit root or no cointegration
Panel stats are weighted by long run variances
(see OBES reference for details)
This is the output from the cointegration test.
Cointegration Test-With Time Dummies
panel v-stat 1.05
panel rho-stat -0.93
panel pp-stat -2.76
panel adf-stat -1.55
group rho-stat 0.60
group pp-stat -2.14
group adf-stat -1.23
Individuals 29
Time Periods 43
Regressors 1
All reported values are distributed N(0,1)
under null of unit root or no cointegration
Panel stats are weighted by long run variances
(see OBES reference for details)
Copyright © 2025 Thomas A. Doan