Project CEMAPRE internal
|Title||A Nonlinear Spatial Lag Model for Fractional Responses (continuation)|
|Participants||Isabel Proença (Principal Investigator)|
|Summary||Some of the variables modelled with econometric tools are fractional in nature. Examples are pension|
plan participation rates, industry market share, television ratings, fraction of land area allocated
to agriculture, poverty rates, and test pass rates. The usual procedure for dealing with these
variables is to transform them, for example using the logit transformation, to apply linear models
to the transformed variable. However, the literature points out several drawbacks to this procedure
in the context of cross-sectional data and time series. On the other hand, when the observed values
of the variables are in the limits (0 or 1) the logit transformation cannot be applied. The
alternative is to resort to nonlinear models. In spatial econometrics, nonlinear models are little
studied in terms of estimation and inference of unknown coefficients and marginal effects, because
they involve greater complexity due to the presence of spatial dependence of the observations.
This project aims to study the use of nonlinear models to explain fractional spatial variables
taking into account the spatial dependence of the data. A nonlinear autoregressive model is
specified, marginal effects are deduced, and estimation procedures are proposed. Their performance
is evaluated through a detailed simulation study and compared with the traditional approach based on
the application of the linear model. To illustrate the approach introduced in this study, an
empirical application explaining the poverty rate in European regions is carried out.
The project has the collaboration of Ludgero Glórias from Banco de Portugal.