STATA COMMAND FOR PANEL DATA ANALYSIS

Declaring panel data

xtset id year

How to fill missing data for panel time series

bysort countryname: ipolate x time, gen(xi) epolate

Suppose you want to describe data:

xtsum y x1 x2 x3 x4

How to run Im-Pesaran-Shin Unit-root test (IPS)

Command for ips unit root for constant and no trend

xtunitroot ips x

For constant and trend:

xtunitroot ips x, trend

Panel unit-root test for constant after taking first difference:

xtunitroot ips d.x

For constant and trend variable after differencing:

xtunitroot ips d.x, trend

How to run correlation matrix:

pwcorr y x1 x2 x3 x4, sig

How to run pooled OLS:

reg y x1 x2 x3 x4

Command for regression with Driscoll-Kraay standard errors:

xtscc y x1 x2 x3 x4 x5, lag(4)

Command for fixed effect (FE) model:

xtreg y x1 x2 x3 x4, fe

Use the following command to store FE result

est sto fe

How to run random effect (RE) model:

xtreg y x1 x2 x3 x4, re

Command for storing random effect result

Now run hausman test to choose suitable model between fixed and random effect

If the probability value <0.05, then FE otherwise RE model should be selected.

Command for Breusch and Pagan Lagrangian multiplier (LM) test to choose b/w RE model and
POLS

xttest0

After estimating RE, If the probability value <0.05, then RE is better, otherwise POLS model should be
selected

Command for detecting heteroskedasticity problem:

estat hottest

prob value of more than 10% indicates presence of heteroscedasticity

Command for checking specification error test:

estat ovtest

How to check multicollinearity:

estat vif

Command for detecting serial correlation problem:

xtserial y x1 x2 x3 x4

Command for checking heteroskedasticity in fixed effect

xttest3

How to solve heteroskedasticity problem in FE model

xtreg y x1 x2 x3 x4, fe robust

Command for solving serial correlation in FE model:

xtregar y x1 x2 x3 x4 , fe

How to solve serial correlation when RE model is selected:

xtregar y x1 x2 x3 x4, re

Command for removing outlier data:

winsor x gen(Wx) p(0.1)

Command for fully modified ordinary least square

Command for dynamic ordinary least square (DOLS)

xtdolshm y x1 x2 x3 x4 x5

Command for canonical cointegration regression:

cointreg smd rq bd lep lpvr un ir, est(ccr) vlag(1) kernel(qs) bmeth(andrews) eqtrend(1) nodivn

Command for Westerlund panel cointegration test:

xtwest y x1 x2 x3 x4, constant lags(1 3) leads(0 3) lrwindow(3)

Command for mean group (MG) model estimation:

xtpmg d.y d.x1 d.x2 d.x3, lr(l.y l.x1 l.x2 l.x3) replace mg

est sto mg

Command for pooled mean group (PMG) model

Xtpmg d.y d.x1 d.x2 d.x3, lr(l.y l.x1 l.x2 l.x3) replace pmg

est sto pmg

The coefficient of the speed of adjustment should be negative, less than one and statistically significant as
a condition for correcting the disequilibrium.

Now use hausman test to compare MG PMG models as follows:

hausman mg pmg sigmamore

If P-value>0.05 then run PMG otherwise MG should be estimated

Choosing between MG and DFE

Run dynamic fixed effect (DFE) model as follows:

xtpmg d.y d.x1 d.x2 d.x3, lr(l.y l.x1 l.x2 l.x3) replace dfe

Assuming you already run and stored MG result then:

Run hausman test to choose between MG and DFE

hausman mg dfe sigmamore

If P-value>0.05 use DFE otherwise use MG

Running PMG model (individual)

xtpmg d.y d.x1 d.x2 d.x3, lr(l.y l.x1 l.x2 l.x3) replace replace full pmg

Running MG model (individual)

Command for estimating cross sectional dependency (CD)

pescadf x, lags(1)

pescadf d.x, lags(1)

How to estimate panel unit root test by using CIPS test (2nd generation)

pescadf x, lags(1)

pescadf d.x, lags(1)

How to run augmented mean group

xtmg y x1 x2 x3, trend aug or xtmg y x1 x2 x3, aug

How to run common correlated effect mean group

xtpmg y x1 x2 x3, cce

xtpmg y x1 x2 x3 x4, cce full

Command for dynamic common correlated effect estimator -pooled mean group

xtdcce2 d.y d.L.y d.x1 d.x2, cr(_all) reportc lr(y x1 x2 ) p(y x1 x2)

d.yl d.l.y d.x1 d.x2 y x1 x2

Command for Instrumental variable (2SLS) regression

ivreg lncons (l.lncons = l2.lncons) lnrp lnyd lnps

Command for difference (GMM)

1st step Difference System GMM

xtabond2 y L.y L2.y x1 L.x1 L(0/2).(x2 x3) i.year, gmm(L.y) iv(x1 L.x1 L(0/2).(x2 x3) i.year) nolevel
robust

xtabond2 y L.y L2.y x1 L.x1 L(0/2).(k ys) yr*, gmm(L.(n w k)) iv(L(0/2).ys yr*) nolevel robust small

xtabond2 y L.y L2.y x1 L.x1 L(0/2).(x2 x3) i.year, gmm ( L.(y x1 x2)) iv( L(0/2).x2 i.year) nolevel
robust

Two-step Difference System GMM

xtabond2 y L.y L2.y x1 L.x2 L(0/2).(x2 x3) i.year, gmm(L.y) iv(x1 L.x1 L(0/2).(x2 x3) i.year) nolevel
twostep

Two step difference GMM with Orthogonal sub-option command

Where xtabond2 stands for gmm command, the variables in the first bracket represents endogenous
variables while the second bracket contains instrumental variables of the model.

Command for System GMM:

To run system gmm, we remove “nolevel” in the ‘difference gmm’ command above

One step System GMM

xtabond2 y L.y L(0/1).(x1 x2) yr*, gmmstyle(L.(y x1 x2)) ivstyle(yr*, equation(level)) robust
small
xtabond2 y L.y L(0/1).(x1 x2) i.year , gmmstyle(L.(y x1 x2)) ivstyle(i.year, equation(level))
robust small
Two step System GMM:
xtabond2 y L.y L(0/1).(x1 x2) i.year , gmmstyle(L.(y x1 x2)) ivstyle(i.year, equation(level))
twostep small
xtabond2 y L.y L(0/1).(x1 x2) yr1978-yr1984, gmm(L.(y x1 x2), lag(1 1)) iv(yr1978-yr1984,
eq(level)) h(2) robust twostep
Command for checking Endogeneity
The first step is to regress using the following command:
ivregress gmm dv X1 X2 X3 (X4=X5 X6)
Then test for endogeneity as follows:
estat endog
A lower probability value is a sign of the endogeneity problem existing in the model
The next step is to test for validity of the instruments used as follows:
estat overid
A higher probability value of the Sagan/Hansen test signifies model is correctly specified
otherwise instruments are invalid.