13
subsection). The country fixed effects also have the merit of soaking up unobserved variables that
may lead to permanent differences in measured net foreign asset positions across countries.16 To
control for common global movements, in particular of world GDP per capita, demographics and
public debt, we also include time dummies in all the regressions.
As a precursor to the regression analysis, we explored the univariate time series properties of
the data. We tested for nonstationarity in our series for net foreign assets, demographic variables,
government debt and log GDP per capita using the NPT1.1 econometric package--see Chiang and
Kao (2000). The tests were performed separately on the industrial and the developing country
samples, using the panel unit root test of Hadri (2000) (allowing for fixed effects and no time trend).
For all series in the four samples, the test rejects the null hypothesis of stationarity.17 In light of the
evidence on the presence of unit roots, we subsequently tested for panel cointegration among our
variables using tests suggested by Kao (1999) and Pedroni (1999). Both are residual-based tests for
which the null hypothesis is lack of cointegration (nonstationarity of residuals). These test statistics
are reported in Table 1 and strongly suggest the existence of a cointegrating relation among net
foreign assets and our fundamentals.
Having ascertained that the variables display a common trend, we follow Stock and Watson
(1993) and estimate their long-run relation using a dynamic ordinary least squares (DOLS [-1,1])
specification.18 We report estimates for the 1970-98 and 1980-98 intervals. The data set is more
complete for the post-1980s period and, in addition, this latter period may better reflect an
environment of open capital accounts.19
With respect to the specification, we want to allow the entire age structure to influence the net
16 This may capture both country-specific determinants of net foreign asset positions and permanent
measurement errors in our estimates of national net foreign asset positions.
17 Other panel unit root tests gave broadly similar results. The unit root test results are available from the
authors.
18 A DOLS[-2,2] specification gave similar results. Only leads and lags of output growth and changes in public
debt are included (including changes in demographic variables makes no difference). Standard errors are
corrected for heteroskedasticity.
19 In future work, we plan to explicitly look at measures of capital account liberalization.