11
log form), the stock of public debt as a ratio of GDP and the shares of population under 14, over 65
and between 15 and 64 (in 5-year cohorts).12
Public debt is defined as the sum of external public debt, net of foreign exchange reserves,
and gross domestic public debt.13 For industrial countries, the main source of data for public debt is
the OECD (general government definition); for developing countries, the data have been constructed
using the World Bank’s Global Development Finance, the IMF’s Government Financial Statistics and
national sources. Unfortunately the definition of government for developing countries is not
homogeneous—it can refer to central government, general government or nonfinancial public sector.
When data availability was not a constraint, we have used the broadest definition of government. A
data Appendix detailing sources and definitions for the debt data is available from the authors.
Finally, population shares were constructed using the United Nations’ Demographic
Yearbook (Historical Supplement 1948-1997), supplemented by data from Herbertsson and Zoega
(1999).14
3.3.1 Bivariate Relations As a precursor to the multivariate econometric work, we begin in Figures
6-8 by showing the bivariate relations between net foreign asset positions on the one side and output
per capita, public debt and demographic structure on the other. In these graphs, the data are measured
in terms of average changes between the 1980-89 and 1990-98, capturing the medium- or long-term
movement in country positions.15 In each figure, Panels A and B contain observations from the
industrial and developing countries respectively.
Panel A of Figure 4 shows a quite striking positive bivariate relation between growth in
12 Ideally, we would like to measure net foreign assets relative to a country’s total wealth but this would require
data on land values, natural resources, human capital and the value of domestic assets. In any event, it is
plausible that GDP may serve as a reasonable proxy for wealth.
13 We would of course prefer to use net domestic public debt, but data availability for such a measure is much
more limited. Since we focus on time series behavior, and given the strong co-movement between the two
measures for those countries for which they are both available, we are confident that this choice still allows us
to capture the right long-run relation. As we will discuss later, obstacles are more serious when undertaking
cross-sectional analysis because of cross-country differences in the definitions of “government.”
14 We thank these authors for kindly sharing their data.
15 This “cross-section in first differences” is essentially a country fixed-effects specification, picking up intra-
country time variation. We get similar graphs if we also employ data from the 1970s but the more recent period
offers more complete data and may better capture behavior under integrated capital markets.