THE COMPOSITION OF GOVERNMENT SPENDING AND THE REAL EXCHANGE RATE
Before turning to the panel DOLS estimates that focus on the time series dimension,
we show graphically the relation between changes in government spending and changes
in the real exchange rate along the cross-sectional dimension. Figure 1 shows the relation
between the cross-country relation between changes in the fiscal variables and changes
in the real exchange rate over 1980-2004.23 The real exchange rate is positively correlated
with government consumption (the correlation is 0.52), while it is negatively correlated
with government investment (the correlation is -0.26). Accordingly, the cross-country
data supports the basic thrust of the theoretical model, by highlighting the real exchange
rate is differentially sensitive to government consumption and government investment.
The DOLS results for the real exchange rate are presented in Table 2. We run the
regressions for a variety of country groups. In addition to the full sample of OECD coun-
tries, we also differentiate between larger and smaller economies by running separate
panel regressions for the G3 and non-G3 samples.24 We also consider the sample of EMU
member countries. This group is especially interesting, since real exchange rates for these
countries have been much less volatile since the formation of European Monetary Union
in 1999 - the elimination of bilateral nominal exchange rate shocks among the member
countries may make it easier to capture the long-run patterns that are the focus of this
paper. In turn, we allow for differences in country size within the EMU bloc by run-
ning separate panel regressions for the E4 group of the four largest countries (France,
Germany, Italy and Spain) and the group of smaller member countries (non-E4 group).
Column (1) shows the estimates for the full sample. In relation to the fiscal variables,
an 1 percentage point increase in relative government consumption is associated with a
1.2 percent appreciation in the real exchange rate and this estimate is significant at the 1
percent level. In contrast, relative government investment has a negative sign but is not
significant. Accordingly, the general pattern is in line with predictions of the theoretical
model, in that the composition of government spending matters: an increase in govern-
ment consumption appreciates the real exchange rate in the long-run, but government
investment does not. The control variables also are highly significant in the expected
variable is important. If the dependent variable were the difference in the relative price of nontradables
acrosss countries, then the fiscal variables should be entered in relative terms. We have run this alternative
specification - these results are available upon request.
23More precisely, Figure 1 plots the residual of the change in the real exchange rate against the residuals
of the changes in government investment and government consumption, after each of these variables has
been regressed on the set of controls: changes in the trade balance, relative GDP per capita and the omitted
fiscal variable. Accordingly, the scatter plot captures the conditional comovement between these variables.
24The G3 is defined as Germany, Japan and the United States. See Lane and Milesi-Ferretti (2002, 2004)
and Galstyan (2007) on the reasons why the sensitivity of relative prices to fundamentals should systemically
vary with country size. In the simplified model in Section 2, the share of nontradables in consumption (γ)
serves as a proxy for country size, in that we may expect larger economies to have relatively larger nontraded
sectors.