16
apply OLS to equation (13) and the within-group estimator is BLUE (Best Linear Unbiased Estimator)
(Hsiao, 1986, p. 32).
Next, we turn to the estimation results. The sample consists of 11 EU countries for which data are available:
Austria, Belgium, Denmark, Finland, France, Germany, Italy, Netherlands, Spain, Sweden and the United
Kingdom. The intention was to use publicly available data sources (OECD and IMF) (see Appendix B).
The sample contains 216 observations (231 less 11 due to the lagged dependent variable less 4 due to the
lack of forward exchange rate data for Italy). The results are reported in Table 2.
Regressions A and B explain closed interest differentials by their fundamental determinants xi,t, not including
the lagged dependent variable:
(i-i^)ijt = ζi÷β2xijt÷ei,i (18)
In regression A all determinants except the lagged dependent variable are included. In regression B we
apply the general-to-specific approach (see Charemza and Deadman, 1992). The specification started with
the inclusion of all determinants except the lagged dependent variable, finally arriving at the significant
regressors only.33
We attribute remaining differences from closed interest parity to country risk premia imposed by the
international financial community on particular countries. The generalized country-specific effect γi reflects
the effects of expectations that controls might be tightened or eased in the future, or the effect of controls
not currently binding might become a constraint in the future, or the effect of expectations of new controls.
The generalized country-specific effect may also include the risk as assessed by the market that the policy
will not maintained (peso problems) and country-specific risk that cannot be explained by its fundamental
determinants (news). In addition, these country premia also capture the other determinants which cannot
be included in the right-hand side of the regression because they are time-invariant such as the index of
exchange rate flexibility (EXR), the Eijffinger-Schaling index of central bank independence (ES), the proxy
for the political leaning of the government (LEFT) and the measure for political instability (SIGGOV).
Because these variables are (more or less) constant over time, the coefficients of these variables cannot
be identified in a panel data approach. Moreover, since both EXR and ES are indirectly measured by the
depreciation of the exchange rate (DEP) and the rate of inflation (INF) respectively, the country-specific
effect is mainly attributed to political risk variables such as LEFT and SIGGOV. Furthermore, possible
other sources of political conflict such as high general government gross debt ratios (DEBT) and current
account imbalances (CA) are already included in the right-hand side of the regression.
The first question, then, is to address whether the signs of the individual coefficients conform to theoretical
expectations (see discussion in section 3). By now, considerable agreement exists across studies (Alesina,
Grilli and Milesi-Ferretti, 1994 and Milesi-Ferretti, 1995) that the realized inflation rate (INF) and related
attributes such as the inflation tax (INF TAX) and seigniorage revenue (SEIGN) account for an important
part of closed nominal interest rate differentials (compare the magnitude of the estimated coefficients in
regression A). Generally, high inflation rates are associated with more capital export restrictions. Realized
depreciations of the exchange rate (DEP) are positively associated with the intensity of capital export controls.
Both INF and DEP encourage capital outflows, so more capital export restrictions are expected to be in
place. Strong evidence is found for the intensity capital controls to depend negatively on the size of general
government deficit (DEF).
33
We apply a joint F-test of zero restrictions on the coefficient of the deleted variables. We omit the insignificant variables.