Home country interest rates and international investment in U.S. bonds

That the empirical literature on how low interest rates change investors’ portfolios is thin, and what exists covers narrow investor sets, while the capital-flows literature has “largely relied on aggregate balance-of-payments data” (Sections 1-2, pp. 1-4). The authors place themselves at the intersection of two strands: the push/pull factors literature on capital flows (Forbes and Warnock 2012; Fratzscher 2012; Broner et al. 2013), and the newer literature on portfolio risk-taking when rates are low. They list the existing micro evidence as covering corporate bond mutual funds (Choi and Kronlund), U.S. money market funds (di Maggio and Kacperczyk), German insurance companies (Domanski, Shin and Sushko), and – their own companion paper – private foreign investors’ holdings of U.S. corporate bonds (Ammer et al. 2018). Their claimed advance is combining “long time series of portfolio holdings with a cross-section of several dozen investor countries that exhibit significant heterogeneity in the dynamics of their home interest rates” (p. 1). The companion paper showed reach-for-yield within the U.S. corporate bond portfolio (toward lower-rated, longer-dated securities); this paper adds the aggregate flow and the corporate-versus-Treasury allocation.

Annual TIC surveys of foreign holdings of U.S. securities for 31 countries, 2003-2016, at end-June each year, aggregated from confidential security-level records to country-by-bond-type, restricted to private holdings and measured at face value (Section 3, pp. 4-6). Three design choices carry weight. The private/official split matters most for Treasuries, “since those constitute a large share of foreign official reserves and could thus be driven by different motives” (p. 4). Face value is used “in order to isolate the effect of active new investments and portfolio shifts, thus abstracting from the effect of price changes” (p. 5), which means the year-over-year change in holdings can be read as an inflow or outflow. And the reporting structure – U.S.-resident custodians, mandatory reporting of all U.S. securities held for foreign residents – makes the coverage comprehensive at country level, at the cost of a known limitation: “the TIC data are reported on a resident basis rather than on the basis of the ultimate owner,” so intermediaries in custodian countries and financial centres hold securities on behalf of investors elsewhere (p. 5). Home interest rates are Bloomberg local-currency sovereign yields at 1-year and 5-year maturities, computed as June averages to align with the end-June holdings date.

A large share of their foreign bond holdings but a small share of their total bond holdings – and, from the U.S. side, each country is small (Section 4.3, pp. 7-8). Drawing on the IMF’s Coordinated Portfolio Investment Survey, as of end-2015 U.S. bonds were just under 30 percent of the sample countries’ holdings of foreign bonds on average, but only just under 7 percent of their total bond portfolios once home holdings are included – the home-bias wedge. For some countries the foreign share is dominated by the United States: “for Mexico and Canada, the CPIS data show that the shares of overseas investment allocated to the United States are some 94 percent and 68 percent, respectively,” which the authors offer as a reason the U.S. slice may be representative of those countries’ international behaviour (p. 8). Running the other way, “the average country from our sample holds just 0.7 percent of the outstanding U.S. bonds,” a fact the authors later lean on when arguing that these flows are unlikely to have moved U.S. issuers’ financing conditions (p. 8).

Corporate holdings collapsed in the crisis and then rebounded strongly as foreign rates fell after 2012, lifting the corporate share of the foreign U.S. bond portfolio by about 10 percentage points to roughly 60 percent by the end of the sample (Section 4.2, pp. 6-7). Foreign holdings of U.S. corporate debt rose sharply into the global financial crisis, fell during the crisis and the euro sovereign debt crisis “reflecting the ‘flight home’ and search-for-safety during that period that is documented in the literature,” then recovered as foreign rates declined – “suggesting that investors from these countries were compensating for declining returns on safe assets at home by purchasing U.S. corporate debt” (p. 7). Treasury holdings were stable pre-crisis, fell during it, and resumed afterward; the authors note that late in the sample the increase in corporate holdings outpaced that of Treasuries. On the identifying variation, Section 4.1 stresses that the interquartile range of sovereign yields across countries “remains substantial throughout, even as the median approached zero toward the end of the period,” with Japan low throughout, many European rates falling only late, and most emerging markets never reaching very low levels (p. 6).

A country-year panel of U.S. bond holdings scaled by home GDP, regressed on the home 5-year sovereign yield plus country financial and macroeconomic conditions and links to the United States, with country and time fixed effects, estimated by weighted least squares (Section 5, pp. 8-11). The controls are the country’s share in total U.S. trade, an analogous bank-exposure measure of financial linkage, the home 5-year sovereign CDS premium (to strip out risk-driven moves in the home yield), the standard deviation of the bilateral exchange rate against the dollar, the change in that exchange rate, and expected corporate earnings growth from IBES as a proxy for domestic investment opportunities. Weights are countries’ holdings of U.S. corporate and Treasury bonds. Time fixed effects do real work here: they mean “changes in the overall U.S. and global economic and financial environment, including changes in the U.S. safe interest rate, are already accounted for” (p. 9), so the estimates are driven by relative movements in home rates. The trade-off the authors accept is that country fixed effects absorb the standard time-invariant gravity variables – distance, common language – so those cannot be examined. Crisis timing enters as an interaction of the home yield with a 2008-2012 dummy; a separate specification drops the crisis years entirely and interacts the yield with a 2013-2016 post-crisis dummy.

They are mostly excluded, and the two that remain are given a composite euro-area yield rather than their own (Section 5, pp. 9-10; Section 6.1, p. 12). Caribbean banking centres are dropped both because they lack meaningful sovereign debt and therefore reliable sovereign rates, and because “their investments are predominantly held on behalf of non-residents, for whom the interest rate to use is ambiguous” given the paper’s focus on home investment opportunities. Luxembourg drops out of the main specifications for lack of data on all controls. Belgium and Ireland are retained but assigned “the average sovereign rates of four larger euro zone countries: Netherlands, France, Italy, and Spain” – Germany deliberately excluded because its rate “might reflect Germany’s safe haven status, rather than investment opportunities in Germany” (fn. 17, p. 9). This turns out to matter for magnitudes: when Belgium and Ireland are further excluded, the corporate-bond result survives “but the coefficient does decline considerably, suggesting that investment from the excluded countries is especially sensitive to the composite European interest rate we assigned them” (p. 12). For Treasuries the authors present only the sample excluding financial centres, because Treasury Department reports note that “some foreign official holdings likely are misclassified as private holdings because they are held through private intermediaries,” and such misclassified official money is disproportionately in Treasuries (p. 12).

A coefficient of -0.036 in the baseline, meaning a home rate 100 basis points lower goes with U.S. corporate bond holdings 3.6 percent of GDP higher, rising to as much as 5.3 percent of GDP in the specification that drops the crisis years (Section 6.1, pp. 11-12; Conclusion, p. 16). The sign is negative and statistically significant, which the authors read as “consistent with a ‘push factor’ that would induce flows in the context of a portfolio balance framework” (pp. 11-12). Two non-results are reported alongside: the interaction with the 2008-2012 crisis dummy is not statistically significant for corporate bonds, so the relationship is not detectably different during the crisis; and the post-crisis interaction is also insignificant, so despite rates being unusually low in most advanced economies after 2012, “we do not find that in the post-crises period … the relation is much different” (p. 12).

Much weaker, and with opposite signs in the crisis and post-crisis periods (Section 6.1, pp. 12-13). Over the full sample the home-yield coefficient for Treasuries is not significant, which the authors attribute to offsetting period effects. The crisis interaction is positive and significant – “indicating a decline in Treasury holdings for foreign investors with declining home interest rates,” which they read as flight-home behaviour during crises. The post-crisis interaction is strongly negative, indicating that after the crisis, with rates low across advanced economies, foreign investment in Treasuries rose, “likely suggesting that investors seeking safe assets were drawn to the relatively higher yielding U.S. Treasuries, compared to home sovereign bonds.” The authors treat this too as search-for-yield, just operating within the safe-asset class. The magnitude reported in the introduction is an order smaller than the corporate effect: a 100 basis point drop in the home rate is associated with a rise in Treasury investment of 0.2 percent of GDP, and only in the post-crisis period.

Mostly little, with two interpretable exceptions (Section 6.1, pp. 13-14). Stronger bank lending ties to the United States are associated with more investment in U.S. corporate bonds in one specification, but the effect disappears once financial centres are dropped. The share in U.S. trade is insignificant for corporate bonds and, counterintuitively, negative for Treasuries – though the authors report that redefining trade as bilateral trade over the investor country’s own GDP, rather than as a share of total U.S. trade, makes the coefficient positive but insignificant for both. Higher expected domestic corporate earnings growth reduces investment in U.S. Treasuries, which fits the yield-seeking story: when domestic risky prospects improve, the safe foreign asset is less attractive. The authors offer a candid reason the rest come up empty: “there is no sufficient time variation and the country fixed effects we include in these regressions fully capture the cross-country variation” (p. 13).

It isolates the portfolio tilt from the overall pull of U.S. bonds, and it is where the search-for-yield reading is strongest (Section 6.2, pp. 13-14). The dependent variable is a country’s U.S. corporate bond holdings as a share of its total U.S. corporate-plus-Treasury holdings on a given survey date, which “implicitly controls for the general incentives to invest in U.S. bonds.” The home-yield coefficient is negative – statistically stronger once financial centres are excluded – and implies that “a 100 basis points decline in the home interest rate entails an estimated shift toward corporates of about 2.3 - 2.7 percentage points” (p. 14). The crisis interaction is positive and significant, so “the effects may be smaller and not even present in crises period,” while the post-crisis effects do not differ significantly from pre-crisis. The expected-earnings-growth coefficient flips sign relative to the Treasury level regression, “consistent with Treasuries being a defensive investment when prospects for riskier investments are weaker.”

On the unhedged home rate; the hedged synthetic dollar yield is statistically insignificant in every specification (Section 7.2, pp. 15-16). The authors construct the synthetic dollar yield as the home-currency 1-year sovereign yield minus the 12-month U.S. dollar forward premium – the dollar return an investor would realize by hedging a home sovereign bond into dollars – and enter it both alone and alongside the unhedged rate. Alone, it “does not appear to affect investment in U.S. corporate bonds in a statistically significant way.” Entered together, the hedged rate remains insignificant while the unhedged coefficient matches its earlier magnitude, and the same holds for Treasuries. The authors give two readings, both stated as interpretations rather than tests: this is “consistent with institutional incentives not to hedge exchange rate exposure,” and “consistent with investors being aware of the limited predictive power of forward premiums for future exchange rates,” citing the forward premium puzzle in the tradition of Fama (1984). Their summary line is deliberately behavioural: “investors do not appear to take hedging costs into account nor expect uncovered interest rate parity to hold. Rather, they appear to compare nominal promised rates of return among investment choices” (p. 16).

Robust across sample cuts, interest-rate maturities, control sets, and an autocorrelation correction, with magnitudes varying as expected (Section 7.1, pp. 14-15; Section 5, p. 10). Dropping the country controls entirely leaves the sign and significance intact on a substantially larger sample, though the coefficient is larger – an effect the authors attribute specifically to Luxembourg re-entering. Excluding observations where the country’s sovereign CDS spread is in the top fifth percentile, which guards against the home yield picking up distress rather than opportunity, leaves coefficients “very similar to the ones reported in baseline Table 3.” Substituting the 1-year for the 5-year home sovereign yield preserves the results “with slightly smaller estimated effects.” Adding the lagged dependent variable produces the expected positive and significant coefficient without altering the main results, and adding banking-sector CDS premia leaves results unchanged at the cost of a much smaller sample. Including corporate asset-backed securities in the corporate category does not change the results either.

It argues that reverse causality is implausible in both direction and magnitude, but the defence is an argument about the setting rather than an identification strategy (Conclusion, p. 16). Two points are made. On direction: an increase in the supply of U.S. bonds “would likely draw in more cross-border investment,” but it “would tend to raise rates” abroad, while the paper’s finding is that more U.S. investment goes with lower investor-country rates – so the confound would work against the observed sign. On magnitude: because each sample country’s holdings are small from the U.S. perspective (0.7 percent of outstanding U.S. bonds on average), “they are unlikely to have affected the financing conditions of the issuers.” The abstract’s verbs (“low(er) interest rates … lead to greater investment”) are stronger than the regression tables, which report associations under country and time fixed effects; readers should hold the weaker reading the results sections themselves use.

Whether these investors’ total portfolios actually became riskier (Conclusion, pp. 16-17). Because comparable data on domestic portfolios do not exist alongside the TIC holdings, “we cannot say whether the investor behavior we observe is the same or differs from the domestic investment patterns. It could be that these investors invest more aggressively abroad and more conservatively at home, and as such their overall portfolio need not be more risky” (p. 17). The authors set out the alternative conjecture – that these investors “likely have made risk-increasing shifts elsewhere in their portfolios that could pose financial stability risks abroad, particularly if the low-rate environment persists” – explicitly as a conjecture. What they do claim as established is narrower and directional: “our findings suggest that there are spillover effects from low interest rates through cross-border capital flows,” and the results “are also consistent with central banks’ balance sheet policies having a significant effect on demand for foreign financial assets” (p. 17).