Flatliners: Ideology and Rational Learning in the Diffusion of the Flat Tax

verted distance) matrix; WYi,t-1 is the spatial lag, which gives a weighted sum of Yj,t-1
for each Y_i,_t-1; ρ is the spatial-autoregressive coefficient; nY_t-1 is the ‘diffusion’ lag,
where n is the number of ‘flat’ countries in a given year; γ is the temporal-diffusion coef-
ficient; X is a matrix of observations on the independent variables and β is the vector of
its coefficients; e is the residual vector. The temporal-autoregressive coefficient α shifts
depending on the value of Yi,t-1 (1 or 0). We assume that all variables X have the same
independent effect on the dependent variable irrespective of the value of Yi,t-1 , but α
captures the difference between effects arising from the transition from 0 to 1 (flat tax
adoption) and those from 1 to 1 (flat tax persistence), by α. Thereby we can distinguish
between the impact of diffusion (adoption) and attraction, the probability of flat tax
adoption given its persistence.

We construct W, a standardized inverted distance matrix, to specify diffusion via
distance, with zeros along the diagonal and elements Wij reflecting the degree of con-
nectedness from country j to i.³⁰ Due to the diversity of flat tax countries, we forsake
alternative specifications of this matrix on the basis of EU membership status or regions
and construct a distance matrix instead.

We estimate the spatial lag with no substantive weights, and we also weigh WY_i,t-1
by a variable that we expect to register the attractiveness of ‘flat’ countries to ‘non-flat’
countries: competition for foreign direct investment (measured as the ratio of a country’s
FDI to that of its immediate region). We estimate separately several models: without
spatial effects, with spatial effects, and with spatial effects weighted by FDI (all including
³⁰Each value in the matrix represents the distance between all capital cities in Eastern Europe. The
values are inverted so that neighboring countries register higher values of Wij and each row is standardized
by dividing each cell in a row by that row’s sum, as common in the literature. Since we have 20 countries
in the dataset, the NxN matrix is 20x20. To compare and discuss short-run spatial effects in neighboring
countries (Table 3), we also specify a border-contiguity matrix, with bordering countries coded as 1, 0
otherwise, and row-standardized as above. These two specifications produce very similar results.

20

<<   21   22   23   24   25   26   27   >>

More intriguing information