regression coefficient) it cannot be judged to be significantly different from zero. By carrying out
the same analysis using a dependent variable the cyclically-adjusted government revenues and
primary expenditures one notices that most of the deterioration of the primary CAB in the
aftermath of pension reforms is associated with a reduction in revenues rather than with increased
expenditures. Again, the impact on revenues is however not statistically significant. It is to notice
that, although the analysis does not permit to distinguish whether the budgetary impact of pension
reforms is a director an indirect one, interpreting the impact of pension reforms on government
revenues as being mostly direct does not seem consistent with the evidence illustrated in table 5:
on average, social security contributions do not change significantly in the aftermath of pension
reforms.
Finally, alternative specifications of fiscal reaction functions have been estimated with the purpose
of analysing the impact on budget deficits arising from the interaction between different types of
reforms. In table 11, specifications (1) - (3) include each one of the reform dummies in isolation
but impose a sample size such that there are no missing observations for any of the reform
dummies. This permits to compare these results with those in specification (4), which include all
reform dummies simultaneously, and with those in specifications (5) - (8) which instead include
dummies taking value 1 when, respectively, are carried out: labour market and product market
reforms; labour market and pension reforms; product market and pension reforms; all three type of
reforms. Results show that when the sample is restricted to year/country combinations for which
there are no missing observations for any reform dummy the impact of labour market and product
market reforms taken in isolation appears less negative (and less statistically significant) while that
of pension reforms more negative and statistically different from zero (compare table 11 with
tables 7 and 10). By including all reform dummies in the specification, the dummy for product
market reforms turns slightly positive, becoming largely statistically insignificant, whereas the
coefficient for labour market reforms and pension reforms becomes more negative compared with
the case in which they are included separately in the regressions (specifications (1) and (3) in table
11). The lack of robustness of reform regression coefficients signals a possible problem of
multicollinearity which affects especially the coefficient of the product market reform dummy.
This seems consistent with the evidence from correlation analysis, which reveal a quite significant
correlation across the sample between product market reforms and pension reforms. Does it matter
if different types of reforms occur at the same time? Comparing results from specifications (5) -
(8) with those in specification (4) in table 11 could help answering this question. In all cases, the
113