References
Cortes, F., and L. Sanfilippo. 2020. “Do Multi-Sector Bond Funds Pose Risks to Emerging Markets?” IMF Working Paper No. 20/152, International Monetary Fund, Washington, DC.
Cremers, M., and A. Petajisto. 2009. “How Active is Your Fund Manager? A New Measure that Predicts Performance.” Review of Financial Studies, 22(9): 3329–65.
Miyajima, K., and I. Shim. 2014. “Asset Managers in Emerging Market Economies.” BIS Quarterly Review, September.
Annex 1. Regression Analysis: Methodology
The sample is extended through the second quarter of 2020 to include the COVID-19 shock and the ensuing overall emerging market flows bust and nascent recovery in hard currency.
To discuss the impact of multi-sector bond funds (MSBFs) reallocations on underlying markets—currencies, local currency bonds, and hard currency bonds—this note builds on the panel regression analysis first presented in Cortes and Sanfilippo (2020) (C&S)1. The sample here is extended through the second quarter of 2020, to include the COVID-19 shock. That said, it estimates the same parsimonious three-dimensional panel regressions with interaction effects. The two cross-section identifiers are (1) the MSBF and (2) the country it is invested in. The third dimension is time. The main regression is specified as follows:
yijt= xijt’β+αij+uijt ,
where the left-hand scale is the dependent variable for fund i in country j at time t, xijt’ is a vector of covariates, p is a vector of coefficients, aij are unobserved cross-section (fund-country) specific fixed effects2 and uijt represents time variant unobservables. The covariates include an interaction term. The motivation for the interaction term in the model is to test the hypothesis that the relationship between changes in MSBF holdings and emerging market currencies/bonds is different in a risk-on environment (overall emerging market inflows) versus a risk-off environment (overall emerging market outflows). The regressions account for cross-sectional dependence and heteroscedasticity.3 The variables included in the estimation are defined as follows:
Dependent variables
Performance Spreads. Performance is determined over the relevant emerging market benchmark and expressed in percent. A positive value indicates outperformance of the domestic asset compared to the benchmark, and vice versa for a negative value.
1. Currency Spread (CrncySP) = (Quarter-over-quarter change in local currency vs. US dollar) – (Quarter-over-quarter change in the J.P. Morgan Emerging Market Currency Index)
2. Local Currency Government Bond Index Spread (GBISP) = (Quarter-over-quarter change in domestic Local Currency Total Return Bond index) – (Quarter-over-quarter change in the total return overall on the J.P. Morgan Government Bond Index-Emerging Markets)
3. Hard Currency Government Bond Index Spread (EMBIGSP) = (Quarter-over-quarter change in domestic Hard Currency Total Return Bond index) – (Quarter-over-quarter change in the total return overall on the J.P. Morgan Emerging Market Bond Index)
Sources: Bloomberg and J.P. Morgan.
Explanatory variables
MSBFs Flows. Changes in MSBF emerging market allocations (dlog(msbf)) per fund per country are defined as the market value of the portfolio allocations at the end of each quarter, adjusted for price changes. To adjust portfolio allocations for the changes solely due to the changes of portfolio assets’ value, we assume that the asset returns derived from price changes are approximated by country index returns (GBI for local currency and EMBIG for hard currency).4 We use first difference log transformations of the (price-adjusted) dollar value to allow for a more intuitive interpretation. Furthermore, we differentiate between local currency holdings and hard currency holdings; the motivation is that changes in hard currency holdings should not, at least directly, impact the performance of the domestic currency, or local currency bonds. A positive relationship between changes in (price-adjusted) MSBF holdings and the dependent variable is expected. That is, with larger MSBF purchases (or sales) of assets, the domestic currency or bonds are expected to be associated with outperformance or underperformance versus their emerging market benchmark.
Risk Appetite. We create a dummy variable (DRA) that takes the value 1 in quarters when money is flowing into the emerging market space (risk-on), and 0 in periods of outflows. The values are determined using data on all dedicated emerging market bond funds (local, hard, and blend). In a period of emerging market portfolio inflows, the emerging market asset class as a whole performs well, so a strongly positive relation between the dummy and the dependent variable is expected. The risk appetite dummy is included as an interaction term to test if the impact of a change in MSBF holdings on emerging market currencies/bonds is different in a risk-on environment compared with a risk-off environment. In a so-called “risk-on” environment, the effect of changes in MSBFs holdings is expected to be more moderate. For instance, MSBF outflows could weigh less on a currency/bonds as capital inflows from other institutions or investors can compensate for MSBF outflows; or, along the same reasoning, local outperformance due to MSBF inflows could be dampened as capital inflows boost the emerging market asset class as a whole. Meanwhile in a “risk-off environment, the effect of changes in MSBF holdings is expected to be amplified. The impact of MSBF outflows could be exacerbated when all investors simultaneously are running for the exits, or inflows could lead to greater local outperformance when the emerging market benchmark is performing badly.
The following key results can be gleaned from the estimations (see table 2):
For the emerging market currency and local currency fixed income regressions, the results show the expected signs and are (statistically) significant (even more so than in C&S). There is a positive relation between changes in MSBF holdings and (i) currency and (ii) local currency bond performance. This implies that, in a period of outflows (DRA=0), an increase of MSBF inflows by 1 percent in an emerging market is associated with an improvement in the performance of the (i) currency and (ii) local currency bonds with an outperformance over the emerging market (i) EMCI, and (ii) GBI benchmark by (i) 1.65 percent (vs. 1.09 percent in C&S) and (ii) 1.87 percent (vs. 1.23 percent in C&S). In a period of inflows (DRA=1), this impact is moderated to (i) 0.30 percent (vs. 0.44 percent in C&S) and (ii) 0.47 percent (vs. 0.41 in C&S).
For the hard currency fixed income regression, coefficients are significant, but the sign of the MSBF flows shows a negative relation between changes in MSBF holdings and hard currency bond performance. This suggests that, in a period of outflows (DRA=0), an increase of MSBF inflows in an emerging country allocation of 1 percent is associated with underperformance over the emerging market EMBIG benchmark by –0.78 percent (vs. +0.17 percent in C&S). In a period of inflows (DRA=1), this impact is moderated to –0.06 percent (vs. +0.02 percent in C&S).
The opposing developments in local currency versus hard currency flows, and the proposed explanation are in line with the argument in the literature that posits that although borrowing in local currency from foreign lenders mitigates the currency mismatch for the borrower, it transfers the currency mismatch to lender’s balance sheets—a phenomenon dubbed “original sin redux” (Carstens and Shin 2019; BIS 2019). During periods of risk aversion, when local currencies weaken and domestic assets sell off, foreign investors typically reduce their exposure, triggering outflows. Even in the absence of outflows, increased foreign currency hedging could exert substantial pressure on the exchange rate and the cost of funding. Moreover, there is significant pro-cyclicality to flows as emerging market exchange rates play a key amplifying role in the portfolio adjustment of global investors—as local currency spreads and exchange rates move in lockstep, see Carsten and Shin 2019. So, while local currency issuance has its benefits, as it helps sovereigns develop their own local currency denominated rates markets, it has also certain risks, particularly when a large share of the investor base is made of opportunistic and concentrated foreign investors.
Estimation Results: Sensitivity of Currency and Bond Indices to Changes in MSBF Holdings
Estimation Results: Sensitivity of Currency and Bond Indices to Changes in MSBF Holdings
| Panel A. | Panel B. | Panel C. | |
|---|---|---|---|
| CrncySP | GBISP | EMBIGSP | |
| dlog(msbf) | 1.65 *** | 1.87 *** | -0.78 *** |
| (0.30) | (0.39) | (0.26) | |
| DRA | 0.06 | -0.76 | 0.14 |
| (0.42) | (0.57) | (0.13) | |
| dlog(msbf) x DRA | -1.35 *** | -1.41 *** | 0.73 *** |
| (0.44) | (0.55) | (0.30) | |
| Obs. | 4,668 | 4,339 | 19,769 |
| R2 | 0.205 | 0.163 | 0.049 |
Estimation Results: Sensitivity of Currency and Bond Indices to Changes in MSBF Holdings
| Panel A. | Panel B. | Panel C. | |
|---|---|---|---|
| CrncySP | GBISP | EMBIGSP | |
| dlog(msbf) | 1.65 *** | 1.87 *** | -0.78 *** |
| (0.30) | (0.39) | (0.26) | |
| DRA | 0.06 | -0.76 | 0.14 |
| (0.42) | (0.57) | (0.13) | |
| dlog(msbf) x DRA | -1.35 *** | -1.41 *** | 0.73 *** |
| (0.44) | (0.55) | (0.30) | |
| Obs. | 4,668 | 4,339 | 19,769 |
| R2 | 0.205 | 0.163 | 0.049 |
The authors are grateful for the useful comments and contributions from Antonio Garcia Pascual.
These arguably are still conservative estimates. Public accounts of MSBFs likely underestimate the true extent of their exposures. Reported regulatory data by MSBFs relates to their public or co-mingled (open-ended) funds in which fund managers have discretion over the underlying investments. MSBF families, however, do not just manage “discretionary” funds but are also responsible for so-called unregistered advisory (or “managed”) accounts, often run for large institutional investors. Moreover, the estimated foreign investor base also includes the foreign official sector, in addition to nonbanks and banks (see Arslanalp and Tsuda 2014). The former is typically less sensitive to market dynamics, so the relative importance, or put differently, the marginal market impact, of changes in MSBF behavior could be much larger than implied by exposure estimates.
To quantify the degree of active management (that is, deviation from benchmarks), the note applies a method introduced by Cremers and Petajisto (2009) and used by Miyajima and Shim (2014). This method determines the “active share” of a fund by comparing its holdings to those of its benchmark index, using the J.P. Morgan emerging markets debt indices—hard currency: the Emerging Markets Bond Global Diversified (EMBIG) Index, and local currency: Government Bond Index-Emerging Markets (GBI-EM) Global Diversified Index. The paper then compares the benchmark diversion metrics of MSBFs with EPFR emerging market-dedicated funds and EPFR global bond funds. As expected, MSBFs are found to be very active, on average, and do not adhere to the benchmark index. In contrast, EPFR emerging market-dedicated funds have a low active share, while EPFR global funds have a higher active share, although significantly lower than MSBFs.
Estimation results for the equivalent regression in C&S were (i) 1.1 and (ii) 1.2 percent. In a period of inflows, this impact was moderated to 0.4 percent in both cases.
Estimations account for cross-sectional dependence and heteroscedasticity.
An F-test between the pooled ordinary least squares and an (unrestricted) fixed effect model support panel fixed effects. The authors also conducted estimations including time-fixed effects as a robustness test. The inclusion of time-fixed effects, however, did not change materially the (signs or significance of) coefficients of the explanatory variables (results available on demand). At the same time, the time fixed effects (FE) were broadly insignificant. Hence, the authors present the estimation results without it.
Standard errors and covariances in the panel regressions are estimated using a (degree-of-freedom corrected) variant of the Panel Corrected Standard Error (PSCE) methodology (Beck and Katz 1995). The PSCE cross-section seemingly unrelated regression (SUR) equations method handles cross-section correlation (period clustering) by replacing the outer product of the cross-section residuals in the coefficient covariance estimator with an estimate of the (contemporaneous) cross-section residual covariance matrix.
The price-adjusted series is determined as follows:
(1) Determine the quarter-over-quarter growth rate in the market value of the MSBF position: ΔMVQ2–1 = ((MVQ2 / MVQ1) – 1);
(2) Subtract the quarter-over-quarter return in the relevant index from the change in MV: (ΔMVQ2–1 – RiQ2–1);
(3) Apply the return-adjusted growth rate to the MV in Q1. Such that the price-adjusted MV in Q2 becomes: MVQ2 = MVQ1 * (1+(ΔMVQ2–1 – RiQ2–1))
(4) Repeat in every quarter: MVQ3 = MVQ2 * (1+(ΔMVQ3–2 – RiQ3–2)); and so on.