Sie sind auf Seite 1von 1

4088 Ann Harrison and Andrés Rodríguez-Clare

importance of air and sea distance across countries to vary over time. In the first stage of
his estimation, Feyrer allows the coefficient on air and sea distance to be time-varying
in explaining bilateral trade. He then uses those time varying coefficients to con-
struct time and country-specific instruments for the first stage. Feyrer shows that
trade has important, robust effects on growth using a panel of countries for 1960
through 1995. Donaldson (2009) uses historical data for 1861 through 1930 across
Indian districts to estimate the impact of the introduction of railroads on trade costs,
income volatility, price convergence, and economic growth. He traces out the
mechanisms through which railroads affected income and shows that the reduction
of trade costs through the introduction of railroads had large and important effects on
economic growth.
(3) Which variables to include in the set of controls Z.
The third major area of controversy in this literature is which variables to include in
the set of controls Z. There is a growing literature which claims that two key omitted
variables from the Z vector are institutions and geography. Indeed a recent literature
has sought to distinguish between institutions, economic geography, and trade as
sources of economic growth, including Alcalá and Ciccone (2004), Easterly and Levine
(2003), and Rodrik et al. (2004). Only Alcala and Ciccone find that openness matters;
the other two studies find that “institutions rule.” There are several reasons why Alcala
and Ciccone get very different results from Rodrik et al. First, Alcala and Ciccone use
real trade shares while Rodrik et al. use nominal trade shares as their measure of
openness. Second, Alcala and Ciccone improve upon the Frankel and Romer measure
by expanding the first stage and using more countries, improving the first-stage F and
reducing the fragility of the instrument.
None of these three studies, which have been extensively cited in the empirical
literature on the determinants of growth, uses trade policy as a measure of openness.
Easterly and Levine (2003) use the Sachs and Warner (1995) and Dollar (1992) measures
to proxy for openness; the flaws of these two measures are discussed above. Rodrik et al.
use the average of nominal trade shares for 1950-1998 as their openness measure. All
three papers focus on a pure cross section of countries. As pointed out by Harrison
(1996), trade policies and trade shares have changed too much over the last 40 years to
make long-run averages very meaningful.
Given the problems inherent in the openness measures, and the reliance on pure
cross-sectional estimation, it is not surprising that openness is trumped by institutions
in two of these three studies. This research also highlights the tremendous problems
associated with measuring institutions in a way which is distinct from trade policy.
The correlation between the openness and institutions measures in Easterly and
Levine (2003) is 0.68, which suggests that multicollinearity is likely to be a significant
problem. Both Rodrik et al. and Alcala and Ciccone use the Kaufmann, Kraay, and