The analysis is motivated by the following considerations. If domestic entry regulation
is strong, interactions with foreign countries should be weak. In other words, we would
expect that the interaction pmr × b has a positive sign and the interaction pmr × b* a
negative one. If foreign regulation pmr* is high, domestic firms can rely very little on
foreign demand. Hence, whenever b goes up, they have to bear most of the induced reduc-
tion in demand themselves; we therefore expect that the effect of the interaction pmr * × b
on domestic unemployment is positive. However, domestic unemployment would depend
less on foreign distortions since the foreign economy plays a smaller role for domestic
firms. Therefore, the coefficient on pmr* × b* should be negative. Column (6) in Table
3 tests these predictions in a model with all four potential interaction terms. Interaction
terms with domestic regulation come out with the right sign while those for foreign regu-
lation do not. Column (7) focuses on domestic regulation and the respective interaction
terms. They are statistically significant and show up with the right signs: the more closed
the domestic economy is, the more important are domestic institutions and the less the
foreign ones. This is in line with our theory. Column (8) concentrates on foreign regula-
tion. Interestingly, the more closed the foreign economy is the stronger are the domestic
unemployment-creating effects of foreign labor market distortions.
4.6 Robustness checks
Tables A3 and A4 in the Appendix contain a number of robustness checks on our preferred
specifications. Columns (1) to (6) in Table A3 refer to regressions that include the foreign
unemployment rate on the right-hand-side; columns (7) and (8) use foreign exogenous
variables. The regressions in columns (1) and (2) use the log of unemployment lnu as
the dependent variable instead of the level in a regression of domestic unemployment on
foreign one, but are otherwise perfectly similar to the regressions (4) and (5) presented in
Table 1. Compared to the benchmark case where the level of u is used, this transformation
ensures that the dependent variable takes values on the entire real line. There is no clear
consensus in the empirical cross-country unemployment literature as to whether ln u or u
is to be preferred. In the case of our regressions, the log specification has the drawback
that our IV strategy does not work well here; see the overidentification test associated to
the regression in column (2). However, qualitatively, our main result holds up in the OLS
and the IV model.
Column (3) in Table A3 reverts to the level of the unemployment rate as the dependent
variable and uses contemporaneous instruments rather than the lagged ones. This does
not change the qualitative findings relative to the benchmark of Table 1. Column (4)
presents an OLS model, using ui*,t-1 as the dependent variable. Results change very little.
Columns (5) and (6) add an EU dummy to the regressions, but otherwise leaves the
regressions identical to those in Table 1. Inclusion of the dummy does not change the
results of the OLS and the IV model. Comfortingly, the EU dummy is only marginally
significant in the OLS model and insignificant in the IV specification, so that the accession
to the EU does not have any effects on the unemployment rate other than those already
42