gains; rather, productivity only improves significantly when firms are serving
advanced, high-wage export markets. Lastly and most importantly, there are also
crucial methodological issues involved when testing for a productivity effect from
exporting. A problem usually encountered in microeconometric evaluation studies is
sample selectivity. This arises when making comparisons between a ‘treatment group’
(e.g. export-market entrants) and the rest of the population, when it is suspected that
the treatment group are not randomly drawn from the whole population. This issue is
of paramount importance when interpreting the results obtained from comparing
exporters and non-exporters, upon which policy conclusions are then based (see
Section IV for more details; also Blundell et. al., 2005, for a recent overview).9,10
Several standard approaches have been proposed in the literature to combat this
selection problem, such as ‘matching’ techniques to select a valid ‘control’ group to
compare the performance of exporters with only those non-exporters with similar
characteristics (Girma et. al., 2004); and the difference-in-difference estimator to
eliminate selectivity bias (Greenaway and Kneller, 2004).
III. Data and Descriptive Statistics
The FAME dataset is used for this study, which includes all firms operating in the UK
that are required to make a return to Companies House. It contains basic information
on firm-specific characteristics, such as turnover, intermediate expenditure,
employment, assets, and most importantly, overseas sales. Apart from financial
9 See Section IV for more details; also Blundell et. al. (2005) for a recent overview.
10 Another potential econometric problem may arise since most empirical studies tend to pool
information across all firms with heterogeneous export histories to examine these learning effects of
exporting. In fact, distinct learning effects are uncovered amid firms of different ages (Kraay, 1999;
Delgado et. al., 2002; Baldwin and Gu, 2003; Greenaway and Yu, 2004).
10