Page 19 - Contributed Paper Session (CPS) - Volume 8

CPS2151 Sarah B. Balagbis

Estimating a panel data model with structural
change and panel heterogeneity
Sarah B. Balagbis
Supervising Statistical Specialist, Philippine Statistics Authority, Quezon City, Philippines

Abstract
The forward search algorithm and nonparametric bootstrap are used in the
context of the back fitting algorithm to estimate a panel data model with
structural change and panel heterogeneity. Simulated data with two
covariates are used to illustrate the procedure. The method is comparable to
time series cross section regression (estimated using generalized least
squares) with respect to predictive ability in scenarios where there is actually
no perturbation or when there is structural change in the data. The method,
however, is superior when there is panel heterogeneity and both panel
heterogeneity and structural change in the data. The proposed procedure
yields robust covariate parameter estimates. Further, it yields efficient and
reliable covariate parameter estimates which are comparable to the time series
cross section regression estimated using generalized least squares when there
are no real perturbations or when there is structural change in the data.

Keywords
Forward search algorithm; nonparametric bootstrap; back fitting algorithm

1. Introduction
Panel data enable the study of the dynamics of a phenomenon better than
either a cross-section or time series alone. Gujarati (2003) as cited by Yaffee
(2003) noted that the combination of time series with cross-sections can
enhance the quality and quantity of data. Panel data control the heterogeneity
of the units and gives more informative data, more variability, and less
collinearity among variables. Panel data analysis can also provide a rich and
powerful study of a set of units, if one is to consider both the space and time
dimension of the data (Yaffee, 2003).
This paper proposes an estimation procedure for panel data modelling in
the presence of structural change or panel heterogeneity. The method takes
advantage of the benefits from the forward search algorithm and the
bootstrap to hopefully come up with robust estimates. It adapts the spatial-
temporal model proposed by Landagan and Barrios (2007) but omits the
spatial component in the system. A nonparametric bootstrap and the forward
search algorithm from Campano (2008) are incorporated into the back fitting
procedure proposed by Dumanjug (2007).

8 | I S I W S C 2 0 1 9