Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Caitlin Dannhauser is at the Villanova School of Business, Villanova University, 800 Lancaster Avenue, Villanova, PA
19085. She can be reached at [email protected]. Phone: 610‐519‐4348. Saeid Hoseinzade is at the
Sawyer Business School, Suffolk University, 120 Tremont Street, Boston, MA 02108. He can be reached at Email:
[email protected] . We thank Itay Goldstein (discussant), Antti Petajisto (discussant), Pierluigi Balduzzi,
Jeffrey Pontiff, Jonathan Reuter, Ronnie Sadka, Hassan Tehranian, and participants at the 2017 ICI and Darden
Academic and Practitioner Symposium on Mutual Funds and ETFs, the Second Annual “Philly Five” Finance
Symposium, and Suffolk University Seminar Series for helpful comments.
The Transformation of Corporate Bond Investors and Fragility:
Evidence on Mutual Funds and ETFs
Caitlin Dannhauser
Villanova University
Saeid Hoseinzade
Suffolk University
The liquidity transformation of mutual funds and exchange‐traded funds (ETFs) is examined as
a potential source of corporate bond market fragility. Using differential exposure among bonds
from the same issuer to plausibly exogenous outflows, ETFs cause temporary flow‐driven yield
pressure that reverts after seven months. Outflows from active and index mutual funds have no
impact. The differential effects are attributed to short‐term positive feedback traders’ presence in
ETFs and delayed arbitrage propagating shocks. A one‐month effect is found for active mutual
funds in the lowest quartile of liquid assets, suggesting their significant liquidity buffers may be
insufficient in prolonged shocks.
1
1. Introduction
Fragility resulting from the unintended consequences of financial innovation is a recurring
theme in market history (Gennaioli, Shleifer, and Vishny, 2012).1 With a heightened interest in
vulnerabilities created by various non‐bank financial institutions, Schmidt, Timmermann, and
Wermers (2016) observe that pooled vehicles for which the liquidity mismatch becomes magnified in
times of turmoil are particularly susceptible to run‐like behavior and its consequences. A more recent,
yet prevalent investment theme is the transformation of the systemically important corporate bond
market. Between 2008 and 2016 the amount of corporate debt outstanding grew 56% according to the
Securities Industry and Financial Markets Association (SIFMA). Over the same period, Federal
Reserve Board statistics reveal that broker‐dealer corporate and foreign bond holdings decreased 78%,
while those of nontraditional investors, exchange‐traded funds (ETFs) and open‐end mutual funds,
more than tripled. With combined holdings nearly matching the levels of traditional long‐term bond
investors, insurance companies, historically familiar concerns regarding the intraday and daily
liquidity provisions of ETFs and mutual funds relative to illiquid corporate bonds have emerged.2 For
example, the 2015 Financial Stability Oversight Council annual report lists the expansion of these
vehicles as a potential threat and Bill Gross warned of a mass exodus through ETFs.3 Conversely, such
risks are deemed limited by the Investment Company Institute due to mutual funds use of a liquidity
buffer and by ETF sponsors due to structural features that reduce their interaction with constituent
bonds.4,5 Given the importance of corporate debt to firms and the growing relevance of these
investment vehicles to investors’ portfolios, this paper contributes to the literature by cleanly
identifying the asset price implications of flows from corporate bond ETFs and mutual funds during
a period of turmoil.
1 Collateralized mortgage obligations in the 1980s and 1990s, mortgage backed securities in the 2000s, and money market funds in
2008 motivate the theory that excessive issuance of innovations sparked by virtues of diversification, tranching, or insurance prior to
the revelation of the securities’ vulnerability to unattended risk results in market fragility. 2 Bloomberg characterize the change in the post‐crisis bond market model as a transition of power from banks to mutual funds,
hedge funds, and exchange‐traded funds. They state that regulations have paved the way for the buy‐side “to exert more influence
than ever on markets.” See: https://www.bloomberg.com/news/features/2016‐08‐15/the‐rise‐of‐the‐buy‐side 3 “The obvious risk—perhaps better labeled the ‘liquidity illusion’—is that all investors cannot fit through a narrow exit at the same
time,” Bill Gross. See: https://www.wsj.com/articles/goldman‐sachs‐joins‐bond‐etf‐party‐1496925000 4 ICI 2016 Factbook states “There are many reasons to believe [concerns that outflows to bond funds could pose challenges for fixed‐
income markets] are overstated.” See: http://www.icifactbook.org/deployedfiles/FactBook/Site%20Properties/pdf/2016_factbook.pdf 5 Bill McNabb, CEO of Vanguard, states “This discovery that most ETF share trading does not lead to any activity in the ETF portfolio
means the impact of ETFs — and the possibility of disruption or volatility — on the primary market is limited.” See:
https://www.ft.com/content/53054716‐e6e9‐11e6‐893c‐082c54a7f539
2
Although ETFs and mutual funds are both pooled investment vehicles that provide diversified
exposure to the difficult to access over the counter (OTC) underlying market, their distinguishing
features make the mechanism through which outflows from the funds can impact the pricing of the
underlying distinct. For mutual funds, investors can redeem shares for cash at the end of day net asset
value (NAV) leaving the remaining investors to bear the consequences of another’s outflow, thus
creating the potential for runs induced by strategic complementarities. Goldstein, Jiang, and Ng (2017)
find that active corporate bond mutual funds are particularly susceptible due to a concave flow to
return relationship. The authors conclude that for these funds to be a major systemic concern evidence
of an effect on market prices and potentially real economic activity is needed.
The distinguishing features of ETFs mitigate concerns of similar runs because the redeeming
shareholder bears the cost of his own actions. Nevertheless, recent theoretical works suggests that the
absence of these externalities does not preclude ETFs, from being a potential source of market fragility.
Focusing on ETFs backed by hard to trade assets, Bhattacharya and O’Hara (2016) claim that ETFs are
now a preferred trading vehicle capable of affecting underlying asset prices. Their model predicts that
ETFs can exacerbate market instability and herding due to imperfect learning and delayed price
synchronization. The authors, along with a broader ETF model from Malamud (2016) shows that the
creation and redemption mechanism of ETFs can serve as a shock propagation channel. Finally, Pan
and Zeng (2017) theorize that the dual role of authorized participants (APs) as corporate bond market
makers and arbitrageurs creates conflicting incentives that increase fragility.
The challenge of addressing the effect of turmoil period outflows from nontraditional
corporate bond investors on market yields is that the majority of asset growth has occurred during
the post‐crisis bull market, limiting the number of unanticipated exogenous shocks. However, in the
summer of 2013 the Federal Reserve unexpectedly proposed ending its bond buyback program known
as quantitative easing (QE). The change in expectations about monetary policy led investors to alter
their perception of risk and to sell their bond fund holdings, in an episode commonly referred to as
the Taper Tantrum. According to Lipper, the uncertainty in timing and scope of the Fed’s QE
withdraw led to $8.6 billion in outflows from taxable bond mutual funds and ETFs in the week
following the announcement and $23.7 billion over four weeks, the sharpest four‐week exodus since
2008. The turmoil and outflows continued throughout the summer, forcing the Fed to officially
announce in September that it would delay any QE slowdown.
3
Using this plausibly exogenous shock as a quasi‐natural experiment, we follow Coval and
Stafford (2007), Lou (2012), and Mitchell, Pulvino, and Stafford (2004) by studying if outflows from
either vehicle during a period of turmoil create yield pressure that subsequently reverses. Beyond
controlling for observable bond characteristics and liquidity, the inclusion of issuer level fixed effects
is key to our identification similar to Manconi, Massa, and Yasuda (2012). Relying on within issuer
variation, endogeneity concerns regarding changes in fundamental risk are addressed by effectively
comparing two bonds from the same firm exposed to different levels of outflows. We find significant
yield pressure created by ETFs, with a one standard deviation increase in ETF Tantrum outflows
leading to a 12.4 basis point greater increase in the yield spread of corporate bonds in September 2013.
Economically, this implies a 10.8% (8.5%) increase in the yield spread of the median (mean) corporate
bond held by ETFs prior to the onset of the Taper Tantrum. The effect is transient lasting seven months
before reverting, thus implying that bonds temporarily traded at non‐fundamental values due to ETF
outflows. In contrast, we document that outflows from active mutual funds have no significant effect
on corporate bond yield spreads in the months following the Taper Tantrum. Since the index strategy
of most ETFs reduces the flexibility of a manager’s response to flows (Christoffersen, Keim, and
Musto, 2008; Elton, Gruber, and Busse, 2004), we also consider the impact of index mutual funds. We
find that Taper Tantrum outflows from index mutual funds do not have a significant impact on the
yield spread of their holdings. The insignificance of an index mandate may reflect the use of
representative sampling or mutual funds benchmarking to aggregate bond indices, rather than to the
corporate bond dedicated indices followed by ETFs.
Next, we attempt to identify which characteristics of ETFs contribute to the distinctive yield
impact. In particular, we examine the appeal of intraday trading to different investor types and the
implications of the in‐kind creation and redemption mechanism. First, as described by Chordia (1996),
Deli and Varma (2002) and Nanda, Narayanan, and Warther (2000) funds adopt different structures
to appeal to the stochastic liquidity needs of investors. Following Amihud and Mendelson (1986),
Constantinides (1986) and Poterba and Shoven (2002), we posit that the intraday liquidity of ETFs is
likely to attract short‐term traders who are theorized to focus on the behavior of other investors, rather
than on long‐term fundamentals (Allen, Morris, and Shin, 2006; De Long, Shleifer, Summers, and
Waldmann, 1990a; Froot, Scharfstein, and Stein, 1992; Stein, 2005). As described by Cella, Ellul, and
Giannetti (2013), during normal markets short‐horizon investors in ETFs should not affect underlying
4
bonds because other investors readily provide liquidity. However, in turmoil periods short‐horizon
investors are expected to sell en masse (Bernardo and Welch, 2004; Morris and Shin, 2004).
The distribution of fund flows provides preliminary evidence of distinct investor bases in the
different investment vehicles. We follow the literature to formally test if the underlying ETF investors
have shorter investment horizons by considering the volatility of fund flows (Bollen, 2007; Sialm,
Starks, and Zhang, 2015) and the churn of institutional investors (Cella, Ellul, and Giannetti, 2013) . In
both the cross‐sectional regressions of Sialm, Starks, and Zhang (2015) and panel regressions with date
fixed effects, we find that the volatility of ETFs is statistically significantly greater than mutual funds
after controlling for lagged fund characteristics. Next, we focus on a subset of investors in the
corporate bond funds in our study, institutional investors. We find that the churn of institutions that
use corporate bond ETFs is statistically significantly shorter than those that use mutual funds. Having
established that short‐term investors self‐select into ETFs, we then examine if they engage in positive
feedback trading, that is investors increasing flows in periods of lower interest rates and decreasing
flows in periods of higher interest rates. Specifically, we regress flows on lagged changes in Treasury
rates, an ETF dummy, and their interaction, as well as, controls for lagged fund flows, returns,
turnover, expense ratio, and the average rating and duration of its holdings. The coefficient on the
interaction term is negative and significant, implying that ETF investors are more sensitive to common
market shocks and trade on past price trends, which is theorized by De Long, Shleifer, Summers, and
Waldmann (1990b) to be potentially destabilizing. Overall, these results provide evidence that bonds
exposed to ETFs are subjected to distinct pressures associated with short‐horizon positive feedback
traders.
Second, while mutual funds deal directly with all investors through fund share and cash
transactions, ETFs interact only with APs through the in‐kind creation and redemption mechanism,
known as the primary market. ETF creation (redemption) occurs when an AP buys the pre‐specified
basket (ETF shares) and exchanges it for a block of ETF shares (underlying basket). The reliance of
ETFs on this mechanism has important implications for both portfolio construction and arbitrage.
Because ETFs generally do not need to provide cash on demand they may invest more in benchmark
securities. In contrast, for various types of mutual funds Chen, Goldstein, and Jiang (2010), Chernenko
and Sunderam (2016) and Hoseinzade (2015) show that managers have an incentive maintain a
liquidity buffer to mitigate the adverse effect of investor flows, particularly during periods of
5
volatility. Comparing the asset allocation in the reporting period prior to the Taper Tantrum, we find
that the median investment in cash and government bonds is 21.2% for active mutual funds and 47.7%,
for index mutual fund, compared to just 2.74% for ETFs, a difference that is both statistically and
economically significant.
The primary market also enables APs to engage in arbitrage between the ETF market price
and NAV. While arbitrage should be instantaneous, when ETFs are backed by hard to trade assets it
may not be the case (Bhattacharya and O’Hara, 2016). However, a difference in the two prices may
occur and persist if limits to arbitrage are present (Abreu and Brunnermeier, 2003; De Long, Shleifer,
Summers, and Waldmann, 1990a; Ofek and Richardson, 2003; Pontiff, 1996; Shleifer and Vishny, 1997).
A number of theories further suggest that in the presence of limits arbitrage may amplify rather than
stabilize fundamental shocks (Greenwood and Thesmar, 2011; Hong, Kubik, and Fishman, 2012;
Hugonnier and Prieto, 2015; Kyle and Xiong, 2001). We begin by documenting that ETF arbitrage was
impaired. Splitting corporate bond ETFs into two groups based on the average percentage price to
NAV deviation during the turmoil period we show that the high deviation ETF group, those for which
ETF sellers pushed market prices significantly below the underlying NAV, traded at slight premiums
prior to the event. Comparing the characteristics of the ETF groups immediately before the Taper
Tantrum shows that both are similar size and hold bonds with similar credit quality, volatility of flows
and institutional ownership. The high deviation ETFs hold bonds with greater effective duration for
which we would expect more selling pressure for in expectation of higher interest rates and have
higher bid ask spreads suggestive of known limits to arbitrage (Gromb and Vayanos, 2010).
Interestingly, there is an insignificant difference in the AP arbitrage activity proxy from Da and Shive
(2013) prior to the event, but arbitrage in the high deviation ETFs is statistically significantly lower
during the Taper Tantrum.
The existence of these arbitrage opportunities translates into potential selling pressure for the
individual bonds from arbitrageurs, who are concerned about relative price efficiency, rather than the
absolute price efficiency of the individual assets (Brown, Davies, and Ringgenberg, 2016; Mitchell,
Pedersen, and Pulvino, 2007). We first calculate the holdings‐weighted exposure of each bond held by
ETFs to price‐NAV deviations during the tantrum, denoting those with below median measures high
arbitrage exposure as they are held by ETFs trading at Taper Tantrum induced discounts. Then we
effectively implement a strategy that sells high arbitrage exposure bonds and buys a portfolio of ETF
6
bonds without similar exposure. To formally test the impact of delayed arbitrage, we regress the yield
spread for each month of the Taper Tantrum through December 2014 on a high arbitrage exposure
dummy and in the multivariate regression, we control for observable bond characteristics. We find
that the yield spreads of the two groups are not significantly different during the Taper Tantrum.
However, as the turmoil mitigates and arbitrageur reenter to restore the relative price efficiency
between the ETF and the underlying, the yield spreads of high arbitrage exposure bonds are
statistically significantly higher for nine months. The timeline for the convergence of the two portfolios
coincides with the closing of the arbitrage opportunity. These results support the theories of
Bhattacharya and O’Hara (2016) and Malamud (2016) that predict that ETF arbitrage can serve as a
shock propagation channel from the liquid ETF to prices of the illiquid underlying.
Our main sample includes all open‐end mutual funds and ETFs with an average allocation to
corporate bonds greater than 20%. We use this threshold because most corporate bond funds,
particularly index mutual funds, are benchmarked to aggregate bond indices that have sizeable
government and securitized investments.6 In contrast, the majority of ETFs follow corporate bond
dedicated indices with the median ETF holding 93% of their assets in corporate bonds. To address
comparability concerns as robustness tests we execute our main tests for active mutual funds with a
corporate bond allocation greater than 80%. An emerging literature focuses on the importance of
corporate bond fund cash buffers with Jiang, Li, and Wang (2016) finding in times of volatility mutual
funds sell their holdings proportionally rather than relying on the liquidity buffer and Morris, Shim,
and Shin (2017) showing in a global setting that cash hoarding is the norm. Therefore, we also consider
active mutual funds in the lowest quartile of combined liquid investments and the highest quartile of
flows volatility. The only evidence of flow‐induced yield pressure is in active mutual funds in the
lowest quartile of liquid assets lasting just one month. Given these findings, we cannot rule out mutual
fund outflows as a possible source of fragility in a prolonged market shock. Rather than lessen fears
of strategic complementarities in mutual funds (Goldstein, Jiang, and Ng, 2017), the findings of this
paper may exacerbate them as we highlight that short‐horizon positive feedback ETFs investors may
have the true first‐mover advantage through intraday trading.
6 In 2013 Bloomberg Barclays US Aggregate Bond that had a 22.3% corporate bond allocation See:
https://www.bbhub.io/indices/sites/2/2016/08/2017‐02‐08‐Factsheet‐US‐Aggregate.pdf
7
This paper adds to a broader literature that studies asset price movements induced by
financial institutions and arbitrageurs, by cleanly identifying the short‐horizon investors and
imperfect arbitrage of ETFs as a new source of price pressure in corporate bonds. In the stock market,
equity funds are shown to move underlying stock prices in a variety of settings (Ben‐Rephael, Kandel,
and Wohl, 2011; Coval and Stafford, 2007; Greenwood and Thesmar, 2011; Jotikasthira, Lundblad, and
Ramadorai, 2012; Lou, 2012). For instance, studies by Brunnermeier and Nagel (2004) and Griffin,
Harris, Shu, and Topaloglu (2011) of the technology bubble document that traditional arbitrageurs,
hedge funds, amplified rather than stabilized the market. While, Cella, Ellul, and Giannetti (2013)
show that stocks held by short‐horizon investors experience greater price pressure during the Lehman
bankruptcy. In fixed income markets, Feroli, Kashyap, Schoenholtz, and Shin (2014) motivated by the
Taper Tantrum use nineteen years of data to show potentially destabilizing feedback from fund flows
to some market returns. Specifically focused on corporate bonds, broad studies of insurance
companies by Ambrose, Cai, and Helwege (2012) and of mutual funds by Hoseinzade (2015) find no
evidence of yield pressure. In contrast, Ellul, Jotikasthira, and Lundblad (2011) and Manconi, Massa,
and Yasuda (2012) find that constrained insurance companies and investors holding securitized bonds
in the crisis, respectively, engage in corporate bond fire sales. Finally using the measure of Coval and
Stafford (2007), Jiang, Li, and Wang (2016) show that bonds subjected to trading by extreme flow
funds, have abnormal returns in the next two quarters and in a fund family network setting Falato,
Hortacsu, Li, and Shin (2016) find price pressure in the quarter of peer‐induced sales.
The results of this paper also contribute to the growing research on ETFs, by identifying
another possible unintended risk. The arbitrage response to noise traders is shown by Ben‐David,
Franzoni, and Moussawi (2014) to increase stock volatility and by Brown, Davies, and Ringgenberg
(2016)to lead to predictable returns. Additional studies of equity ETFs find that they increase co‐
movement (Da and Shive, 2013) and lower information acquisition benefits and liquidity (Israeli, Lee,
and Sridharan, 2017). In corporate bond ETFs, Dannhauser (2017) documents that ETF constituency
lowers bond yields due to the migration of liquidity traders from the underlying market to ETFs.
2. Background
This section provides a detailed discussion of the corporate bond market and its participants.
In particular, we first focus on the evolution of the market and the emergence of ETF and mutual fund
8
investors. Then we detail the Taper Tantrum of the summer of 2013, which we use as a plausibly
exogenous shock to fund flows.
2.1. The transformation of the corporate bond market
Since the financial crisis, the corporate bond market has undergone a radical change. First,
incentivized by unprecedentedly low interest rates corporations have increased their debt issuance
leading to a 56% increase in corporate debt outstanding between 2008 and 2016. Panel A of Figure 1
presents the growth in corporate bond assets outstanding and the annual issuance.
[Insert Figure 1]
Second, regulatory pressure has altered the dynamics of the historically opaque market
(Bessembinder, Jacobsen, Maxwell, and Venkataraman, 2016), in which broker‐dealers held large
inventories to facilitate trades with institutional investors. Traditionally, the majority of investors were
insurance companies and pension funds who are broadly characterized by Bessembinder, Maxwell,
and Venkataraman (2006) as long‐horizon investors. Despite the July 2002 introduction of delayed
trade reporting through the Trade Reporting and Compliance Engine (TRACE), Bessembinder,
Maxwell, and Venkataraman (2006), Edwards, Harris, and Piwowar (2007) and Goldstein, Jiang, and
Ng (2017) all document that corporate bonds trade infrequently, particularly relative to other asset
classes. Pressured by increased capital requirements broker‐dealer inventory of corporate and foreign
bonds has decreased 35% according to Federal Reserve Board data as plotted on the right axis of Panel
B of Figure 1. The impact of their retreat on liquidity is uncertain with Anderson and Stulz (2017) and
Bao, O’Hara, and Zhou (2016) finding evidence of greater illiquidity in periods of stress. Conversely,
Bessembinder, Jacobsen, Maxwell, and Venkataraman (2016) find that average transaction costs are
not significantly higher since the crisis, but that the traditional commitment structure is changing.
Third, as the corporate bond market structure has evolved, nontraditional investors: mutual
funds and ETFs; have emerged as increasingly important investors. The accumulation of assets by
these funds, whose investment strategy and investor base is significantly different than traditional
investors has implications for the underlying market because of the liquidity they provide their
investors and thus the liquidity they may demand. The left axis of Panel B of Figure 1 shows that the
combined holdings of the nontraditional investors are nearly equal to those of insurance companies.
Figure 1 also present details on the growth of ETFs in Panel C and mutual funds in Panel D, including
9
the number of funds and assets under management of all long‐term taxable bond funds from ICI
FactBook, as well as, corporate bond specific assets from the Federal Reserve Board. From these figures
it is evident that a large portion of growth in these vehicles has occurred since 2008 with the corporate
bond assets of mutual funds tripling and ETF corporate bond assets growing by eleven times.
Given the focus of this paper on ETFs and mutual funds, an extended discussion of their
structures is needed. At the most basic level, mutual funds and ETFs are investment vehicles backed
by a basket of corporate bonds. Mutual funds are distinguished by their investment mandate as either
active mutual funds, who attempt to outperform their benchmark, or passive mutual fund, who
attempt to replicate their benchmark. Regardless of the type of mutual fund, investors are provided
with daily liquidity. That is an investor can submit a buy or sell order for the mutual fund shares at
any point throughout the day with all transactions occurring at the closing NAV.
Corporate bond ETFs, the first of which was introduced in June 2002, are a hybrid between
traditional open‐end mutual funds and closed end funds (CEFs). The unique features of ETFs allows
them to offer lower management fees, greater transparency, and tax efficiencies (Poterba and Shoven,
2002). ETFs provide liquidity to investors in two venues, the primary and secondary markets. The
primary market, which is the direct channel linking ETFs to the underlying, is used by ETFs to handle
liquidity shocks in the secondary market, to ensure that orders are filled, and to arbitrage excessive
market price deviations from NAV. This market involves transactions between APs and the fund
sponsor that exchange ETF units for a basket of the underlying securities through the in‐kind creation
and redemption process. In contrast, the creation and redemption process of mutual funds occurs
between the fund and individual investors as an exchange of cash for fund shares. The secondary
market of ETFs, is where most buyers and sellers of the ETF transact directly on the equity exchange
without sponsor involvement.
In Table 1 we present summary statistics for ETFs in Panel A, active mutual funds in Panel B
and index mutual funds in Panel C. While the holdings of active mutual funds and ETFs have similar
credit quality and maturity, the holdings of index mutual funds have lower average ratings and
longer‐term to maturity. Further, the distribution of fund characteristics validates the common refrain
surrounding ETFs relative to mutual funds. For instance, ETFs have expense ratios and turnover that
are more than half of the mean level of active mutual funds reflecting the relative simplicity of ETF
10
management. The average index fund expense ratio is slightly higher than the average index mutual
fund, but the variation is greater for index mutual funds.
[Insert Table 1]
2.2. The Taper Tantrum as an exogenous shock to fund flows
An empirical challenge of identifying potential vulnerabilities created by this new corporate
bond market regimen is the absence of market shocks and resultant fund flows during the post crisis
bond bull market. The exception is the Taper Tantrum of the summer of 2013 which serves as our
plausibly exogenous market‐wide shock. Prior to this event, corporate bond funds saw
disproportionate inflows relative to other asset classes (DiMaggio and Kacperczyk, 2017) as the
Federal Reserve employed unprecedented policy initiatives, including open market purchases of
government bonds and mortgage‐backed securities through its QE program. Initiated in November
of 2008, QE was intended to lower long‐term interest rates in an effort to bolster the economy.
Following positive economic developments, Chairman Ben Bernanke testified on May 22, 2013 that
the Federal Reserve would likely begin slowing or tapering the pace of its bond purchases conditional
on continued economic stability. On June 19, 2013 the Chairman held a news conference to document
the economic justification for the purchase slowdown. Convinced that the end of QE was near, the
market responded by selling assets, in an episode commonly compared to a tantrum. Between May
21, 2013 and the beginning of September both the five‐ and ten‐year Treasuries increased over 100
basis points as depicted in Figure 2. The sharp reevaluation of risk lead investors to utilize the liquidity
provisions of ETFs and mutual funds by swiftly withdrawing assets. According to Trim Tabs
Investment Research, a record $69 billion was withdrawn from bond ETFs and mutual fund in June
2013 alone, with outflows continuing throughout July and August. The period of uncertainty was
resolved on September 18, 2013 when the Fed officially announced it would wait to begin scaling back
the QE program.
[Insert Figure 2]
3. Data
Our corporate bond fund data begins in January 2010, in order to study a period in which the
assets managed by these two investment vehicles is no longer negligible, and ends in March 2015. We
11
begin by identifying corporate bond funds using the Center for Research in Security Prices (CRSP)
Survivor‐Bias‐Free US Mutual Fund Database. In particular, we denote any mutual fund or ETF that
as an average corporate bond weighting of twenty percent or greater as a corporate bond fund. This
low threshold is used to ensure that we have a sufficient number of index mutual funds, which most
frequently are benchmarked to the Bloomberg Barclays US Aggregate Bond Index that had a 22.3%
corporate bond allocation in 2013. We also collect fund characteristics from the Morningstar Mutual
Fund Database, including the number, concentration, and the average credit quality of the bond
holdings. Matching on fund cusip, we then use CRSP to obtain fund holdings, monthly returns, assets,
turnover, expense ratio, and share class data. Furthermore, we use the CRSP database and fund name
search to distinguish ETFs and index mutual funds. Using this data we follow the literature to
compute the flow to each fund, , in month, , as
,, , ∗ 1 ,
,
For mutual funds with multiple share classes we aggregate data at the portfolio level by summing
fund assets, and value weighting other characteristics. Further, we account for the VETF structure of
Vanguard in which the ETF is a share class of the larger index mutual fund.
The Trade Reporting and Compliance Engine (TRACE) database filtered for possibly
erroneous trades with the method of Dick‐Nielsen (2009) is used to obtain bond transaction data. The
monthly spread of bond from issuer in month , , , is computed as the volume weighted
average yield reported for all transactions in a month over the maturity‐matched Treasury rate. The
TRACE database is also used to calculate our liquidity proxy, , , developed by Amihud
(2002). Specifically, the proxy is computed as the median of the daily measure
,1 | |
∗ 10 ,
where is the number of returns on day d, rj is the return of consecutive transactions, and is the
dollar volume of a trade. This measure can be interpreted as the basis points price movement per one
million dollars of traded volume. Bond characteristics come from the Merchant Fixed Income
Securities Database (FISD) and are merged on eight‐digit CUSIP. In addition, for each bond we create
a numerical conversion of Standard & Poor’s (S&P) ratings.
(2)
(1)
12
Finally, we obtain the daily shares outstanding of ETFs from Bloomberg. We use the shares
outstanding to compute the Da and Shive (2013) creation and redemption intensity measure, ,
for ETF in month as
, ,
,.
The activity associated with CRI reflects the direct involvement of the affiliates of the ETF in a bond.
4. Methodology and results
This section details the methodology and presents the results on the impact of turmoil
outflows from ETFs and mutual funds on the yield spreads of associated bonds. Then we investigate
potential explanations for our main results.
4.1. The yield spread effect of mutual funds and ETFs
Our empirical analysis begins by examining if outflows during the summer of 2013 following
the Federal Reserve announcement, have an impact on the pricing of constituent bonds. We follow
the methodology of Coval and Stafford (2007) and Mitchell, Pulvino, and Stafford (2004), by looking
at cumulative returns and subsequent reversals to provide evidence of a non‐fundamental shock to
prices. Specifically, for each investment vehicle in each of the t months following the summer of 2013,
we run the regression
,
, , , ,
, , , , , , , ,
, , ∗ ,
∗ , , , .
The dependent variable, ,, is the change in the volume‐weighted average
yield of bond, i, from issuer j over the maturity‐matched Treasury rate. The change is measured
relative to the bond’s volume‐weighted yield in May for all transactions prior to May 22, the day of
Bernanke’s initial testimony.
(3)
(4)
13
The covariate of interest is , , , the weighted‐average monthly flow for
either all ETFs, all active mutual funds, or all index mutual funds that report bond i as a holding prior
to the onset of the event. To be included in the regression we require the weighted average flow to be
negative, i.e. an outflow. The covariate is multiplied by ‐1 to help with interpretation of the coefficient
of interest, , which measures the impact of Tantrum outflows to the different investment vehicles
on a bond’s yield spread relative to pre‐event levels. The square of , , , controls
for any potential nonlinearities in the relationship of interest.
As in Manconi, Massa, and Yasuda (2012), the inclusion of issuer fixed effects, addresses
endogeneity concerns associated with changes in the fundamentals of an issuer. The use of issuer fixed
effects controls for managers selecting to sell bonds during the turmoil event for fundamental reasons,
thus we are effectively comparing the change in yield spreads of bonds from the same issuer subjected
to different levels of fund outflows. Therefore, if a firm has only one outstanding issue subjected to
outflows, it automatically is excluded from the analysis.
We also include a vector of bond specific characteristics set to their pre‐event levels to control
for differences in issues from the same firm. We control for the log of the issue size in millions of
dollars, Log(Size); the number of years left to maturity and its square, Time to Maturity; the coupon,
Coupon; and the liquidity of the issue using the Amihud illiquidity measure, Amihud. The yield spread
of the bond prior to the event, is also used to account for differences in bonds from the
same issuer that are not captured by our other controls, for instance covenants. Finally, we include
interactions to account for the likelihood that investors sell bonds with higher interest rate exposure,
∗ ,, and greater liquidity, ∗
,.
Using Eq. (4), we run the regression for ETFs, active mutual funds, and index mutual funds
separately to identify the impact of each investment vehicle on its underlying bonds. Table 2 presents
the results for ETFs in Panel A. The results show that ETF outflows drive spreads significantly higher
for up to seven months beyond the event. In September 2013, as the Taper Tantrum subsided, bonds
with a one percentage increase in exposure to ETF Tantrum outflows had yield spreads 7.2 basis points
higher than a bond from the same issuer. For a one standard deviation greater exposure to ETF
Tantrum outflows this corresponds statistically to a 12.4 basis point increase in yield spread relative
to bonds from the same issuer and economically to a 10.8% (8.5%) increase in the yield spread of the
14
median (mean) corporate bond held by ETFs prior to the onset of turmoil. The statistical significance
in the difference diminishes after seven months with yield spreads reverting to their pre‐crisis level.
[Insert Table 2]
Panel B of Table 2 presents the results for all active mutual funds in our sample. Although the
coefficient of interest is positive for seven months, it is neither statistically nor economically
significant. These findings suggest that after controlling for fundamental risk and observable bond
characteristics there is no significant impact of active mutual fund outflows in a period of turmoil on
exposed corporate bonds.
Panel C of Table 2 shows the results for corporate bond index mutual funds, which like the
majority of ETFs focus on replicating the returns of a pre‐specified benchmark rather than
outperformance. While an index strategy reduces the analysis required of a manager, it also constrains
her response to flows, by mandating her to buy and sell index bonds in amounts determined by their
benchmark weighting (Christoffersen, Keim, and Musto, 2011; Elton, Gruber, and Busse, 2004). In
contrast to most equity index funds that utilize a strict indexing strategy, fixed income index funds
generally employ a representative sampling strategy, selecting only a subset of bonds to best match
the characteristics of the index. While representative sampling increases the manager’s ability to
selectively respond to flow, if the index‐based nature of ETFs is responsible for the flow‐induced yield
pressure we would expect similar yield pressure from index mutual funds. The results of Panel C
show that outflows from index mutual funds do not have a significant impact on the yield spread
change of their constituent corporate bonds. Therefore, we conclude that the reduced flexibility of an
index mandate is not responsible for the yield pressure of ETFs.
According to Coval and Stafford (2007) if these impacts are due to informed trading, the yields
should remain permanently higher. In Figure 3 we plot the impact of a one standard deviation outflow
from the different investment vehicles on the yield spread of bonds relative to a another bond from
the same issuer. The Panel A shows the impact of ETF Tantrum outflows peaking in September and
remaining statistically significant through March of 2014 suggestive of temporary pressure from ETFs
pushing yields beyond their fundamental levels. Panel B and Panel C show that there is little statistical
significance between Taper Tantrum outflows from active mutual fund and index mutual fund,
respectively, and the yield spread of their bonds.
15
[Insert Figure 3]
4.2. Examining the differential effect of mutual funds and ETFs
The main results of this paper are particularly intriguing given the size and youth of the ETF
market relative to the larger and more mature mutual fund market. The differential effect identified
suggests the distinguishing features of ETFs may be a source of potential fragility for the corporate
bond market. In this subsection we focus on two distinctive ETF features: the appeal of intraday
trading to short‐horizon investors and the portfolio construction and arbitrage implications of the in‐
kind creation and redemption mechanism as potential explanations of the main results.
4.2.1. The investment horizon of fund investors
ETFs are often described as mutual funds that trade intraday on an exchange. This key
structural feature has appealed to investors of all types as evidenced by the outsized secondary market
volumes of corporate bond ETFs. Nanda, Narayanan, and Warther (2000) and Chordia (1996) show
that different fund structures are used to screen out high liquidity demand investors, whose frequent
trading exposes more stable investors to trade induced externalities. Moreover, Amihud and
Mendelson (1986) and Constantinides (1986) predict that short‐term investors self‐select into more
liquid assets. Considering these theories and the predictions of Poterba and Shoven (2002), we
hypothesize that ETFs are more likely to attract short‐horizon traders. Cespa and Vives (2015) show
that in normal market conditions, where liquidity shocks are persistent there is no impact on asset
prices from the presence of short‐horizon investors, who are theorized to specialize in investment
strategies that focus not on fundamentals, but instead on the expected behavior of other investors
(Allen, Morris, and Shin, 2006; De Long, Shleifer, Summers, and Waldmann, 1990a; Dow and Gorton,
1994; Froot, Scharfstein, and Stein, 1992; Morris and Shin, 2004; Stein, 2005). However, their presence
can be destabilizing in periods of turmoil if as predicted by Allen, Morris, and Shin (2006), Bernardo
and Welch (2004) and Morris and Shin (2004) short‐horizon investors are expected to trade
simultaneously and place a disproportionate weight on a public signal. Bhattacharya and O’Hara
(2016) specifically show that ETFs can expose the underlying to rational herding, where speculators
trade on the same systematic signal, which may lead to market fragility in the underlying assets.
To examine if ETFs appeal to a different investor base we first describe the distribution of
flows to the different investment vehicles during both the Taper Tantrum and non‐Taper Tantrum
16
periods in Table 3. The table provides preliminary evidence that ETFs attract a more active investor
base with volatility of flows over 1.6 times that of both active and index mutual funds. During the
Taper Tantrum event ETFs in the lowest flow decile experienced outflows more than 3.5 times greater
than those in normal market conditions compared to 1.8 times for active mutual funds and 1.6 times
for index mutual funds.
[Insert Table 3]
As further evidence of intraday trading appealing to a distinct short‐term investor base we
follow the literature by examining the standard deviation of fund flows and the churn of institutional
investors in the sample funds. Foremost, Bollen (2007) and Sialm, Starks, and Zhang (2015) use the
volatility of monthly fund flows as a proxy for investor behavior. As described by the former, the
benefits of this measure are that it captures the net effect of investors’ decisions without forcing a
structure on the process and it captures the burden short‐term investors place on managers and long‐
term investors through frequent trading. Using the volatility of fund flows as our initial proxy for the
investment horizon of all fund investors, we follow Sialm, Starks, and Zhang (2015) by running the
following regression
∗ .
We first execute Eq. (5) as a cross‐sectional regression similar to Sialm, Starks, and Zhang (2015).
In these specifications the dependent variable is the average twelve month volatility of fund flows for
all investment vehicles in our sample between January 2010 and March 2015. , is an indicator
variable equal to one if a fund is an ETF and zero otherwise. The covariates are the expense ratio,
turnover ratio, log of total net assets, log of fund age in years, log of total fund family assets, and in
some specifications their interaction with . We use the demeaned average value of all continuous
variables in the cross‐sectional tests rather than their initial values as in Sialm, Starks, and Zhang
(2015) to account for the tremendous growth of ETFs over our sample period. We also run Eq. (5) in a
panel setting with date fixed effects. In these regressions the dependent variable is twelve month
volatility of fund flows, one‐month lagged covariates, and standard errors clustered at the firm level.
Table 4 presents the results of the regressions on the relation between flow variability and fund
characteristics. Regardless of the specification the coefficient on the ETF variable is positive and
significant suggesting that the monthly volatility of ETF flows are 2.2% to 3.3% greater than mutual
(5)
17
funds after controlling for basic fund characteristics. These results indicate that corporate bond ETFs
attract more short‐term traders than their mutual fund peers.
[Insert Table 4]
As further evidence that ETF investors have a shorter investment horizon, we look at the
subsample of fund holders — institutional investors. Using the Thomson Reuters holdings database
we identify any institutional investors that report holding one of the corporate bond mutual funds or
ETFs in our study. Of the 4,528 distinct investors in the Thomson Reuters s34 database, 1,367 hold at
least one of the funds in our sample between January 2010 and March 2015. Of those investors
recording a holding of a corporate bond fund, 144 hold both an ETF and mutual fund, 125 hold just a
mutual fund, and 1,098 hold an ETF. For the investors that hold either just a mutual fund or just an
ETF, we compute the churn measure of Cella, Ellul, and Giannetti (2013) for institutional investor, ,
holding an investment set of firms, , as,
,∑ | , , , , , , , , Δ , |∈
∑ , , , , , ,2∈
,
where , and , , are the price and number of shares of stock held by institution in quarter .
We focus on the stock holdings of institutional investors because of the subjectivity in reporting bond
prices (Cici, Gibson, and Merrick, 2011). Table 5 presents the results of the statistical significance of
the difference in means and medians of the churn measure over the entire sample period and for the
quarter immediately preceding the period of turmoil. The tests of the median throughout this paper
perform a nonparametric k‐sample test with the null hypothesis that the k samples were drawn from
populations with the same median. For two samples, the chi‐squared test statistic is computed both
with and without a continuity correction. Regardless of the measure or the period, institutional
investors that hold ETFs have both statistically and economically significantly greater churn than
institutional investors that hold mutual funds. Consistent with the results above and those of Ben‐
David, Franzoni, and Moussawi (2017) that compare the churn of equity ETFs to stocks, the churn
ratio analysis suggests that for this subset of fund investors those that use ETFs have shorter trading
horizons than those that use mutual funds.
[Insert Table 5]
(6)
18
Finally, the literature suggests that one way in which short‐horizon investment can be a source
of fragility is if these investors engage in positive feedback trading, buying when the market is moving
higher and selling when the market is moving lower. De Long, Shleifer, Summers, and Waldmann
(1990b) present a model of investors who trade on past price trends and show that a result of their
presence is price destabilization. To test if these new corporate bond investors utilize a positive
feedback strategy, we follow Edelen and Warner (2001), Goetzmann and Massa (2003) and Warther
(1995) by examining the sensitivity of investors to lagged market returns over the period from January
2010 to March 2015. Since interest rates dictate the price of the underlying corporate bonds, we test if
ETFs are more likely to respond using the following specification
, ∆ ⁄ ∗ ∆ ⁄ , , .
The dependent variable in the above regression, , , is the monthly flow to fund in month .
Δ / is the change in either the one‐ or five‐year treasury rate in the prior month normalized by its
standard deviation. We also control for the average effective duration and rating of the fund’s holding,
as well as, the log of fund assets, expense ratio, turnover, average fund return over the last three
months, and one‐, two‐, and three‐month lagged fund flows. The coefficient of interest, , on the
interaction Δ ⁄ ∗ , measures the differential behavior of ETF investors. Table 6 shows the
results of these tests with standard errors clustered at the date level.
[Insert Table 6]
The coefficient on Δ ⁄ shows that after controlling for other factors lagged changes in interest
rates are not a determinant of mutual fund flows. Meanwhile, the coefficient on the ETF dummy
shows that this investment vehicle has larger inflows over the sample period. Interestingly, the
coefficient on the interaction is negative and significant for both one‐ and five‐year interest rate
changes after controlling for fund characteristics suggesting that ETF investors are more likely to
demand liquidity when interest rates increased in the previous month. In particular, a one‐standard
deviation increase in the one‐year Treasury rate leads to 0.098% higher monthly outflows for ETFs
than for mutual funds. These results show that ETF investors are more likely to trade in response to
interest rate changes.
Overall, this subsection finds that ETF investors have short‐horizons and engage in potentially
destabilizing positive feedback trading strategies. In normal markets, the secondary market feature of
(7)
19
ETFs should prevent the introduction of this new investor type from impacting the underlying.
Nevertheless, the presence of these traders, combined with the main findings suggest that the liquidity
illusion of ETFs is a potential source of fragility for the corporate bond market.
4.2.2. The implications of the creation and redemption mechanism
The key operational difference between mutual funds and ETFs is the reliance of the latter on
in‐kind creation and redemption, which has important portfolio construction and arbitrage
implications. First, despite having an investor base of liquidity demanders ETFs may have a reduced
incentive to manage a liquidity buffer because they do not need to produce cash on demand to satisfy
investor flows. In contrast, mutual fund managers rely on liquid holdings during volatile times to
reduce externalities as documented by Chen, Goldstein, and Jiang (2010). For corporate bond
managers, the liquidity buffer consists of cash, cash equivalents, and the more frequently traded
government bonds (Goldstein, Jiang, and Ng, 2017).
In Table 7, we present the mean and median assets allocated to cash, government bonds,
corporate bonds, and securitized bonds. Panel A presents the comparison of active mutual funds and
ETFs and of index mutual funds and ETFs in Panel B. The evidence suggests that there is a statistically
significant difference in the portfolio construction techniques used by the two investment vehicles.
The median active mutual fund holds 21.19% of its assets in cash and government bonds, and for
index mutual funds 47.40% of their holdings are in these liquid assets. In sharp contrast, the median
ETF holds just 2.40% in cash and 0.34% in government bonds, allowing a 93% investment in corporate
bonds. For index funds, the liquidity buffer has a greater representation of government securities than
active mutual funds. In fact, the difference between the cash held by the mean and median index fund
and ETF is not statistically different. This significant allocation to government securities reflects the
broad benchmarks of index mutual fund investing in corporate bonds. Of the 75 distinct investment‐
grade bond index funds identified by Morningstar, 22 are benchmarked to either international
markets, government securities, or mortgage‐backed securities. Of the remaining 53 funds, 45 follow
the Bloomberg Barclays US Aggregate or Bloomberg Barclays US Government/Credit indices and five
follow the BlackRock CoRI indices. The final three funds that follow corporate bond dedicated indices
are offered by Vanguard and have an ETF‐specific share class. Further there are no high yield index
20
mutual funds. While there are ETFs benchmarked to these broad indices the majority of products
focus specifically on corporate bond indices (Dannhauser, 2017).
First, the main tests use all funds with an allocation to corporate bonds of 20% or greater to
incorporate index mutual funds. In Panel C, we present the portfolio allocations for active mutual
funds and ETFs with over 80% of the fund invested in corporate bonds. For these funds there is no
statistical difference in the level of government bond holdings, but the cash holdings of the median
(mean) active mutual fund is 2.5 (1.6) times greater than that of comparable ETFs, a difference that
remains statistically significant.
[Insert Table 7]
Mutual fund managers may utilize their liquidity buffer to meet redemption requests without
selling corporate bonds as found by Chernenko and Sunderam (2016) and Hoseinzade (2015).
Critically, we do not rule out that the liquidity buffer of mutual fund managers would be sufficient in
a prolonged market shock, during which Jiang, Li, and Wang (2016) find active corporate bond
managers are more likely to sell all assets in proportion to their target allocations.
Another consequence of the creation and redemption mechanism is that APs can engage in
arbitrage between the market price of the ETF and the NAV of the underlying. In normal market
conditions and for ETFs backed by assets that are not hard to trade, arbitrage occurs nearly
instantaneously. However, the persistence of any price deviation between two identical assets is
predicted by the limits to arbitrage literature to be the result of noise trader risk (De Long, Shleifer,
Summers, and Waldmann, 1990a), synchronization risk (Abreu and Brunnermeier, 2003), liquidation
risk (Shleifer and Vishny, 1997), transaction costs (Pontiff, 1996), or short sale constraints (Ofek and
Richardson, 2003). Further theories from Greenwood (2005), Hong, Kubik, and Fishman (2012),
Hugonnier and Prieto (2015) and Kyle and Xiong (2001)
We begin our study of the impact of imperfect arbitrage during a period of turmoil on
constituent bonds, by documenting the existence of persistent arbitrage opportunities. First, we split
our ETF sample into two groups based on the three month Taper Tantrum average of the percentage
price to NAV deviation for ETF, , computed monthly as
,, ,
,. (8)
21
We then denote ETFs with the largest average Tantrum period discount, for which selling pressure
pushes the ETF below the NAV, as the high deviation ETF group. Figure 4 plots the average deviation
of this group over time. The Figure shows that these ETFs previously traded above their NAV before
the ETF price fell significantly below during the Taper Tantrum, reflecting the mass exodus
experienced in anticipation of unexpectedly higher interest rates. The ETF discount bottoms in August
2013 before converging in March 2014.
[Insert Figure 4]
In Table 8 we compare the difference between observable characteristics for the high deviation
ETFs and the other half of the ETF sample that remained on average at a premium during the Taper
Tantrum. The average duration of the high arbitrage exposure group is significantly higher, reflecting
a retreat of ETF holders from funds with greater exposure to potentially higher interest rates. The
credit quality of ETF holdings, the volatility of fund flows, and institutional ownership are not
significantly different for the two groups. Representative of common limits to arbitrage, the high
deviation ETF group has higher bid‐ask spreads. Despite these differences, the creation and
redemption intensity measure, which proxies for the amount of arbitrage undertaken by APs is
insignificantly different during normal market conditions. However, during the period of turmoil
arbitrage in the high deviation group drops while it is nearly unchanged in the low deviation group,
suggesting that the arbitrage mechanism of these ETFs was impaired.
[Insert Table 8]
The existence of these arbitrage opportunities translates into potential selling pressure for the
individual bonds from arbitrageurs, who are concerned about relative price efficiency, rather than the
absolute price efficiency of the individual assets at fundamental level (Brown, Davies, and
Ringgenberg, 2016; Mitchell, Pedersen, and Pulvino, 2007). To test if impaired arbitrage is a potential
source of the yield pressure and reversion, for each bond i we compute the ownership‐weighted
average deviation of the set of K ETFs that hold the bond during the Taper Tantrum as,
,∑ , ∗ ,
∑ ,,
where , is the amount of bond i held by ETF k and , is the average deviation of
the ETF during the Taper Tantrum months.
(9)
22
Using this measure we divide the entire sample of ETF bonds into high and low arbitrage
exposure portfolios. The high arbitrage exposure portfolio includes all bonds with an exposure
measure from Eq. (9) less than the median value and the low arbitrage portfolio are bonds with
exposure measures greater than the median. Effectively at the conclusion of the turmoil period our
strategy sells bonds held by ETFs trading at significant discounts and buys a portfolio of ETF bonds
without similar exposure. Panel A of Figure 5 tracks the average yield spread of these two portfolios
through March 2015. The yield spreads of the two portfolios are close during the period of turmoil,
but begin to diverge as the Taper Tantrum concludes and limits to arbitrage dissipate. As arbitrageurs
reentered the market by selling the underlying and buying the ETF, the yield spreads of bonds with
high exposure to ETF discount bonds are pushed above those without similar arbitrage pressure.
Panel B of Figure 5 plots the difference in yield spreads for the two portfolios. The yield spreads of
high exposure bonds increase significantly before reverting in July 2014. The timeline for the
convergence of the two portfolios coincides with the closing of the arbitrage opportunity shown in
Figure 4 supporting theoretical predictions that arbitrageurs serve as a shock propagation mechanism
from the liquid ETF to the underlying bonds.
[Insert Figure 5]
To formally test the impact of impaired ETF arbitrage on constituent corporate bonds, we run
both univariate and multivariate regressions. Corresponding to Figure 5, we first run a univariate
specification of the yield spread level of bond i in month t on the high exposure dummy,
, . We also run the following multivariate specification to control for
bond level characteristics,
, , , ,
, , , .
Table 9 presents the results of the univariate test in Panel A and the multivariate test in Panel
B. The multivariate specification shows that after controlling for observable characteristics of bonds,
there is no significant difference between the yield spread levels of bonds with high exposure to ETF
arbitrage during the tantrum period, indicating that the yield spreads of exposed constituents did not
react significantly different to the change in interest rates expectations. Rather, the yields of bonds
most exposed to arbitrage opportunities created by extreme selling pressure in the ETF moved higher
(10)
23
after the turmoil and limits to arbitrage subsided. Specifically, the yield spreads of high arbitrage
exposure bonds are 10.1 basis points higher in September than those from the low arbitrage exposure
group. The difference in yield spreads peaks in November 2013 and completely reverts by May 2014.
[Insert Table 9]
The results of this subsection show that the unique creation and redemption mechanism of
ETFs has significant implications for the asset prices of constituent bonds. First, we document that
mutual funds have access to a liquidity buffer to respond to temporary periods of turmoil, such as the
Taper Tantrum studied in this paper. For ETFs the in‐kind mechanism reduces the funds’ reliance on
the buffer and allows for greater investment in constituent bonds. However, when arbitrage is limited
the liquidity of the ETF relative to the illiquidity of the underlying may create persistent arbitrage
opportunities that allow for a market‐wide shock to propagate to exposed bonds as predicted by
Bhattacharya and O’Hara (2016) and Malamud (2016). In particular, we show that bonds held by ETFs
subjected to selling pressure during the Taper Tantrum have significantly higher yield spreads when
arbitrageurs reenter the market to restore the relative pricing of the ETF price and NAV, but the
difference reverts as arbitrage pressure abates.
5. Robustness tests
In this section we conduct a variety of robustness tests to see if there is evidence of any flow‐
induced yield pressure for mutual funds. In particular, we repeat the analysis from Eq. (4) and Table
2 on different subsets of active mutual funds. Given the lack of heterogeneity in index mutual fund
benchmarks, we do not conduct the tests for subsets of these funds. The results of our robustness
checks are reported in Table 10.
[Insert Table 10]
First, the main tests use all funds with an allocation to corporate bonds of 20% or greater to
incorporate index mutual funds. Panel A reports the results for active mutual funds with a corporate
bond allocation greater than 80% to focus on funds that have similar corporate bond allocations as the
ETFs in our sample. The coefficients for the first two months are positive, but are neither statistically
nor economically significant indicating that the cutoff used is not responsible for the distinctive effects.
24
Next, there is an emerging literature focused on the liquidity management techniques of
corporate bond fund managers. In global game model of investor runs, Morris, Shim, and Shin (2017)
find that cash hoarding, rather than a liquidity pecking order is the norm. In their empirical study of
42 global bond funds they find that the incidence of cash hoarding is significantly lower for funds
with relatively more liquid assets, such as US corporate bonds. In a panel study, Jiang, Li, and Wang
(2016) find dynamic cash buffer management by active corporate bond mutual funds. In normal times,
funds use liquid holdings to meet flows but in high volatility periods they sell assets proportionally
to maintain their liquidity buffer. The authors then use the pressure measure of Coval and Stafford
(2007) to show evidence of extreme flow exposure leading to price pressure that lasts two quarters,
with the effect doubling in volatile periods. In contrast, Chernenko and Sunderam (2016) study of
active bond funds finds that they use substantial cash holdings to accommodate flows rather than
selling illiquid underlying, particularly in times of low market liquidity. Given the potential
importance of liquid assets in the period studied in this paper, we consider active mutual funds in the
lowest quartile of liquid assets, government and cash. Panel B documents evidence of statistically
significant flow‐induced yield pressure by active mutual funds with low levels of liquid assets, lasting
one month before reverting. In the month following the Taper Tantrum bonds exposed to 1% greater
outflows from these active mutual funds have yields 4.2 basis points higher than a bond from the same
issuer, an effect that is approximately 60% the size of the ETF effect.
Finally, in Panel C active mutual funds in the highest quartile of fund flow volatility over the
year prior to the onset of the period of turmoil are considered. As discussed in subsection 4.2.1, funds
with more variable flows are believed to be subjected to the negative effects of frequent trading by
short‐term investors. The results show no impact for either active or index funds with a more volatile
investor base.
6. Conclusion
The historically illiquid OTC corporate bond market long dominated by long‐term investors,
such as insurance companies, has undergone a radical transformation since the crisis. As broker‐
dealers retreat, the assets, and perhaps influence, of nontraditional investors offering a liquidity
transformation have increased. The growth of mutual funds and ETFs has democratized access to this
asset class, while also raising concerns about market fragility implications justified by financial market
25
history. This paper uses the unexpected revision in interest rate expectations and subsequent outflows
from ETFs and mutual funds during the Taper Tantrum, to cleanly identify the impact of outflows
from these investment vehicles on bond yields. Comparing bonds from the same issuer to control for
fundamental risk, we find that ETF outflows lead to significantly higher yield spreads for seven
months following the shock. The significance and pattern of the coefficients of our regressions indicate
that ETFs contribute to flow‐induced yield spread pressure in the corporate bond market. In contrast,
there is no significant impact of Tantrum period outflows to active mutual funds or index mutual
funds on subsequent changes in yield spreads.
Additional tests show that the contrasting findings can be attributed to the structural and
operational features of ETFs. First, we show that the intraday trading of ETFs attract a distinct
short‐horizon positive feedback investor base. Specifically, we show that corporate bond ETFs have
statistically significantly higher flow volatility than mutual funds and are held by institutional
investors with greater levels of portfolio churn. Further, we then provide evidence that ETF flows are
more sensitive to lagged changes in interest rates suggesting these new investors in corporate bond
assets trade past price trends, which is theorized to destabilize prices (De Long, Shleifer, Summers,
and Waldmann, 1990b). Second, we consider the portfolio construction and arbitrage implications of
the in‐kind creation and redemption mechanism used by ETFs. We document that both active and
index mutual funds have a significant liquidity buffer, which may be used to meet redemption
requests without selling corporate bonds at potentially distressed prices. For ETFs the in‐kind
mechanism reduces the funds’ reliance on the buffer and allows for arbitrage between the ETF market
price and its NAV. However, when arbitrage is limited the liquidity of the ETF relative to the
illiquidity of the underlying may create persistent arbitrage opportunities that allow for a market‐
wide shock to propagate to exposed bonds (Bhattacharya and O’Hara, 2016; Malamud, 2016). We
show that bonds held by ETFs that trade at significant discounts to their NAV due to selling pressure
during the Taper Tantrum temporarily have significantly higher yield spreads as arbitrageurs focused
on the relative price efficiency between the ETF price and NAV, rather than absolute price efficiency,
reenter the market.
In this paper we highlight the short‐horizon positive feedback trading strategy of ETF
investors and trading by arbitrageurs focused on the relative, not fundamental, price efficiency are
potential sources of corporate bond market fragility. Further robustness tests find evidence of flow‐
26
induced yield pressure for active managers in the lowest quartile of liquid assets lasting a month.
Therefore, we conclude that the liquidity management techniques of mutual fund managers may not
be sufficient in a prolonged market shock. Rather than ease concerns about runs induced by strategic
complementarities of corporate bond mutual funds, these findings exacerbate them because the
intraday liquidity transformation may give ETF investors the true first‐mover advantage. Overall the
findings of this paper and those emerging on corporate bond mutual funds, raise concerns that must
be balanced with the exposure these investment vehicles provide investors to a systemically important
asset class that previously was difficult for non‐institutional investors to access.
27
References
Abreu, D., Brunnermeier, M., 2003. Bubbles and Crashes. Econometrica 71, 173–204.
Allen, F., Morris, S., Shin, H.S., 2006. Beauty contests and iterated expectations in asset markets.
Review of financial Studies 19, 719–752.
Amihud, Y., 2002. Illiquidity and stock returns: cross‐section and time‐series effects. Journal of
Financial Markets 5, 31–56.
Amihud, Y., Mendelson, H., 1986. Asset pricing and the bid‐ask spread. Journal of Financial
Economics 17, 223–249.
Anderson, M., Stulz, R.M., 2017. Is Post‐Crisis Bond Liquidity Lower? National Bureau of Economic
Research.
Bao, J., O’Hara, M., Zhou, X.A., 2016. The Volcker rule and market‐making in times of stress.
Ben‐David, I., Franzoni, F.A., Moussawi, R., 2017. Do ETFs Increase Volatility? Journal of Finance
Forthcoming.
Bernardo, A.E., Welch, I., 2004. Liquidity and financial market runs. The Quarterly Journal of
Economics 119, 135–158.
Bessembinder, H., Jacobsen, S.E., Maxwell, W.F., Venkataraman, K., 2016. Capital commitment and
illiquidity in corporate bonds.
Bessembinder, H., Maxwell, W., Venkataraman, K., 2006. Market transparency, liquidity
externalities, and institutional trading costs in corporate bonds.
Journal of Financial Economics 82, 251–288.
Bhattacharya, A., O’Hara, M., 2016. Can etfs increase market fragility? effect of information linkages
in etf markets. Working Paper.
Bollen, N.P., 2007. Mutual fund attributes and investor behavior. Journal of Financial and
Quantitative Analysis 42, 683–708.
Brown, D.C., Davies, S.W., Ringgenberg, M., 2016. ETF Arbitrage and Return Predictability.
Cella, C., Ellul, A., Giannetti, M., 2013. Investors’ horizons and the amplification of market shocks.
Review of Financial Studies 26, 1607–1648.
Cespa, G., Vives, X., 2015. The beauty contest and short‐term trading. The Journal of Finance 70,
2099–2154.
Chen, Q., Goldstein, I., Jiang, W., 2010. Payoff complementarities and financial fragility: Evidence
from mutual fund outflows. Journal of Financial Economics 97, 239–
262.
Chernenko, S., Sunderam, A., 2016. Liquidity Transformation in Asset Management: Evidence from
the Cash Holdings of Mutual Funds.
Chordia, T., 1996. The structure of mutual fund charges. Journal of Financial Economics 41, 3–39.
Christoffersen, S.E.K., Keim, D.B., Musto, D.K., 2011. Valuable Information and Costly Liquidity:
Evidence from Individual Mutual Fund Trades. SSRN Working Paper
Series.
Cici, G., Gibson, S., Merrick, J.J., 2011. Missing the marks? Dispersion in corporate bond valuations
across mutual funds. Journal of Financial Economics 101, 206–226.
Constantinides, G.M., 1986. Capital market equilibrium with transaction costs. Journal of Political
Economy 94, 842.
Coval, J., Stafford, E., 2007. Asset fire sales (and purchases) in equity markets. Journal of Financial
Economics 86, 479–512.
Da, Z., Shive, S., 2013. When the bellwether dances to noise: Evidence from exchange‐traded funds.
28
Dannhauser, C.D., 2017. The impact of innovation: Evidence from corporate bond exchange‐traded
funds (ETFs). Journal of Financial Economics 125, 537–560.
De Long, J.B., Shleifer, A., Summers, L.H., Waldmann, R.J., 1990a. Noise trader risk in financial
markets. Journal of Political Economy 703–738.
De Long, J.B., Shleifer, A., Summers, L.H., Waldmann, R.J., 1990b. Positive feedback investment
strategies and destabilizing rational speculation. Journal of Finance
45, 379–395.
Deli, D.N., Varma, R., 2002. Contracting in the investment management industry: Evidence from
mutual funds. Journal of Financial Economics 63, 79–98.
Dick‐Nielsen, J., 2009. Liquidity biases in TRACE. Journal of Fixed Income 19, 43.
DiMaggio, M., Kacperczyk, M., 2017. The unintended consequences of the zero lower bound policy.
Journal of Financial Economics 123, 59–80.
Dow, J., Gorton, G., 1994. Arbitrage chains. Journal of Finance 49, 819–849.
Edelen, R.M., Warner, J.B., 2001. Aggregate price effects of institutional trading: a study of mutual
fund flow and market returns. Journal of Financial Economics 59,
195–220.
Edwards, A.K., Harris, L.E., Piwowar, M.S., 2007. Corporate bond market transaction costs and
transparency. Journal of Finance 62, 1421–1451.
Ellul, A., Jotikasthira, C., Lundblad, C.T., 2011. Regulatory pressure and fire sales in the corporate
bond market. Journal of Financial Economics 101, 596–620.
Elton, E.J., Gruber, M.J., Busse, J.A., 2004. Are investors rational? Choices among index funds. the
Journal of Finance 59, 261–288.
Falato, A., Hortacsu, A., Li, D., Shin, C., 2016. Fire‐Sale Spillovers in Debt Markets.
Froot, K.A., Scharfstein, D.S., Stein, J.C., 1992. Herd on the street: Informational inefficiencies in a
market with short‐term speculation. The Journal of Finance 47, 1461–
1484.
Gennaioli, N., Shleifer, A., Vishny, R., 2012. Neglected risks, financial innovation, and financial
fragility. Journal of Financial Economics 104, 452–468.
Goetzmann, W.N., Massa, M., 2003. Index funds and stock market growth. Journal of Business 76, 1–
1.
Goldstein, I., Jiang, H., Ng, D.T., 2017. Investor flows and fragility in corporate bond funds. Journal
of Financial Economics 126, 592–613.
Greenwood, R., 2005. Short‐and long‐term demand curves for stocks: theory and evidence on the
dynamics of arbitrage. Journal of Financial Economics 75, 607–649.
Greenwood, R., Thesmar, D., 2011. Stock price fragility. Journal of Financial Economics 102, 471–490.
Gromb, D., Vayanos, D., 2010. Limits of Arbitrage. The Annual Review of Financial Economics is 2,
251–75.
Hong, H., Kubik, J.D., Fishman, T., 2012. Do arbitrageurs amplify economic shocks? Journal of
Financial Economics 103, 454–470.
Hoseinzade, S., 2015. Do Bond Mutual Funds Destabilize the Corporate Bond Market? Unpublished
working paper. Boston College.
Hugonnier, J., Prieto, R., 2015. Asset pricing with arbitrage activity. Journal of Financial Economics
115, 411–428.
29
Israeli, D., Lee, C.M., Sridharan, S.A., 2017. Is there a dark side to exchange traded funds? An
information perspective. Review of Accounting Studies 22, 1048–
1083.
Jiang, H., Li, D., Wang, A., 2016. Dynamic Liquidity Management by Corporate Bond Mutual Funds.
Kyle, A.S., Xiong, W., 2001. Contagion as a wealth effect. The Journal of Finance 56, 1401–1440.
Lou, D., 2012. A flow‐based explanation for return predictability. Review of Financial Studies 25,
3457–3489.
Malamud, S., 2016. A dynamic equilibrium model of ETFs.
Manconi, A., Massa, M., Yasuda, A., 2012. The role of institutional investors in propagating the crisis
of 2007–2008. Journal of Financial Economics 104, 491–518.
Mitchell, M., Pedersen, L.H., Pulvino, T., 2007. Slow Moving Capital. The American Economic
Review 215–220.
Mitchell, M., Pulvino, T., Stafford, E., 2004. Price pressure around mergers. Journal of Finance 59, 31–
63.
Morris, S., Shim, I., Shin, H.S., 2017. Redemption risk and cash hoarding by asset managers. Journal
of Monetary Economics.
Morris, S., Shin, H.S., 2004. Liquidity black holes. Review of Finance 8, 1–18.
Nanda, V., Narayanan, M., Warther, V.A., 2000. Liquidity, investment ability, and mutual fund
structure. Journal of Financial Economics 57, 417–443.
Ofek, E., Richardson, M., 2003. Dotcom mania: The rise and fall of internet stock prices. The Journal
of Finance 58, 1113–1137.
Pan, K., Zeng, Y., 2017. Arbitrage under liuquidity mismatch: Theory and evidence from authorized
participants of bond ETFs.
Pontiff, J., 1996. Costly Arbitrage: Evidence from Closed‐End Funds. The Quarterly Journal of
Economics 111, 1135–1151.
Poterba, J.M., Shoven, J.B., 2002. Exchange‐Traded Funds: A New Investment Option for Taxable
Investors. American Economic Review 92, 422–427.
Schmidt, L., Timmermann, A., Wermers, R., 2016. Runs on money market mutual funds. The
American Economic Review 106, 2625–2657.
Shleifer, A., Vishny, R.W., 1997. The limits of arbitrage. The Journal of Finance 52, 35–55.
Sialm, C., Starks, L.T., Zhang, H., 2015. Defined contribution pension plans: Sticky or discerning
money? The Journal of Finance 70, 805–838.
Stein, J.C., 2005. Why are most funds open‐end? Competition and the limits of arbitrage. The
Quarterly Journal of Economics 120, 247–272.
Warther, V.A., 1995. Aggregate mutual fund flows and security returns. Journal of financial
economics 39, 209–235.
30
Panel A: The growth of the corporate bond market Panel B: The holdings of corporate bond market participants
Panel C: The growth of the corporate bond ETFs Panel D: The growth of the corporate bond mutual funds
Fig. 1: The evolution of the corporate bond market and its participants
Panel A plots the growth of the corporate bond market since 2000. The left hand axis and black line represent the total dollar amount outstanding in corporate bonds.
The right hand axis and the red bars document the total amount of new issuance each year. Panel B plots corporate bond assets held by different market participants
since 1990. The data comes from the Federal Reserve Board. The left axis plots the sum of assets held by life and property‐casualty insurers using the dashed line and
ETFs and mutual funds with the straight line. The right axis plots the holdings by broker‐dealers with the dotted line. Panel C plots the growth in assets under
management and the number of corporate bond ETFs. The left axis document the total assets under management. The solid line represents all taxable long‐term bond
fund assets from ICI Factbook, while the dashed line depicts the total corporate bond assets held by long‐term ETFs from the Federal Reserve Board. The right axis
and red bars plot the number of long‐term taxable bond mutual fund offerings from ICI. Panel D plots the growth of assets and number of open‐end mutual funds as
in Panel C.
31
Fig. 2: The yield on government bonds during 2013
This figure presents the daily closing yield on the five‐ and ten‐year Treasury bond during 2013. Important dates related to our period
of turmoil, the Taper Tantrum are denoted by the vertical lines. The first line represents the day that the episode began, May 22, 2013,
when the Federal Reserve Chairman, Ben Bernanke, first mentioned slowing down the bond buyback program, known as Quantitative
Easing (QE). The second line demarks, June 19, 2013, the date that Chairman Bernanke held a press conference documenting the
economic reasoning behind the tapering of the QE program.
32
Panel A: The change in yield spread of bonds exposed to ETF Taper Tantrum outflows
Panel B: The change in yield spread of bonds exposed to active mutual fund Taper Tantrum outflows
Panel C: The change in yield spread of bonds exposed to index mutual fund Taper Tantrum outflows
Fig. 3: The impact of a one standard deviation change in Taper Tantrum outflows on yield spreads
This figure presents the impact of a one standard deviation increase in turmoil outflows to ETFs in Panel A, active mutual funds in
Panel B, and index mutual funds in Panel C on the cumulative yield spread of constituent bonds exposed to outflows relative to bonds
from the same issuer. The line plots the change in the volume‐weighted average yield of a bond over the maturity‐matched Treasury
rate. The change is measured relative to the bond’s volume‐weighted yield in May of all transaction prior to May 22, the day to initial
Bernanke’s testimony. The bars represent the t‐statistics from the regression described in Table 2.
33
Fig. 4: The arbitrage opportunity of high discount ETF group during the Taper Tantrum.
This figure plots the time series of average monthly deviation, computed as computed as / , of all
ETFs falling in the lower half of the distribution during the Taper Tantrum.
34
Panel A: The average yield spread of bonds in the high and low ETF exposure portfolios
Panel B: The difference in yield spreads of bonds in the high and low ETF exposure portfolios
Fig. 5: The impact of ETF arbitrage on exposed bonds
This figure presents the impact of high ETF arbitrage exposure on portfolios of bonds. In Panel A we present the
average yield spread and Panel B the difference and relevant t‐statistics. For each bond i we compute the ownership‐
weighted average deviation of the set of K ETFs that hold bond during the Taper Tantrum as,
,∑ , ∗ ,
∑ ,,
where , is the amount of bond i held by ETF k in the holdings period prior to the onset of the tantrum episode
and , is the average deviation of ETF k during the Taper Tantrum months. Using this measure we
divide the entire sample of bonds into high and low discount exposure portfolios. The high discount exposure portfolio
includes all bonds with exposure measure less than the median value, i.e. held by ETFs trading at a discount.
35
Table 1
Summary Statistics
Summary statistics by investment vehicle type for the quarterly holdings data released in March 2013 immediately
before the Taper Tantrum are presented below. Panel A presents the distribution of observable summary statistics for
corporate bond exchange‐traded funds (ETFs), Panel B for active mutual funds (AMFs), and Panel C for index mutual
funds (IMFs). Total net assets is the dollar value in millions for all share classes of the fund. # of holdings is the
number of unique bonds held by the fund. % in top 10 holdings documents the percentage of assets concentrated in
the largest holdings of the fund. Turnover is a yearly measure defined by CRSP. Expense ratio is the asset‐weighted
percent expense ratio of the fund for all share classes of the fund. Average duration and average credit quality are the
value‐weighted characteristics of all bonds holdings.
Panel A: ETFs Mean STD 10% 25% 50% 75% 90%
Total net assets ($ mln) 1792.36 4400.34 10.60 30.80 169.00 771.70 5385.60
No of holdings 636.70 1765.45 33.00 106.00 231.00 681.00 1351.00
% in top 10 holdings 21.81 22.35 4.83 8.67 13.78 24.87 43.63
Turnover 0.49 0.74 0.05 0.10 0.19 0.65 1.10
Expense ratio (%) 0.34 0.26 0.12 0.16 0.24 0.45 0.60
Average maturity 6.58 5.68 1.92 2.90 4.91 7.50 12.26
Average effective duration 4.98 3.10 1.84 3.25 4.82 6.05 7.81
Average credit quality 2.81 1.29 1.00 2.00 3.00 4.00 4.00
Panel B: AMFs Mean STD 10% 25% 50% 75% 90%
Total net assets ($ mln) 1549.99 4708.65 25.00 85.60 338.30 1139.75 3104.60
No of holdings 365.18 424.85 20.00 85.00 259.00 473.00 849.00
% in top 10 holdings 31.56 51.17 6.67 11.22 19.05 36.55 85.02
Turnover 1.27 1.76 0.21 0.37 0.69 1.39 3.15
Expense ratio (%) 0.87 0.49 0.35 0.59 0.81 1.07 1.33
Average maturity 6.23 2.90 2.61 4.46 6.38 7.56 9.08
Average effective duration 4.00 1.85 1.42 2.97 4.20 5.10 5.87
Average credit quality 2.62 1.22 1.00 2.00 3.00 3.00 4.00
Panel C: IMFs Mean STD 10% 25% 50% 75% 90%
Total net assets ($ mln) 5782.76 17366.64 121.10 232.90 783.85 2598.10 12650.90
# of holdings 2359.78 3394.15 43.50 346.50 1277.50 2670.50 7494.50
% in top 10 holdings 26.89 33.33 4.03 6.24 11.96 34.38 95.67
Turnover 1.55 2.32 0.23 0.41 0.79 1.32 4.36
Expense ratio (%) 0.62 0.59 0.10 0.13 0.32 1.15 1.37
Average maturity 7.61 4.98 2.99 5.82 6.96 7.50 11.79
Average effective duration 5.28 2.49 2.89 4.42 5.08 5.31 6.47
Average credit quality 4.32 1.01 3.00 4.00 5.00 5.00 5.00
36
Table 2
Regressions of Changes in Bond Yield Spreads Relative to Pre‐Event Levels on Tantrum Outflows
Panel A reports results for exchange‐traded funds (ETFs), Panel B for active mutual funds (AMFs), and Panel C for
index mutual funds (IMFs) of the following monthly regressions
,
, , , , , ,
, , , , , , , ,
∗ ,
∗ , , , .
The dependent variable, ,, is the change in the volume‐weighted average yield of bond of bond
from issuer over the maturity‐match Treasury rate in month . The change relative to the bondʹs volume‐weighted
yield in May prior to May 22, 2013, the day of Bernankeʹs testimony. The months run from 1M (September 2013) to
16M (December 2014) after the end of the turmoil period in the summer of 2013. is an issuer fixed effect, whose
inclusion allowing us to effectively compare the change in yield spreads of bonds from the same issuer subjected to
different levels of fund induced pressures. , is the weighted average monthly flow for all mutual
funds or ETFs that reported bond as a holding immediately before the event. Only bonds with negative flow exposure
are included. The coefficient is multiplied by ‐1. Bond issue level controls included are the log of the issue size in
millions of dollars, ; the number of years remaining until maturity, ; the bond coupon, ,
the median monthly Amihud illiquid proxy, ; and the yield spread of the bond prior to the event, . All
issue level controls are set to their May level. t‐statistics are clustered at the issuer level and presented below the
coefficients. * indicates significance at the 10% level; **, at the 5% level; and *** at the 1% level.
Panel A: ETFs
Variable 1M 4M 7M 10M 13M 16M
Fund Outflow 0.072*** 0.051** 0.042** 0.009 0.022 0.003
(3.69) (2.46) (2.08) (0.52) (1.29) (0.19)
Fund Outflows2 ‐0.004* ‐0.001 ‐0.000 0.001 0.000 0.003
(‐1.81) (‐0.31) (‐0.11) (0.58) (0.00) (1.24)
Log(Size) 0.012 0.004 0.002 0.008 0.012 ‐0.002
(0.73) (0.19) (0.09) (0.55) (0.70) (‐0.12)
Time to Maturity 0.010*** 0.014*** 0.017*** 0.022*** 0.024*** 0.030***
(2.87) (3.33) (4.88) (5.67) (5.21) (6.17)
Coupon ‐0.017** ‐0.040*** ‐0.039*** ‐0.032*** ‐0.013 0.004
(‐2.04) (‐4.17) (‐4.20) (‐4.05) (‐1.43) (0.57)
Amihud ‐0.250 0.340 ‐0.642 ‐0.147 0.584 1.811
(‐0.12) (0.19) (‐0.35) (‐0.11) (0.36) (0.42)
SpreadMay ‐0.194** ‐0.169 ‐0.229** ‐0.316*** ‐0.391*** ‐0.447***
(‐2.10) (‐1.62) (‐2.33) (‐3.72) (‐4.31) (‐4.50)
Time to Maturity * Fund Outflow ‐0.002*** ‐0.002*** ‐0.002*** ‐0.001** ‐0.001 ‐0.001
(‐3.19) (‐4.08) (‐3.74) (‐2.00) (‐1.35) (‐1.30)
Amihud * Fund Outflow 0.309 ‐0.033 0.196 0.075 ‐0.079 ‐0.399
(0.53) (‐0.07) (0.44) (0.25) (‐0.21) (‐0.37)
Issuer Fixed Effects Yes Yes Yes Yes Yes Yes
Number of Observations 2,855 2,799 2,765 2,748 2,673 2,664
R‐Squared 0.666 0.707 0.774 0.854 0.886 0.942
37
Panel B: AMFs
Variable 1M 4M 7M 10M 13M 16M
Fund Outflow 0.021 0.022 0.021 ‐0.022 ‐0.034 0.011
(1.55) (1.05) (0.93) (‐0.80) (‐1.10) (0.45)
Fund Outflows2 ‐0.000 ‐0.001 ‐0.001 0.001 0.001 ‐0.001
(‐0.28) (‐0.87) (‐0.49) (0.82) (0.65) (‐0.49)
Log(Size) 0.054* 0.027 0.028 ‐0.010 0.029 ‐0.002
(1.74) (0.75) (0.73) (‐0.23) (0.66) (‐0.08)
Time to Maturity ‐0.003 0.001 0.004 0.007 0.012 0.028***
(‐1.48) (0.27) (0.99) (0.99) (1.24) (9.40)
Coupon ‐0.041*** ‐0.053*** ‐0.047*** ‐0.077*** ‐0.042 ‐0.011
(‐4.40) (‐3.12) (‐3.14) (‐3.14) (‐1.31) (‐1.06)
Amihud ‐1.900 ‐1.847 ‐1.078 ‐4.995** ‐3.969 ‐2.547
(‐0.64) (‐0.73) (‐0.57) (‐2.37) (‐1.48) (‐1.25)
SpreadMay 0.179*** ‐0.009 ‐0.002 0.094 ‐0.143 ‐0.340***
(6.04) (‐0.08) (‐0.03) (0.43) (‐0.47) (‐4.77)
Time to Maturity * Fund Outflow ‐0.000 ‐0.000 ‐0.000 0.000 0.001 ‐0.000
(‐0.63) (‐0.40) (‐0.61) (0.06) (0.88) (‐0.55)
Amihud * Fund Outflow ‐0.719 0.799 0.505 2.093** 1.723 1.237
(‐0.45) (0.69) (0.59) (2.37) (1.54) (1.50)
Issuer Fixed Effects Yes Yes Yes Yes Yes Yes
Number of Observations 3,744 3,611 3,494 3,380 3,249 3,192
R‐Squared 0.72 0.65 0.72 0.77 0.75 0.90
Panel C: IMFs
Variable 1M 4M 7M 10M 13M 16M
Fund Outflow ‐0.003 ‐0.018 ‐0.011 ‐0.040 ‐0.097** ‐0.089**
(‐0.10) (‐0.50) (‐0.36) (‐1.31) (‐2.49) (‐2.00)
Fund Outflows2 ‐0.005 ‐0.000 ‐0.000 0.003 0.003 0.008*
(‐0.87) (‐0.03) (‐0.05) (0.54) (0.34) (1.82)
Log(Size) 0.044** 0.015 0.008 ‐0.004 0.000 0.038
(2.53) (0.57) (0.42) (‐0.22) (0.02) (1.47)
Time to Maturity 0.000 0.004 0.007*** 0.015*** 0.017*** 0.025***
(0.14) (1.29) (3.18) (6.37) (7.03) (6.83)
Coupon ‐0.015** ‐0.019** ‐0.020*** ‐0.018** ‐0.012 ‐0.006
(‐2.25) (‐2.42) (‐2.67) (‐2.37) (‐1.51) (‐0.72)
Amihud 2.696* 2.735* 2.334* 1.703 1.443 0.892
(1.96) (1.67) (1.74) (1.21) (0.97) (0.32)
SpreadMay ‐0.103*** ‐0.254** ‐0.223** ‐0.317*** ‐0.354*** ‐0.384***
(‐2.95) (‐2.22) (‐2.34) (‐3.33) (‐4.56) (‐2.68)
Time to Maturity * Fund Outflow 0.002 0.004* 0.003 0.003* 0.005** 0.005*
(1.49) (1.79) (1.58) (1.96) (2.51) (1.72)
Amihud * Fund Outflow ‐2.683 ‐3.604 ‐3.576* ‐2.428 ‐1.726 ‐1.047
(‐1.24) (‐1.45) (‐1.79) (‐1.15) (‐0.77) (‐0.25)
Issuer Fixed Effects Yes Yes Yes Yes Yes Yes
Number of Observations 1,703 1,652 1,601 1,599 1,567 1,559
R‐Squared 0.58 0.57 0.58 0.70 0.71 0.83
38
Table 3
The Distribution of Fund Flows by Fund Type
This table shows the statistics for flows to funds of different corporate bond investment vehicles, including exchange
traded funds (ETFs), all mutual funds, index mutual funds, and active mutual funds. The full sample period is from
January 2010 to March 2015. Also presented are flow characteristics in normal periods, Non‐Taper Tantrum, and those
during the turmoil event studied in this paper, the Taper Tantrum. Taper Tantrum months are June, July, and August
of 2013.
Mean Std. Dev. 10% 25% 50% 75% 90%
ETFs
Full Sample 4.75 12.47 ‐2.11 ‐0.36 1.21 6.53 16.62
Non‐Taper Tantrum 4.85 12.35 ‐1.74 ‐0.33 1.37 6.56 16.56
Taper Tantrum 3.51 13.75 ‐6.15 ‐1.16 ‐0.09 6.38 17.21
Active Mutual Funds
Full Sample 1.33 7.73 ‐2.71 ‐0.98 0.19 1.97 5.81
Non‐Taper Tantrum 1.38 7.72 ‐2.55 ‐0.92 0.22 1.99 5.80
Taper Tantrum 0.49 7.93 ‐4.54 ‐2.08 ‐0.37 1.44 5.99
Index Mutual Funds
Full Sample 0.98 6.75 ‐2.27 ‐0.52 0.46 1.81 4.47
Non‐Taper Tantrum 1.04 6.73 ‐2.10 ‐0.46 0.50 1.88 4.67
Taper Tantrum ‐0.07 7.06 ‐3.38 ‐1.83 ‐0.29 0.95 2.85
39
Table 4
Relation between Flow Variability and Fund Characteristics
This table presents the results of regressions of the standard deviation of fund flows in percent on fund characteristics.
Columns (1)—(3) show the relations from cross‐sectional regressions with lifetime standard deviation as the dependent
variable. Columns (4)—(6) show the results of panel regressions with month‐fixed effects using twelve month standard
deviation of flows as the dependent variable. The independent variables include a dummy if a fund is an ETF, ; a
fundʹs expense ratio, ; turnover ratio, ; the log of a fundʹs assets under management,
; the log of total family assets, , and interactions of these variables with the ETF
dummy. In the cross‐sectional regressions all continuous independent variables are set to their average value over our
sample period and are then demeaned. In panel regressions the controls are lagged one month and standard errors are
clustered at the fund level. t‐statistics are presented below the coefficients. * indicates significance at the 10% level; **,
at the 5% level; and *** at the 1% level.
Cross‐Section Panel
Variable (1) (2) (3) (4) (5) (6)
ETF 5.618*** 3.331*** 10.698*** 5.043*** 2.262*** 9.820***
(9.99) (6.44) (2.85) (8.96) (4.66) (3.58)
Expense Ratio 1.136*** 1.216*** 0.371 0.400*
(3.64) (3.88) (1.55) (1.66)
Turnover Ratio ‐0.024** ‐0.023** ‐0.004 ‐0.007
(‐2.38) (‐2.20) (‐0.59) (‐0.96)
Log(Assets) ‐0.078 ‐0.095 ‐0.310*** ‐0.308***
(‐0.94) (‐1.08) (‐5.01) (‐4.85)
Log(Age) ‐2.278*** ‐2.219*** ‐1.637*** ‐1.552***
(‐16.34) (‐15.83) (‐14.55) (‐13.80)
Log(Family Assets) 0.064 0.083 0.103** 0.117**
(0.98) (1.23) (2.04) (2.26)
ETF * Expense Ratio ‐5.207** ‐2.545
(‐2.08) (‐1.45)
ETF * Turnover Ratio ‐0.079 0.053***
(‐1.27) (4.03)
ETF * Log(Assets) 0.528* 0.451
(1.81) (1.64)
ETF * Log(Age) ‐3.434*** ‐3.802***
(‐3.40) (‐3.95)
ETF * Log(Family Assets) ‐0.491* ‐0.437**
(‐1.78) (‐2.04)
Date Fixed Effects No No No Yes Yes Yes
Number of Observations 1,428 1,121 1,121 73,850 54,193 54,193
Adjusted R‐Squared 0.058 0.324 0.334 0.047 0.164 0.175
40
Table 5
The Investment Horizon of Institutional Investors
This table presents information on the investment horizon of the institutional owners of the corporate bond
mutual funds and exchange‐traded funds (ETFs) in our sample. The churn ratio, computed as the sum of
quarterly absolute changes in dollar holdings over average assets under management, is our measure of
institutional investor horizon. The mean and median church ratio is shown for investors that hold corporate
bond ETFs or mutual funds at any point for the entire sample period, January 2010 to March 2015, and for
the period immediately preceding the Taper Tantrum. The tests of the median perform a nonparametric
k‐sample test with the null hypothesis that the k samples were drawn from populations with the same
median. For two samples, the chi‐squared test statistic is computed both with and without a continuity
correction. The difference in means and medians is also presented with the associated p‐values. * indicates
significance at the 10% level; **, at the 5% level; and *** at the 1% level.
Whole Sample Pre‐Taper Tantrum
Holder of Mean Median Mean Median
ETFs 0.176 0.098 0.185 0.088
Mutual Funds 0.107 0.071 0.088 0.058
Difference 0.069*** 0.027*** 0.097*** 0.030***
p‐value 0.000 0.000 0.000 0.000
41
Table 6
The Sensitivity of Fund Flows to Changes in Interest Rates
This table reports the results from the following panel regression over the entire data sample from January 2010 to
March 2015
, ∆ ⁄ ∗ ∆ ⁄ , , .
, is the monthly percentage flow of fund in month . / is the lagged change in either the one‐ or five‐
year Treasury rate normalized by its standard deviation. is a dummy equal to one for ETFs and zero for mutual
funds. , are fund level covariates including the effective duration and credit rating of the fundʹs holdings, one‐, two‐
and three‐month lagged flows, the average monthly fund return over the previous three months, as well as, the expense
log of fund assets, and turnover ratios of the funds. t‐statistics are clustered by date and presented below the
coefficients. * indicates significance at the 10% level; **, at the 5% level; and *** at the 1% level.
Variable I/R=One‐Year Treasury I/R=Five‐Year Treasury
ΔI/Rt‐1*ETFf ‐0.090 ‐0.098** ‐0.114** ‐0.085*
(‐1.60) (‐2.23) (‐2.07) (‐1.74)
ΔI/Rt‐1 ‐0.086 ‐0.047 ‐0.101 ‐0.022
(‐0.90) (‐0.44) (‐0.95) (‐0.26)
ETFf 3.670*** 1.111*** 3.710*** 1.164***
(7.51) (5.32) (7.57) (5.48)
Effective Duration ‐0.440*** ‐0.220** ‐0.440*** ‐0.220**
(‐3.43) (‐2.56) (‐3.43) (‐2.56)
Average Rating ‐0.036 ‐0.039
(‐0.39) (‐0.42)
Flowt‐1 1.642*** 1.642***
(9.25) (9.26)
Flowt‐2 1.301*** 1.296***
(8.30) (8.28)
Flowt‐3 0.920*** 0.923***
(8.25) (8.29)
Fund Rett,t‐3 0.581*** 0.570***
(4.83) (4.68)
Log(Size) ‐0.108** ‐0.110**
(‐2.08) (‐2.11)
Expense Ratio ‐0.125** ‐0.126**
(‐2.10) (‐2.13)
Turnover ‐0.017 ‐0.017
(‐0.35) (‐0.35)
Number of Observations 48,882 31,172 48,882 31,172
R‐Squared 0.016 0.15 0.016 0.16
42
Table 7
Portfolio Composition by Investment Vehicle Prior to the Onset of the Taper Tantrum Relative to Exchange‐Traded Funds (ETFs)
This table presents statistics on the percentage of fund assets held in different investment categories by active mutual funds (AMFs), index mutual funds (IMFs) and
exchange‐traded funds (ETFs) for the period immediately before the onset of the period of turmoil. Panel A shows the statistics for all AMFS and ETFs in our study,
Panel B for AMFs with greater than 80% allocation to corporate bonds and Panel C for IMFs and ETFs only. To test the difference between medians we use
nonparametric two‐sample test. In each panel the mean and median, as well as the difference between the mutual fund and ETF statistics are presented. * indicates
significance at the 10% level; **, at the 5% level; and ***, at the 1% level.
Asset Allocation Mean Median
Panel A: All AMFs AMFs ETFs Difference p‐Value AMFs ETFs Difference p‐Value
Cash 14.10 8.83 5.27*** 0.00 9.07 2.83 6.24*** 0.00
Government bonds 15.99 12.12 3.21** 0.05 10.61 0.34 10.27*** 0.00
Corporate bonds 51.54 74.45 ‐22.91*** 0.00 45.39 93.06 ‐47.67*** 0.00
Securitized bonds 14.87 3.12 11.75*** 0.00 9.31 0.03 9.28*** 0.00
Panel B: All IMFs IMFs ETFs Difference p-Value IMFs ETFs Difference p-Value
Cash 9.36 8.83 0.53 0.84 5.20 2.83 2.37 0.17
Government bonds 36.03 12.12 23.91*** 0.00 42.25 0.34 41.91*** 0.00
Corporate bonds 33.73 74.45 ‐40.72*** 0.00 22.27 93.06 ‐70.79*** 0.00
Securitized bonds 18.71 3.12 15.59*** 0.00 24.51 0.03 24.48*** 0.00
Panel C: AMFs & ETFs with 80% in Corporate AMFs ETFs Difference p-Value AMFs ETFs Difference p-Value
Cash 4.55 2.78 1.77*** 0.00 3.64 1.47 2.17*** 0.00
Government bonds 0.93 1.30 ‐0.37 0.37 0.00 0.00 0.00 0.36
Corporate bonds 92.63 95.58 ‐2.95*** 0.00 93.93 97.22 ‐3.29*** 0.00
Securitized bonds 1.43 0.40 1.03*** 0.00 0.49 0.00 0.49*** 0.00
43
Table 8
Exchange‐Traded Funds (ETFs) Summary Statistics based on Taper Tantrum Arbitrage Exposure
This table presents statistics on exchange‐traded funds (ETFs) based on the average arbitrage opportunity during the months of the Taper Tantrum: June, July, and August of 2013.
For all corporate bond ETFs, we compute the monthly deviation as the average daily percentage difference between the ETF market price and its NAV. ETFs with below median
deviations trade at significant discounts during the turmoil episode and are denoted . ETFs with average Tantrum deviation above the median are called
. The characteristics of ETFs considered are the average credit rating and effective duration of the fundsʹ holdings, the portion of ETF shares outstanding held by
institutional investors, IO, the volatility of fund flows over the prior twelve months, Flow Vol, and the log of ETF assets. Using data from January to April 2013 the average monthly
ETF percentage bid‐ask spread and the creation and redemption intensity measure of Da and Shive (2017), , are shown. is computed as the standard deviation of ETF shares
outstanding in a month divided by the average shares of the ETF outstanding that month, proxies for the level of arbitrage conducted by authorized participants. The average Taper
Tantrum months is also presented. t‐statistics for the difference in means are shown. * indicates significance at the 10% level; **, at the 5% level; and *** at the 1% level.
Pre‐Taper Tantrum
Taper
Tantrum
Group
Credit
Quality
Effective
Duration IO (%) Flow Vol
Bid Ask
Spread (%) Log(Size) CRI
CRI
High Deviation ETFs 2.94 5.82 48.33 7.47 0.360 5.00 0.031 0.017
Low Deviation ETFs 2.97 3.9 44.05 8.15 0.105 5.67 0.040 0.037
Difference ‐0.03 1.92*** 4.28 ‐0.68 0.255*** ‐0.67 ‐0.009 ‐0.020**
t‐statistic ‐0.09 2.48 0.77 ‐0.41 2.71 ‐1.38 ‐0.81 ‐2.43
44
Table 9
The Impact of Arbitrage Opportunities on Bond Yield Spreads
This table presents the results of our tests for the impact of turmoil induced arbitrage ETF on constituent bonds. For all corporate bond ETFs, the monthly deviation is computed as
the average daily percentage difference between the ETF market price and its NAV. For an individual bond we compute its exposure to ETF arbitrage by calculating the ownership‐
weighted average deviation of the ETFs that hold the bond during the Taper Tantrum. The measure is computed as
,∑ , ∗ ,
∑ ,.
The sample of bonds is divided into high and low arbitrage exposure portfolios. Bonds denoted high arbitrage exposure have an , measure below
the median, indicative of being held by ETFs trading at a discount. In Panel A, we present the results of the univariate regression
, , , , where , is the monthly‐volume weighted yield of bond over the maturity‐matched Treasury rate in month and , is a dummy set to one if
the bondʹs exposure measure is below the median. In Panel B, the results of the multivariate regression that controls for bond characteristics, including the years remaining until
maturity, ; the bond’s coupon, ; a proxy for illiquidity, ; and a numerical conversion of the credit rating of the bond, .
, , , , , , , .
t‐statistics with standard errors clustered by issuer are presented next to the coefficient. * indicates significance at the 10% level; **, at the 5% level; and ***, at the 1%.
Panel A: Univariate analysis Month High Arbitrage Exposure t‐stat Constant t‐stat Number of Observations
Jun‐13 ‐0.055 ‐1.29 1.805*** 59.17 4,963 Jul‐13 ‐0.018 ‐0.44 1.662*** 56.27 4,917 Aug‐13 0.025 0.61 1.615*** 53.91 5,039 Sep‐13 0.121*** 2.80 1.563*** 50.50 4,874 Oct‐13 0.150*** 3.30 1.506*** 46.46 4,845 Nov‐13 0.193*** 4.09 1.419*** 42.00 4,748 Dec‐13 0.200*** 4.57 1.287*** 41.39 4,711 Jan‐14 0.177*** 3.76 1.239*** 36.86 4,675 Feb‐14 0.185*** 4.04 1.236*** 37.76 4,578 Mar‐14 0.198*** 4.27 1.157*** 34.82 4,560 Apr‐14 0.197*** 4.41 1.078*** 33.51 4,481 May‐14 0.210*** 5.64 1.032*** 38.10 4,416 Jun‐14 0.203*** 5.54 0.996*** 37.22 4,418 Jul‐14 0.147*** 4.00 1.063*** 39.37 4,342 Aug‐14 0.109*** 2.79 1.179*** 41.07 4,218 Sep‐14 0.039 0.92 1.266*** 40.45 4,219 Oct‐14 0.015 0.30 1.444*** 40.12 4,220 Nov‐14 0.091* 1.77 1.462*** 38.54 4,165 Dec‐14 ‐0.044 ‐0.71 1.722*** 36.94 4,125
45
Panel B: Multivariate analysis
Month
High Arbitrage
Exposure t‐stat
Time to
Maturity t‐stat Rating t‐stat Log(Size) t‐stat Coupon t‐stat Amihud t‐stat
Number of
Observations R‐Squared
Jun‐13 ‐0.021 ‐0.55 0.003* 1.66 0.154*** 38.62 0.112*** 4.57 0.275*** 32.07 10.659*** 8.97 4,583 0.532
Jul‐13 0.028 0.69 0.001 0.33 0.134*** 32.96 0.072*** 2.81 0.256*** 28.89 9.531*** 7.84 4,493 0.463
Aug‐13 0.002 0.06 0.007*** 2.78 0.126*** 29.14 0.072*** 2.62 0.254*** 26.35 8.373*** 7.00 4,512 0.406
Sep‐13 0.101** 2.23 0.005* 1.90 0.132*** 28.82 0.117*** 4.03 0.239*** 24.06 12.989*** 9.09 4,477 0.385
Oct‐13 0.141*** 2.86 0.006** 2.32 0.141*** 26.86 0.087*** 2.81 0.228*** 20.82 13.780*** 7.59 4,412 0.354
Nov‐13 0.200*** 3.69 0.004 1.26 0.132*** 23.19 0.025 0.75 0.233*** 19.36 9.739*** 5.72 4,223 0.313
Dec‐13 0.199*** 4.09 0.006** 2.32 0.129*** 25.03 0.031 1.01 0.203*** 18.84 9.378*** 6.11 4,232 0.326
Jan‐14 0.149*** 2.73 0.008** 2.57 0.128*** 22.61 0.046 1.35 0.211*** 17.68 14.334*** 7.52 4,240 0.290
Feb‐14 0.140*** 2.64 0.011*** 3.70 0.142*** 25.22 0.048 1.48 0.195*** 16.76 10.621*** 6.20 4,092 0.320
Mar‐14 0.101* 1.87 0.013*** 4.34 0.127*** 22.19 0.064* 1.90 0.207*** 17.47 10.206*** 6.39 4,124 0.286
Apr‐14 0.110** 2.13 0.010*** 3.70 0.126*** 22.55 0.074** 2.25 0.182*** 15.86 12.225*** 6.96 4,070 0.280
May‐14 0.083** 2.00 0.012*** 5.23 0.100*** 22.45 0.011 0.42 0.194*** 21.23 10.215*** 7.41 3,974 0.333
Jun‐14 0.058 1.40 0.013*** 6.06 0.103*** 22.01 0.040 1.52 0.170*** 18.25 8.648*** 5.99 3,967 0.305
Jul‐14 ‐0.053 ‐1.29 0.017*** 7.52 0.106*** 23.36 0.005 0.19 0.188*** 20.49 7.751*** 5.63 3,862 0.342
Aug‐14 ‐0.088** ‐2.06 0.022*** 9.01 0.142*** 28.42 0.033 1.22 0.190*** 19.86 6.672*** 5.06 3,648 0.412
Sep‐14 ‐0.122*** ‐2.64 0.016*** 6.30 0.141*** 26.32 0.050* 1.71 0.212*** 20.48 12.522*** 7.49 3,736 0.381
Oct‐14 ‐0.129** ‐2.49 0.012*** 4.35 0.181*** 28.58 0.056* 1.68 0.223*** 18.98 18.599*** 9.58 3,779 0.397
Nov‐14 ‐0.047 ‐0.84 0.008*** 2.65 0.169*** 24.72 0.029 0.78 0.250*** 19.36 10.994*** 6.27 3,690 0.353
Dec‐14 ‐0.255*** ‐3.82 0.015*** 4.29 0.230*** 27.95 0.064 1.49 0.276*** 17.97 18.498*** 10.44 3,701 0.386
46
Table 10
Robustness Regressions of Changes in Bond Yield Spreads Relative to Pre‐Event Levels for Active Funds
This table presents the results of robustness regressions. For various subsets of mutual funds the main specification in Table 2 is repeated with issuer
fixed effects, with only the coefficient and t‐statistics clustered at the issuer level are presented. The months run from 1M (September 2013) to 16M
(December 2014) after the end of the turmoil period in the summer of 2013. Panel A reports the results for active mutual funds with an allocation to corporate
bonds greater than 80%. The results for active mutual funds in the lowest quartile of liquid asset allocation, the sum of cash and government bond
allocation are shown for active mutual funds in Panel B. Panel C shows the results for active mutual funds in the top quartile of fund flow volatility,
respectively. * indicates significance at the 10% level; **, at the 5% level; and *** at the 1% level.
1M 4M 7M 10M 13M 16M
Panel A: Active mutual funds with corporate bond ownership greater than 80%
Fund Outflow 0.009 0.016 ‐0.000 0.003 0.024 0.021
(0.35) (0.45) (‐0.01) (0.10) (1.15) (1.20)
Number of Observations 2,325 2,255 2,155 2,038 1,961 1,917
R‐Squared 0.664 0.542 0.711 0.785 0.791 0.918
Panel B: Active mutual funds in the lowest quartile of liquid assets
Fund Outflow 0.046** 0.017 0.019 ‐0.009 0.020 0.011
(2.18) (0.77) (0.71) (‐0.36) (0.74) (0.69)
Number of Observations 3,023 2,907 2,795 2,651 2,550 2,481
R‐Squared 0.695 0.649 0.734 0.842 0.836 0.934
Panel C: Active mutual funds in the top quartile of flow volatility
Fund Outflow 0.009 0.007 ‐0.003 ‐0.008 ‐0.009 0.001
(0.92) (0.55) (‐0.36) (‐0.87) (‐0.79) (0.13)
Number of Observations 2,986 2,910 2,802 2,718 2,603 2,558
R‐Squared 0.715 0.622 0.737 0.794 0.742 0.883