Many reforms do not occur in a single year but rather are spread over several consecutive
years, in which case the indicator takes the value 1 throughout the whole period. In principle, this
implies a risk that major, but gradual, reforms weigh too heavily in the data sample as compared
with major, one-off reforms. As well, such gradual reforms might be implemented in years and
under economic conditions that are significantly different from when they were decided, thereby
possibly making it more difficult to identify the drivers of reform. In practice, however, protracted
reform processes are rare in the policy areas and for the countries covered here.
The key explanatory variables considered in the regressions are the level of and the change
in the fiscal balance. These and the control variables are defined as follows:
- Situation of public finances: cyclically-adjusted general government fiscal balance as a share of
GDP (Source: OECD, OECD Economic Outlook 76, December 2004). 93
- Change in the situation of public finances: first difference of the cyclically-adjusted general
government primary fiscal balance as a share of GDP (Source: OECD, OECD Economic
Outlook 76, December 2004).
- Initial labour market performance: unemployment rate of 15-64 year-olds (Source: OECD,
OECD Employment Outlook, June 2004).
- Economic crisis situation: dummy variable which takes value 1 when actual GDP is at least
4 percentage points below potential and 0 otherwise (Source: OECD, OECD Economic Outlook
76, December 2004).
- Small open economy: dummy variable which takes value 1 for the following countries: Austria,
Belgium, Denmark, Finland, Greece, Ireland, Netherlands, New Zealand, Norway, Portugal,
Sweden, Switzerland and 0 otherwise. These are the smallest countries in the sample in terms
of GDP size.94
4.3 Probit regression results
The five policy reform indicators can be used in two alternative ways. One option is to
merge them so as to obtain an aggregate policy reform indicator which takes value 1 for those
pairs (country, year) that correspond to major reform in at least one policy area and 0 otherwise.
Alternatively, the individual policy reform indicators can be stacked up in order to expand
dramatically the number of observations available for the econometric estimates. One problem
with this option is that it rests implicitly on the assumption that structural reforms undertaken in a
given pair (country, year) in different fields are independent from one another. However, this can
93. The cyclically-adjusted budget balance is conceptually closer to the political and financial capital available to the government than the
actual budget balance. The same holds for the change in the cyclically-adjusted budget balance, which should in principle be more representative
of the political capital spent in consolidation of public finances than the change in the actual budget balance. In addition, fiscal variables that are
calculated in cyclically-adjusted terms are much less likely to be correlated with other explanatory variables such as the unemployment rate or
the economic crisis dummy. This contributes to minimise multicollinearity risks. One important difference between the regressions below and
those presented in Duval and Elmeskov (2005) is that both the level of and the change in the fiscal balance are calculated in cyclically-adjusted
terms here -while actual values were used for the level of the budget balance in Duval and Elmeskov (2005).
94.Setting a cut-off point is not straightforward, considering for instance that the economic size of the smallest “large” country, Australia, exceeds
only slightly that of the largest “small” country, the Netherlands (the gap between the two countries’ GDP in PPPs was about 15% in 2000).
193