In (4) the coefficient α3 measures the impact of cross-border regulatory interactions and the β-
coefficients the influence of the further control variables. By construction, α1=k*REG*i,t-1 and
is expected to be positive, whereas α2= - k would be negative if the status quo bias really loses
power due to learning from deregulation. From this reduced form, the implicit “optimum” level of
regulation equals α1∕(-α2).
The two baseline specifications include following control variables among the x: GDP growth, the
extent of FDI flows as a measure of globalization constraints57 (sum of absolute values of in- and
outflows in % GDP), an EU dummy, dummies for governments led by centre and left parties
(compared to governments led by right parties).58 Besides these control variables a second
baseline specification includes business and consumer confidence indicators, which come at the
cost of a substantial loss in the number of observations and which, therefore, will not be used in
the subsequent consecutive inclusion of fiscal variables. The benchmark variable is constructed as
the difference of the regulatory situation vis-à-vis the EU average for European countries and vis-
à-vis the USA for non-European OECD countries.
Any dynamic fixed effect estimation faces econometric problems due to the correlation of the
error term with the “within” transformed regressors and a resulting bias. The frequently applied
solution with small T is the application of GMM techniques along the lines of Arellano and Bond
(1991). However, this solution has been shown to come at the costs of imprecise estimates
(Attanasio et al., 2000). Since the time dimension of the regressions is relatively long (the
specifications include up to 26 years) the trade-off hints towards the application of standard OLS
estimates. Endogeneity problems related to a possible impact of reforms on some of the control
variables such as growth or the confidence indicators are limited by the use of lagged right hand
variables.
The baseline regression reveal marked differences in reform dynamics among different policy
fields. For product and financial market reforms both variants of the baseline robustly point to a
stable implied optimum level of deregulation being close to unity (which is defined as the fully
liberalized status). In contrast to that, no stable optimum level emerges for labour and tax reforms.
The same structure applies to the cross-border interaction term: Whereas for product and financial
57 In Heinemann (2005), the FDI variable appeared to be a more important measure of globalization constraints of regulatory policies compared to
other frequently used variables such as trade openness.
58 For data sources, see Table A-1 in the appendix.
151