Website of
Examples of Dynamic Panel Data Estimation Using TSP:
Realization of the Estimation Using Ox Version DPD
by Yoshitsugu Kitazawa, Faculty of Economics, Kyushu Sangyo University, Japan
First design 2003.11.30. Last update 2003.12.04.
In this website, some TSP example scripts are depicted for carrying out dynamic panel data estimations. The TSP scripts generate the same estimation results as Ox version DPD scripts by Professor Doornik, Professor Arellano, and Professor Bond, except for the serial correlation tests. Firstly, the Ox scripts are introduced. Next, the TSP scripts are presented. The estimation models and the estimation methods are described in “Simple Explanation” below.
Ox scripts for estimating the dynamic panel data models
Note: Run each action script after putting the data files in the same directory as the action script. The scripts below do not use the small sample correction of the estimated variances for the two-step estimators, by Windmeijer (2000, IFS Working Paper).
<GMM (DIF) and GMM (SYS) without Time
Dummies>
Action Script: ds1_n.ox
Output: ds1_n.out
<GMM (DIF) and GMM (SYS) with Time
Dummies>
Action Script: ds1_t.ox
Output: ds1_t.out
<Data Files for the Ox scripts>
TSP scripts generating the same results with the Ox scripts
Note: Run each action script after putting the including files and the
data files in the same directory as the action script. These scripts are
written on the accumulation of the past technical supports by Professor Bronwyn
Hall and Dr. Clint Cummins.
<GMM(DIF) without Time Dummies>
Action Script: b1_dif_n.tsp
Output: B1_DIF_N.OUT
<GMM(SYS) without Time Dummies>
Action Script: b1_sys_n.tsp
Output: B1_SYS_N.OUT
<GMM(DIF) with Time Dummies>
Action Script: b1_dif_t.tsp
Output: B1_DIF_T.OUT
<GMM(SYS) with Time Dummies>
Action Script: b1_sys_t.tsp
Output: B1_SYS_T.OUT
<Automatic Routine> [New 2003.12.01]
This action script carries out the four estimations above by specifying the options. We can change the periods used in the estimations in an easy way by specifying the options.
Action Script: b1_opt.tsp
Output DN (GMM(DIF) without Time Dummies): b1_601.out
Output SN (GMM(SYS) without Time Dummies): b1_602.out
Output DT (GMM(DIF) with Time Dummies): b1_611.out
Output ST (GMM(SYS) with Time Dummies): b1_612.out
<Including Files for the TSP
Scripts>
<Data Files for the TSP Scripts>
Simple Explanation of the dynamic panel data models and the estimation
methods used in the scripts above
A series of the estimations above are
implemented for all time periods used for the estimations 
and number of
individuals 
.
The estimation model without time dummies in this website is as follows:
, (1)
where 
is the dependent
variable for individual 
at time 
, 
is the
explanatory variable, 
is the fixed
effect, 
is the
disturbance, and the parameters 
and 
are the
parameters of interest to be estimated. In this model, 
and 
. In this case, we suppose that 
for 
, and 
is endogenously
determined (
for 
and 
and 
for 
and 
).
Two types of moment conditions to
consistently estimate the parameters 
and 
by GMM are
considered on the basis of some assumptions.
The first type is the Standard moment conditions:
, (2)
for 
and 
, where 
is the first
difference operator.
The second type is the additional Stationarity moment conditions:
, (3)
for 
.
Under the estimation model without time dummies (1), the GMM (DIF) estimator in Arellano and Bond (1991, Review of Economic Studies) uses the set of moment conditions (2), while the GMM (SYS) estimator in Arellano and Bover (1995, Journal of Econometrics) and Blundell and Bond (1998, Journal of Econometrics) uses both the set of moment conditions (2) and (3).
Further, when we estimate the following model with time dummies 
for 
:
, (4)
the time dummies 
s are the further parameters of interest.
Under the estimation model with time dummies (4), the GMM (DIF) estimator uses the set of moment conditions (2) plus the set of moment conditions with respect to the time dummies:
, (5)
for 
and 
s for 
are the
parameters of interest to be estimated instead of 
s, while the GMM (SYS) estimator uses both the set of moment
conditions (2) and (3) plus the set of moment conditions with respect to the
time dummies:
, (6)
for 
and 
s for 
are the
parameters of interest to be estimated.
A zipped file of the scripts in this website and this index file is downloadable from here.
On Query, E-mail: kitazawa@ip.kyusan-u.ac.jp