The baseline model (model 1) has an

The baseline model (model 1) has an R-squared of 0.52,
indicating that 52% of the variation in the dependent variable
is explained in the empirical model. The test statistics for heteroskedasticity
and autocorrelation indicated their presence in
the model, so we used the Newey–West estimation procedure
to compute robust HAC standard errors. 8 The coefficient of
transfer dependency is statistically significant at the 5% level;
a one percentage point increase in the ratio of central transfers
to provincial expenditure lowers the share of sub-provincial
expenditure in total expenditure by 0.164 percentage point.
In other words, an average province with 50.69% in transfer
dependency will have an 8.31 percentage point drop in expenditure
decentralization, a substantial 11.82% drop from the
average ratio of 70.63%. Economic development, or real
GDP per capita, has a coefficient of 6.421, but it is not statistically
significant at the 10% level. Population density is statistically
significant at the 10% level, and a 1% increase in
population density leads to a drop of 0.320 percentage point
in the dependent variable. A one percentage point increase
in the share of FDI in GDP is associated with a 0.279 percentage
point drop in the dependent variable, and the coefficient is
significant at the 5% level. The share of trade in GDP has a
positive coefficient, but it is not statistically significant.
In model 2, we added two control variables – the share of
state-owned enterprise (SOE) employment in the total population
and the share of secondary industry output in the total
GDP. The inclusion of the two additional variables does not
significantly change our empirical results.
In models 3 and 4, the economic development variable is
treated as endogenous, 9 and the results are similar to those
in models 1 and 2, having no flip of sign in coefficient estimates
or a substantial change in the level of significance. Comparing
the results in models 2 and 4, the coefficient on transfer dependency
changes from 0.156 to 0.169. The coefficients of population
density and share of FDI in GDP drop from 0.316
and 0.270 to 0.230 and 0.203, respectively. The coeffi-
cients of other variables are not statistically significant at the
10% level. 10
The linear models reported above do not ensure that the expected
value of the dependent variable lies on the unit interval.
We used pooled QMLE to estimate two fractional probit models;
one model (model 5) assumes that all explanatory variables
are strictly exogenous, and the other (model 6) allows
the economic development variable to be endogenous. In both
models, we used 1,000 bootstrap replications to obtain bootstrap
standard errors, which are robust to general heteroskedasticity
and autocorrelation. The transfer dependency
variable remains statistically highly significant in both models.
Its APE estimates, 0.158 in model 5 and 0.159 in model 6,
are very similar in magnitude to the estimates of fixed effects
and 2SLS models. So the three sets of estimates tell a consistent
story: transfer dependency has a negative and statistically
significant effect on fiscal decentralization.
5. DIS