Does Liquidity Management Induce Fragility in Treasury Prices? Evidence from Bond Mutual Funds

Abstract

While investors conventionally view the U.S. Treasury market as a safe haven, regulators have concerns about the increasing fragility of the Treasury market. In 2016, Jerome Powell, the current chair of the Federal Reserve, pointed out that “spikes in volatility and sudden declines in liquidity have become more frequent in both Treasury and equity markets … [t]here is also evidence that liquidity shifts more rapidly and hence is less predictable in these markets” (Powell 2016). Several recent episodes in the Treasury market exemplify this statement, including the “taper tantrum” in 2013, the “flash rally” in 2014, and the COVID-19 turmoil in March 2020.¹ It is not completely clear what economic mechanism drives increasing fragility in the Treasury market—the most liquid market in the world.

In this paper, we argue that the common practice of liquidity management among open-end corporate bond mutual funds contributes to increased fragility in the Treasury market. Mutual funds that invest in corporate bonds (or other illiquid assets) perform liquidity transformation—holding illiquid securities but issuing liquid claims (ie, fund shares) to investors. These funds often face run risks arising from strategic complementarities among investors (e.g., Chen, Goldstein, and Jiang 2010; Goldstein, Jiang, and Ng 2017). To mitigate the risk, bond mutual funds actively engage in liquidity management, that is, they maintain a large amount of cash-like or highly liquid assets as a buffer for investor withdraws (see Choi et al. 2020; Jiang, Li, and Wang 2021). As a result, their tradings on liquid assets—mostly Treasuries—appear to be excessively sensitive to investors’ demands for liquid claims. We argue that this liquidity management can potentially transmit and concentrate the nonfundamental shocks, which are driven by fund flows, onto the price of Treasuries held by the funds, generating excessive volatility in the Treasury market. Such an effect has been particularly pronounced in recent years as the total size of open-end funds that invest in illiquid assets has increased manifold.²

To test the asset pricing implications of liquidity management on Treasuries, we focus on global open-end corporate bond mutual funds (for brevity, labeled as “bond funds” hereafter) from 2002 through 2021.³ Corporate bond funds are ideal for testing our argument because they have several unique features. First, bond funds usually trade two major asset classes with distinct liquidity levels: U.S. Treasuries and corporate bonds. Second, detailed data on fund holdings is available at the quarterly frequency for a long period, which allows us to analyze funds’ trading behaviors directly. Third, we can measure investors’ demand for liquid claims precisely by fund flows. Last, including global corporate bond funds is important when we exploit the 2017 Liquidity Risk Management Rule to pin down the causal impact of liquidity management.

Before testing the asset pricing implications, we first confirm our premise that bond funds indeed use U.S. Treasuries as a buffer to manage liquidity. Specifically, we examine whether bond funds disproportionately adjust their holdings of Treasuries and corporate bonds in response to fund flows. We find that, for example, with a 1% fund inflow, funds increase their holdings in U.S. Treasuries by about 1.42%, but increase their holdings in corporate bonds by only 0.84%. Moreover, the excessive purchase of Treasuries to contemporaneous fund flow tends to revert in the subsequent quarter. To corroborate the evidence, we take advantage of the global coverage of our bond fund holding data and further examine the trade-to-flow sensitivity of other government bonds.⁴ We find that the trade-to-flow sensitivity of other government bonds is 1.18, which is between that of U.S. Treasuries and corporate bonds. This pecking order, as suggested by trade-to-flow sensitivities, is consistent with liquidity management and increases in asset liquidity.

Then, we examine how liquidity management affects the prices of Treasuries, which are arguably the most liquid assets in the world. As bond funds aggressively trade Treasuries as a liquidity buffer in response to fund flows, this could generate strong price pressure on Treasuries. The fund flow–induced price pressure can cause excessive price fluctuations in Treasuries with high bond fund ownership (see argument in, e.g., Greenwood and Thesmar 2011). Moreover, such exposure becomes a defining factor for Treasury prices as mutual funds are now a major participant in the Treasury market. Liang (2020) estimates that the marketable Treasury shares held by long-term mutual funds increased from 3% in 2008 to 8% in 2019—more than the amount held by banks and broker-dealers, and that bond funds trade much more actively than other major market participants, such as pension funds and insurance companies.

To this end, we first hypothesize that there should be a positive association between the excess return volatility of Treasuries and bond fund ownership.⁵ We exploit the cross-sectional variations of Treasuries’ bond fund ownership; these variations can rule out confounding effects from the time series and better pin down the underlying economic mechanism. Excess volatility is calculated as the standard deviation of Treasuries’ daily risk-adjusted returns, which are the residuals from either a market model that adjusts for average returns on Treasuries (following Frazzini and Pedersen 2014) or a three-factor model that adjusts for average returns on Treasuries, investment-grade corporate bonds, and junk bonds (following Choi, Kronlund, and Oh 2022).⁶

To examine our conjecture, we conduct Fama-Macbeth regressions controlling for Treasuries characteristics, including time-to-maturity, coupon rate, on-the-run status, bid-ask spread, and issue size. We indeed find that Treasuries’ excess return volatility is positively associated with bond fund ownership. Such an effect is statistically significant and economically meaningful. We find that a one-standard-deviation increase in bond fund ownership is associated with a 0.0029% higher volatility of market-adjusted daily returns. For reference, during our sample period from 2002 through 2021, the average market-adjusted excess return volatility of Treasuries was 0.065%, so the effect is about 4.4% relative to our sample mean.⁷

Next, we link liquidity management directly to Treasury excess return volatility. To gauge the extent to which a fund uses Treasury positions to buffer flow shocks, we first construct a fund-level measure called liquidity management intensity (LMI). At each quarter, LMI equals the coefficient estimate from a univariate regression of a fund’s net purchase of U.S. Treasuries on contemporaneous fund flows over the previous 12 quarters. Intuitively, when one bond fund conducts more liquidity management with Treasuries, its LMI is greater. Consistent with the results on the trade-to-flow sensitivity on the Treasuries, the average LMI across funds is greater than 1 (equals 1.25). In this way, LMI can capture and take into account all prior factors that motivate liquidity management, such as cash holdings, portfolio liquidity, and fund flow volatility (Chernenko and Sunderam 2016; Jiang, Li, and Wang 2021). We empirically confirm that this is the case in the data: funds with low cash holding, high flow volatility, and high fund illiquidity tend to exhibit strong LMI.

We then split our sample of bond funds based on liquidity management intensity in each quarter: the top one-third are high-LMI funds, whereas the rest are low-LMI funds. We hypothesize that ownership of high-LMI funds induces stronger excess volatility of Treasury returns than does ownership of low-LMI funds. The evidence confirms this conjecture: the effect of ownership on volatility for high-LMI funds is more than four times that for low-LMI funds.⁸ Meanwhile, we confirm that different proportions of Treasury holdings in fund portfolios do not drive the differential effects of high- and low-LMI fund ownership. We further show that our results are robust after excluding the recent COVID-19 crisis period (ie, 2020Q1 and 2020Q2), suggesting that our findings can be generalized beyond the COVID-19 Treasury market turmoil.

While the above findings are consistent with our argument that liquidity management of bond funds induces excess volatility in Treasuries, we are aware of potential endogeneity issues. For example, bond funds may prefer to hold Treasuries with certain characteristics that could lead to higher excess return volatility in the future. Fund characteristics (such as fund performance, particularly during the stress periods) can potentially endogenously determine fund trading behavior.⁹ While such an argument is unlikely from the perspective of liquidity management, we exploit two empirical designs to provide evidence for a causal interpretation.

In our first empirical design, we exploit the effect of the U.S. Securities and Exchange Commission’s (SEC) Liquidity Risk Management Rule adopted in 2017. The Rule sets a minimum requirement for liquid assets for U.S. mutual funds. Such a rule forces funds investing in illiquid corporate bonds to conduct liquidity management with liquid assets; the impact of this policy should be more pronounced for U.S. funds that do not actively manage liquidity beforehand. We use this policy and conduct a difference-in-differences (DiD) analysis. In the DiD analysis, the treatment group consists of low-LMI U.S. funds, and the control group consists of other funds, including high-LMI U.S. funds and all non-U.S. funds. We first verify our premise that low-LMI U.S. funds conduct more liquidity management with Treasuries after the policy is implemented. Specifically, low LMI U.S. funds significantly increase their trade-to-flow sensitivity for Treasuries. In contrast, other funds do not appear to increase their liquidity management intensity after the rule. Furthermore, we examine the effect on Treasuries’ volatility. We find that Treasuries heavily held by these low-LMI U.S. funds experience an increase in excess volatility in the post-event period, consistent with the increased liquidity management with Treasuries of these funds after introducing the Liquidity Risk Management Rule in 2017.

The second empirical design is based on the turmoil in the Treasury market in March 2020. Bond mutual funds experienced large fund outflows during the COVID-19 outbreak period; this provides an ideal opportunity to study whether liquidity management induced price declines in Treasuries. Consistent with prior studies (e.g., Falato, Goldstein, and Hortaçsu 2021; Ma, Xiao, and Zeng 2022), from the beginning of March 2020 (shortly before the World Health Organization announced the global pandemic), bond funds experienced significant fund outflows until the market was stabilized with the Fed’s intervention and bailout (see Figure 1). In our sample, the total fund outflow during this period was more than 5% of bond funds’ pre-event total assets under management (AUM). More importantly, we find that Treasuries heavily held by high-LMI funds (as measured before the event) experienced larger price declines during this period. Also, such price declines were fully reversed after the Fed’s intervention, suggesting that the price declines during the COVID period were not fundamentally driven but arose from liquidity-management-induced price pressure.¹⁰

Fig. 1

Fund flows during COVID-19 pandemic announcement and Fed intervention

This figure plots the weighted average daily fund flows (panel A) and cumulative fund flows (panel B) for all bond funds in our sample from 2020Q1 through 2020Q2.

Open in new tabDownload slide

The two tests based on the 2017 Liquidity Risk Management Rule and the COVID-19 market turmoil not only buttress the identification but also offer new insights into the puzzling pattern of Treasury prices during the COVID-19 crisis. As we know, during periods of financial market turmoil, the prices of Treasuries usually increase due to flight-to-safety and flight-to-liquidity. The COVID-19 crisis, however, was dramatically different, as Treasuries experienced sharp price declines. This phenomenon is puzzling to academics and practitioners, raising concerns about the safe-haven status of U.S. Treasuries (see, e.g., Duffie 2020; He, Nagel, and Song 2022). Our results suggest that this can be an unintended consequence of the 2017 Liquidity Risk Management Rule, as the policy intensified liquidity management of bond funds, concentrating flow shocks onto the price of Treasuries with the COVID-19 outbreak.¹¹

We further extend our study and examine the return comovement of Treasuries for completeness. Return correlations among individual assets within a particular asset class largely determine the total return variance of the asset class, which is another natural measure of fragility. Thus, the study of return comovement of Treasuries can further shed light on the systematic risk in the Treasury market. Specifically, we follow Anton and Polk (2014) to calculate common ownership of bond funds for each pair of Treasuries and find that common ownership positively forecasts comovement among Treasuries, consistent with the studies of Greenwood and Thesmar (2011) and Anton and Polk (2014) on equities. More relevant for market fragility, we show that the association between common ownership and the Treasury return comovement is stronger during downside markets than during upside ones. On the contrary, there is no such asymmetric pattern on corporate bonds. These findings highlight the uniqueness of Treasuries in liquidity management and are consistent with the intuition on the urgency of selling Treasuries to meet fund redemption during downside markets.

Related Literature. Our study contributes to several strands of literature. First, it is closely related to the growing literature on financial fragility and liquidity management of mutual funds. When mutual funds perform liquidity transformation—holding illiquid assets but issuing liquid claims to investors—they are often subject to financial fragility due to strategic complementarities among investors (e.g., Chen, Goldstein, and Jiang 2010; Goldstein, Jiang, and Ng 2017); this was particularly true during the COVID-19 pandemic (for detailed evidence, see Falato, Goldstein, and Hortaçsu 2021). To mitigate this financial fragility, mutual funds use cash or cash-like assets to manage their liquidity needs (see Chernenko and Sunderam 2016; Aragon, Ergun, and Girardi 2022; Jiang, Li, and Wang 2021). In addition, Jotikasthira, Lundblad, and Ramadorai (2012) show that emerging market funds prefer to trade assets in more liquid markets when accommodating fund flow shocks. The main focus of these studies is to identify the strategic complementarities among investors or to document the liquidity management of mutual funds. Our study incrementally extends this literature by systematically studying how global corporate bond mutual funds trade different asset classes in response to fund flows and by examining the pricing impact of liquidity management on the buffer assets, that is, Treasuries. We show that liquidity management can induce sizable price pressure and volatility among Treasuries. Our finding based on the SEC’s Liquidity Risk Management Rule of 2017 is also novel to the literature, as it helps to pin down the causal impact of liquidity management on the excess volatility of Treasuries.¹²

Ma, Xiao, and Zeng (2022) is the contemporaneous work most related to our study. While we share a similar motivation (liquidity management with Treasuries), our paper has different focuses. Ma, Xiao, and Zeng (2022) look solely at the COVID-19 period, documenting the liquidation behavior of various types of fixed-income mutual funds and showing evidence of price pressure from imputed fund outflow. By comparison, our study focuses on a more general setting by using a direct measure of fragility (excess return volatility) and a longer sample period (including both normal and crisis times). Our study has at least three advantages. First, while Ma, Xiao, and Zeng (2022) find that outflow-induced trades generate selling pressures on Treasuries, the role of liquidity management during the COVID-19 crisis is unclear. The differential impacts of high- and low-LMI bond funds on the price declines among Treasuries during the COVID-19 period clearly show the role of liquidity management, complementing the work of Ma, Xiao, and Zeng (2022). Second, one of our important innovations is exploiting the 2017 Liquidity Risk Management Rule as a quasi-natural experiment to pin down the causal impact of liquidity management on Treasuries. One of the biggest challenges to the literature on liquidity management is to find an exogenous shock to liquidity management and then identify its causal effect on asset prices. We push forward the literature on liquidity management by providing evidence for causal interpretation. Third, our findings based on the 2017 Liquidity Risk Management Rule and the COVID-19 period suggest that the COVID-19 Treasury market turmoil may have been an unintended consequence of the 2017 Liquidity Risk Management Rule, shedding light on the puzzling and unconventional Treasury market during the COVID-19 period (see, Duffie 2020; He, Nagel, and Song 2022).

Our study also has policy implications and sheds light on the debate over liquidity management policies. The extension to the Secondary Market Corporate Credit Facility, a policy introduced by the Federal Reserve to boost the liquidity of corporate bonds, together with our study on the 2017 Liquidity Risk Management Rule, can provide policy guidance on how the Federal Reserve can prevent Treasury fragility in the future.

Our study is also related to some contemporaneous studies on the economic mechanisms underlying the COVID-19 Treasury market turmoil, particularly on why the Treasury market failed to follow the crisis playbook in March 2020. For example, Duffie (2020) emphasizes the frictions in the market-making mechanism, whereas Schrimpf, Shin, and Sushko (2020) and Kruttli et al. (2021) highlight the role of margin spirals and hedge funds. He, Nagel, and Song (2022) focus on the interaction between leveraged investors financing with repo and dealers subject to balance sheet constraints. We complement this strand of literature by providing a novel perspective from liquidity management of bond funds and the 2017 Liquidity Risk Management Rule. Specifically, we argue that when bond funds, particularly those with large illiquid corporate bond holdings (e.g., Falato, Goldstein, and Hortaçsu 2021), experienced large fund outflows during the COVID-19 pandemic, liquidity management induced dramatic fire sales on Treasuries from those funds, which at least partially contributed to the Treasury market turmoil during the COVID-19 pandemic. Also, the 2017 Liquidity Risk Management Rule intensified bond funds’ liquidity management before the COVID-19 crisis, which in turn strengthened the sales of Treasuries during the crisis. Another important difference between our study and this strand of literature is that we focus on a long sample period and use data with rich information (e.g., detailed bond holdings and fund characteristics). The more generalized setting not only allows us to conduct cross-sectional tests to pin down the underlying mechanism but also helps demonstrate that liquidity management, together with the fast-growing bond fund sector, has contributed to increased fragility in the Treasury market over the past decades.

A few papers also examine the COVID-19 episode and the corporate bond market rather than the Treasury market (our focus). For example, Kargar et al. (2021) and O’Hara and Zhou (2021) examine the liquidity and micro-structure of the corporate bond market. Haddad, Moreira, and Muir (2021) find that corporate bonds with better credit ratings tended to exhibit more severe price crashes, which were likely driven by the selling pressure from mutual funds. Jiang et al. (2022) find that corporate bonds with higher latent fragility, measured by the asset illiquidity of their mutual fund holders, experienced more negative returns in March 2020.

Finally, our paper is related to the large body of literature on the role of institutional trading in generating price impacts and financial fragility. Edmans, Goldstein, and Jiang (2012) and Lou (2012) show that fund flow–induced trading has a significant price impact on stock markets. Jiang (2023) highlights the usage of high leverage, which can generate crash risk in stock prices. Anton and Polk (2014) show that fund common ownership forecasts the return correlation between stocks. Greenwood and Thesmar (2011) estimate the correlation between fund flows among mutual funds and link the correlated fund flows to the stock return comovement. Huang, Song, and Xiang (Forthcoming) document that the correlation between mutual fund flows contributes to a large portion of the variance-covariance in anomaly returns. Our study contributes to this literature by focusing on the role of liquidity management and bond funds. We find that liquidity management may lead to excess volatility and exacerbate the contagion effect during market turmoil, even in the most liquid market.

1 Background, Data, and Methodology

1.1 Background

Recent decades have witnessed the fast growth of open-end mutual funds that invest in relatively illiquid assets (e.g., corporate bonds). According to a report by the Investment Company Institute (2022), the total assets under management of worldwide regulated open-end bond mutual funds increased from 6.1 trillion USD in 2010 to about 13.7 trillion USD in 2021.

At the same time, the total mutual fund ownership of Treasuries also increased dramatically, making mutual funds an important component in the Treasury market. Within the entire Morningstar bond fund universe, the total ownership of outstanding Treasury securities rose from about 5% in 2002 to more than 13% in 2021. This pattern is also confirmed by Liang (2020), who documents that the number of marketable Treasury shares held by long-term mutual funds in 2019 exceeded the amount held by banks and broker-dealers.

Unlike other major investors in the Treasury market—such as insurance companies, pensions, or sovereign wealth funds, who tend to be passive and to buy and hold—open-end mutual funds trade actively in the Treasury market to accommodate fund flows. Such trade-to-flow sensitivity could be further amplified by liquidity management. Thus, the trading of open-end mutual funds could generate price impacts on Treasuries. This argument echoes several recent studies, such as that of Brooks, Katz, and Lustig (2020).

1.2 Sample construction

We obtain data on global actively managed open-end corporate bond mutual funds (termed “bond funds” hereafter) from Morningstar, which include detailed information on fund characteristics, such as fund returns and total net assets (TNA), as well as bond funds’ portfolio holdings from 2002Q3 at the quarterly frequency. To obtain a full list of global bond funds, we start with the entire Morningstar universe and apply the following sample filters: (1) we keep funds with Morningstar Global Board Category Group of “Fixed Income”; (2) we exclude exchange-traded funds (ETFs) and keep only open-end funds (Morningstar universe code equals “FO”); (3) we keep funds focusing on corporate bond investment by dropping funds with Morningstar categories that contain the following keywords: “mortgages,” “backed,” “government,” “money market,” “equity-biased,” “inflation,” “banking,” “bank loan,” “muni,” and “preferred stock”; (4) we drop funds with legal structures that are not considered to be open-end mutual funds (ie, FCPE, open-ended protected cell company, and unit trusts);¹³ and (5) we keep funds with US Treasury holding histories.

Note that our sample includes passive index funds. The motivation is based on recent work of Koont et al. (2023) and Choi, Cremers, and Riley (2024). Both studies show that passive funds in fixed income markets actively manage their portfolios, trading off index tracking against liquidity transformation. In this sense, index funds also have a liquidity management motive, and thus, we include index bond funds in our sample. We find that index funds and actively managed bond funds in our sample have similar turnover in Treasuries. Also, our main results are robust to excluding index funds (see Appendix Table A9).¹⁴

We summarize how each sample filter affects the number of funds in our sample in Appendix Table A2, panel A. Our final sample includes 5,667 unique bond funds worldwide from 2002Q3 through 2021Q4 (identified by a Morningstar portfolio identifier, ie, masterportfolioid). Note that after applying these filters, Treasury bond funds and money market funds are excluded from the sample.

In Appendix Table A3, we provide descriptive statistics on our bond fund sample across different dimensions, including domicile countries (panel A), base currencies (panel B), Morningstar major holding types (panel C), and Morningstar fixed-income style (panel D). In a nutshell, bond funds in our sample mainly come from the United States and other developed countries, are mainly based on USD and EUR, invest mostly in fixed-income assets, are less interest-rate sensitive, and are more exposed to credit risks.

We obtain data on U.S. Treasuries from the Center for Research in Security Prices (CRSP). During our sample period of 2002Q4 through 2021Q4, there were 2,435 Treasuries in CRSP.¹⁵ At each quarter, we require each Treasury security to have (1) at least 30 nonmissing daily returns, (2) nonmissing price and shares outstanding to compute mutual fund ownership, and (3) more than 6 months remaining time-to-maturity. Our final sample contains 1,623 unique U.S. Treasuries (identified by bond CUSIP).¹⁶ The effect of each sample filter on the number of U.S. Treasuries in our sample is summarized in Appendix Table A2, panel B.¹⁷

The exclusion of Treasuries with time-to-maturity of less than 6 months is motivated by practice among practitioners and prior studies. Particularly, well-known fixed income indices, such as J.P. Morgan fixed-income indices, include only bonds with remaining maturity of 6 months or more (e.g., J.P. Morgan 2020). This choice is unsurprising as bonds near maturity (within 6 months) are rarely traded and lack liquidity. The lack of liquidity makes these bonds behave oddly in yields and more subject to erroneous prices (e.g., Gebhardt, Hvidkjaer, and Swaminathan 2005; Gürkaynak, Sack, and Wright 2007). As a result, prior studies (e.g., Harris and Piwowar 2006; Todorov 2020; Huynh and Xia 2021; Bali, Subrahmanyam, and Wen 2021; Ma, Xiao, and Zeng 2022) often exclude bonds maturing within 6 months or 1 year. After carefully considering the choice between 6 months and 1 year, we selected 6 months as the cutoff in our main analysis to keep our sample as close as possible to the entire CRSP sample. Nevertheless, we also consider alternative cutoffs (ie, 9 months and 1 year) for robustness and find similar results as reported in Appendix Table A6.

1.3 Fund flows, net buy, and ownership

Next, using bond fund holding data, we calculate how bond funds trade three different asset classes: U.S. Treasuries, other government bonds, and corporate bonds. We identify U.S. Treasuries by matching holding details to the CRSP by CUSIP. For other government bonds, we first identify holdings with the following Morningstar detailed holding type identifiers: Government/Treasury (BT), Government Inflation Protected (TP), Supranational (BZ), Subsovereign Government Debt (DS), and Short-Term Government Bills (GS); then we exclude U.S. Treasuries. In our sample, the majority of these other government bonds are issued by developed countries. For corporate bonds, we include holdings with the following Morningstar detailed holding type identifiers: Corporate bond (B), Corporate Information Protected (IP), Bank Loans (BR), Capital Contingent Debt (CT), Convertibles (BC), Contingent Convertible (CN), and Short-Term Corporate Bills (BB).

1.4 Liquidity management intensity and fund characteristics

Here, Net Buy (US Treasuries)|$_{f,q}$| is fund f’s trading on U.S. Treasuries in quarter q; |$Fund \ Flow_{f,q}$| is fund f’s net flows in quarter q. The regression coefficient, β, captures the sensitivity of fund f’s Treasury trading in response to fund flows; LMI equals the coefficient estimate of β. Intuitively, if funds use Treasuries for buffer and smooth flow shocks, β should be greater than one. For each quarter, we sort all funds on LMI and label the top tercile high-LMI funds, the rest are labeled low-LMI funds.

To investigate the determinants of LMI, we consider the following fund characteristics that could be associated with liquidity management intensity: (1)  Cash Weight, the percentage of cash-type assets in a fund’s portfolio;²⁰  (2)  Fund Flow, the quarterly net flow of the fund; (3)  Fund Returns, the quarterly gross return of the fund; (4)  Fund Flow Volatility, the standard deviation of quarterly fund flows from the past 12 quarters; (5)  US Fund, a dummy variable that equals one if the domicile of the fund is the United States, and zero otherwise; and (6)  Log(Age), the natural logarithm of fund age.

1.5 Risk-adjusted returns, volatility, and correlation

Here, |$P_{i,d}$| is the clean price (or the average of bid and ask, if the clean price is missing) at the day-end from the CRSP.

Third, with the daily risk-adjusted returns, we compute the standard deviation of daily risk-adjusted returns (either market-adjusted or three-factor-adjusted) in a quarter for each U.S. Treasury, denoted as Volatility, as our key proxy for Treasury fragility. To capture the excess return comovement, we calculate the pairwise correlation between a pair of U.S. Treasuries using daily risk-adjusted returns, denoted as |$Corr\_all$|⁠. To examine the asymmetry in the excess return comovement, within each quarter, we split all trading days into two groups (downside markets and upside markets) based on the aggregate Treasury market returns (ie, TRY). We then calculate the excess return comovement among U.S. Treasuries using daily risk-adjusted returns in each group, denoted as |$Corr\_down$| and |$Corr\_up$|⁠. Finally, we take the difference in the excess return comovement between downside and upside markets, denoted as Down-minus-up. These pairwise correlation measures are also calculated for corporate bond pairs for comparison.

1.6 Summary statistics

Table 1 reports summary statistics. As shown in panel A, the market size of bond funds has expanded quickly over time. The number of bond funds increased from 1,070 in 2002 to 3,436 in 2021. The total AUM of all bond funds multiplied nearly 10 times, from 637 billion USD in 2002 to more than 6.3 trillion USD in 2021. Panel B reports summary statistics for bond funds that are included in the trade-to-flow analysis. The average quarterly fund flow is about 1% and the average quarterly fund return is about 0.6%. Panel C reports summary statistics on U.S. Treasuries. The average volatility of market-adjusted daily returns is about 0.065% (or 1.03% annualized), and the average quarterly bond fund ownership is 4.6%. Panel D reports summary statistics of the variables for Treasury pairs in the excess return comovement analyses. The average market-adjusted excess return correlation is 15.3% for pairs of Treasuries.

2 Main Results

2.1 Liquidity management with U.S. Treasuries

We first examine the liquidity management of bond funds. Our analysis is similar to prior studies on liquidity management, for example, Chernenko and Sunderam (2016), Choi et al. (2020), and Jiang, Li, and Wang (2021). While those studies examine cash and cash-like securities, we focus on the role of U.S. Treasuries in liquidity management. As we will show later, using U.S. Treasuries as a tool of liquidity management has important and testable asset pricing implications.

Like banks, bond funds perform liquidity transformation and are subject to potential investor run. That is, while bond funds invest heavily in illiquid assets (e.g., corporate bonds), they issue liquid claims (fund shares) that investors can redeem at the net asset value (NAV) on a daily basis. This liquidity mismatch between fund shares and the underlying assets can generate strategic complementarities among fund investors, leading to financial fragility of funds (e.g., Chen, Goldstein, and Jiang 2010; Goldstein, Jiang, and Ng 2017). To mitigate this fragility, bond funds actively manage their liquidity (see Chernenko and Sunderam 2016; Choi et al. 2020; Jiang, Li, and Wang 2021). We argue that U.S. Treasuries play an important role in liquidity management because they are the most liquid assets and trading them has a limited effect on price. In fact, liquidity management is common not only among bond funds but also among other open-end funds holding illiquid assets (e.g., bank loan funds and real estate funds) and even among commercial banks (e.g., Chen et al. Forthcoming; Ma, Xiao, and Zeng 2022).

In addition to Treasuries, it is also worth examining flow-induced trading of other government bonds. In our sample, the other government bonds are issued mostly by developed countries and have better liquidity than corporate bonds, but are not as liquid as U.S. Treasuries. Based on our hypothesis and on liquidity management practices, bond funds may also prioritize trading other government bonds to accommodate fund flows, but to a lesser extent than in the case of U.S. Treasuries. In this sense, the position of other government bonds can serve as the middle layer between Treasuries and corporate bonds to buffer liquidity shocks.

To verify and quantify the role of U.S. Treasuries in liquidity management, we examine how bond funds trade U.S. Treasuries, other government bonds, and corporate bonds in response to fund flows. We illustrate our test design with the following simplified example. Suppose that a fund has total net assets of $100 at the beginning of the quarter and it allocates $20 to U.S. Treasuries, $10 to other government bonds, and $70 to corporate bonds. Then, suppose there is a 10% outflow during the quarter. If the fund manager does not engage in liquidity management, they will proportionally liquidate the holdings in all three asset classes. That is, the fund will sell $2 of U.S. Treasuries, $1 of other government bonds, and $7 of corporate bonds. As a result, the positions in all three asset classes will decrease by 10%, where the trade-to-flow sensitivity is one on U.S. Treasuries, other government bonds, and corporate bonds.

In contrast, if the fund wants to avoid large price impacts in trading corporate bonds, it will prioritize selling government bonds, especially U.S. Treasuries. That is, the fund is likely to liquidate relatively more Treasuries and other government bonds than corporate bonds, for example, selling $7 of U.S. Treasuries, $2 of other government bonds, and $1 of corporate bonds. As a result, the fund’s total holding of U.S. Treasuries and other government bonds will decrease by more than 10%, while that of corporate bonds will decrease by less than 10%. In this case (with liquidity management), the trade-to-flow sensitivity is greater than one on U.S. Treasuries and other government bonds but is less than one on corporate bonds.

Table 2 reports the results. Columns (1)–(2) are for U.S. Treasuries, columns (3)–(4) are for other government bonds, and columns (5)–(6) are for corporate bonds. We find that the trade-to-flow sensitivity is greater than one for both U.S. Treasuries and for other government bonds. More importantly, bond funds’ trading of U.S. Treasuries is more sensitive to fund flows than it is for other government bonds. For example, as shown in columns (1) and (3), a 1% fund inflow is associated with a 1.42% increase in U.S. Treasury holdings (t-statistic = 48.6) and a 1.18% increase in the holding of other government bonds (t-statistic = 47.7). In contrast, for corporate bonds, the trade-to-flow sensitivity is less than one. As shown in column (5), a 1% fund inflow is associated with an increase in corporate bond holdings of only 0.84% (t-statistic = 57.2). Overall, results presented in Table 2 demonstrate that bond funds prioritize the trading of U.S. Treasuries to actively manage their liquidity conditions. More importantly, the pecking order—as suggested from trade-to-flow sensitivities on U.S. Treasuries, other government bonds, and corporate bonds—is consistent with liquidity management and assures us that the high trade-to-flow sensitivity on Treasuries is not a coincidence. In Appendix Table A4, we construct |$Net \ Buy_{f,q}$| as the proportional change in the market values of bonds held by a fund on a quarterly basis, and we find similar results.

2.2 Bond fund ownership and excess return volatility

Having confirmed that bond funds use U.S. Treasuries in liquidity management, we study the asset-pricing implications of such liquidity management in this subsection. As bond funds trade U.S. Treasuries in response to fund flows, they will transmit nonfundamental demand shocks from fund flows into the prices of Treasuries. Thus, the prices of Treasuries held heavily by bond funds should experience excessive volatility. As mutual funds have become major active traders in the Treasury market, such exposure becomes an important driving force for Treasury prices. In our empirical tests, we examine the cross-sectional association between bond fund ownership and U.S. Treasuries’ return volatility, which is viewed as a conventional measure of fragility. Also, since we focus on the cross-sectional studies, we can avoid confounding effects in time-series tests and shed some light on the increasing Treasury fragility.

Table 3 reports the results. We find that bond fund ownership is indeed positively associated with subsequent excess return volatility for U.S. Treasuries. For example, column (3) shows that, after controlling for all Treasury characteristics, the coefficient on Ownership is 0.070 (t-statistic = 3.7). Such an effect is both statistically significant and economically meaningful: a one-standard-deviation increase in bond fund ownership is associated with a 0.0029% higher volatility of market-adjusted daily returns. For reference, during our sample period from 2002 through 2021, the average market-adjusted excess return volatility of Treasuries was 0.065%. So the effect is about 4.4% relative to our sample mean. Such an effect implies a sizable economic magnitude, which is evident from the comparison between the Treasury and stock markets. Greenwood and Thesmar (2011) find that an increase in mutual fund ownership of 10% leads to an increase in daily stock volatility of about 10% of the sample mean. Column (3) in Table 3 suggests that an increase in mutual fund ownership of 10% is associated with a rise in Treasury return volatility of 10.8% (⁠|$=0.070\times 10\%/0.065$|⁠) of the sample mean. In this sense, the effect of mutual fund ownership on Treasury return volatility is comparable to, if not greater than, the effect of mutual fund ownership on stock return volatility.²²

Finally, we have explored the cross-sectional heterogeneity in liquidity at the Treasury level. We measure Treasury liquidity using the average daily bid-ask spread in each quarter. As shown in Appendix Table A5, the positive association between bond fund ownership and Treasury return volatility is most pronounced among Treasuries with mild liquidity conditions (columns (2) and (5)). For the subsample of Treasuries with the worst liquidity conditions (columns (3) and (6)), the coefficient on fund ownership is large in magnitude but statistically insignificant. For the most liquid Treasury group (columns (1) and (4)), the coefficient is of a small magnitude. These results are not surprising and are consistent with the intuition: bond funds tend to use liquid Treasuries for liquidity management, but bond fund trading exerts large (small) price impacts on Treasuries with low (high) liquidity.

2.3 Liquidity management as the mechanism

To investigate the liquidity management mechanism behind the positive association between bond fund ownership and Treasuries’ volatility, we link bond funds’ liquidity management directly to Treasuries’ excess return volatility. Specifically, we take the following steps. First, we estimate a fund-quarter-level proxy of liquidity management intensity as described in Section 1.4. Second, we decompose bond fund ownership into two parts—ownership from the high-LMI funds versus low-LMI funds—and rerun our analyses in Table 3. If the aforementioned findings in Table 3 indeed arose from liquidity management, our baseline results should come mainly from the ownership of high-LMI funds.

To set the stage, for each fund f in each quarter q, we run a univariate regression of Net Buy (US Treasuries)|$_{f,q}$| on |$Fund \ Flow_{f,q}$| using the samples from the past 12 quarters (at least five nonmissing observations required). The regression coefficient on |$Fund \ Flow_{f,q}$| is denoted |$LMI_{f,q}$|⁠, and it captures how aggressively fund f trades Treasuries in response to fund flows. Intuitively, the more a bond fund conducts liquidity management with Treasuries, the greater its LMI.

We acknowledge that our LMI measure is a statistic of various dimensions of liquidity management motives. Inspired by the literature on mutual fund liquidity management, we conduct Fama-MacBeth regressions to examine the determinants of LMI. We first consider the following fund characteristics that might be correlated with liquidity management intensity: (1)  Cash Weight, (2)  Fund Flow, (3)  Fund Returns, (4)  Fund Flow Volatility, (5)  US Fund dummy, and (6)  Log(Age). In addition, inspired by Jiang et al. (2022), we also construct three proxies for fund illiquidity: Fund Amihud, Fund IRC, and Fund Spread.²³  Table 4 shows that funds with low cash holding, high flow volatility, and high fund illiquidity tend to exhibit strong liquidity management intensity, consistent with results reported in prior studies (e.g., Chernenko and Sunderam 2016; Choi et al. 2020; Jiang, Li, and Wang 2021).

After justifying the LMI measure, we use it to examine the mechanism behind the positive association between bond fund ownership and Treasury excess return volatility. Specifically, at the end of each quarter, we classify bond funds with LMI in the top tercile as high-LMI funds and the rest as low-LMI funds. Then, for each U.S. Treasury, we calculate its fund ownership separately for high- and low-LMI bond funds, denoted as Ownership (High LMI) and Ownership (Low LMI), respectively. Finally, we conduct the same Fama-MacBeth regressions as the ones reported in Table 3 and replace Ownership with Ownership (High LMI) and Ownership (Low LMI).

Table 5 reports the results and confirms our conjecture. Across all specifications, both high- and low-LMI ownership measures are positively and significantly associated with U.S. Treasuries’ excess return volatility. More importantly, the effect is significantly stronger for ownership from high-LMI funds. For example, column (3) shows that the coefficient on Ownership (High LMI) is 0.182 (t-statistic = 2.9) and the coefficient on Ownership (Low LMI) is only 0.042 (t-statistic = 2.9). The difference, 0.140, is statistically significant at 1% (t-statistic = 3.5). In other words, the effect of ownership on volatility for high-LMI funds is more than four times that for low-intensity funds. These results provide direct evidence supporting the liquidity management mechanism.

We further check the robustness of results from Tables 3 and 5 across various alternative empirical specifications. First, we consider an alternative way to estimate daily excess returns for U.S. Treasuries. In particular, Treasury returns could be affected by term-structure of government bond yields. To address this issue, we include three additional term-structure variables of government bond yields and their respective two lags in Equation (8) when estimating daily excess returns for each Treasury in each quarter. These variables are based on Chen, Ferson, and Peters (2010) and include a short-term interest rate (proxied by the 3-month Treasury rate), a measure of the term slope (proxied by the difference between the 10-year yield and the 1-year yield), and a measure of the curvature of the yield curve (proxied by |$y_{3}-(y_{7}+2y_{1})/3$|⁠, where y_i is the i-year fixed-maturity yield). Column (1) of Appendix Table A6 shows that our results are highly similar to our main results.

Second, we consider different sample filters based on the time-to-maturity. In our baseline analyses, we exclude Treasuries with a time-to-maturity of less than 6 months because Treasuries close to maturity are usually illiquid. In columns (2)–(5) of Appendix Table A6, we consider two alternative maturity filters: 9 months and 1 year. Our results remain the same using these alternative maturity cutoffs.

Third, we consider alternative filters on sample periods and bond funds. In Appendix Table A7, we exclude the first two quarters of 2020; in Appendix Table A8, we exclude non-U.S. funds; and in Appendix Table A9, we exclude index funds. Our results remain robust using these alternative filters.

Fourth, we consider one alternative measure of bond fund ownership—the residual fund ownership from regressing on lagged Treasury return volatility. This is to tease out the component of lagged Treasury return volatility in determining fund ownership. Specifically, in each quarter, we conduct a cross-sectional regression of Ownership on the past four-quarter average Treasury excess return volatility and obtain the residuals as an alternative measure of fund ownership, denoted Ownership Residual. Ownership (High LMI) Residual and Ownership (Low LMI) Residual are computed in the same way. Then, we reconduct our analyses in Tables 3 and 5. As shown in Appendix Table A10, we find that our main results remain unchanged using residual fund ownership.