(labour) productivity29 in period t - 1 are significantly more likely to sell overseas in
period t, although the strength of this relationship varies across industry sub-groups.
Generally, the impact of productivity on the probability of exporting is larger in the
manufacturing sector.
Firms with non-zero intangible assets are also generally much more likely to export,
and this again points to a need to invest in highly productive resources that lead to a
greater ability to internalise external knowledge in order to overcome barriers to
exporting (the absorptive capacity argument). The average effect across all the
industry sub-groups suggests that having intangible assets increases the likelihood of
exporting by some 7%; however, in some industries (e.g. food, beverages and tobacco;
metals; and other manufacturing) the impact is much larger (around 19% on average),
while in some industries there is no significant effect (covering repair/sale of motor
vehicles; transport services; post/telecoms; and finance) or even a negative impact (in
agriculture; and retail, hotels and restaurants). It is also worth noting that the age of
the firm in t-1 is usually found to be a major determinant of exporting, supporting
process-based incremental models of internationalisation (cf. Johanson and Vahlne,
1977).
Thus, in line with the majority of previous studies, we also find that there was strong
self-selection by UK firms during 1996-2004, in all of the 16 industry sub-groups
examined. Turning to our results from estimating Equation (5) in order to test whether
there is also a ‘learning-by-exporting’ effect associated with post-entry sales to
overseas markets, we have employed the three approaches discussed above. As stated
above, we include ln Ageit and Intang it as part of the instrument set when estimating
the production function, since we find that these variables are not significant
29 Labour productivity (rather than TFP) is used in estimating the probability of exporting as we need
the results from the probit model (i.e. the selectivity terms) when estimating the ‘control function’
production function model.
21