GALSTYAN AND LANE
namic OLS (DOLS) estimator to establish the long-run relation between the explanatory
variables and the real exchange rate and the relative price of nontradables. In general,
our empirical specification is based on equation (14).
The specification for DOLS(-1,1) is
j=1
yit = αi + θt + β0xit + X ∆xit-j + it (17)
j=-1
where yit is the dependent variable and x is the set of explanatory variables.17 The β
coefficients in equation (17) capture the long-run impact on y of long-run changes in
the x variables. We are interested in explaining the long-run behavior of: (a) the real
exchange rate; and (b) the relative price of nontradables.18
For the real exchange rate, we plausibly assume that the productivity differential be-
tween the traded and nontraded sectors is increasing in the level of GDP per capita.19
Accordingly, we expect a positive long-run relation between the relative level of GDP
per capita (measured relative to trading partners) and the real exchange rate. In relation
to the positive impact of international investment income on the real exchange rate, we
exploit the steady-state relation that the trade balance surplus should equal the inter-
national investment income deficit (TB = -rB) and thereby include the trade balance
surplus as a regressor (the expected sign is negative).20 Finally, government consumption
and government investment (both measured relative to trading partners) are included,
where the model suggests that an increase in government consumption should be asso-
ciated with real appreciation but the sign on government investment is not tied down.21
The specification is quite similar for the relative price of nontradables, with the exception
that GDP per capita and the fiscal variables are measured in absolute terms, rather than
relative to trading partners.22
17The choice of one lead and lag is dictated by the sample length.
18Although these two variables are isomorphic in the simplified model that was presented in Section 2,
we recognize that various factors can generate substantial differences in the empirical behavior of these
variables.
19Since government investment operates via a productivity channel, we do not want to directly include
a productivity measure in the regression specification. Rather, we include GDP per capita as a general
correlate of the level of productivity.
20See also Lane and Milesi-Ferretti (2002) and Galstyan (2007). An alternative is to include the level of net
foreign assets (Lane and Milesi-Ferretti 2004 and Ricci et al 2008). However, the long-run relation between
net foreign assets and the trade balance depends on the composition of the international balance sheet and
the levels of returns earned on foreign assets and liabilities. Accordingly, the long-run trade balance is a
better indicator of the long-run level of net international investment income.
21Although the model is written in terms of the level of public capital Z, we assume a close correspon-
dence between the long-run level of public investment and the long-run level of public capital.
22The dependent variable is the relative price of nontradables, for which the value of domestic fiscal