The external wealth of nations: measures of foreign assets and liabilities for industrial and developing countries

For three reasons the authors state at the outset, each attaching to a different literature (§I). “First, the level of net foreign assets is a key state variable in many open-economy models of growth and business cycles, and a fundamental determinant of external sustainability. Second, many of the benefits of international financial integration are tied to gross holdings of foreign assets and liabilities, which are not captured by net flow data. Third, the composition of international investment positions between equity (portfolio and FDI) and debt may be important in understanding vulnerability to external shocks and the degree of international risk sharing.” The framing is that flows were never the binding constraint – “current capital flows are tracked on a daily basis” – but that “surprisingly little is known about the magnitudes of the stocks of foreign assets and liabilities held by various countries, especially in the developing world.”

67 industrial and developing countries over 1970-1998, with three explicit exclusions (§I, §III). The sample “does not include transition economies” and excludes “developing countries with per capita income below $1500 in 1985 (according to Summers and Heston) or population below 1 million” – so the coverage is deliberately not global, and the poorest countries are absent. Forty-five of the 67 are developing. The authors also flag that the industrial/developing split is “to some degree arbitrary: Singapore, for example, is classified as a developing country but, by the 1990s, had an income per capita higher than most industrial countries.” One consequence of incomplete coverage is visible in the aggregates: “[t]he fact that the aggregate NFA position for our sample is negative reflects, in addition to our incomplete country coverage, the global discrepancy in the measurement of the current account,” and indeed “the cumulative value of the world current account discrepancy … tracks very closely the aggregate external position of the countries in our sample.”

A decomposition starting from the current account and working towards stocks, which exposes every channel by which positions change without passing through the current account (§II). “The methodological contribution of the paper consists in the presentation of a simple accounting framework, starting from variations in the stock of foreign assets as measured by the current account balance. This framework highlights the link between balance of payments flows and the underlying stocks, as well as the impact of unrecorded capital flight, exchange rate fluctuations, debt reduction, and valuation changes not captured in a ‘crude’ cumulative current account.” The components are grouped into three categories – FDI, portfolio equity and debt – with exceptional financing (arrears, balance-of-payments loans, debt reduction operations) and IMF loans both treated “as sources of changes in the stock of external debt.” The authors also note a definitional fuzziness they cannot resolve: FDI covers a controlling stake “typically set at above 10%” with the remainder counted as portfolio equity, so “in certain cases the distinction between these two categories can de facto be blurred, but the issue cannot be clarified further in the absence of detailed disaggregated data.”

Because they rest on different assumptions about where the missing information lies, and their difference is diagnostic rather than noise (§II.A, §II.D, §III). ACUMCA cumulates the current account with adjustments “to reflect the impact of capital transfers, valuation changes, capital gains and losses on equity and FDI and debt reduction and forgiveness,” and then “estimates debt assets residually as the difference between ACUMCA and the sum of the stock of reserves, net FDI and equity stocks, plus external debt liabilities” – so it absorbs both cumulative errors and omissions and the gap between the World Bank debt stock and cumulated debt inflows. ACUMFL instead “calculates NFA for developing countries as the sum of the various stocks/cumulative flows, estimating debt assets as cumulative recorded debt outflows plus errors and omissions, and hence does not include the difference between the debt stock and the cumulative flow of debt liabilities.” IPNFA is the reported International Investment Position net of gold. The authors note the two constructed measures are equivalent in an accounting sense “but they may cease to be so once we make use of existing stock measures for some of the cumulative capital flows.”

Two different ways, matched to the two constructed measures (§II.A). “Even disregarding valuation effects, measuring net foreign assets of a country with cumulative flows requires some assumption about initial stock values.” For ACUMFL the initial 1970 position is built up component by component: existing stock measures for reserves and, for developing countries, external debt; for inward FDI in developing countries, “the stock of industrial countries’ FDI in developing countries in 1967 (OECD, 1972)” cumulated forward; for initial debt assets, “the value reported by Sinn (1990)”; and cumulative flows or national sources for the rest. For ACUMCA the initial value is “either an existing estimate of NFA (from Sinn (1990) or national sources), or the cumulative current account from the 1950s (with valuation adjustments).” The paper situates itself relative to two precursors: Sinn (1990), “by far the most comprehensive study undertaken on this subject” covering 145 countries but only 1970-87, thereby “missing the large increase in international capital flows and the changes in their composition that took place over the last decade,” and Rider (1994) for 1984-93, “focusing mostly on industrial countries.”

It removes a systematic overstatement of liabilities that the current account cannot see (§II.B). “If a country’s external debt is reduced because of debt forgiveness or because debt restructuring has reduced its face value, the cumulative current account will overstate the size of the country’s liabilities because the reduction in debt liabilities is not reflected in the current account balance.” The correction uses the World Bank’s Global Development Finance database, “which under ‘debt reduction and forgiveness’ reports the total amount of debt reduction, excluding debt-equity swaps, as well as debt forgiven.” The scale is illustrated by Chile: “[d]uring the period 1987-1990, Chile’s cumulative current account deficit was around US$2 billion. However, net external liabilities declined substantially because debt forgiveness and reduction operations reduced external debt by over $8 billion (over 25% of 1990 GDP).”

By marking cumulated dollar flows to market with different indices for assets and liabilities, which requires an explicit and strong portfolio assumption on the asset side (§II.C, Appendix A). “For equity liabilities, stocks are adjusted for changes in the end-year US$ value of the domestic stock market; for equity assets, stocks are adjusted analogously by a ‘world’ portfolio index, the Morgan Stanley Capital Index.” The asset-side assumption is stated plainly in the appendix: “we assume that all countries allocate their investment abroad in the same fashion, and that the composition of their portfolio reflects the Morgan Stanley Composite Index of world stock markets” – so cross-country differences in the geography of outward equity holdings are, by construction, not captured. Flows within the year are assumed to occur uniformly and are converted to end-of-year value by the ratio of the end-year market value to its average over the year. The validation offered is that “[s]tocks estimated with this method track the IIP stock measures for most countries more accurately than unadjusted cumulative flows.”

Because market-value data are unavailable for almost every country, and it matters a great deal – which the authors demonstrate rather than assert (§II.C, Appendix B). “Given that the difference between portfolio equity investment and FDI can be blurred in some cases, we would ideally want to estimate both stocks according to the same methodology. However, estimating the market value of FDI would require data which is unavailable for all but very few industrial economies, and in particular a breakdown between reinvested earnings and new direct investment flows.” The defence of book value is twofold: “[m]ost countries reporting IIP estimates of FDI stocks do so based on book value … and indeed our adjustment seems to track available stock measures of direct investment more accurately than other methods.” The cost is quantified with the US: in 1998 “the stock of US FDI abroad … increased by $119 bn at current cost and $356 bn at market value, with the underlying flow measuring $133 bn,” while “[o]ur estimate of the increase in the stock of US FDI abroad for 1998 is US$105 bn.” The book-value method itself assumes “that the relative price of capital goods across countries follows relative CPIs,” and deliberately omits the inflation-adjustment term because write-offs are not allowed for and nominal depreciation allowances mean “part of reinvested profits are simply offsetting real capital depreciation.” Historical-cost and replacement-cost variants are supplied in the dataset.

That they represent unrecorded capital flows – specifically residents’ debt assets held abroad – which is what separates the two constructed measures for several large developing countries (§II, §IV). The authors set out the fork explicitly: “[i]f it reflects unrecorded trade transactions, we should adjust the current account accordingly. If it reflects unrecorded financial account transactions, we should add it to capital flows.” They take the second branch “given the prevalence of capital flight in several developing countries for long periods of our sample,” with the further assumption that errors and omissions reflect changes in debt assets abroad. The consequence is visible in the results: “[i]n Argentina, Indonesia, Mexico and Venezuela net external liabilities measured with ACUMFL are significantly larger than ACUMCA, reflecting unrecorded capital outflows.” And for industrial countries the assumption receives indirect support: “the difference between the two estimates is strongly correlated with cumulative errors and omissions (0.75 for industrial countries).” Because the assumption is contestable, the dataset reports cumulative errors and omissions separately, “which can be used to obtain alternative estimates of the official NFA position.”

They disagree substantially for several developing countries, and the gap is interpreted as unrecorded capital outflows over and above errors and omissions (§II.D). Using World Bank external debt data instead of cumulated debt inflows changes the estimate, and “[i]n most cases, this difference is positive, and it is substantial for several developing countries, even after controlling for the impact of cross-currency fluctuations and debt forgiveness.” The Argentine example: “[d]uring 1977-81, cumulative debt inflows in Argentina measured US$7.9 bn, while the debt stock (net of the effect of currency fluctuations) increased by US$24.8 bn, a difference of over 20% of Argentina’s 1981 GDP.” The inference is stated conditionally, as it must be: “[a]ssuming that debt stocks are measured correctly, this discrepancy implies that the capital inflows reported in the balance of payments statistics underestimate actual inflows. If the current account and net flows are also measured correctly, changes in indebtedness can exceed the recorded flow of new external liabilities by an amount equivalent to unrecorded capital outflows.” So “the difference between the debt stock DWB and the cumulative sum of [debt liability flows] … plus cumulative errors and omissions give a measure of the stock of unrecorded assets held abroad by domestic residents,” connecting the exercise to the capital-flight literature.

On levels and, more demandingly, on year-to-year changes (§IV, Table 2). On levels: “ACUMCA gives a similar overall picture of trends in net foreign asset positions when compared to IPNFA, which is a direct estimate of the stock position. Nevertheless, there are some significant differences. For instance, ACUMCA is well below the Swiss IPNFA, while it substantially exceeds the Canadian position.” On short-run variability – the harder test, since positions “can fluctuate quite sharply on a year-on-year basis, due to valuation changes induced by exchange rate and asset market fluctuations, not reflected in the current account” – the paper compares the current account and the first difference of its own measure against the first difference of the official position. “For countries such as Germany, Italy and Spain, all correlations are high. For others (Australia, Netherlands, Switzerland, United Kingdom, US) the correlation between the current account and changes in IPNFA is low or even negative, but ACUMCA tracks changes in IPNFA much more closely.” That asymmetry is the paper’s strongest evidence that the valuation adjustments are doing real work rather than adding noise. For developing countries, correlations with the current account are “generally high, but significantly below unity for several countries, in particular for the ACUMFL measure, confirming the importance of valuation adjustments.”

Persistent creditors are few, persistent large debtors exist without apparent penalty, and most developing countries are debtors (§IV). Among industrial countries, “[r]elatively few countries have maintained positive net foreign asset positions throughout the 1970-98 period (Germany, Japan, Netherlands and Switzerland); the rest of the group are almost evenly split between persistent debtors and ‘switchers,’” the best-known switcher being the United States. The authors draw one inference from the persistence of large negative positions: “[t]he fact that some countries have maintained permanently negative NFA positions that are quite large (e.g. Canada, Australia, New Zealand) suggests open access to international credit for these countries over a sustained interval.” Among developing countries as of 1998, “[m]ost of the countries in our sample are debtors, the most notable exceptions being Botswana, Kuwait, Singapore and Taiwan,” while “[t]he countries with the largest net external liabilities in our sample are Cote d’Ivoire, Jamaica and (with the ACUMFL measure) Indonesia.” Regional patterns are described rather than modelled: many Latin American countries share “a sharp worsening during the 1982 debt crisis and an improvement starting in the late 1980s,” whereas “[t]here is more heterogeneity among Asian countries.”

A marked rise in gross equity and FDI stocks relative to GDP, later and steeper in developing countries on the liability side (§V, §VI). “The stock of FDI in relation to GDP [has] been relatively stable in industrial countries during the 1970s and the early 1980s, but has shown a substantial increase since then, a trend which is common across countries. A similar trend has occurred for the stock of equity capital, fuelled by both larger equity flows and increasing stock market valuations. The stocks of FDI and equity liabilities show a similar rapid increase in developing countries, especially since 1990, while the stock of FDI and equity assets increase more slowly.” The conclusion characterises this as “the increasing degree of equity diversification during the past decade, with rising gross stocks of equity and FDI in relation to GDP in both industrial and developing countries, but especially in the former.” As of end-1998 the highest FDI liabilities relative to GDP were in “Chile, Costa Rica, Jamaica, Malaysia, Panama, Singapore and Trinidad and Tobago,” while “Chile, Korea, Mexico, Singapore and South Africa have the largest portfolio equity liabilities.”

Development, openness and size all matter, but the regressions are offered as descriptive first cuts and several coefficients are read as ambiguous (§IV.C, §V, Tables 5-6). Using the average of the adjusted cumulative current account over the 1990s, with GDP per capita, trade openness (exports plus imports over GDP) and 1989 population from the Penn World Tables, “[t]he results show a generally positive relationship between net foreign assets and GDP per capita, in line with the ‘stages’ hypothesis,” though the weaker industrial-country estimate “suggests that the true relationship may be nonlinear.” Openness is “positive and significant for the full sample and the industrial countries” but weaker for developing ones, which the authors attribute either to collinearity with GDP per capita or to two offsetting mechanisms they had set out in advance: “vulnerability encourages open countries to accumulate foreign assets as a buffer stock in anticipation of external shocks while the positive impact on credit risk enables a more open country to borrow more overseas.” Size is “positive and significant across sub-samples.” On composition: richer and more open countries hold more FDI assets, but “there is a strong correlation between openness and FDI liabilities for developing countries, so that their net FDI position is negatively correlated with openness” – so “among the industrial nations, the relatively less developed are net recipients of FDI; among developing nations, it is the countries most open to international trade.” Openness and size raise gross equity assets and liabilities across subsamples, “smaller and poorer developing countries are found to have greater debt liabilities but openness is not significant,” and for the equity-to-debt ratio of developing countries’ liabilities, “[t]rade openness explains a high fraction of the cross-country variation in this ratio: in line with our theoretical priors, the mix of liabilities shifts from debt to equity in more open developing countries.” The authors are explicit about what these are not: “[t]he regression results in Tables 5-6 should be viewed as initial attempts to establish some basic stylized facts.”

It lists the specific weaknesses before stating the claim, and the claim is comparative rather than absolute (§VI). “Clearly, the data we constructed have ample margins for error. Our estimates of FDI are based on book values, while our equity estimates are adjusted to reflect market value. Estimates of the gross debt position for industrial countries are hampered by the lack of data comparable to the external debt statistics for developing countries, and are not adjusted for the impact of cross-currency fluctuations. Measures of debt assets for developing countries are subject to the caveats well known from the capital flight literature. Nevertheless, our estimates are constructed on a consistent basis across countries, they match existing stock data quite closely and they fill an important gap.” Note the first two sentences together: the paper’s own FDI and equity stocks are valued on different bases, which is a limitation on comparing the two components within a country’s balance sheet, not merely across countries.

Questions about how external positions affect macroeconomic behaviour, which they describe as unexplored (§VI). “[T]he data we have assembled can be used to address several interesting questions in international economics. The preliminary results discussed above are an initial step in investigating the determinants of countries’ external wealth. In addition, the impact of stocks of foreign assets and liabilities on macroeconomic behavior is an important question that has not been empirically explored. For instance, these data allow us to revisit the classic ’transfer problem’, investigating the long-run relation between real exchange rates and net foreign assets for a large set of countries.” The dataset itself was released alongside the paper, which is a substantial part of what the paper contributed.