results obtained, those parameter values that are significant (at the 15% level or better)
are weighted by their shares in total (real) gross output to obtain an overall estimate
for the UK economy (Table 5). Two sets of results are presented, with the second
omitting the retail and wholesale sectors due to the generally atypical results these
large sectors have. Overall, the second set of results in Table 5 show that there is a
fairly substantial post-entry productivity effect for firms that are new to exporting (e.g.
based on the IV model, a 34% long-run increase in TFP in the year of entry, and only
a small effect of around 5% in the year following entry); firms exiting overseas
markets overall experience negative productivity effects in the year they exit and
subsequently (around 7-8% on average for the economy); while firms that enter and
then exit experience large productivity gains whilst exporting (some 19% in the year
of entry, but with a 5% decline the following year).
Our results differ in both approach and outcome to others for the UK. Besides
weighting our data to ensure it is representative of the population of firms, and having
a more extensive dataset (in terms of the number of observations and industries
covered), we also use a dynamic GMM systems approach to directly estimating TFP
within a production function model that attempts to control for both sample selection
and endogeneity. Girma et. al. (2004) used unweighted matched data and a difference-
in-differences approach32, but TFP is obtained using a growth accounting model and
thus there is no direct estimation of an economic model where causality can be
consistently dealt with. Also constraining the underlying production function to
exhibit constant returns-to-scale is likely to further bias any estimates of the
exporting-productivity relationship, as the exporting variable(s) in the model have to
32 Hence, their dependent variable is the growth of output (∆ln Yit), or productivity, depending on the
different specifications they use. Such a model cannot provide an estimate of the long-run impact of
exporting on productivity levels, as long-run impacts by definition are omitted. This is not a trivial
issue, as Equation (5) used here encompasses both short- and long-run impacts.
30