Abstract
This study compares two alternative approaches–the traditional ordinary least square (OLS) and the Poisson Quasi Maximum Likelihood (PQML) procedures–to estimation of the home market effects for the heterogeneous, the reference priced and the homogeneous goods in the gravity model derived in Feenstra et al. (Citation2001). Despite our findings that the PQML estimation results oftentimes (significantly) differ from those obtained by the OLS procedure, both estimation procedures largely lead to the qualitatively similar conclusions on the existence of the home market effect for the different types of goods with several exceptions.
Notes
1 The data set file has been downloaded from the homepage of Andrew Rose, http://faculty.haas.berkeley.edu/arose/
2 The reported estimation results for the log-linear model are slightly different from those reported in of Feenstra et al. (Citation2001), but they exactly match the corresponding results posted on the website of Andrew K. Rose in the STATA output file may1.log located in FMRout.zip. The difference can be traced back to the minor errors in the data set that have been corrected, as discussed therein.