highly significant, supporting the view that major reforms are typically “bunched” into packages
and/or that policy changes in certain areas can pave the way for reforms in others.
4.4 Linear regression results
Insofar as the reform process is inherently a binary one, the probit approach adopted above
should be appropriate. However, in practice, distinguishing “reforms” from “small policy changes
going in the right direction” necessarily involves some degree of subjectivity, which the
methodology adopted here to identify major reforms could only at best minimise. Moreover, the
probit approach does not allow studying the dynamics of the reform process, and may even
unduly reward slow movers as opposed to overnight reformers, even though this problem is
probably limited in the context of this paper. For instance, an interesting issue which can not
easily be tackled within the probit framework is whether the propensity to undertake reform is
greater the stricter the initial policy stance in the field considered.
Against this backdrop, the following linear equation is estimated on the aggregate dataset
as a robustness check for previous findings:
Δ(X )=αX 1 +∑βkZk +δ +ε ,
it it -1 k it i it
k
X
where it is the overall policy stance in country i in period t, calculated as the sum of normalised
-so as to allow comparisons of policy stances across fields99- OECD policy indicators in each of
the five fields covered (unemployment benefits, taxes, EPL, retirement incentives, PMR). In other
Xt, =∑ (Yj-Y)/stdev(Yj) Y
words, j , where it is the OECD policy indicator in field j (e.g. the
αX j
measure of unemployment benefits or the strictness of EPL). The term it-1 captures the lagged
impact of the policy stance on the change in the policy stance, i.e. on the propensity to undertake
reform. A negative coefficient would indicate that reforms -measured here as a decline in the
overall policy stance- are likely to be greater the more there is room for reform, i.e. the stricter the
initial policy stance. The country fixed effect δi captures in a crude way possible cross-country
differences in policy preferences. TheZit ’s denote the set of explanatory variables used
previously, including the level and the change of the cyclically-adjusted fiscal balance, but
excluding the country size dummy which is now captured by the country fixed effects. The crisis
variable is refined and now denotes the number of crisis years over the past three years.
99.It acknowledged, however, that there is no straightforward way to normalise the various policy stance indicators. The choice of the normalisation
method could have some influence on the results. For instance, here, dividing by the standard deviation of policy stances implies that the weight
of a given policy reform in the estimation is inversely related to the cross-country variation in policy stances in the field considered, ceteris
paribus.
196