Home Country Interest Rates and International Investment in U.S. Bonds

K.7 Home Country Interest Rates and International Investment in U.S. Bonds Ammer, John, Stijn Claessens, Alexandra Tabova, and Caleb Wroblewski Please cite paper as: Ammer, John, Stijn Claessens, Alexandra Tabova, and Caleb Wroblewski (2018). Home Country Interest Rates and International Investment in U.S. Bonds. International Finance Discussion Papers 1231. https://doi.org/10.17016/IFDP.2018.1231 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1231 June 2018

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1231 June 2018 Home Country Interest Rates and International Investment in U.S. Bonds John Ammer, Stijn Claessens, Alexandra Tabova, and Caleb Wroblewski NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Home Country Interest Rates and International Investment in U.S. Bonds John Ammera, Stijn Claessensb, Alexandra Tabovaa, Caleb Wroblewskia June 19, 2018 Abstract Weanalyzehowinterestratesaffectcross-borderportfolioinvestments. Dataon U.S.bondholdingsbyforeigninvestorsfrom31countriesfortheperiod2003-2016 and a large variety in movements in interest rates in these countries provide for a unique way to analyze shifts in investment behavior in response to interest rates. Wefindthatlow(er)interestrates,nowprevailinginmanyadvancedcountries,lead to greater investment in general into the United States, with the effects generally driven by investment in (higher yielding) corporate bonds, rather than in Treasury bonds. In addition to affecting overall investments, lower interest rates at home are associated with a greater weight on corporate bonds, consistent with searchfor-yield. The results are economically important and robust to controlling for a number of country-specific macroeconomic and financial conditions as well as to sample restrictions and choices of interest rate. Our findings have important policy implications in that they suggest that low interest rates can lead to shifts in the volume and composition of overseas investments. JEL Classification: F21, F34, G11, G20 Keywords: low interest rates, search-for-yield, portfolio choice, safe and risky assets, U.S. bonds aBoard of Governors of the Federal Reserve System, Washington, DC 20551, USA bBank for International Settlements, 4051 Basel, Switzerland We thank Joshua Aizenman, Pierre-Olivier Gourinchas, Galina Hale, Zheng Liu, Mark Spiegel, Colin Weiss, and participants at the 2017 Asia Economic Policy Conference for helpful comments. We thankViktorsStebunovsandWenxinDuforhelpwithsomeofthedata. Correspondingauthor: alexandra.m.tabova@frb.gov. Wroblewski is currently at the University of Chicago Booth School of Business, most of the work was completed while he was at the Board of Governors. The views in this paper are solelytheresponsibilityoftheauthorsandshouldnotbeinterpretedasreflectingtheviewsoftheBoard of Governors of the Federal Reserve System or any other person associated with the Federal Reserve System or the Bank for International Settlements.

1 Introduction Using data on foreign private investment in U.S. bonds from 31 countries for the period 2003-2016, the paper studies how portfolio investment is affected by investors’ home country macroeconomic and financial conditions. In particular, we explore how home investment opportunities, proxied by the home country sovereign yield, affect bond investment into the United States in general, as well as the composition of this investment in terms of riskier (corporate bonds) and safer (Treasuries) securities. We find that, in response to a lower interest rate at home, foreign investors increase their aggregate bond investment in the United States, and they also increase risk-taking in their U.S. portfolios through an increased weight on corporate bonds, consistent with search-for-yield. Our work relates to two strands of the literature: that on the push and pull drivers of capital flows; and that on the role of country characteristics in international portfolio allocation and the related, but more recent work on portfolio risk-taking in a low interest rate environment. In this paper we expand on these literatures in several ways. First, we use data on foreign countries’ holdings of U.S. bonds that distinguish between private and official investors’ portfolios. This allows us to focus on portfolio shifts by private investors in response to domestic macroeconomic and financial conditions. The distinction is useful because the motivations of official investors (e.g., central bank reserve managers) for holding U.S. securities likely differ from those of private investors. Second, our empirical identification is strengthened because we are able to combine 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 and other financial and macroeconomic conditions. Thus, these data allow us to study how investor-country conditions interact with their investment choices. Related, we contribute to the limited empirical work on the effects of interest rates on the composition of investors’ debt securities portfolios (Domanski, Shin, and Sushko (2017), Choi and Kronlund (2017), di Maggio and Kacperczyk (2017), Ammer et al. (2018)).1 The paper complements Ammer et al. (2018), who using security-level data, find evidence of search-for-yield behavior within foreign investors’ portfolio of U.S. corporate bonds in that when investing in U.S. corporate bonds, investors facing declining home investment opportunities prefer higher yielding (but riskier) securities. Here, we show that both aggregate flows and the allocation between corporate bonds and safer Treasuries are also affected by home country interest rates. We explore these advantages in the data by comparing the drivers of investment in the two largest classes of U.S. debt securities: corporate and Treasury bonds. This allows us 1Most of the empirical literature on risk-taking related to interest rates have focused on the effects on either bank lending and bank loan portfolios or mutual fund flows to broad asset categories. 1

toinvestigatebothflight-for-safetyandsearch-for-yield. Thedataonportfolioinvestment in U.S. bonds is derived from the detailed data underlying the Treasury International Capital (TIC) annual surveys. While much of our focus is on home country investment opportunities, proxied mainly by the home country sovereign yield, we also analyze the role of other country macroeconomic and financial conditions. We also control for certain “gravity” type characteristics for investment, such as countries’ trade and financial links with the United States. We find that the lower the interest rate in the investor’s home country, the more investors increase their investments in the United States as a ratio to their home GDP, with the effects generally coming through investment in U.S. corporate sector bonds, rather than in Treasury bonds. These regression results are consistent with international capital flows responding to relative investment opportunities, as well as shifts in portfolio composition reflecting search-for-yield motives. Importantly, regression results using hedged and unhedged sovereign rates show that the incentives to invest in both the more risky U.S. corporate bonds and in Treasuries depend on the (nominal) home, not on the equivalent hedged dollar interest rate. This finding suggests that investors’ incentives lead them to place more weight on the unhedged local rate as a measure against which to compare the gross return on a U.S. dollar debt investment. Put differently, investors do not appear to take hedging costs into account. Rather, they appear to compare nominal promised rates of return among investment choices. The effects are economically important. We estimate that when a country’s home interest rate is 100 basis-points lower, its investment in U.S. corporate bonds rises by 3.6 to 5.3 percent of GDP. The effects for investment in Treasuries are much smaller and only evident in the post-crises period: an equivalent drop in the home interest rate is associated with a rise in investment of 0.2 percent of GDP. Analyzing further the portfolio allocation within a country’s U.S. bond portfolio, specificallytheshareofcorporatebonds, wefindfurtherevidencesuggestiveofsearch-foryield in terms of foreign investors taking on relatively more credit risk. More specifically, alowerhomeinterestrategenerallyincreasesportfolioweightsforcorporatebondswithin countries’ portfolio of U.S. bonds, although we find less evidence of this during periods of financial crises, when investors shift more toward Treasuries. The results are robust to different choices for the domestic interest rate with which we proxy home investment opportunities as well as to country sample choices. Results are also robust to other controls related to the dollar exchange rate and home investment opportunities. Overall, our findings suggest that foreign investors’ U.S. bond portfolios gravitate toward corporate securities, as opposed to the safe Treasury bonds, when their home interest rates reach low levels. Although, apart from investment in U.S. assets, we do 2

not know how private investors in these countries allocate their overseas investments, our finding that a lower interest rate at home increases U.S. portfolio debt investment disproportionately in corporate bonds, suggests that countries’ external investments may rebalance toward riskier assets when their domestic interest rates are low. Thepaperproceedsasfollows. Insection2, wereviewtherelatedliterature. Insection 3, weprovideanoverviewofthesecuritiesholdingsdatasetandotherdatasourcesweuse. Section 4 presents some stylized facts and summary statistics. In section 5 we outline our empirical methodology and in section 6 we present the empirical results on how countries’ U.S. bond portfolios vary in response to changes in home interest rates and other country conditions and characteristics. Section 7 includes robustness tests. Section 8 concludes and discusses possible policy implications. 2 Related Literature Our work adds to two main strands in the literature on capital flows. The first is the literature on push and pull factors, which has explored the role of source and destination countryconditionsforcapitalflows(amongothersForbesandWarnock(2012), Fratzscher (2012), and Broner et al. (2013)). Among the push factors, an important one has been (low) interest rates, especially their effect on capital inflows to emerging markets, and morerecentlytheeffectsoftheuseofunconventionalmonetarypolicybyseveraladvanced countries.2 Otherpapershaveanalyzedhowbanksgloballyreallocateloansinresponseto changes in interest rates (e.g., Aramonte, Lee, and Stebunovs, 2015; Morais, Peydro, and Ruiz (2017)). Much of this literature, however, has largely relied on aggregate balance-ofpayments data to assess international portfolio composition and capital flows.3 Research using more granular data on investment choices typically has been limited to a narrower set of investors for which data are available. And rarely have studies covered a broad cross-section of investor countries. In addition, our paper is related to the growing literature on search-for-yield. A number of papers have pointed out that there could be a search-for-yield effect for institutions with long-term liabilities and shorter-term assets, such as life insurance companies and pension funds.4 Incentives to reach for yield among asset managers could be greater at low levels of the interest rate (Rajan (2010) and Stein (2013)). 2See, e.g., Chari, Dilts Stedman, and Lundblad (2017); Fratzscher, Lo Duca, and Straub (2016, 2017); Ahmed and Zlate (2014); and Bowman, Londono, and Sapriza (2015). 3See also Cerutti et al. (2018) and Rey (2013). Using loan level data, Baskaya et al. (2018) link improving external financial conditions to capital inflows, increased local bank credit, and lower loan rates. 4Rajan (2005), Dell’Ariccia and Marquez (2013), Domanski, Shin, and Sushko (2017). 3

The empirical literature on the effect of low interest rates on investors’ portfolio holdings is scarce (Choi and Kronlund (2015) focus on corporate bond mutual funds, di Maggio and Kacperczyk (2017) on U.S. money market funds, Domanski, Shin and Sushko(2017)onGermaninsurancecompanies; Ammeretal. (2018)onprivateinvestors’ holdings of U.S. corporate bonds).5 Our paper is also related to Ammer et al. (2018) which focuses on the risk distribution specifically of the portfolio of U.S. corporate bonds held by foreign investors to study reach-for-yield behavior within this asset class. The authors use security level TIC data for U.S. corporate bond holdings by foreign investors, which are the underlying data for the countries’ aggregate holdings of U.S. corporate bonds that we use in the current paper. They find that declines in safe interest rates push international investors toward lower-rated and longer-dated securities within their portfolios of U.S. corporate bonds to increase yield, which is consistent with a search-foryield behavior. The paper also relates to the literature on the international (bilateral) allocation of securities, using aggregate data, typically the IMF Coordinated Portfolio Investment Survey (CPIS) but more recently also the newly available data on euro-area security-level holdings (Portes and Rey (2005), Boermans and Vermeulen (2016)). But these papers do not investigate the role of time-varying country conditions, including (low) interest rates. 3 Data We use the annual U.S. Treasury International Capital (TIC) surveys of foreign holdings of U.S. securities for the period 2003 - 2016. Data are (confidentially) reported at the security level for each country holder of that security as of end-June of each year, and for the analysis in this paper we aggregate the holdings to the country and bond type level.6 This means that for each year in the period 2003 - 2016 we observe the total holdings per country of each bond type. We focus on the two main classes of U.S. bonds: Treasuries and corporate bonds. Importantly for our analysis, the detailed nature of the data allow us to distinguish between private and official investors’ holdings of U.S. bonds. The paper studies the holdings of foreign private investors only, because the motivations of official investors (e.g., central bank reserve managers) for holding U.S. securities may differ from those of private investors. This distinction between private and official investors’ portfolios of U.S. bonds is particularly important when we analyze 5To assess a search-for-yield behavior, Hau and Lai (2016) focus on equity and money market fund flows. Neely (2015) and Koijen et al. (2017) analyze the portfolio rebalancing channel of the Federal Reserve’s quantitative easing and of the ECB asset purchase program, respectively. 6Ammer et al. (2018) use the security-level TIC survey data to study investors’ choices specifically within their corporate bond portfolios. 4

foreign holdings of Treasury bonds since those constitute a large share of foreign official reserves and could thus be driven by different motives. The other advantage of the TIC surveys is that the data include both the face and market value of holdings. In order to isolate the effect of active new investments and portfolio shifts, we use in our analysis the face value of holdings, thus abstracting from the effect of price changes.7 For the TIC surveys the main reporters are U.S.-resident custodians which must reportallU.S.securitiestheyholdonbehalfofforeignresidentsandreportingismandatory. Due to the mandatory reporting of holdings by custodians, the data are comprehensive, capturing countries’ entire portfolios of U.S. securities at the country level. Country-level holdings data are published on the Treasury Department’s website, although without the split between holdings by private investors and holdings by official institutions. Because the TIC data are reported on a resident basis rather than on the basis of the ultimate owner, this creates some data challenges because intermediaries in major custodian countries and financial centers hold securities on behalf of investors from other countries.8 Anotherkeycomponentofthedatawedeployisinvestors’homeinterestrates. Weuse data on foreign countries’ local currency sovereign yields at 1-year and 5-year maturities. The underlying data are from Bloomberg, and the annual rates are calculated as average yields for the month of June of each year so the data are aligned with the holdings data that are reported as of end-June of each year. Since the sovereign interest rate is our key variable for evaluating country-level incentives for risk-taking, the sample excludes bonds held by investors in Caribbean and other financial centers for which we do not observe sovereign interest rates. In robustness checks we also include the U.S. dollar equivalent of the home sovereign rate, which we construct using Bloomberg data on 12-month forward premiums for the U.S. dollar against the investor countries’ home currencies and calculate the synthetic dollar yields foreign investors would obtain if they hedged their home-currency 1-year sovereign bonds into the U.S. dollar. For the other investor-country characteristics we draw on a variety of data sources. WeusedatafromtheIMF’sDirectionofTradeStatistics(DOTS)forimportsandexports betweenthe investorcountryandthe United Statesbased on thenotion that the intensity of trade is a good proxy for economic and other ties as well as the degree of information asymmetry between the investor country and the United States (Portes and Rey, 2005; Aviat and Coeurdacier, 2007; Okawa and Van Wincoop, 2012). To take into account countries’ financial linkages to the United States we include in our specifications the share 7Estimates of monthly positions can be constructed (see Bertaut and Judson, 2014), and have been used recently by Chari et al. (2017) to study the effect of U.S. monetary policy on emerging market asset returns and capital inflows. We use the annual data to study adjustments to (low) interest rates over longer periods, and because the annual survey data are more precise. 8SeeBertaut,Griever,Tryon(2006)formoredetailsabouttheTICdataanddatacollectionprocess. 5

of U.S. dollar bank claims and liabilities for each foreign country relative to total bank claims and liabilities, drawing on a different component of the TIC data. Countries’ exchange rates versus the U.S. dollar may influence their cross-border investments, in part through carry-trade related motivations and deviations from (un-)covered interest rate parity. To address this, we collect exchange rate data from the IMF’s International Financial Statistics and include in the regressions the volatility as well as the change in the real bilateral exchange rate. In our specifications we also control for countries’ riskiness using their sovereign CDS spread.9 Weusetheexpectedearningsgrowthofcountries’corporatesectorasaproxyfor the attractiveness of domestic investment opportunities; the data are from IBES. Finally, we obtain GDP data from the World Bank WDI database; in some of the specifications we use home-country GDP to scale foreign countries’ bond investments in the United States. 4 Stylized facts 4.1 Sovereign yields Our analysis relies on the variation in sovereign yields over time and across countries. Figure 1 captures the range of home country rates in our data panel, showing the median, maximum and minimum for each country. For example, rates in Japan have been low for most of the 2003 - 2016 period, while rates in many European countries have varied considerably, falling only more recently to low levels in most cases. Rates have not been as low in the majority of emerging markets. In addition, Figure 2 shows the evolution of cross-sectional quartiles of sovereign rates over the sample period. Importantly, not only is there significant variation in sovereign yields over time, but the interquartile range remains substantial throughout, even as the median approached zero toward the end of the period. This heterogeneity in the panel helps us considerably for identifying the effects of low rates on risk-taking. [Insert Figure 1 Here] [Insert Figure 2 Here] 9In a robustness check we also used the CDS spread for the country’s banking system to control for overall riskiness. Since the sample size declines due to availability of bank CDS data and our main results are unchanged, we do not report these results. 6

4.2 Foreign holdings of U.S. bonds Figure 3 shows how foreign holdings of U.S. bonds have evolved over the period 2003 - 2016.10 After a sharp increase in the years leading up to the GFC, foreign holdings of U.S. corporate debt declined during the GFC and the subsequent euro sovereign debt crisis, reflectingthe“flighthome”andsearch-for-safetyduringthatperiodthatisdocumentedin the literature.11 However, as interest rates in many foreign countries declined after 2012, inflows into corporate bonds rebounded, leading to a sharp increase in holdings especially towards the end of the sample period suggesting that investors from these countries were compensating for declining returns on safe assets at home by purchasing U.S. corporate debt. In terms of the U.S. safe asset, holdings of Treasuries were stable pre-GFC, but declined during the GFC, which is consistent with the “flight home” documented in the literature. After the GFC, foreign flows into Treasuries resumed. While during the European debt crisis the pick-up of foreign flows into Treasuries combined with a decline inflowsintocorporatebondsisconsistentwithageneralsenseofrisk-aversioninturbulent times, towards the latter part of the sample period when interest rate in many advanced countries have been low for some time, the increase in corporate bond holdings surpasses that of Treasury holdings. Figure 4 shows the composition of the foreign portfolio of U.S. bonds. The corporate share increased approximately 10 percentage points over the period and stood at roughly 60 percent for the last two years of the sample period.12 [Insert Figure 3 Here] [Insert Figure 4 Here] 4.3 U.S. bonds in foreign portfolios Because of the data availability, the paper focuses on just one part of foreign countries’ portfolios: theirholdingsofU.S.bonds. UsingdatafromtheIMF’sCoordinatedPortfolio Investment Survey (CPIS), columns (1)-(2) in Table 1 give an overview of how important U.S. bonds are in foreign countries’ aggregate bond portfolios.13 U.S. bonds constitute a significant share of countries’ holdings of foreign bonds: as of end-2015, on average for the 10ThefigureplotsthefacevalueofholdingsasreportedintheTICsurveys,thusabstractingfromthe effect of price changes. Therefore, the change in holdings from one year to the next can be interpreted as investment inflows (or outflows). 11See, for example, Giannetti and Laeven (2012) and De Haas and Van Horen (2012 and 2013). As also noted by Becker and Ivashina (2015), incentives to reach for yield were likely lower during the GFC for several, related reasons: investors were likely more risk averse as owners and regulators exercised more oversight; there was a high general uncertainty making risk assessments more challenging; and spreads on many instruments were high in the first place, making reaching for yield less attractive. 12In Appendix we show that the share of corporate holdings that include corporate asset backed securities (ABS) follows a similar pattern. 13The holdings data in Table 1 include holdings of all types of U.S. bonds. The majority of countries reporting to the CPIS do not distinguish holdings of corporate and sovereign bonds. 7

sample, thatshareisjustunder30percent. Becauseofhomebiasthatiswell-documented in the international finance literature, this share is much smaller in the countries’ overall portfolios that include domestic bond holdings: bond investment into the United States constitutes just under 7 percent of total bond portfolios on average for the sample. Also in terms of individual countries, Table 1 shows that for many countries, investment in the United States is a large part of their overall portfolios, and in particular of their international investments. For example, 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. As such, foreign investment in the United States constitutes a relatively large share of these countries’ international investments and could thus be representative of their international behavior. [Insert Table 1 Here] While the United States can be a large part of overseas investment for many countries, from the perspective of the United States, the fraction of U.S. bonds held by foreign investors tends to be small. The average country from our sample holds just 0.7 percent of the outstanding U.S. bonds.14 15 The last column in Table 1 shows countries’ holdings of U.S. bonds as a share of their GDP. Major custodians and financial centers have oversized holdings relative to their GDP. As discussed in detail in the next section, for these countries, instead of their national sovereign yield we use a composite European sovereign yield in the regressions. In addition, in some specifications, we exclude these countries from the sample. 5 Methodology We consider first the determinants of a country’s total private investment in U.S. corporate and Treasury bonds. We use the face values of holdings reported in the TIC surveys, thus abstracting from the effect of price changes, in order to isolate the effect of active new investments and portfolio shifts. To allow for differences in investor-country size, we scale countries’ holdings by their GDP. In country-year panel regressions separately for U.S. corporate bonds and for Treasuries we regress these scaled holdings on countries’ home interest rates and a number of other variables that reflect their financial and macroeconomic conditions, including sovereign CDS spreads, exchange rate changes, 14The CPIS data are not always directly comparable to the TIC data since major custodians for U.S. securities serve as the most important sources for the data, while in the CPIS, holdings are measured from the investors’ perspective and therefore are less subject to custodial biases, although they could in some cases suffer from other measurement idiosyncrasies. 15The columns using TIC data are based on publicly available TIC data that do not distinguish private from official holdings of U.S. securities. 8

expected earnings growth of these countries’ domestic corporate sector. We also include in the specifications countries’ ties to the United States as an investment destination: their share in U.S. trade as well as a measure of banking sector ties that is a proxy for the financial link between the investor country and the United States. In all regressions we include country fixed effects, which allows us to focus on time-varying country developments. We also include time fixed effects, which means 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. FollowingthegeneralapproachinAmmeretal. (2018),weuselocalcurrencysovereign bond rates to represent investment opportunities in investors’ home markets. Low rates can drive residents to invest more abroad, including in risky securities. Our sample excludes financial centers such as the Caribbean banking centers for two reasons. First, these countries do not have significant sovereign debt outstanding and therefore lack reliabledataonsovereignrates, whichisoneofourmainvariables. Second, theirinvestments are predominantly held on behalf of non-residents, for whom the interest rate to use is ambiguous, given our focus on the effect of home investment opportunities on private investors’ behavior. In our main specifications that include all country variables specified above, Luxembourg, which is an important financial center, drops out as it lacks data for all variables. That said, our baseline regressions do include the bond holdings of Belgium and Ireland, two European financial centers that largely cater to investors from other European countries. In terms of investors’ home interest rate for these countries, a composite European yield is likely to be a better choice than the national sovereign yield.16 Accordingly, using the approach in Ammer et al. (2018), instead of using their own sovereign rate, we assign these countries the average sovereign rates of four larger euro zone countries: Netherlands, France, Italy, and Spain. One could also use the German rate in the calculation of an average European rate, but one concern with this approach is that it might reflect Germany’s safe haven status, rather than investment opportunities in Germany.17 The empirical specification of the model we estimate is then: 16While entities resident in other countries in our sample may also hold some bonds on behalf of ultimate investors in different countries, Luxembourg, Belgium, and Ireland stand out for having TIC holdings in excess of home country data on their investors’ U.S. investments. In addition, Table 1 shows thatTICholdingsasapercentageofinvestor-countryGDParebyfarthehighestinthesethreecountries, suggesting that mismeasurement of investor nationality is much less of an issue elsewhere. 17Alternatively, and as a robustness check, we kept the European financial centers’ observations, but treated the entire euro zone as one country, instead of individual euro zone countries’ bilateral holdings. This approach might be fine for corporate bonds, but is likely less appropriate for U.S. Treasuries since, as discussed in the text, custodian countries, such as Belgium and other financial centers, could disproportionately influence the reported country holdings of U.S. Treasuries, especially the breakdown ofprivateandofficialinvestors’holdings. Nonetheless,usingthisapproachdoesnotqualitativelychange our results. 9

H /GDP = α+βSOV5y +γX +c +v +ϵ (1) j;t j;t j;t j;t j t j;t wherethedependentvariable, H /GDP , istheshareofU.S.bondsheldbyresidents j;t j;t of country j in year t in country j’s GDP. We explore separately the holdings (H) of U.S. corporate bonds and the holdings of Treasuries. SOV5y is country j’s sovereign yield j;t measured as the year-end 5-year sovereign yield. X are time varying home country j;t controls that may affect investment: (cid:15) theshareintotaltrade(exportsplusimports)withtheUnitedStates(Tradeshare), (cid:15) an analogous measure of bank exposure links to the United States (Bank link), (cid:15) the home country 5-year sovereign CDS premium (Sov CDS), to control for riskrelated fluctuations in home interest rates, (cid:15) the standard deviation of the bilateral (versus U.S. dollar) exchange rate (StDev FX), (cid:15) the change in the bilateral exchange rate (Delta FX), (cid:15) expected corporate earnings growth (ExpEarnGr), to control for domestic investment opportunities. In a robustness check, we also controlled for domestic financial risks using the average CDS premium of investor countries’ banks. Since the results are similar while the sample size declines quite a bit due to the availability of data on CDS premiums for the banking sector, the results we present in the paper do not include the bank CDS premiums as a control variable. In another robustness check we also included the lagged dependent variable as a control variable to account for autocorrelation. In most specifications the coefficient on the lagged dependent variable is as expected positive and significant, but its inclusion does not alter our main results. Because we include country fixed effects, we do not capture cross-country relations arising from other, time-invariant variables such as distance to and common language with the United States that are commonly used in the literature as proxies for transaction costs and familiarity with the foreign market. ToisolatetheeffectoftheGFCandtheEuropeandebtcrisisweincludeaninteraction term of the sovereign yield with a dummy variable for the period 2008 - 2012. In addition, in specifications that aim to compare the effects in the pre- and post-crises periods, we exclude the crises period of 2008 - 2012 and include an interaction of the sovereign 10

yield with a dummy for the post-crises period of 2013 - 2016. All regressions include country and time fixed effects, denoted by c and v , respectively. We estimate the model j t parameters by Weighted Least Squares, using for weights countries’ holdings of U.S. corporate and Treasury bonds. Figures 1 and 2 show details on the country sovereign rates. Table 2 presents summary statistics for the variables used in the regressions. [Insert Table 2 Here] Our second set of empirical exercises examines the determinants of the portfolio allocations that foreign investors choose within their U.S. bond holdings. In particular, the unit of analysis here is a particular country’s holdings of U.S. corporate bonds as a share of its total U.S. bond holdings on a given survey date. The setup is similar to the one above: we again include the home country sovereign yield and its interaction term with the dummy for the 2008 - 2012 crises period; as well as an interaction with a post-crises dummy variable excluding the crises period altogether in order to compare the period effects. We also include the same country determinants and fixed effects. Since all regressions include time dummies, they absorb all common factors, including changes in financial conditions in the United States. The empirical specification of the model we estimate is then: ∑ HCorp/ Hi = α+βSOV5y ++γX +c +v +ϵ j;t j;t j;t j;t j t j;t (2) i ∑ Where HCorp/ H is the country’s holdings of corporate bonds (HCorp) as a share j;t i j;t ∑ j;t of its total U.S. bond holdings on a given survey date Hi . i j;t 6 Empirical Results 6.1 Determinants of International Investment in U.S. Bonds We begin by investigating foreign private investment in U.S. corporate and Treasury bonds following specification shown in equation (1). The first four columns of Table 3 show the results for U.S. corporate bonds. In column (1) we report the results from the panel regression of country-level holdings of U.S. corporate bonds (scaled by home GDP) for the entire period 2003 - 2016 on the home country 5-year sovereign bond yield and its interaction with a 2008 - 2012 period dummy. The regression includes country and time fixed effects, along with the other country-specific control variables as discussed above. We find a statistically significant negative sign for the coefficient on the sovereign yield, consistent with a “push factor” that would induce flows in the context of a portfolio 11

balance framework. More specifically, the -0.036 coefficient means that when a country’s homeinterestrateis100basis-pointslower,itsinvestmentinU.S.corporatebondsrisesby 3.6 percent of its GDP. We do not find evidence that this relationship is different during the crises period as the coefficient on the interaction of the investor-country sovereign yield with the crises dummy is not statistically significant. [Insert Table 3 Here] In column (2), we consider whether the home sovereign yield effects differ in the postcrises period compared to the pre-crises period. To do this, we exclude the crises years from the sample completely, but retain a period-interaction. We find that the main result is maintained, and with a larger coefficient than in column (1). However, we do not find that in the post-crises period, when interest rate were generally low in most advanced countries, the relation is much different, as this interaction variable is not statistically significant. The sample in columns (1) and (2) already excludes the Caribbean banking centers andLuxembourgastheylacksovereignyieldandothercountry-leveldata, butincolumns (3) and (4) we further exclude Belgium and Ireland from the sample. As shown in the last column of Table 1, these countries have oversized holdings of U.S. bonds relative to their GDP. Columns (3) and (4) show that when we exclude financial centers and major custodians the main regression result on the relationship between investors’ home interest rates and their investment in U.S. corporate bonds is maintained, 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. In Appendix Table A1 we show that these results are unchanged if we include holdings of corporate asset backed securities (ABS) in the sample of corporate bonds. Important custodian countries such as Belgium and financial centers could have even stronger influence on reported country holdings of U.S. Treasuries, especially when it comes to the breakdown of private and official investors’ holdings. As pointed out in Treasury Department reports regarding the TIC survey data, some foreign official holdings likely are misclassified as private holdings because they are held through private intermediaries, and therefore data on private holdings in major custodians may reflect some holdings of foreign official institutions, which are usually disproportionately allocated to Treasuries.18 Since in this paper we focus on private investors’ behavior, for foreign holdings of Treasury bonds in the last two columns of Table 3 we present the results just for the sample excluding the financial centers. We use the same two period samples as before, the full sample and the sample excluding the crises period that com- 18AnnualTreasuryDepartmentreportsontheTICsurveydata: seeTreasuryDepartmentTICwebsite https://www.treasury.gov/resource-center/data-chart-center/tic/Pages/index.aspx 12

pares the pre- and post-crises effects of the home sovereign rates. We find a positive and statistically significant effect for the crises period (column (5)), indicating a decline in Treasury holdings for foreign investors with declining home interest rates. This is consistent with a “flight-home” investor behavior during crises. The insignificant effect of the sovereign yield for the normal times period in column (5) stems from the offsetting effects present pre- and post-crises, shown in column (6). The strong negative coefficient on the post-crises interaction of the sovereign yield indicates that post-crises, when rates were low in a lot of advanced economies, foreign investment in Treasury bonds rose, likely suggesting that investors seeking safe assets were drawn to the relatively higher yielding U.S. Treasuries, compared to home sovereign bonds. We, therefore, view this result as another indication of search-for-yield behavior. Turning to the other country-specific variables, the positive sign of the bank link coefficient in column (2) suggests that stronger bank lending ties between the investor country and the United States are associated with increased investment in U.S. corporate bonds, but we do not find this effect once we exclude financial centers from the sample. The coefficient on the countries’ share in U.S. trade is not statistically significant for corporatebondinvestment, andthenegativesignforTreasurybondinvestmentindicates, somewhatcounterintuitively, thatanincreaseinacountry’stradevolumewiththeUnited States is associated with less investment. However, if we define the trade variable as countryi’stradewiththeUnitedStatesrelativetocountryi’sGDPratherthanasashare in total U.S. trade, the coefficient on this variable is statistically insignificant but positive for both the corporate and Treasury investment specifications. Less surprising, however, is that higher expected earnings growth at home (ExpEarnGr), which we take as a proxy for risky domestic investment opportunities, also reduces investment in U.S. Treasuries as investors seek higher yielding opportunities. This results could be interpreted as another manifestation of yield searching behavior. Generally, the other country variables do not appear to significantly affect holdings of U.S. bonds. One reason could be that there is no sufficient time variation and the country fixed effects we include in these regressions fully capture the cross-country variation. 6.2 Determinants of Composition of International Investment in U.S. Bonds We next follow specification 2 to study the composition of private investors’ portfolio of U.S. bonds, which implicitly controls for the general incentives to invest in U.S. bonds. The idea is to explore how the share of the riskier, and therefore higher yielding, U.S. assets (U.S. corporate bonds) relative to the safe U.S. asset (U.S. Treasury) is affected by 13

investors’ home country interest rates. As outlined in the methodology section above, our variable of analysis here is holdings of U.S. corporate bonds as a share of a country’s total holdings of U.S. corporate and Treasury bonds on a given survey date.19 Table 4 reports the results. In column (1) we include all countries in the sample; in (2)-(3) we exclude the financial centers. The negative coefficient on the home interest rate in columns (1)- (2) is consistent with international bond investors relatively shifting their U.S. portfolios towardtheriskiercorporatebondswhendomesticinvestmentopportunitiesarelackluster. The effects are statistically stronger for the sample excluding financial centers. Effects are somewhat muted during the crises period, as the coefficient on the interaction of the interest rate with the crises dummy is statistically significant positive. As such, the effects may be smaller and not even present in crises period. [Insert Table 4 Here] In column (3), we find that post-crises the effects do not differ in a statistically significant way from those for the pre-crises period. Mirroring the positive sign on the expected earnings growth for Treasury holdings in Table 3, the expected earnings growth effect is negative in Table 4, consistent with Treasuries being a defensive investment when prospects for riskier investments are weaker. The estimated effects imply that international bond investors increasingly shift their U.S. bond portfolios toward corporate securities as home rates reach low levels, with economically meaningful effects: a 100 basis points decline in the home interest rate entails an estimated shift toward corporates of about 2.3 - 2.7 percentage points. 7 Robustness 7.1 Sample restrictions and choice of interest rate Table 5 shows that the baseline results on the effect of countries’ sovereign rates on their holdings of U.S. corporate and Treasury bonds are similar when the regressions exclude the country controls. The specifications are the same as in the baseline Table 3, i.e. using investment in U.S. corporate and Treasury bonds as ratio to the home country GDP as the dependent variable, but the sample is much larger than in Table 3 as some countries lack data for all controls. The larger coefficient on the sovereign yield variable in Table 5 is driven by the inclusion of Luxembourg in the sample. [Insert Table 5 Here] In Table 6 we report robustness checks to further restrictions to the country sample 19Results are similar if we instead scale by country’s total holdings of all types of U.S. bonds, which include U.S. Agencies. 14

and to using a different maturity for the home country interest rate. The sample in Table 6 excludes the financial centers, and it therefore corresponds to the results in columns (3)-(6)ofbaselineTable3. Asafirstrobustnesstest, incolumns(1)-(2)forU.S.corporate bonds and (4)-(5) for Treasuries, we exclude observations where countries’ sovereign CDS spread is in the top 5th percentile over the sample period.20 We find that that main regression results are confirmed and the size of the coefficients is very similar to the ones reported in baseline Table 3. As another robustness check, to explore the sensitivity to the specific interest rate used, columns (3) and (6) of Table 6 report estimates of total private investment in U.S. corporate and Treasury bonds, respectively, with regard to the 1-year home sovereign rate, instead of the 5-year rate we use in the baseline. The main results carry through, with slightly smaller estimated effects on investment from a lower home interest rate.21 [Insert Table 6 Here] Table 7 presents the same robustness checks but now for the Table 4 baseline results on the effects of home country drivers of the corporate share. The sample excludes the financial centers so it is comparable to the one used in columns (2)-(3) of Table 4. In columns (1)-(2) we exclude high sovereign CDS countries, while in column (3) we use the shorter 1-year maturity home sovereign yield. The main story is preserved. As we saw with total investment as a percent of GDP, the coefficient on the 1-year sovereign yield is smaller than on the 5-year yield. [Insert Table 7 Here] 7.2 Hedged and Unhedged Investment in U.S. Bonds So far, we have used countries’ local currency sovereign yields. Now, we examine if and to what extent the country’s private investment in U.S. corporate and Treasury bonds also responds to the U.S. dollar equivalent of the home sovereign rate (Y$), and how the effectsdiffercomparedtothehomesovereignyieldmeasuredinlocalcurrency(YHC).We test for this by calculating home country yields in U.S. dollar terms. For this, we collect 12-month forward premiums (FP) for the U.S. dollar against the home currencies of each of the investor countries in our sample. We calculate the synthetic dollar yields foreign investors wouldobtain if each of them hedged their home-currency 1-year sovereign bonds into the common currency, the U.S. dollar (Y$ (cid:17) YHC (cid:0) FP). This allows us to test for the role of the synthetic dollar yields (Y$) on foreign investments, while at the same 20Results are similar if we instead altogether exclude countries whose sovereign CDS spread has ever been in the top 5th percentile over the sample period. Results are also similar for a different cut-off for sovereign CDS spreads. 21Resultsarealsorobustiffortheweightsintheestimationweusecountries’totalU.S.bondholdings rather than holdings of U.S. corporate and Treasuries bonds only. 15

also being able to include both the home-currency yield (YHC) and the synthetic dollar yield (Y$) as two distinct competing factors affecting international investment choices. As before, in this set of results we control for the standard set of linkages and country economic and financial conditions. Table 8 presents the results for the sub-sample excluding the financial centers. Since here we use the 1-year synthetic dollar rate, these results have to be compared to the results in Table 6 where as a robustness check we use countries’ 1-year local currency sovereign rates instead of their 5-year rates. In column (1) of Table 8 we include only the synthetic dollar yield (Sov1y synth. USD), i.e. the hedged sovereign rate, along with our set of controls, and find that it does not appear to affect investment in U.S. corporate bonds in a statistically significant way. This is in contrast to the findings so far that the local currency sovereign yield negatively and in a statistically significant way affects investment in U.S. corporate bonds. However, we get more insights when we include both the hedged and the unhedged rate. In column (2) we report the results including both the hedged and unhedged rate in a specification that excludes the crises period. The hedged rate (Sov1y synth. USD) is again statistically insignificant, and the coefficient on the unhedged rate (Sov1y) is similar to the one reported in Table 6 column (3) where we use the 1-year unhedged rate on its own in the same type of specification. The story is similar in the case of investment in Treasury bonds. As reported in columns (3)-(4) of Table 8, the hedged rate does not appear to affect investment in Treasuries in a statistically significant way, while the effect of the unhedged rate is preserved. These results imply that investment is more sensitive to the unhedged rate, consistent with institutional incentives not to hedge exchange rate exposure. We also interpret these results as consistent with investors being aware of the limited predictive power of forward premiums for future exchange rates. Put differently, 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.22 [Insert Table 8 Here] 8 Conclusion We analyze how interest rates affect international investments and cross-border risktaking by examining the extent to which investors have shifted toward riskier assets overseas in response to low interest rates at home. Data on foreign investors’ holdings of 22Focusing on unhedged interest rate differential is arguably rational for informed investors who are tolerantofexchangeraterisk; theforwardpremiumpuzzlehasbeenwellknownatleastsincethe1980s. See, for example, Fama (1984). 16

U.S. bonds for 31 countries for the period 2003 - 2016 and a large variety in movements in interest rates in these countries provide for a unique way to analyze risk-taking behavior of investors in response to their home interest rates. Our analytical framework largely avoids concerns that the results might be driven instead by reverse causality from U.S. credit markets to foreign interest rates. For example, although an increase in the supply of U.S. bonds would likely draw in more cross-border investment, it likely would have only second-order effects on rates in other countries, and would tend to raise rates, while our finding is that increased U.S. investment is associated with lower investor-country rates. And, while these foreign investments are likely to have been affected by economic and financial conditions at home, because they are small from a U.S. perspective, they are unlikely to have affected the financing conditions of the issuers. We find that the lower the interest rate in the investor’s home country, the more investors increase their investments in the United States as a ratio to their home GDP, with the effects generally coming through investment in U.S. corporate sector bonds, rather than Treasury bonds. The results show that when a country’s home interest rate is 100 basis-points lower, its investment in U.S. corporate bonds rises by 3.6 to 5.3 percent of GDP. Furthermore, in terms of portfolio composition, international bond investors increasingly shift their U.S. bond portfolios toward corporate securities as home rates reach low levels, with economically meaningful effects. Since increased foreign investment in the United States is disproportionately allocated to corporate bonds, the results imply that low rates at home induce international investors to assume more credit risk in their U.S. bond portfolios. We find further evidence that the search-for-yield is a function of the home interest rates and not of the hedged dollar equivalent rates. Our findings have important policy implications in that they suggest that low interest rates can lead to shifts in overseas investments. Although we control for investor country characteristics, since we do not have comparable data on domestic investment portfolios as we do for holdings of U.S. bonds, 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. Or, extrapolating from the small part of their behavior we observe, one could conjecture that foreign 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. Our results are also consistent with central banks’ balance sheet policies havinga significanteffect on demand for foreign financial assets. Regardless, our findings suggest that there are spillover effects from low interest rates through cross-border capital flows. 17

References Ahmed, Shaghil, and Andrei Zlate. 2014. “Capital Flows to Emerging Market Economies: A Brave New World?” Journal of International Money and Finance, 48: 221 – 248. Ammer, John, Stijn Claessens, Alexandra Tabova, and Caleb Wroblewski. 2018. “Searching for Yield Abroad: Risk-Taking through Foreign Investment in U.S. Bonds.” International Finance Discussion Papers 1224, Board of Governors of the Federal Reserve. Aramonte, Sirio, Seung Jung Lee, and Viktors Stebunovs. 2015. “Risk-Taking and Low Longer-Term Interest Rates: Evidence from the U.S. Syndicated Loan Market.” Finance and Economics Discussion Series 2015-068. Aviat, Antonin, and Nicolas Coeurdacier.2007.“TheGeographyofTradeinGoods and Asset Holdings.” Journal of International Economics, 71(1): 22 – 51. Baskaya, Yusuf Soner, Julian di Giovanni, Sebnem Kalemli-Ozcan, and Mehmet Fatih Ulu. 2018. “International Spillovers and Local Credit Cycles.” NBER Working Paper No. 23149. Becker, Bo, and Victoria Ivashina. 2015. “Reaching for Yield in the Bond Market.” Journal of Finance, 70(5): 1863–1902. Bertaut, Carol, and Ruth Judson. 2014. “Estimating U.S. Cross-Border Securities Positions: New Data and New Methods.” International Finance Discussion Papers 1113, Board of Governors of the Federal Reserve. Bertaut, Carol, William Griever, and Ralph Tryon. 2006. “Understanding U.S. Cross-Border Securities Data.” Federal Reserve Bulletin. Boermans, Martijn, and Robert Vermeulen. 2016. “International Investment Positions Revisited: Investor Heterogeneity and Individual Security Characteristics.” DNB Working Papers 531, Netherlands Central Bank. Bowman, David, Juan M. Londono, and Horacio Sapriza. 2015. “U.S. Unconventional Monetary Policy and Transmission to Emerging Market Economies.” Journal of International Money and Finance, 55(C): 27–59. Broner, Fernando, Tatiana Didier, Aitor Erce, and Sergio L. Schmukler. 2013. “Gross Capital Flows: Dynamics and Crises.” Journal of Monetary Economics, 60(1): 113 – 133. Carnegie-NYU-Rochester Conference. 18

Cerutti, Eugenio, Stijn Claessens, and Andrew K. Rose. Forthcoming. “How Important is the Global Financial Cycle? Evidence from Capital Flows.” IMF Economic Review. Chari, Anusha, Karlye Dilts Stedman, and Christian Lundblad. 2017. “Taper Tantrums: QE,itsAftermathandEmergingMarketCapitalFlows.”NBER Working Paper No. 23474. Choi, Jaewon, and Mathias Kronlund.2018.“Reachingfor YieldinCorporateBond Mutual Funds.” The Review of Financial Studies, 31(5): 1930–1965. De Haas, Ralph, and Neeltje Van Horen. 2012. “International Shock Transmission after the Lehman Brothers Collapse: Evidence from Syndicated Lending.” American Economic Review, 102(3): 231–37. Dell’Ariccia, Giovanni, and Robert Marquez. 2013. “Interest Rates and the Bank Risk-Taking Channel.” Annual Review of Financial Economics, 5(1): 123–141. Domanski, Dietrich, Hyun Song Shin, and Vladyslav Sushko. 2017. “The Hunt for Duration: Not Waving but Drowning?” IMF Economic Review, 65(1): 113–153. Fama, Eugene F. 1984. “Forward and Spot Exchange Rates.” Journal of Monetary Economics, 14(3): 319 – 338. Forbes, Kristin J., and Francis E. Warnock. 2012. “Capital Flow Waves: Surges, Stops, Flight, and Retrenchment.” Journal of International Economics, 88(2): 235 – 251. Fratzscher, Marcel. 2012. “Capital Flows, Push versus Pull Factors and the Global Financial Crisis.” Journal of International Economics, 88(2): 341–356. Fratzscher, Marcel, Marco Lo Duca, and Roland Straub. 2016. “ECB Unconventional Monetary Policy: Market Impact and International Spillovers.” IMF Economic Review, 64(1): 36–74. Fratzscher, Marcel, Marco Lo Duca, and Roland Straub. 2018. “On the International Spillovers of US Quantitative Easing.” The Economic Journal, 128(608): 330– 377. Giannetti, Mariassunta, and Luc Laeven. 2012. “Flight Home, Flight Abroad, and International Credit Cycles.” American Economic Review, 102(3): 219–24. 19

Hau, Harald, and Sandy Lai. 2016. “Asset Allocation and Monetary Policy: Evidence from the Eurozone.” Journal of Financial Economics, 120(2): 309 – 329. Koijen, Ralph S.J., Francois Koulischer, Benoit Nguyen, and Motohiro Yogo. 2017. “Euro-Area Quantitative Easing and Portfolio Rebalancing.” American Economic Review: Papers & Proceedings, 107(5): 621–627. Maggio, Marco Di, and Marcin Kacperczyk. 2017. “The Unintended Consequences of the Zero Lower Bound Policy.” Journal of Financial Economics, 123(1): 59 – 80. Morais, Bernardo, Jose-Luis Peydro, and Claudia Ruiz Ortega. Forthcoming. “The International Bank Lending Channel of Monetary Policy Rates and QE: Credit Supply, Reach-for-Yield, and Real Effects.” Journal of Finance. Neely, Christopher J.2015.“UnconventionalMonetaryPolicyHadLargeInternational Effects.” Journal of Banking and Finance, 52: 101–111. Okawa, Yohei, and Eric van Wincoop. 2012. “Gravity in International Finance.” Journal of International Economics, 87(2): 205 – 215. Portes, Richard, and H(cid:19)el(cid:18)ene Rey. 2005. “The Determinants of Cross-Border Equity Flows.” Journal of International Economics, 65(2): 269–296. Rajan, Raghuram G. 2005. “Has financial development made the world riskier?” Proceedings - Economic Policy Symposium - Jackson Hole, August, 313–369. Rajan, Raghuram G. 2010. Fault Lines: How Hidden Fractures Still Threaten the World Economy. Princeton University Press. Rey, H(cid:19)el(cid:18)ene. 2013. “Dilemma not Trilemma: The Global Financial Cycle and Monetary Policy Independence.” Proceedings - Federal Reserve Bank of Kansas City Economic Symposium - Jackson Hole, 285–333. Stein, Jeremy. 2013. “Overheating in Credit Markets: Origins, Measurement, and Policy Responses.” At the “Restoring Household Financial Stability after the Great Recession: Why Household Balance Sheets Matter” research symposium sponsored by the Federal Reserve Bank of St. Louis, St. Louis, Missouri. 20

Figure 1: Sovereign Yields by Country (2003-2016) JAPAN SWITZERLAND SINGAPORE HONG KONG GERMANY DENMARK UNITED KINGDOM FINLAND NETHERLANDS SWEDEN FRANCE CANADA AUSTRIA BELGIUM NORWAY CHINA, MAINLAND ITALY MALAYSIA SPAIN IRELAND PORTUGAL KOREA, SOUTH ISRAEL GREECE AUSTRALIA CHILE PERU MEXICO COLOMBIA PHILIPPINES BRAZIL 0 4 8 12 16+ 5−year sovereign yield, range: 2003 − 2016 The figure plots the 5-year sovereign yield for the countries in our sample. For each country we plot the median (dot) and the min and the max (boundaries of the box) of the sovereign yield for the period 2003-2016. Last value on x-axis refers to values of 16+ for the sovereign yield. Chart includes the countries in the baseline sample. Authors’ calculations using data from Bloomberg. Figure 2: Sovereign Yields over time (2003-2016) dleiy ngierevos raey−eviF 6 4 2 0 2004 2006 2008 2010 2012 2014 2016 Year (end−June) 75th Percentile 25th Percentile Median Thefigureplotstheinterquartilerangeofthe5-yearsovereignyield(in%)forthecountriesinthebaseline sample. Authors’ calculations using data from Bloomberg. 21

Figure 3: Foreign Holdings of U.S. Bonds (2003-2016) 2,000 1,500 1,000 500 0 DSU fo noilliB ,sgnidloH Treasuries Corporate Bonds 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 The figure plots annual holdings of U.S. corporate bonds (black bars) and U.S. Treasury bonds (gray bars) by foreign private investors. We include all countries in the baseline sample. Corporate bonds exclude corporate ABS. The figure plots the face value of holdings as reported in the TIC surveys. Authors’ calculations using data from the Treasury International Capital annual surveys. Figure 4: Corporate Share in Foreign Holdings of U.S. Bonds (2003-2016) .9 .8 .7 .6 .5 erahS dnoB etaroproC 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Year The figure plots the corporate bond share of foreign private investors’ holdings of Treasuries and U.S. corporate bonds. We include all countries in the baseline sample. Corporate bonds exclude corporate ABS. Authors’ calculations using data from the Treasury International Capital annual surveys. 22

Table 1: U.S. share in foreign countries’ bond portfolios (as of end-2015) CPIS: U.S. bonds % of: TIC: U.S. bonds % of: Foreign Total Out- GDP Portfolio Portfolio standing Australia 38.8 5.3 0.15 4.0 Austria 8.3 4.3 0.02 2.4 Belgium 6.2 3.2 1.50 107 Brazil 57.4 0.1 0.65 13.6 Canada 68.3 10.6 0.56 13.6 Chile 44.4 8.0 0.07 9.6 China 49.4 0.6 4.08 13.2 Colombia 61.5 6.1 0.11 13.5 Denmark 16.9 4.3 0.12 14.5 France 10.0 4.2 0.27 4.0 Germany 10.4 5.7 0.49 5.2 Greece 0.7 0.3 0.00 0.8 Ireland 20.7 13.2 1.31 178 Israel 50.5 10.7 0.06 8.7 Italy 10.4 2.1 0.11 2.4 Japan 43.7 7.4 3.89 31.3 Luxembourg 23.5 19.8 1.84 1175 Malaysia 10.8 0.8 0.06 6.7 Mexico 94.2 8.0 0.27 6.5 Netherlands 14.6 7.9 0.29 14.8 Norway 27.2 17.2 0.30 28.7 Peru 34.0 4.2 0.04 9.1 Philippines 29.2 2.1 0.11 14.6 Portugal 3.6 1.2 0.01 2.1 Singapore 33.2 20.2 0.47 56.9 Spain 8.1 1.6 0.10 3.2 Sweden 22.5 4.5 0.12 9.1 Switzerland 23.9 18.2 0.91 50.1 United Kingdom 19.8 6.1 1.55 20.8 Average 29.0 6.8 0.7 62.7 Median 23.5 5.3 0.3 13.2 Authors’ calculations based on IMF CPIS, BIS, TIC data. CPIS: total bond holdings by private residents of that country as reported in the CPIS data. TIC: total bond holdings (private and official) as reported in TIC. Outstanding: U.S. bond 23 market capitalization. Table includes countries from baseline sample for which CPIS data are available. Luxembourg is included in the table for illustration.

Table 2: Summary Statistics (2003-2016) Median Mean St.Dev 1st p. 10th p. 95th p. 99th p. Sovereign 5-year LC Yield (%) 3.39 3.77 2.99 -0.35 0.57 10.70 13.70 Sovereign 1-year LC Yield (%) 2.33 2.85 2.83 -0.54 0.07 8.17 13.23 Sovereign 1-Year USD Yield (%) 1.38 2.27 2.23 -0.13 0.28 5.69 7.83 Expected Earnings Growth 0.12 0.16 0.32 -0.30 0.01 0.38 1.61 BankExposureShare(Banklink) 0.01 0.03 0.07 0.000 0.001 0.08 0.43 Trade Share 0.01 0.03 0.04 0.000 0.002 0.14 0.18 Sovereign CDS spread 5-year 0.43 1.03 2.16 0.01 0.03 3.63 9.98 Sovereign CDS spread 1-year 0.18 0.81 3.98 0.01 0.01 1.94 10.85 Exchange Rate St. Dev. 0.06 0.07 0.12 0.01 0.03 0.13 0.15 Exchange Rate Change -0.02 -0.003 0.10 -0.23 -0.14 0.19 0.24 Treasury Holdings/GDP (%) 0.51 2.41 7.40 0.00 0.02 8.82 40.32 Corporate Bond Holdings/GDP 0.85 5.62 14.2 0.01 0.08 38.8 75.9 (%) Corporate Share in Holdings 0.66 0.62 0.24 0.05 0.25 0.95 0.98 Notes: The table reports summary statistics for the country specific variables. The statistics are over the entire sample period 2003-2016. The exchange rate statistics refer to countries’ bilateral (versus the U.S. dollar) exchange rate. The exchange rate change is calculated as the log difference. The bank and trade exposure are reported as shares. Sovereign 1-Year USD Yield refers to the 1-year synthetic dollar yield calculated by the authors as explained in the text. In this table bond holdings to GDP statistics are reported in percent; in the regressions they are included as shares. 24

Table 3: Holdings of U.S. Bonds as a Share of Countries’ GDP The table shows the estimated coefficients for equation (1) in the text. The dependent variable in columns (1)-(4) is country j’s holdings of U.S. corporate bonds at time t scaled by country j’s GDP. The dependent variable in columns (5)-(6) is country j’s holdings of U.S. Treasuries at time t scaled by countryj’sGDP.Luxembourgdropsoutofthesampleasitlacksdataforallcontrols. Incolumns(1)-(2) weincludeallcountriesinthesample; inadditiontoLuxembourg, columns(3)-(6)alsoexcludeBelgium and Ireland. The sample period in columns (1), (3), and (5) is 2003-2016; in columns (2), (4), and (6) we exclude the crises years 2008-2012. 2008-2012 is a dummy variable equal to 1 for the crises period 2008-2012. Postisadummyvariableequalto1fortheperiod2013-2016and0fortheperiod2003-2007. Countries’ sovereign rates are the year-end 5-year sovereign local currency yields. The sovereign rate for Belgium and Ireland is the average of the sovereign rates of the Netherlands, France, Italy, and Spain. Weighted regression where the weights are countries’ holdings of U.S. corporate and Treasury bonds. For brevity the coefficients for fixed effects and the constant are not reported. * p < 0:10, ** p < 0:05, *** p<0:01. Heteroscedasticity-consistent standard errors are reported in parentheses. U.S. Corporate Bonds / GDP Treasuries / GDP All countries Excl. Fin. Centers Excl. Fin. Centers (1) (2) (3) (4) (5) (6) Sov5y -0.036(cid:3)(cid:3)(cid:3) -0.053(cid:3)(cid:3)(cid:3) -0.010(cid:3)(cid:3)(cid:3) -0.008(cid:3)(cid:3) 0.002 0.006(cid:3)(cid:3) (0.011) (0.015) (0.003) (0.004) (0.002) (0.003) 2008-12=1 (cid:2) Sov5y -0.003 -0.000 0.002(cid:3)(cid:3) (0.006) (0.003) (0.001) Post=1 (cid:2) Sov5y -0.009 -0.003 -0.007(cid:3)(cid:3)(cid:3) (0.015) (0.009) (0.002) Bank link 0.358 0.687(cid:3)(cid:3) 0.001 -0.073 0.010 -0.074(cid:3)(cid:3) (0.229) (0.281) (0.070) (0.090) (0.030) (0.032) Trade share 1.222 1.860 -0.145 -0.257 -0.403(cid:3)(cid:3)(cid:3) -0.337(cid:3)(cid:3) (1.166) (1.214) (0.178) (0.266) (0.135) (0.137) Sov CDS spr (5-year) -1.975 5.951(cid:3) 0.586 0.387 -0.341 0.094 (1.260) (3.215) (0.475) (1.376) (0.212) (0.308) StDev FX 0.426 0.428 0.062 -0.131 0.078(cid:3) 0.019 (0.328) (0.457) (0.117) (0.198) (0.042) (0.048) Delta FX 0.111 0.163 0.043(cid:3) 0.062(cid:3) 0.007 -0.006 (0.082) (0.103) (0.024) (0.036) (0.013) (0.017) ExpEarnGr 0.066(cid:3) -0.016 -0.010 -0.020 -0.015(cid:3)(cid:3)(cid:3) -0.011(cid:3)(cid:3) (0.039) (0.041) (0.009) (0.015) (0.004) (0.005) Observations 376 236 349 218 349 218 Adj. R-sq 0.92 0.91 0.89 0.89 0.92 0.93 Country FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes 25

Table 4: Corporate Bond Share Thetableshowstheestimatedcoefficientsforequation(2)inthetext. Thedependentvariableiscountry j’s holdings of U.S. corporate bonds at time t relative to its holdings of U.S. Treasuries and corporate bonds. Luxembourgdropsoutofthesampleasitlacksdataforallcontrols. Incolumn(1)weincludeall countries in the sample; in addition to Luxembourg, columns (2)-(3) also exclude Belgium and Ireland. The sample period in (1)-(2) is 2003-2016; in column (3) we exclude the crises years 2008-2012. 2008- 2012 is a dummy variable equal to 1 for the crises period 2008-2012. Post is a dummy variable equal to 1 for the period 2013-2016 and 0 for the period 2003-2007. Countries’ sovereign rates are the year-end 5-year sovereign local currency yields. The sovereign rate for Belgium and Ireland is the average of the sovereign rates of the Netherlands, France, Italy, and Spain. Weighted regression where the weights are countries’ holdings of U.S. corporate and Treasury bonds. For brevity the coefficients for fixed effects and the constant are not reported. * p<0:10, ** p<0:05, *** p<0:01. Heteroscedasticity-consistent standard errors are reported in parentheses. All countries Excl. Fin. Centers (1) (2) (3) Sov5y -0.027(cid:3) -0.023(cid:3)(cid:3)(cid:3) -0.027(cid:3)(cid:3) (0.014) (0.009) (0.013) 2008-12=1 (cid:2) Sov5y 0.018(cid:3)(cid:3) 0.016(cid:3)(cid:3) (0.009) (0.007) Post=1 (cid:2) Sov5y 0.001 (0.013) Bank link -0.135 0.199 0.212 (0.253) (0.128) (0.182) Trade share -5.556(cid:3)(cid:3)(cid:3) -0.918 -1.518 (1.407) (0.950) (1.161) Sov CDS spr (5-year) 0.094 0.498 -0.200 (1.085) (1.132) (2.602) StDev FX 0.514 -0.054 -0.057 (0.353) (0.179) (0.245) Delta FX -0.117 -0.032 -0.073 (0.093) (0.049) (0.068) ExpEarnGr 0.035 0.084(cid:3)(cid:3)(cid:3) 0.134(cid:3)(cid:3)(cid:3) (0.031) (0.032) (0.045) Observations 376 349 218 Adj. R-sq 0.83 0.91 0.90 Country FE Yes Yes Yes Time FE Yes Yes Yes 26

Table 5: Robustness: Holdings of U.S. Bonds as a Share of Countries’ GDP The table shows the estimated coefficients for equation (1) in the text. The dependent variable in columns (1)-(4) is country j’s holdings of U.S. corporate bonds at time t scaled by country j’s GDP. The dependent variable in columns (5)-(6) is country j’s holdings of U.S. Treasuries at time t scaled by country j’s GDP. In addition to the baseline sample, the sample here includes the following countries: Hungary, Jamaica, Luxembourg, New Zealand, Sri Lanka, and Thailand. In columns (1)-(2) we include allcountriesinthesample;columns(3)-(6)alsoexcludeBelgium,Ireland,andLuxembourg. Thesample period in columns (1), (3), and (5) is 2003-2016; in columns (2), (4), and (6) we exclude the crises years 2008-2012. 2008-2012 is a dummy variable equal to 1 for the crises period 2008-2012. Post is a dummy variable equal to 1 for the period 2013-2016 and 0 for the period 2003-2007. Countries’ sovereign rates are the year-end 5-year sovereign local currency yields. The sovereign rate for Belgium, Ireland, and LuxembourgistheaverageofthesovereignratesoftheNetherlands, France, Italy, andSpain. Weighted regression where the weights are countries’ holdings of U.S. corporate and Treasury bonds. For brevity the coefficients for fixed effects and the constant are not reported. * p<0:10, ** p<0:05, *** p<0:01. Heteroscedasticity-consistent standard errors are reported in parentheses. U.S. Corporate Bonds / GDP Treasuries / GDP All countries Excl. Fin. Centers Excl. Fin. Centers (1) (2) (3) (4) (5) (6) Sov5y -0.157(cid:3)(cid:3) -0.162(cid:3)(cid:3) -0.013(cid:3)(cid:3)(cid:3) -0.016(cid:3)(cid:3)(cid:3) 0.002(cid:3) 0.005(cid:3)(cid:3)(cid:3) (0.066) (0.074) (0.003) (0.004) (0.001) (0.002) 2008-12=1 (cid:2) Sov5y -0.018 0.004 0.003(cid:3)(cid:3) (0.033) (0.002) (0.001) Post=1 (cid:2) Sov5y -0.016 -0.002 -0.009(cid:3)(cid:3)(cid:3) (0.063) (0.006) (0.001) Observations 466 297 424 270 424 270 Adj. R-sq 0.93 0.92 0.88 0.88 0.88 0.91 Country FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes 27

Table 6: Robustness: Holdings of U.S. Bonds as a Share of Countries’ GDP Thedependentvariableincolumns(1)-(3)iscountryj’sholdingsofU.S.corporatebondsattimet scaled by country j’s GDP. The dependent variable in columns (4)-(6) is country j’s holdings of U.S. Treasury bonds at time t scaled by country j’s GDP. The sample excludes Belgium, Ireland, and Luxembourg. The sample period in columns (1) and (4) is 2003-2016. 2008-2012 is a dummy variable equal to 1 for the crises period 2008-2012. Columns (2)-(3) and (5)-(6) exclude the crises period 2008-2012. Post is a dummy variable equal to 1 for the period 2013-2016 and 0 for 2003-2007. In columns (1)-(2) and (4)-(5) weexcludecountrieswhosesovereignCDSspreadisinthetop5thpercentileoverthesampleperiodand use countries’ 5-year sovereign rate. In columns (3) and (6) countries’ sovereign rates are the year-end 1-year sovereign local currency yields. Weighted regression where the weights are countries’ holdings of U.S. corporate and Treasury bonds. For brevity the coefficients for fixed effects and the constant are not reported. * p < 0:10, ** p < 0:05, *** p < 0:01. Heteroscedasticity-consistent standard errors are reported in parentheses. U.S. Corporate Bonds / GDP Treasuries / GDP Excl. High CDS 1-year Sov. Excl. High CDS 1-year Sov. (1) (2) (3) (4) (5) (6) Sov5y -0.010(cid:3)(cid:3)(cid:3) -0.006(cid:3)(cid:3) 0.003 0.007(cid:3)(cid:3)(cid:3) (0.003) (0.003) (0.002) (0.003) 2008-12=1 (cid:2) Sov5y -0.001 0.002(cid:3)(cid:3) (0.003) (0.001) Post=1 (cid:2) Sov5y -0.001 -0.008(cid:3)(cid:3)(cid:3) (0.009) (0.002) Sov1y -0.007(cid:3) 0.005(cid:3)(cid:3)(cid:3) (0.004) (0.001) Post=1 (cid:2) Sov1y 0.003 -0.004(cid:3)(cid:3)(cid:3) (0.011) (0.002) Bank link 0.005 -0.091 0.329(cid:3) 0.009 -0.076(cid:3)(cid:3) 0.173(cid:3)(cid:3)(cid:3) (0.075) (0.094) (0.181) (0.030) (0.033) (0.047) Trade share -0.156 -0.245 -0.220 -0.380(cid:3)(cid:3)(cid:3) -0.323(cid:3)(cid:3) -0.380(cid:3)(cid:3)(cid:3) (0.181) (0.291) (0.396) (0.139) (0.144) (0.137) Sov CDS spr (5-year) 0.496 0.012 -0.595 0.086 (1.433) (2.025) (0.447) (0.558) StDev FX 0.068 -0.110 -0.111 0.077(cid:3) 0.014 0.041 (0.110) (0.165) (0.180) (0.043) (0.049) (0.043) Delta FX 0.043 0.056 0.126(cid:3)(cid:3)(cid:3) 0.009 -0.005 0.027(cid:3)(cid:3) (0.026) (0.037) (0.050) (0.013) (0.017) (0.014) ExpEarnGr -0.010 -0.017 -0.027 -0.017(cid:3)(cid:3)(cid:3) -0.012(cid:3)(cid:3) -0.018(cid:3)(cid:3)(cid:3) (0.009) (0.013) (0.019) (0.004) (0.006) (0.005) Sov CDS spr (1-year) 0.408 0.152(cid:3)(cid:3)(cid:3) (0.344) (0.061) Observations 332 209 198 332 209 198 Adj. R-sq 0.89 0.89 0.89 0.92 0.93 0.95 Country FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes 28

Table 7: Robustness: Corporate Bond Share The table shows the estimated coefficients for equation (2) in the text. The sample excludes Belgium, Ireland, and Luxembourg. The sample period in columns (1) is 2003-2016. 2008-2012 is a dummy variable equal to 1 for the crises period 2008-2012. Columns (2)-(3) exclude the crises period 2008-2012. Post is a dummy variable equal to 1 for the period 2013-2016 and 0 for 2003-2007. In columns (1)-(2) we exclude countries whose sovereign CDS spread is in the top 5th percentile over the sample period andusecountries’5-yearsovereignrate. Incolumn(3) countries’sovereignratesare theyear-end1-year sovereign local currency yields. Weighted regression where the weights are countries’ holdings of U.S. corporate and Treasury bonds. For brevity the coefficients for fixed effects and the constant are not reported. * p < 0:10, ** p < 0:05, *** p < 0:01. Heteroscedasticity-consistent standard errors are reported in parentheses. Excl. High CDS 1-year Sov. (1) (2) (3) Sov5y -0.026(cid:3)(cid:3)(cid:3) -0.031(cid:3)(cid:3) (0.010) (0.015) 2008-12=1 (cid:2) Sov5y 0.016(cid:3)(cid:3) (0.007) Post=1 (cid:2) Sov5y -0.003 (0.015) Sov1y -0.042(cid:3)(cid:3)(cid:3) (0.013) Post=1 (cid:2) Sov1y 0.024 (0.015) Bank link 0.193 0.175 -0.532(cid:3) (0.130) (0.190) (0.302) Trade share -1.013 -1.658 -5.069(cid:3)(cid:3)(cid:3) (0.963) (1.172) (1.170) Sov CDS spr (5-year) 1.569 2.194 (2.383) (4.476) StDev FX -0.020 -0.023 -0.278 (0.180) (0.270) (0.277) Delta FX -0.048 -0.106 -0.108(cid:3) (0.051) (0.077) (0.062) ExpEarnGr 0.091(cid:3)(cid:3)(cid:3) 0.151(cid:3)(cid:3)(cid:3) 0.217(cid:3)(cid:3)(cid:3) (0.034) (0.049) (0.039) Sov CDS spr (1-year) -1.767(cid:3)(cid:3)(cid:3) (0.565) Observations 332 195 198 Adj. R-sq 0.91 0.90 0.90 Country FE Yes Yes Yes Time FE Yes Yes Yes 29

Table 8: Hedged and Unhedged Investment in U.S. Bonds Thedependentvariableincolumns(1)-(3)iscountryj’sholdingsofU.S.corporatebondsattimet scaled bycountryj’sGDP.Thedependentvariableincolumn(4)iscountryj’sholdingsofU.S.Treasurybonds at time t scaled by country j’s GDP. The sample period is 2003-2016. 2008-2012 is a dummy variable equalto1forthecrisesperiod2008-2012. SampleexcludesBelgium,Ireland,andLuxembourg. Weighted regression where the weights are countries’ holdings of U.S. corporate and Treasury bonds. For brevity the coefficients for fixed effects and the constant are not reported. * p<0:10, ** p<0:05, *** p<0:01. Heteroscedasticity-consistent standard errors are reported in parentheses. U.S. Corporate Bonds / GDP Treasuries / GDP (1) (2) (3) (4) Sov1y synth. USD 0.028 0.007 0.002 0.000 (0.020) (0.010) (0.002) (0.004) 2008-12=1 (cid:2) Sov1y synth. USD -0.023 0.001 (0.015) (0.002) Post=1 (cid:2) Sov1y synth. USD 0.031 0.004 (0.027) (0.004) Sov1y -0.010(cid:3) 0.005(cid:3)(cid:3)(cid:3) (0.006) (0.001) Post=1 (cid:2) Sov1y -0.000 -0.004(cid:3)(cid:3)(cid:3) (0.008) (0.001) Bank link -0.030 -0.096 0.073(cid:3)(cid:3) 0.094(cid:3)(cid:3) (0.051) (0.115) (0.035) (0.048) Trade share 0.493 -0.178 -0.595(cid:3)(cid:3)(cid:3) -0.427(cid:3)(cid:3) (0.337) (0.539) (0.136) (0.180) Sov CDS spr (1-year) 0.068 0.203 0.043 0.110(cid:3) (0.207) (0.314) (0.079) (0.063) StDev FX 0.098 0.125 0.098(cid:3)(cid:3)(cid:3) 0.083(cid:3) (0.083) (0.140) (0.040) (0.047) ExpEarnGr -0.005 -0.004 -0.015(cid:3)(cid:3)(cid:3) -0.012(cid:3)(cid:3)(cid:3) (0.007) (0.012) (0.005) (0.005) Observations 325 196 325 196 Adj. R-sq 0.90 0.90 0.93 0.95 Country FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes 30

A Appendix: Corporate Bond Sample Including ABS 31

Table A.1: Holdings of U.S. Corporate Bonds and Corporate ABS as a Share of Countries’ GDP Thetableshowstheestimatedcoefficientsforequation(1)inthetext. Thedependentvariableiscountry j’sholdingsofU.S.corporatebondsandcorporateABSattimet scaledbycountryj’sGDP.Luxembourg drops out of the sample as it lacks data for all controls. Columns (1)-(2) include all countries in the sample; columns (3)-(4) exclude Belgium and Ireland. The sample period in columns (1) and (3) is 2003-2016; incolumns(2)and(4)weexcludethecrisesyears2008-2012. 2008-2012isadummyvariable equal to 1 for the crises period 2008-2012. Post is a dummy variable equal to 1 for the period 2013- 2016 and 0 for the period 2003-2007. Countries’ sovereign rates are the year-end 5-year sovereign local currency yields. The sovereign rate for Belgium and Ireland is the average of the sovereign rates of the Netherlands, France, Italy, and Spain. Weighted regression where the weights are countries’ total holdings of U.S. bonds. For brevity the coefficients for fixed effects and the constant are not reported. * p < 0:10, ** p < 0:05, *** p < 0:01. Heteroscedasticity-consistent standard errors are reported in parentheses. All countries Excl. Fin. Centers (1) (2) (3) (4) Sov5y -0.036(cid:3)(cid:3)(cid:3) -0.049(cid:3)(cid:3)(cid:3) -0.007(cid:3)(cid:3) -0.006(cid:3) (0.011) (0.017) (0.003) (0.003) 2008-12=1 (cid:2) Sov5y -0.003 -0.004 (0.008) (0.003) Post=1 (cid:2) Sov5y -0.004 0.004 (0.016) (0.008) Bank link 0.498(cid:3)(cid:3) 0.837(cid:3)(cid:3)(cid:3) 0.085 0.057 (0.245) (0.314) (0.077) (0.102) Trade share 0.602 0.913 -0.175 -0.403(cid:3) (1.224) (1.365) (0.188) (0.242) Sov CDS spr (5-year) -1.238 5.097 0.448 -0.449 (1.346) (3.705) (0.473) (1.159) StDev FX 0.372 0.214 0.089 -0.120 (0.382) (0.564) (0.101) (0.170) Delta FX 0.129 0.232(cid:3)(cid:3) 0.046(cid:3)(cid:3) 0.070(cid:3)(cid:3) (0.094) (0.110) (0.023) (0.034) ExpEarnGr 0.097(cid:3)(cid:3)(cid:3) 0.027 -0.006 -0.016 (0.035) (0.049) (0.010) (0.012) Observations 376 236 349 218 Adj. R-sq 0.92 0.91 0.89 0.89 Country FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes 32