31. International Asset Pricing

31.1.1 / 31.1.2 Key Points & Data

Maggiori et al. (2020) use security-level data to show: investor holdings are biased toward their own currency; most firms issue only in local currency with no direct access to foreign capital; this holds for most countries except the US; foreign investors who invest globally typically allocate a large fraction of their portfolio to dollar-denominated securities (dollar bias); in recent years the dollar has become the only international currency. Data: Morningstar's complete position-level dataset (one of the largest providers of investment research, with mutual fund and ETF holdings in over 50 countries; as of December 2017, about 5 million positions held by ~9,000 US funds and ~6 million positions held by ~52,000 funds elsewhere). These are open-end funds investing in equities, fixed income, commodities, convertible bonds, and housing; occasionally reported derivative holdings are excluded; the identifier is the 9-digit CUSIP. They also use ICI (Investment Company Institute) data: in 2017 US mutual funds/ETFs had ~22 trillion USD AUM, non-US ~19 trillion (Fig 31.1 US/non-US fixed income fund AUM). For representativeness, the authors focus on developed economies where Morningstar's fixed-income AUM coverage is no less than a quarter of ICI's reported, taking the top countries by AUM. Aggregation uses the Coppola et al. (2020) algorithm (CUSIP, Capital IQ, SDC Platinum, Dealogic, Factset, Orbis) to aggregate firms to their ultimate parent and standardize security characteristics.

31.1.3 Investor Home Currency Bias

Country level (Fig 31.2 the share of each country's investor portfolio denominated in the issuer's currency): most domestic investment is denominated in local currency (e.g. a Canadian company's borrowing from Canadian investors is mostly in Canadian dollars); except for the US, most foreign investment is not denominated in the issuer's local currency (e.g. only 5% of non-Canadian lending to Canadian companies is in Canadian dollars). Fig 31.3: foreign investors tend to hold bonds denominated either in their home currency or in US dollars (red bar = the fraction of foreign bond positions in US dollars, black bar = in home currency).

Security level: let \(s_{j,p,c}\) be the share of corporate bond \(c\) issued by parent \(p\) held by investors from country \(j\) (as of 2017), regression (31.1):

with parent fixed effects \(\alpha_{j,p}\) (capturing firm-level characteristics), \(\mathbf 1\{\cdot\}\) indicating whether the bond is denominated in the investor's currency. Fig 31.4: \(\beta_j\) is significantly positive for all countries — i.e. within-firm, every bond exhibits home currency bias (for the same issuer, investors hold more of the bond denominated in their home currency), robust to other sample selections/specifications. Fund level (Fig 31.5): funds investing in foreign-market corporate bonds mostly invest either all or none in their home currency; most such funds concentrate foreign bond investment in home-currency or USD-denominated bonds; independent of fund size.

31.1.4 / 31.1.5 Firm Home Currency Bias & Rise of Dollar

Foreign-currency debt issuers borrow from foreigners (Fig 31.6): small firms typically borrow from domestic investors in local currency; firms that borrow through foreign-currency bonds borrow almost entirely from foreign investors (along the 45-degree line); the US is a bit different — medium firms also borrow entirely through USD bonds but still have a large share of foreign investors. Local-currency-only firms borrow exclusively from domestic investors (except the US, Fig 31.7): in Canada, EMU, and UK the red circles are always above the blue diamonds, i.e. local-currency-only firms in these countries place most debt in local funds rather than overseas funds; in the US about half the blue diamonds are above the red circles, i.e. US local-currency-only firms place debt indifferently in US funds and funds elsewhere.

Rise of the dollar, fall of the euro (Fig 31.8): after 2008 the USD-denominated share of cross-border bonds rises and the euro-denominated share falls (vertical axis the share of global total cross-border corporate bond positions).

31.1.6 / 31.1.7 Contribution & Discussion

Contribution: a gigantic cross-country open-end fund/ETF position dataset; a detailed analysis of home currency bias, identifying several important patterns on the fund side and firm side. Discussion: when choosing countries, ranking by all-fund-type (equity, fixed income, allocation, money market) AUM with a cutoff of 10 is a bit arbitrary, but since the paper is mainly about fixed income this filter doesn't make sense and should rank by representativeness of the Morningstar data; the crucial assumption "the domicile of a fund equals the country of residence of its investors" isn't necessarily true, and if it breaks down some conclusions won't hold; the (2)(3)(4) columns of (31.1) have no control variables, which is weird; Fig 31.9(b) (a figure that) suddenly switches to equity portfolios, and the equity results aren't really consistent with the bond results. Possible extensions: study what fund characteristics determine the magnitude of home currency bias; if currency risk is the main reason, there should be a huge industry of financial firms providing (and profiting from) currency hedging, so it would be interesting to see if such firms match the magnitude of the potential demand; the methodology could be applied to Stock Connect data between mainland China and Hong Kong from 2014 (after 10 April 2014, mainland investors could buy HK-listed stocks locally with RMB, and vice versa), as a quasi-natural experiment to test whether currency drives home-biased investment in equity markets.

31.2.1 Key Points

Itskhoki-Mukhin (2019): replicate the Mussa (1986) puzzle — after Bretton Woods fixed rates broke down in the early 1970s, the volatility of nominal and real exchange rates surged, while macro variables like inflation (nominal), consumption (real), output (real), and net export (real) showed no simultaneous change (Mussa puzzle). The authors empirically falsify both conventional flexible-price RBC and sticky-price New Keynesian models, and provide a structural model of segmented financial markets — nominal exchange rate risk is held by a small group of financial intermediaries and not smoothly shared throughout the economy. This chapter focuses on the empirical part. Lilley et al. (2019): focus on exchange rate disconnect — it's hard to find economic variables strongly co-moving with the exchange rate (also the empirical focus of Itskhoki-Mukhin). Using a variety of global risk-appetite proxies, they find these had no in-sample explanatory power for the exchange rate prior to 2007 but significant power post-2007; and that US purchases of foreign bonds are highly correlated with these risk proxies and exchange rates since 2008, driven primarily by global investment in dollar-denominated bonds. So the results favor "the dollar became the international currency after the 2008 crisis".

31.2.2 Data

Itskhoki-Mukhin (2019): monthly data on nominal exchange rate, consumer prices, and production index from IFM IFS; the production index is seasonally adjusted; net export = (exports−imports)/(exports+imports); quarterly data on GDP, consumption, imports, and exports from OECD (real and seasonally adjusted); all annualized so volatilities (standard deviations) are comparable. 1960:01–1971:07 is the Bretton-Woods "peg" period, 1973:02–1989:12 the "float" period; the intermediate 1971:08–1972:12 is excluded because the "Nixon shock" (after 1971:08 the direct convertibility of the dollar to gold was limited) took place in 1971:08 and all countries officially declared floating in 1973:02. Lilley et al. (2019): IMF Balance of Payments (BoP) and International Investment Positions (IIP) public data to construct quarterly measures of US capital flows; micro data uses the Maggiori et al. (2020) holdings data (see §31.1.2).

31.2.3 Exchange Rate Disconnect after Peg-to-Float Shock: Itskhoki-Mukhin (2019)

The change in macro variable volatility around the end of Bretton Woods is not comparable to the change in exchange rate volatility. The authors estimate standard deviations of variables before and after 1973:01 using 18-month (or 10-quarter) rolling-window moving averages. From Fig 31.9: nominal and real exchange rate volatility changes dramatically (about 2% → 10%); other macro variables show no comparable magnitude of volatility change — clear evidence of exchange rate disconnect. Itskhoki-Mukhin also break down the rest of the world into Canada, France, Germany, Italy, Japan, Spain, and the UK, showing that in most cases these countries' macro variable volatility didn't significantly increase while the exchange rate volatility did. Finally they find no change in the volatility of relative stock market returns (US vs. rest of the world) before and after the end of Bretton Woods, and that the forward discount puzzle appeared after the end of Bretton Woods.

31.2.4 Exchange Rate Reconnect since 2008 Crisis: Lilley et al. (2019)

Lilley et al. (2019) first replicate the exchange rate disconnect in 1977–2006. They use the broad dollar (trade-weighted US dollar index) to measure the value of the dollar relative to other currencies. The broad dollar \(I_t\) is defined by (31.2):

\(N_t\) is the number of currencies included at \(t\), \(e_{j,t}\) the exchange rate of currency \(j\) at \(t\) (units of currency \(j\) per US dollar), \(w_{j,t}\) the weight of currency \(j\) at \(t\) (updated annually, \(\sum_{j=1}^{N_t}w_{j,t}=1\)). Fig 31.10: the vertical axis is the quarterly log change in the broad dollar defined by (31.2) (evaluated by equal weights for the nine G10-except-US currencies), scattered against the interest rate differential, inflation differential, US corporate bond credit spread (a measure of global risk appetite), and US foreign bond purchases — in 1977–2006 all four have almost no correlation with the broad dollar (even though covered interest rate parity predicts a strong negative relation between the rate differential and the differential, empirically it's near zero).

Reconnect after 2008 (six global risk-appetite measures): regression \(\Delta\ln I_t=\alpha+\beta x_t+\varepsilon_t\), \(x_t\) one of the following risk measures. Fig 31.11 gives the 10-year rolling-window \(R^2\) of this regression — during the 2008 crisis the \(R^2\) surges and stays high. The six measures: the GZ Spread (US corporate bond credit spread), VXO (the monthly log change in implied volatility on the S&P100), the S&P500 (the log total return on the S&P500), the Treasury Premium (the average one-year covered interest parity deviation between developed-country government bonds and US Treasuries, Du et al. 2018), the Global Factor (constructed from world asset prices, Miranda-Agrippino-Rey 2015), and Intermediary Returns (a value-weighted portfolio of holding companies of New York Fed primary dealers, He et al. 2017).

Reconnect with US foreign bond purchases: regression \(\Delta\ln I_t=\alpha+\beta f_t+\varepsilon_t\), \(f_t\) the net US purchases of foreign bonds; also regress \(f_t\) on the six risk measures. Fig 31.12 (\(f_t\) on risk measures' \(R^2\)) and Fig 31.13 (\(\Delta\ln I_t\) on \(f_t\)'s 5-year and 10-year rolling \(R^2\)) both show the \(R^2\) surging after 2008 and staying high for a couple of years — so the broad dollar is also reconnected through US foreign bond purchases.

31.2.5 / 31.2.6 Contribution & Discussion

Contribution: the two papers show empirical patterns of the exchange rate and its relation with fundamentals from two angles, together giving a comprehensive view; both focus on the dramatic change before/after a major international event (IM 2019 the end of Bretton Woods in the early 1970s, Lilley 2019 the 2008 global financial crisis); both conduct thorough empirical analyses with several robustness tests. Discussion: IM 2019's reason for defining the peg period such that 1973:01–1973:02 are excluded (though all countries officially declared floating in 1973:02) isn't clear; many changes are not significant. Lilley 2019: the broad dollar (31.2) is evaluated with equal weights for the nine G10-except-US currencies, but equal weights don't necessarily capture the actual value fluctuation of the dollar (overweighting small currencies, underweighting large ones), and trade-based weights should be used to show robustness; "VXO" is the monthly log change in implied volatility on the S&P100, but it's unclear why the authors use the S&P100 instead of the S&P500 given they also use the S&P500 log return as another risk measure; they primarily use \(R^2\) to show the reconnect, but \(R^2\) may not be the best measure of the relation between a single regressor and the dependent variable.