THE COMPOSITION OF GOVERNMENT SPENDING AND THE REAL EXCHANGE RATE
constructed using relative labor productivities from the same KLEMS database. The real
exchange rate is constructed using the consumer price index and the nominal period-
average exchange rate from the International Monetary Fund’s International Financial
Statistics database, with the trade weights calculated by Bayoumi et al (2005). GDP per
capita (measured in constant 2000 US dollars), is taken from the World Bank’s World
Development Indicators database. The balance on goods and services (in current USD)
is from International Monetary Fund’s International Financial Statistics database. The
current-dollar value of GDP is from International Monetary Fund’s World Economic
Outlook database. Government consumption and government investment in current
US dollars are taken from the OECD.
It is important to measure variables in relative terms in order to capture the forces
driving the structure of international relative prices. Accordingly, variables for each
country are expressed relative to a weighted-average across the set of OECD trading
partners in the database.15
3.2. Empirical Evidence
Next, we turn to more formal econometric evidence. Table 1 shows the statistics from
panel unit root and cointegration tests. The tests do not reject the null of a unit root for
the real exchange rate and the relative price of nontradables. In addition, the tests indi-
cate non-stationarity for several of the explanatory variables (the productivity differen-
tial, relative GDP per capita and the trade balance). The evidence for the fiscal variables
is mixed: while the PP-Fisher unit root test indicates non-stationarity, the other tests re-
ject the unit root null. However, it is clear that the fiscal variables are highly persistent (a
panel estimate of an AR(1) specification gives an estimated coefficient of 0.9), such that
it is pragmatic to also treat the fiscal variables as non-stationary.16 The Kao cointegra-
tion test clearly rejects the null of no cointegration among the variables for both the real
exchange rate and relative price of nontradables equations.
Since non-stationarity is pervasive in the set of variables, we focus on the panel dy-
‘Community Social and Personal Services’. In this case the aggregate price index is the weighted average of
sectoral price indices, where the weights are the value shares in the base year.
15For most members of the euro area (excluding Germany and Greece), the trade share with the set of 19
OECD countries is more than 70 percent. Among the other countries, Canada, Denmark, Norway, Sweden
and the United Kingdom have a trade share of more than 70 percent. Only the United States and Japan have
approximately 50 percent share, while Australia has a share of 60 percent. Accordingly, using the 19 OECD
countries in the database to form trade-weighted averages is more or less representative.
16We also ran the dynamic OLS estimator on artificial data with three non-stationary variables and one
persistent but stationary variable. Our simulation suggests that dynamic OLS is a good estimator for the
scenario.