stressed that these results must be interpreted with care. In particular, they are likely to be affected
significantly by the chosen method for measuring structural reforms.
An issue of robustness arises: to what extent are the results concerning the budgetary impact of
labour and product market reforms driven by the specific way chosen to construct the reform
indicators? Results do not seem to crucially depend upon the chosen benchmark for the change in
structural indexes to define when structural reforms occur, i.e., the median value of positive
changes. In fact, by using the mean value instead, one obtains reform indicators that are still
correlated with those based on the median (with rank correlations equal to 0.85 in the case of
labour market reforms and 0.88 for product market reforms). The estimation of fiscal reaction
functions using reform indicators based on the mean of positive changes in the structural index
yields results similar to those presented in table 7 (see table 8). We have also performed a further
robustness check on the labour market reform indicator used in our analysis. We have constructed
two alternative indicators based on policies actually implemented. Using the FRDB database on
reforms concerning employment protection legislation and unemployment subsidies, two
indicators have been constructed. One indicator takes value 1 in years/countries where reforms in
employment legislation which overall reduce firing costs have been introduced. The other
indicator takes value 1 in case of reforms in unemployment subsidies that improve incentives to
labour market participation. In spite of the fact that the correlation of these new labour market
indicators with the one used in our analysis is rather low, the employment protection reform
indicator so obtained performs in a quite similar way in the estimation of fiscal reaction functions
as compared with our benchmark indicator for labour market reforms (see table 9).42
Table 10 presents the results from the estimations of fiscal reaction functions in the case of
pension reforms. This time, fiscal reaction functions are augmented by a pension reform dummy
that takes value 1 if a pension reform was implemented in the current or previous year. The
analysis in this case refers separately to the determinants of the primary CAB, cyclically-adjusted
revenues and primary expenditures. Results show that the pension reform dummy has a negative
but non-significant impact on primary CABs. The coefficient indicates that a reform implemented
in the current or previous year reduced the value of the primary CAB by about 0.2 GDP points.
However, given the high uncertainty surrounding this estimate (a high standard error of the
42 The rank correlation between the labour market indicator used in our analysis and that based on reforms in
employment protection is 0.02, while the correlation with that with the indicator based on reforms in unemployment
benefits is -0.01.
112