Changes in the Global Investor Base and the Stability of Portfolio Flows to Emerging Markets1
Author: Mr. Luis Brandao Marques1, Mr. R. G Geloshttps://orcid.org/0000-0003-4362-676X, Hibiki Ichiue, and Ms. Hiroko Oura
  • 1 404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: lmarques@imf.org; ggelos@imf.org; hichiue@imf.org; houra@imf.org.
Publication Date:
28 Dec 2015
eISBN:
9781513555560
Language:
English
Keywords:
WP; investment company; hard currency; crossover fund; country flow data; fund manager; global bond; fund's mandate; fund heterogeneity
Download PDF (3.9 MB)

An analysis of mutual-fund-level flow data into EM bond and equity markets confirms that different types of funds behave differently. Bond funds are more sensitive to global factors and engage more in return chasing than equity funds. Flows from retail, open-end, and offshore funds are more volatile. Global funds are more stable in their EM investments than “dedicated” EM funds. Differences in the stability of flows from ultimate investors play a key role in explaining these patterns. The changing mix of global investors over the past 15 year has probably made portfolio flows to EMs more sensitive to global financial conditions.

Abstract

An analysis of mutual-fund-level flow data into EM bond and equity markets confirms that different types of funds behave differently. Bond funds are more sensitive to global factors and engage more in return chasing than equity funds. Flows from retail, open-end, and offshore funds are more volatile. Global funds are more stable in their EM investments than “dedicated” EM funds. Differences in the stability of flows from ultimate investors play a key role in explaining these patterns. The changing mix of global investors over the past 15 year has probably made portfolio flows to EMs more sensitive to global financial conditions.

I. Introduction

The events in emerging markets (EMs), following the “tapering” talk around mid 2013, and subsequent market jitters, served as a reminder that capital flows to and asset prices in EMs are still subject to substantial volatility. For many, this came as a surprise, given that compared to the 1990s, EM financial assets have matured and are now considered an established asset class. Indeed, the landscape of portfolio investment in EM economies has evolved considerably since the late 1990s in terms of investment opportunities and the global investor base.

EM financial markets have become deeper and more interconnected. In the 1990s, EM investment largely meant equity investing through dedicated EM funds (Bekaert and Harvey, 2013). This has changed substantially in the 2000s, and the role of bond flows has grown strongly. The development of local-currency bond markets has meant that many foreign investors participate directly in local markets. Concomitantly, several EMs have overcome the “original sin” problem—the inability for EMs to borrow from foreigners in their own currencies.

The mix of global portfolio investors has also changed. More money around the globe is intermediated by mutual funds, reaching about USD 30 trillion in 2013 (Investment Company Institute, ICI, 2014). Unlike other types of investors, such as pension funds and insurance companies, many of these funds are open ended. End investors can easily and quickly redeem their investments, forcing funds to sell underlying assets.2 Moreover, since the late 2000s, a rising share of EM investment is channeled through exchange traded funds (ETFs).3 The importance of so-called crossover investors (whose portfolio includes both developed market (DMs) and EM assets) has also risen considerably. Insurance companies and pension funds (with about USD 50 trillion assets globally, IMF, 2014a) remain important EM investors, though their relative size has declined somewhat compared to the faster-growing mutual fund industry (IMF, 2011). More recently, many sovereign wealth funds have expanded their portfolio to include EM assets.

Different types of mutual funds, pension funds, and insurance companies are likely to behave differently owing to distinctive mandates, constraints, EM expertise, and incentives. For instance, dedicated EM funds are constrained to invest only in EMs and are more likely to have better EM expertise than crossover funds. Institutional investors—defined in this paper as including among others pension funds, insurance companies, and sovereign wealth funds (but excluding mutual funds or ETFs)—usually are not subject to potential rapid redemptions from their clients.4 However, institutional investors are often required to invest in assets with investment-grade ratings and pursue buy-and-hold strategies. In fact, there is anecdotal evidence that different investors behaved differently during the EM sell-off episode in 2013 and early 2014.5

Against this backdrop, this paper tries to assess the effects that these changes in the composition of mutual fund investors imply for the volatility of capital flow to EMs. We do this by systematically investigating the behavior of different types of mutual funds active in these markets.6 To this end, we first establish how different types of mutual funds react to global financial shocks. Next, we attempt to identify the types of funds that are more likely to engage in return chasing, a behavior that can be destabilizing. Finally, we look into some of the drivers of the differences. In particular, we investigate the role of flows from end investors, and contrast it with the role of portfolio managers.

Our approach sheds new light on key aspects of capital flows to EMs. Most of the existing literature focuses on total capital flows and their aggregated sub components. Although that type of analysis is useful to identify macroeconomic push- and pull factors, it does not allow to assess the impact of structural changes in the investor composition. Our study is one of the few papers that examine the behavior of investors in emerging markets at the micro level.7 For this effect, the study uses monthly data on roughly 1,100 equity funds, from February 1996 to September 2013, and 200 bond funds, from November 2003 to September 2013.

Five key findings stand out. First, bond funds are substantially more sensitive to global financial shocks and engage more strongly in momentum trading (that is, they are more sensitive to past returns) than equity funds. Second, the relatively newer local-currency bond funds do not behave very differently from foreign-currency bond funds. Third, flows from retail-oriented funds, open-end funds and, contrary to conventional wisdom, dedicated EM funds are more sensitive to global financial shocks. The sensitivity of portfolio flows also differs by fund size, domicile, and style. Fourth, momentum trading is more prevalent among funds that also show a high sensitivity to global financial conditions. Finally, differences in volatility of flows from end investors play a key role in explaining these patterns.8

Overall, these results suggest that the rising share of bond flows in recent years may have made total portfolio flows to EMs more sensitive to global financial conditions, and more procyclical.

The rest of this paper is organized as follows. Section II describes the data and overviews fund flows by their types. Section III describes the regression specifications and presents the results. Section IV concludes.

II. Overview of Portfolio Flows from Mutual Funds

A. Data Description

Our mutual fund data source is EPFR Global, with a coverage of US$22 trillion in total assets as of the end of 2013.9 The database covers a very large fraction of U.S. and European investment funds, among others, and provides their basic characteristics such as investment styles, domiciles, benchmarks, and geographic focus.10 We use EPFR Global’s monthly data with information at the fund level, including on assets under management (AUM), inflows and outflows (redemptions), and asset allocation by country. Based on this, we estimate the flow from each fund to each country, adjusting for valuation effects.

The data cover a broad range of EMs and a relatively long time horizon. Our sample includes current and former EMs since many countries currently considered advanced economies were classified as EMs earlier in our sample period (Table 1).11 The data contain around 74,000 fund-country-month observations from November 2003 to September 2013 for bond funds and around 470,000 observations from February 1996 to September 2013 for equity funds.12 Tables 2 and 3 show the distribution of the data by fund type and domicile.

Table 1.

Numbers of Observations by Investment Destination Economy

article image
Note: The table shows the number of funds, the average number of months with data per fund, and the total number of observations (number of funds × average number of months), country by country. The sample period is February 1996—September 2013 for equity funds, and November 2003—September 2013 for bond funds, but actual coverages vary by fund.
Table 2.

Numbers of Observations by Fund Characteristics

article image
Note: The table shows the number of observations (number of funds × average number of months × number of EMs) by type of fund. The sample period is February 1996—September 2013 for equity funds, and November 2003—September 2013 for bond funds, but actual coverages vary by fund and EM. See the Appendix for a definition of mutual fund characteristics.
Table 3.

Numbers of Observations by Fund Domicile

article image
Note: Number of fund-month observations (number of funds × average number of months × number of EMs) by domicile of fund. The shadows correspond to offshore market economies. “Other Europe” does not include the United Kingdom and economies categorized in the offshore market economies. “Other domiciles” includes Australia, Canada, Japan, and the United Arab Emirates. The sample period is February 1996—September 2013 for equity funds, and November 2003—September 2013 for bond funds, but actual coverages vary by fund and EM.

B. Stylized Facts: Changes of Emerging Markets Portfolio Flow Structures

Since the early 2000s, bond flows have grown faster than equity flows. Between 2003 and 2013, on average, annual bond flows to EMs amounted to 12.5 percent of AUM of mutual funds included in the sample, compared to 5.1 percent for equity flows (Figure 1).

Figure 1.
Figure 1.

Cumulative Bond and Equity Flows to Emerging Markets

(1996–2013)

Citation: IMF Working Papers 2015, 277; 10.5089/9781513555560.001.A001

Sources: EPFR Global and authors’ calculations.Note: The lines in the figure represent the log differences of the gross cumulative flows of bond and equity funds to all EPFR-defined emerging markets economies between period t and end 2010, multiplied by 100. EPFR estimates these aggregate flows by taking the product of sample funds’ country allocation weights and end investors’ flows to funds. Cumulative flows are calculated by chaining the flow-to-assets under management ratio in order to adjust the effects from the expansion of the database to cover more funds over time. For the same reason, we use cumulative flows rather than assets under management to see the change of the relative size of various mutual fund segments.

The relative importance of various types of mutual funds and their end investors also has changed. Figure 2 shows that portfolio flows to EMs are increasingly intermediated through open-end funds, passive ETFs, crossover funds, and local-currency bond funds. In particular:

  • The role of closed-end funds relative to that of open-end funds has declined. The growth of (open-end) ETFs, which often are more fee- and tax effective than mutual funds for end investors, is partly behind this trend.

  • The relative importance of passively managed funds (index funds, which track indices) has grown. This again is closely related to the rapid growth of the ETF segment since most of the ETFs are currently managed passively.13

  • The role of global funds (crossover funds) that invest both in EMs and DMs has grown relative to dedicated EM regional funds. Moreover, among dedicated EM funds, global EM funds (which invest in EMs around the world) have grown more rapidly than regional EM funds.

  • Local-currency bond funds have expanded more rapidly than hard-currency bond funds. This is partly the result of efforts by several EMs to overcome the “original sin” problem.

  • Mutual funds have been sold mainly to retail investors, but institutional investors have been purchasing an increasing number of mutual fund shares. Some of this reflects the expansion of defined contribution plans: in the United States, the percentage of defined contribution plans’ assets in mutual funds has risen from 20 percent in 1993 to 60 percent in 2013 (ICI, 2014).14

  • Since 2009, flows to EMs from U.S. domiciled funds have grown relative to those from offshore funds.

Figure 2.
Figure 2.
Figure 2.

Cumulative Portfolio Flows to Emerging Markets by Fund Type: End Investors

(1996–2013)

Citation: IMF Working Papers 2015, 277; 10.5089/9781513555560.001.A001

Sources: EPFR Global and authors’ calculations.Note: The lines in the figure represent the log differences of the gross cumulative flows of bond and equity funds to all EPFR-defined emerging markets economies between period t and end 2010, multiplied by 100. EPFR estimates these aggregate flows by taking the product of sample funds’ country allocation weights and end investors’ flows to funds. Cumulative flows are calculated by chaining the flow-to-assets under management ratio in order to adjust the effects from the expansion of the database to cover more funds over time. For the same reason, we use cumulative flows rather than assets under management to see the change of the relative size of various mutual fund segments. In our database, there are only a few actively managed exchange traded funds and most of the passively managed funds are exchange traded funds, therefore, we bundle passively managed funds and exchange traded funds together in the figure.

Figures 1–2 also show that not all types of fund flows are equally sensitive to global financial shocks. For one, bond flows have been more volatile than equity flows. For instance, bond outflows during the first six months after September 2008 (the collapse of Lehmann Brothers) were 25.5 percent of total AUM, while equity outflows were only 5.1 percent. This pattern was repeated in other episodes, including the one around mid-2013, when the Federal Reserve hinted at the possibility of tapering its asset purchases. In addition, active funds (excluding ETFs), open-end funds, EM regional funds, U.S. domiciled funds, and local currency bond funds seem to be more sensitive to changes in global financial conditions, while passive ETFs, closed-end funds, global or global EM funds, offshore funds, and hard currency bond funds are less sensitive to these changes.

Finally, during some distress episodes, equity flows from institutional investors were more resilient than those from retail investors.15 This is also true for bond funds during most crisis episodes, except for the global financial crisis. These observations are in line with the notion that institutional investors do not change their investment strategies frequently and are therefore less sensitive to short-term market fluctuations.

III. Regressions and Results

In this section, we examine more systematically the sensitivity of country flows to global financial conditions and past returns, by estimating fund-level panel regressions. Since the literature suggests that bond and equity flows behave differently, we run separate regressions for bond and equity funds.16 In a second step, we focus on differences in sensitivities to global shocks across different types of bond- and equity funds.

A. Behavior of Bond and Equity Funds

What, if any, are the differences in the drivers of flows between bond and equity funds? We explore this question by estimating the following model with monthly data:

Flowi,j,t=αi,j+β1⋅Globalt+β2⋅Returni,j,t−1+β3⋅ICRGj,t+β4⋅RIDj,t−1+ɛi,j,t,(1)

where Flowi,j,t is the ratio of the capital flow from fund i to country j to the AUM of fund i (see Appendix). The first explanatory variable, Globalt, is a factor representing global financial conditions. The expected sign of the coefficient β1 is negative because an increase in a global factor (a deterioration of global financial conditions) is likely to result in more capital outflows. Returni,j,t-1 is the lagged return of fund j in country i, and signals return chasing, which could lead to more procyclical and volatile capital flows.17 To identify this effect more clearly, we separate the impact of the global factor from that of lagged country returns by using the residual of a regression of the latter on the global factor. ICRGj,t is the first difference of the ICRG (International Credit Risk Guide) composite country risk rating and is a proxy for changes in local macroeconomic conditions.18 Finally, RIDj,t-1 is the real interest rate differential relative to the 1-month U.S. Eurodollar deposit rate and tries to capture search-for-yield by global investors. The model also includes country-fund fixed effects αi,j. A detailed description of the variables is provided in the Appendix.

We estimate the model using several global factors. We first use the VIX (Chicago Board Options Exchange Market Volatility Index), the implied volatility index for the S&P 500.

Although commonly used (see Forbes et al. 2012, Bruno and Shin, 2012, and Forbes and Warnock, 2012), the VIX is an imperfect measure of global risk. For instance, the VIX remained at a historically low level following the “tapering hint” of U.S. Fed in early 2013. Given these shortcomings, as a robustness check, we use three other variables to capture different dimensions of global risk: the TED spread, a measure of market liquidity; the Merrill Option Volatility Expectations Index (MOVE), which captures uncertainty about future U.S. Treasury rates; and the volatility of Eurodollar futures as a proxy for near-term uncertainty about U.S. monetary policy (see also IMF 2014b). These global factors are positively correlated, but the correlations are not all very large, and are not constant over time (Figure 3 and Table 4).

Figure 3.
Figure 3.

Global Factors

(1996–2013)

Citation: IMF Working Papers 2015, 277; 10.5089/9781513555560.001.A001

Sources: Thomson Reuters Datastream and Federal Reserve Board.Note: VIX is the Chicago Board Options Exchange Market Volatility Index. TED spread is the three-month Eurodollar deposit rate minus the three-month U.S. Treasury bill rate. MOVE is the Merrill Option Volatility Expectations Index.
Table 4.

Correlation Coefficients across Global Factors

(1996–2013)

article image
Note: The correlation coefficients are computed using monthly observations from February 1996 to September 2013. VIX is the Chicago Board Options Exchange Market Volatility Index. TED is the three-month Eurodollar deposit rate minus the three-month U.S. Treasury bill rate. MOVE is the Merrill Option Volatility Expectations Index.

The results from the baseline estimation shown in Table 5 suggest that bond funds are much more sensitive to global financial shocks and engage in “return chasing” more strongly than equity funds. The coefficients on the VIX and the index return are statistically significant in all regressions. However, the estimated coefficients of the regressions for bond funds are much larger than those in the regression for equity funds (in absolute value), and the difference is statistically significant. This result is robust when a common sample is used for the regressions of equity and bond funds. The results are also robust to the use of alternative global factors (Table 6). In addition, although not reported here, very similar results are obtained when adding a measure of financial openness (Chinn and Ito 2006, Lane and Milesi-Ferretti 2007). In line with our estimation results, the bond flows from mutual funds dropped more strongly than equity flows during the market turbulence episode triggered by the uncertainty about U.S. monetary policy in mid-2013 (Figure 1).

Table 5.

Base Estimation Results: Sensitivity of Bond and Equity Flows vis-à-vis the Global Factor by Type of Funds

(VIX as a Global Factor)

article image
Note: “Full” corresponds to the regressions using all available observations (November 2003-September 2013 for bond funds and March 1996-September 2013 for equity funds). “Common” corresponds to the regression for equity funds using the common coverage of countries and observation period with bond funds. The p-values reported in brackets are calculated using Driscoll and Kraay’s (1998) robust standard errors. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. VIX is the Chicago Board Options Exchange Market Volatility Index. Return is the country index return. ICRG is the first difference of the ICRG (International Credit Risk Guide) composite country risk rating. RID is the real interest rate differential relative to the 1-month U.S. Eurodollar deposit rate. See the Appendix for more detail.
Table 6.

Robustness Test: Sensitivity of Bond and Equity Flows vis-à-vis the Global Factor

(Alternative Global Factors)

article image
Note: The p-values reported in brackets are calculated using Driscoll and Kraay’s (1998) robust standard errors. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Global is the global factor. TED spread is the three-month Eurodollar deposit rate minus the three-month U.S. Treasury bill rate. MOVE is the Merrill Option Volatility Expectations Index. Return is the country index return. ICRG (International Credit Risk Guide) is the first difference of the ICRG composite country risk rating. RID is the real interest rate differential relative to the 1-month U.S. Eurodollar deposit rate. See the Appendix for more detail.

The estimated coefficients are economically significant. For instance, the estimates imply that an increase in the VIX similar to that of the last four months of 2008—roughly 50 percent—leads monthly bond flows to drop by 7.5 percent of AUM. This compares to average monthly bond flows of one percent of AUM, between 2003 and 2013.

Given that the share of bond funds is rising, the estimation results imply that portfolio flows to EMs are likely to become more sensitive to global financial shocks. This result has considerable implications for capital flows as a whole for the following reasons. First, portfolio flows, as well as banking flows, are more volatile than foreign direct investment and have played important roles in the volatility of total capital flows. Second, the share of portfolio flows has risen while that of banking flows has declined in recent years, partly because the deleveraging at European banks. In fact, the surge of capital flows to EMs since the global financial crisis has largely been led by portfolio bond flows.19

B. Portfolio Flows and Other Fund Characteristics

Next, we investigate whether different types of funds respond differently to changes in global factors by estimating a regression similar to (1), augmented by one fund characteristic at a time:

Flowi,j,t=αi,j+β1⋅Globalt+β2.Charai,t−1⋅Globalt+β3.Charai,t−1+β4⋅Returni,j,t−1+β5⋅ICRGj,t+β6⋅RIDj,t−1+ɛi,j,t,(2)

where Charai,t−1 is a dummy variable that takes the value of one when fund i has a certain characteristic of interest and zero otherwise. We use the interaction term of Charai,t−1 and Globalt to examine the different sensitivities of portfolio flows to the global factor across types of funds. If a certain type of fund has a higher sensitivity to global financial conditions than all other types of funds (reference group), we would expect the coefficient β2 to be negative.20

A potential drawback of specification (2) is that it does not account for potentially correlated fund characteristics. For instance, crossover funds tend to be larger than other funds and this characteristic could merely reflect a size effect. Therefore, we also estimate a model including multiple fund characteristic dummies simultaneously as follows:

Flowi,j,t=αi,j+β1⋅Globalt+∑kβ2,k.Charak,i,t−1⋅Globalt+∑kβ3,k.Charak,i,t−1+β4⋅Returni,j,t−1+β5⋅ICRGj,t+β6⋅RIDj,t−1+ɛi,j,t.(3)

where Charak,i,t-1 is a dummy variable that takes one when fund i is of type k, and zero otherwise. For equity funds in model (3), the reference group consists of funds that are medium sized, sold to retail investors, open-end, actively managed, not ETFs, domiciled in the United States, and with a global EM focus. For bond funds in model (3), the reference group is further restricted to hard-currency bond funds. However, the geographic-focus and currency-focus dummies are highly correlated, causing a multicollinearity problem.21 Thus, we estimate two variations of the model (3) with multiple fund characteristic dummies for bond funds; one without currency focus dummies, and another without geographic focus dummies.

The analysis reveals interesting differences across types of funds. Table 7 reports the estimated coefficients of the interaction term between the VIX and fund characteristics (results for models (2) and (3) are in the “Single” and “Multiple” columns, respectively). A negative sign indicates that flows from funds with certain characteristic in one type category (for instance, global funds) but otherwise with the characteristics of the reference group decline more than those from funds in the reference group when the VIX increases. Many estimated coefficients are statistically and economically significant and the results are generally robust to alternative global factors (see Appendix Tables 1, 2, and 3).22 We can summarize the main findings as follows.

Table 7.

Estimation Results: Sensitivity of Bond and Equity Flows vis-à-vis the Global Factor by Type of Funds

article image
Note: The table reports the estimated coefficients on the interactions between the VIX (the Chicago Board Options Exchange Market Volatility Index) and fund characteristic dummies. “Single” corresponds to the estimate of each regression with a single characteristic dummy, which is represented in (2). “Multiple” corresponds to the estimates of a regression with multiple characteristic dummies, which is represented in (3). The reference group for model (3) consists of medium-sized open-end funds (excluding exchange traded funds) sold to retail investors, actively managed, domiciled in the U.S., and investing in emerging markets globally (in hard currency in case of bond). This is why the rows for these characteristics in “multiple” columns are empty. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on Driscoll and Kraay’s (1998) robust standard errors.

  • Country flows from funds geared toward institutional investors tend to be less sensitive to changes in global risk aversion than those from retail-oriented funds. This is in line with the results reported in IMF (2014a) and consistent with institutional investors focusing on long-term performance and retail investors being more fickle.

  • As expected, closed-end fund flows are less sensitive to changes in global risk factors than open-end funds, as they are not subject to redemptions by end investors. Hence, the decline of closed-end funds relative to open-end funds over the past two decades has probably increased the sensitivity of EM capital flows to global financial conditions.

  • For bond funds, flows from active non-ETFs are more sensitive to global financial shocks than passive ETFs whereas for equity funds, there is no strong evidence that flows from ETFs react more to global financial shocks than other types of funds.23 Therefore, we do not find strong support for the claim that recent rise of ETFs has increased capital flow volatility or intensified EM sell-offs during 2013 and 2014 (see Financial Times, 2014b).

  • Somewhat surprisingly, global funds are more stable sources of capital flows than dedicated EM funds. This is contrary to the widely-held perception that crossover funds are a more volatile source of funding for EMs, because their fund managers can reallocate their portfolios completely away from EMs. This result holds even after controlling for fund size. It is possible that the behavior of end investors – which could differ for global and dedicated funds – is driving the result. This point is examined in Subsection D.

  • Portfolio flows from local currency bond funds do not seem to be more sensitive to global financial shocks than those of hard currency bond funds. In principle, investment in newer products, such as local currency EM debt, should be more sensitive to global factors than that in more seasoned products like hard currency EM bonds. The fact that investment opportunities in local currency bond markets are still limited to relatively more mature and larger EMs is a likely explanation for this result.

  • Small bond and equity fund flows are more sensitive to the VIX, in line with the notion that the presence of large investors can make smaller traders more aggressive in their selling (Corsetti and others 2004).

  • Flows of funds domiciled in the United Kingdom and the United States are less sensitive to changes in global factors than offshore funds and other European funds. This may reflect the fact that the VIX, the measure of global risk used in these regressions, is more closely related to economic conditions in the United Kingdom and in the United States. In that case, EM assets may provide an important diversification benefit for British and American investors.

The main results listed above hold when we allow for differences across fund types in the reaction to variables other than the global factor (Appendix Table 4). Model (4) includes interaction terms between one fund characteristic dummy and all the other explanatory variables.

Flowi,j,t=αi,j+β1⋅Globalt+β2⋅Charai,t−1⋅Globalt+β3⋅Charai,t−1+β4⋅Returni,j,t−1+β5⋅Charai,t−1⋅Returni,j,t−1+β6⋅ICRGj,t+β7⋅Charai,t−1⋅ICRGj,t+β8⋅RIDj,t−1+β9⋅Charai,t−1⋅RIDj,t−1+ɛi,j,t.(4)

The results are broadly comparable to those reported in Table 7.

C. Which Types of Funds Engage in Return Chasing?

Which funds have a stronger tendency to buy past winners and sell past losers? We use model (4) to investigate this question. A positive coefficient for the interaction term between the fund characteristic dummy and the country’s asset return would imply more return chasing (momentum trading) by those funds.

The results are reported in Table 8. They show that return chasing is often prevalent among funds that are relatively more sensitive to the VIX. For instance, return chasing is more prevalent among open-end funds than among closed-end funds, as well as among EM regional funds (compared to global or crossover funds). In addition, offshore funds and funds from continental Europe engage more in momentum trading than U.K. funds.

Table 8.

Estimation Results: The Extent of Return Chasing by Fund Type

article image
Note: The table reports the estimated coefficients on the interactions between fund characteristic dummies and the lagged excess index return of the regression model (4). ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on Driscoll and Kraay’s (1998) robust standard errors.

There are, however, important exceptions, especially among bond funds. For example, momentum trading seems to be more prevalent in bond funds with institutional investors’ money relative to retail funds. This result does not seem fully consistent with the result reported in IMF (2014a), which shows that institutional investors do not engage in momentum trading. The difference may lie in the source data for funds; the institutional investors reported in EPFR Global’s mutual fund database are not necessarily the same as those captured by the custodian data from Bank of New York Mellon (used in IMF, 2014a).24

D. Roles of End Investors and Fund Managers

Existing investments, their returns, and in- and outflows from mutual fund shareholders determine the size of investible resources for any given fund. The managers of the fund determine how to allocate those resources across assets, within the fund’s mandate. By looking at fund-level country flow data, we can discern the direct influence of ultimate investors and fund managers on the behavior of investment flows.

We apply the analysis of Raddatz and Schmukler (2012) to our data set. They essentially decompose the flow from each fund to each country as:

Flowi,j,t=Flowi,t+(Flowi,j,t−Flowi,t)(5)

where Flowit is the flow from ultimate investors to fund i (expressed as a the ratio to the total AUM of the fund). The first term of the equation corresponds to the contribution of ultimate investors, while the second is the contribution of managers. If the manager always allocates flows from ultimate investors to EMs to keep portfolio weights constant, Flowi,j,t = Flowi,t holds and the flow to each country Flowi,j,t can be explained only with the first term. On the other hand, if the manager changes allocation weights, the second term also plays a role. Raddatz and Schmukler (2012) calculate the share of the total variance of flows that can be attributed to each component for active, passive, equity, and bond funds. We calculate the standard deviation of the ratio of the flow from ultimate investors to the total AUM and the share of the total variance that is attributable to the flow from ultimate investors, for many varieties of funds.25

The results confirm that differences in the stability of flows from ultimate investors are behind the observed cross-fund heterogeneity of portfolio flows. In particular, funds that have more stable ultimate investor flows are generally less sensitive to global financial conditions and display a weaker tendency to engage in return chasing. For instance, when compared to all other funds, global bond funds are less sensitive to global factors (the estimated coefficient β2 in (2) is positive—0.150—and highly significant; see Table 7) and past returns (the estimated coefficient β5 in (4) is negative—0.025—and significant; see Table 8). Global bond funds also have less volatile end-investor flows (a standard deviation of 5 percent compared to over 7 percent for global EM funds and EM regional funds), and have a smaller fraction of their variance explained by such flows (only 17.1 percent Table 9).

Table 9.

Stability of End Investor Flows—Variance Decomposition

article image
Note: “Standard deviation” is the standard deviation of the ratio of the flow from ultimate investors to a fund to the total assets under management of the fund. “Share of variance” is the share of the total variance of the flows from funds to EMs that can be attributed to the flows from ultimate investors. The share is obtained by taking the variance of components corresponding to flows from ultimate investors and to allocation changes of fund managers at the country-fund level, averaging it across country-fund pairs, and dividing the averaged variance corresponding to flows from ultimate investors by the sum of the averaged variances (the covariance term is ignored). The standard deviations and variances are calculated only for country-fund pairs for which at least five time-series data are available. Here, small (large) funds are defined as funds that have been categorized as small (large) funds in more than a half of the available data.

IV. Conclusion

We have examined the behavior and determinants of mutual fund flows to emerging markets in the last two decades, against the backdrop of substantial changes in the landscape of portfolio investment in EM economies, both at the level of the global investor base and in local financial markets. To shed light on the impact of some of these structural changes, we have conducted a detailed examination of the behavior of different types of funds at the microeconomic level. In particular, we have investigated how different types of funds react to changes in global financial conditions, the extent to which they engage in return chasing, and the relative importance of end investors in driving this behavior.

We find that the “maturing” of the EM asset class seems to have made capital flows more volatile, more sensitive to global financial conditions, and more procyclical, despite much more solid macroeconomic fundamentals in EMs. In particular, bond funds, which have substantially gained in importance in the intermediation of funds to EMs, are relatively sensitive to global factors, and engage relatively more in momentum trading.

However, we do not find support for the widely held view that the growth of crossover funds, ETFs, and local currency bond funds are making EM portfolio flows more sensitive to global financial conditions. In part, the lower sensitivity of crossover funds seems to be attributable to lower fluctuations of flows from ultimate investors. Interestingly, funds also differ in their behavior depending on their domicile: those located in the United Kingdom and in the United States are less sensitive to changes in the VIX than funds domiciled elsewhere. These issues warrant further research.

Overall, given the observed differences in behavior across funds, from a recipient’s country perspective, knowing one’s investor base (including end investors) is important. Understanding and diversifying the investor base, possibly by working closely with asset managers, can help EM countries mitigate the volatility risk associated with international capital flows while reaping the benefits from financial globalization.

Appendix I. EPFR Global Mutual Fund Flows Data

EPFR Global covers about 11,000 equity funds and about 4,500 bond funds, with a combined US$22 trillion in total assets as of the end of 2013. According to EPFR Global, its data track more than 95 percent of EM-focused bond and equity funds. Most of them are mutual funds and ETFs, though the database includes limited number of hedge funds. The share of U.S. investment in EMs covered by EPFR Global is around 58 percent for equities, and more than 42 percent for bonds as of the end of 2012.26

The database covers primarily mutual funds, and does not cover all the investment fund flows nor capital flows intermediated directly by institutional investors, and therefore it may not provide the full picture of macro level portfolio flows. However, these coverage issues are of little concern for our analysis. This is because we focus on differences across different types of mutual funds to understand aggregate dynamics. Although funds not covered by EPFR Global may tend to investors that can behave differently, it is unlikely that their inclusion would affect our results.

EPFR Global’s high-frequency reporting (monthly for funds’ asset allocation by country as well as daily and weekly data for selected indicators from sub-samples of funds) is also an advantage compared to quarterly or annual data. Moreover, data recording methods of EPFR Global are better suited for our purpose than the Balance of Payments (BoP) or Coordinated Portfolio Investment Survey (CPIS). EPFR Global records transactions on nationality basis, while the BoP and CPIS report on residency basis.27 The CPIS also misclassifies some bond fund flows as equity flows (Felettigh and Monti, 2008).

A. Portfolio Flows to Each Country from Each Fund

Portfolio flows to each EM from each fund need to be calculated from asset allocation data, adjusting for their change owing to the changes of portfolio assets’ value. For this, we assume that the asset returns are well approximated by country index returns, following Gelos and Wei (2005). The flow-to-AUM ratio of fund i for country j at month t is calculated as:

Flowi,j,t=wi,j,tAi,t−wi,j,t−1Ai,t−1(1+rj,t)(wi,j,tAi,t−wi,j,t−1Ai,t−1)/2

where wi,j,t is the allocation weight at the end of month t, Ai,t is the AUM at the end of month t, and rj,t is the index return from t-1 to t. The estimated flow to a country from a fund in a month is normalized by the average of the assets allocated to the country at the beginning and end of the month. As indexes, we use the MSCI for equity funds, the GBI-EM for local currency bond funds, and the EMBI Global for other bond funds. Since the GBI-EM is a local currency index, we adjust its returns using bilateral exchange rates to obtain U.S. dollars-denominated returns.28

B. Explanatory Variables

Global. The VIX is the implied volatility index for the S&P 500 as reported by the Chicago Board Options Exchange. The TED spread is defined as the three-month Eurodollar deposit rate minus the three-month Treasury bill rate. The MOVE is the average implied volatility across a wide range of outstanding options on U.S. Treasuries. The volatility of Eurodollar futures is the realized volatility calculated using daily changes in the Thomson Reuter’s futures continuous series index of the 9th three-month futures in the corresponding month and annualized by multiplying with the square root of 250. We use the end of month data for the VIX, the TED spread, and the MOVE.

Return. The country index return is calculated as excess dollar returns of relevant indices (namely, bond index for bond flows and equity index for equity flows) over one month U.S. Eurodollar deposit rate. We use the MSCI for equity funds; the U.S. dollars-denominated GBI-EM for local currency bond funds; and the EMBI Global for other bond funds. We use one-month lagged return to mitigate the endogeneity concern. We also take three-month moving averages since it is plausible that flows respond to the past returns with a lag of several months (see, for instance, Jegadeesh and Titman, 1993). Since the individual country index return is somewhat correlated with the global factor, we use its orthogonal component relative to the VIX or the relevant global factor.

ICRG. The ICRG composite country risk rating is based on 22 variables, covering political, financial, and economic risks. A higher rating means a lower risk.

RID. The one-month-lagged real interest rate differential calculated as the difference between the local short term interest rate with a maturity of around one month and the U.S. short term interest rate minus the difference in the one-year-ahead forecasts of CPI inflation from Consensus Forecasts.

C. Definition of Mutual Fund Characteristics

We classify funds according to the following categories:

  • Fund size. We define large and small funds as those above the 80th and below 20th percentiles of AUM in each month, respectively, labeling the remainder as medium funds.

  • End investor type. Mutual funds have been sold mainly to retail investors, but institutional investors have been purchasing an increasing number of mutual fund shares, in part owing to the rise of defined contribution pension plans. EPFR Global provides share-class data for each fund: some shares are targeted to retail investors and the others are to institutional investors. Using these data, we identify whether a fund is sold only to institutional investors or is sold also to retail investors. Many funds do not report the types of ultimate investors for all of their shares. In such a case, we indentify a fund as a retail fund when some of its shares are known to be sold to retail investors. As a result, we limit the unidentified observations to around 7 percent of bond funds and around 14 percent of equity funds.

  • Open-end or closed-end. Investors can flexibly add to or redeem money from open-end funds, but not with closed-end funds.

  • Investment style (active or passive) and ETF. Passive funds usually replicate a given benchmark index. Fund managers of active funds exercise their judgment to over- or underweight certain assets compared to their benchmarks. Most ETFs are passively managed index funds while most non-ETFs (“mutual funds” in a narrow sense) are actively managed. We use four combined characteristics: active, non-ETF; active, ETF; passive, non-ETF; and passive ETF.

  • Fund domicile. Fund domicile roughly corresponds to the residence of ultimate mutual fund shareholders, though the correspondence is not necessarily accurate, especially for funds domiciled in offshore markets. We divide domiciles into five regions: the United States; offshore market economies; the United Kingdom; other European countries; and others. Table 3 reports the numbers of observations by domicile. The table also indicates the definition of offshore market economies (see IMF, 2008).

  • Geographic focus of the fund’s investment destination. We divide funds into four groups: global funds; global EM funds; EM regional funds; and advanced economy regional funds. For instance, global funds correspond to funds that are categorized as “Global” or “Global ex-US” by EPFR Global.

  • Currency focus (for bond funds only). We use five groups: hard currency; local currency; blend currency; unassigned; and other currency focus.

Table 1.

Robustness Test: Sensitivity of Bond and Equity Flows vis-à-vis the Global Factor by Type of Funds

(Ted Spread as the Global Factor)

article image
Note: The table reports the estimated coefficients on the interactions between the TED spread (the three-month Eurodollar deposit rate minus the three-month U.S. Treasury bill rate) and fund characteristic dummies. “Single” corresponds to the estimate of each regression with a single characteristic dummy, which is represented in (2). “Multiple” corresponds to the estimates of a regression with multiple characteristic dummies, which is represented in (3). The reference group for model (3) consists of medium-sized open-end funds (excluding exchange traded funds) sold to retail investors, actively managed, domiciled in the U.S., and investing in emerging markets globally (in hard currency in case of bond). This is why the rows for these characteristics in “multiple” columns are empty. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on Driscoll and Kraay’s (1998) robust standard errors.
Table 2.

Robustness Test: Sensitivity of Bond and Equity Flows vis-à-vis the Global Factor by Type of Funds

(MOVE Index as the Global Factor)

article image
Note: The table reports the estimated coefficients on the interactions between the MOVE (the Merrill Option Volatility Expectations Index) and fund characteristic dummies. “Single” corresponds to the estimate of each regression with a single characteristic dummy, which is represented in (2). “Multiple” corresponds to the estimates of a regression with multiple characteristic dummies, which is represented in (3). The reference group for model (3) consists of medium-sized open-end funds (excluding exchange traded funds) sold to retail investors, actively managed, domiciled in the U.S., and investing in emerging markets globally (in hard currency in case of bond). This is why the rows for these characteristics in “multiple” columns are empty. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on Driscoll and Kraay’s (1998) robust standard errors.
Table 3.

Robustness Test: Sensitivity of Bond and Equity Flows vis-à-vis the Global Factor by Type of Funds

(Volatility of Eurodollar Futures as the Global Factor)

article image
Note: The table reports the estimated coefficients on the interactions between the volatility of Eurodollar futures and fund characteristic dummies. “Single” corresponds to the estimate of each regression with a single characteristic dummy, which is represented in (2). “Multiple” corresponds to the estimates of a regression with multiple characteristic dummies, which is represented in (3). The reference group for model (3) consists of medium-sized open-end funds (excluding exchange traded funds) sold to retail investors, actively managed, domiciled in the U.S., and investing in emerging markets globally (in hard currency in case of bond). This is why the rows for these characteristics in “multiple” columns are empty. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on Driscoll and Kraay’s (1998) robust standard errors.
Table 4.

Robustness Test: Sensitivity of Bond and Equity Flows vis-à-vis the Global Factor by Type of Funds Using the VIX as the Global Factor and with Interactions with Other Factors

article image
Note: This table reports the estimated coefficients on the interactions between the VIX (the Chicago Board Options Exchange Market Volatility Index) and one fund characteristic dummy of the regression model (4), which includes additional interaction terms between the fund characteristic dummy and variables other than the VIX. As in model (2), only one fund characteristic dummy is included in the model at a time. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively. Inference is based on Driscoll and Kraay’s (1998) robust standard errors.

References

  • Bekaert, Greet, and Campbell Harvey, 2013, “Emerging Equity Markets in a Globalizing World,” manuscript.

  • Bollen, Nicolas, and Jeffrey Busse, 2004, “Short-Term Persistence in Mutual Fund Performance,” Review of Financial Studies, Vol. 18, No. 2.

    • Search Google Scholar
    • Export Citation

  • Broner, Fernando A., Gaston Gelos, and Carmen M. Reinhart, 2006. “When in Peril, Retrench: Testing the Portfolio Channel of Contagion,” Journal of International Economics, Vol. 69, No. 1, pp. 203–30.

    • Search Google Scholar
    • Export Citation

  • Chan-Lau, Jorge, and Li Lian Ong, 2005 “U.S. Mutual Fund Retail Investors in International Equity Markets: Is the Tail Wagging the Dog?” International Economics and Finance Journal, Vol. 1, No.1, pp. 73–88.

    • Search Google Scholar
    • Export Citation

  • Chinn, Menzie D., and Hiro Ito, 2006, “What Matters for Financial Development?” Journal of Development Economics, Vol. 81, No.1, pp. 163–92.

    • Search Google Scholar
    • Export Citation

  • Corsetti, Giancarlo, Amil Dasgupta, Stephen Morris, and Hyun S. Shin, 2004, “Does One Soros Make A Difference? A Theory of Currency Crises With Large and Small Traders,” Review of Economic Studies, Vol. 71, No.1, pp. 87–114.

    • Search Google Scholar
    • Export Citation

  • Driscoll, John C., and Aart C. Kraay, 1998, “Consistent Covariance Matrix Estimation with Spatially Dependent Panel Data,” Review of Economics and Statistics Vol. 80, No. 4, pp. 549–560.

    • Search Google Scholar
    • Export Citation

  • Edelen, Roger, and Jerold Warner, 2001, “Aggregate Price Effects of Institutional Trading: A Study of Mutual Fund Flow and Market Returns,” Journal of Financial Economics, Vol. 59, Issue 2, pp. 195–200.

    • Search Google Scholar
    • Export Citation

  • Felettigh, Alberto, and Paola Monti, 2008. “How to Interpret the CPIS Data on the Distribution of Foreign Portfolio Assets in the Presence of Sizeable Cross-border Positions in Mutual Funds. Evidence for Italy and the Main Euro-area Countries,” Questioni di Economia e Finanza (Occasional Papers) 16, Bank of Italy, Economic Research and International Relations Area.

    • Search Google Scholar
    • Export Citation

  • Financial Times, 2014a, “Investors Pull $12 bn from EM Stock Funds,” January 31.

  • Financial Times, 2014b, “EMs are Paying the Price of ETF Liquidity,” February 12.

  • Forbes, Kristin J., Marcel Fratzscher, Thomas Kostka, and Roland Straub, 2012, “Bubble Thy Neighbor: Direct and Spillover Effects of Capital Controls,” NBER Working Paper, No.18052.

    • Search Google Scholar
    • Export Citation

  • Forbes, Kristin J., Marcel Fratzscher, Thomas Kostka, and Roland Straub, and Francis Warnock, 2012, “Capital Flow Waves: Surges, Stops, Flight and Retrenchment,” Journal of International Economics, Vol. 88, No. 2, pp. 235–251.

    • Search Google Scholar
    • Export Citation

  • Fratzscher, Marcel, 2012, “Capital Flows, Push Versus Pull Factors and the Global Financial Crisis,” Journal of International Economics, Vol. 88, No. 2, pp. 341–56.

    • Search Google Scholar
    • Export Citation

  • Frazzini, A, and Owen Lamont, 2008, “Dumb Money: Mutual Fund Flows and the Cross-Section of Stock Returns,” Journal of Financial Economics, Vol. 88, Issue. 2, pp. 299-322.

    • Search Google Scholar
    • Export Citation

  • Gelos, Gaston, 2011, “International Mutual Funds, Capital Flow Volatility, and Contagion–A Survey,” IMF Working Paper, No.11/92.

  • Gelos, Gaston, and Shang-Jin Wei, 2005, “Transparency and International Portfolio Holdings,” Journal of Finance, Vol. 60, No. 6, pp. 2987–3020.