Please click here if you are unable to view this page.
TOPIC:
COMBINED ESTIMATORS IN PANEL MODELS
ABSTRACT
In the analysis of panel data models, a historical question often asked is whether to estimate parameters, or make forecasts, the model used should be fixed effect or random effect. Some researchers make this decision based on the Hausman’s test , but some others often make a decision arbitrarily depending on which one gives “better” results. Similarly, in the empirical analysis of panel data, a question often asked is whether to estimate parameters, or make forecasts, based on the model separately for the different individual cross-section units or to estimate the model by pooling the entire data set. This issue has become more important since Robertson and Symons (1992) and Pesaran and Smith (1993) discussed the biases that are likely to occur if the parameter heterogeneity is ignored and data are pooled. To explore these issues in the above panel models we propose combined estimators which can be used under such uncertainties. These combined estimators are developed for the linear parametric panel models. The estimation of nonparametric functional coefficients panel model, as well as the special case of varying coefficients model, are also proposed.