captured by the institutional variables in the model.43
Finally, columns (7) and (8) carry out two robustness checks relative to Table 2.
First, instead of using the level of the unemployment rate, its logarithm is used. This has
no effect on the signs of the right-hand-side variables nor on their statistical significance.
Second, the EU dummy is introduced into column (8), with the level of the unemployment
rate again as the dependent variable. Again, results do not change qualitatively. Here,
however, we do have some evidence that joining the EU does drive up the equilibrium
rate of unemployment.
Table A4 carries out extensive robustness checks with respect to the choice of nor-
malization of the weights. Columns (1) to (6) vary the elasticities of country size α1
and α2 , as well as that of distance in the bilateral trade flow proxies shown in equation
(29). Odd-numbered columns refer to equation (5) in Table 1, even-numbered columns to
equation (5) in Table 2. First, the coefficient of the distance variable in the computation
of the weights is set to the lower bound of estimates found in the gravity literature, i.e.,
δ = 0.75 (Disdier and Head, 2008). Then a higher bound, i.e., δ = 1.50, is used. Quali-
tatively, these modifications have little effect on the estimates. To achieve a quantitative
comparison, we need to take into account that the sample moments of u* depend on the
weights. The standardized β coefficient of u* in column (1) is 0.096 × 2.47/4.29 = 0.055
and in column (3) 0.065 × 4.63/4.29 = 0.070. which nicely bounds the benchmark results
obtained in Table 1 (column (5), 0.06) from above and from below. Using a finer grid on
δ shows that the obtained standardized beta coefficients systematically fall in δ. In the
limit, when δ is infinite, the effect of u* vanishes. A similar effect is observed in columns
(2) and (4), where the variable of interest is b* instead of u* .
Columns (5) and (6) modify the weights in that they close down the direct size effect:
α1 = α2 = 0. In both models, the signs of the interesting coefficients remain unchanged.
In terms of their economic significance, the standardized beta coefficients are 0.006 ×
36.07/4.29 = 0.050 in the case of column (5) and 0.052 × 8.07/4.29 = 0.098 in the case of
column (6). Hence, taking out country size from the construction of the bilateral weights
reduces the estimated effects, but only in a very limited amount.
The remainder of Table A4 modifies the normalization of bilateral weights. In columns
(7) and (8) the weights are normalized by maxi (ωijt) , for each year t and country j, while
in columns (9) and (10) the weights are normalized such that they add up to one for the
20 OECD countries that our panel regressions draw upon. These different normalizations
do not have any bearing on the qualitative results. The standardized beta coefficient of
u* in column (9) is 0.006 × 32.33/4.29 = 0.045, which is in line with our previous findings.
43 Note that the effect of EU membership is identified using time-variation only since the model features
country fixed-effects.
43