07/15/2021 | Press release | Distributed by Public on 07/15/2021 10:24
July 15, 2021
Beginning in late February 2020, market liquidity for corporate bonds dried up and corporate bond credit spreads soared amid broad financial market dislocations related to the COVID-19 pandemic. The causes of this liquidity dry-up and the spike in corporate bond spreads remain subjects of debate. One question is whether the liquidity dry-up was driven by reduced liquidity supply by dealers, increased liquidity demand by investors, or both. Dealers have a central role in the corporate bond market and use their balance sheets to provide liquidity to corporate bond investors. Thus, one potential explanation of the liquidity dry-up is a reduction in dealers' balance-sheet capacity, or liquidity supply. Another potential explanation of the liquidity dry-up is a rise in liquidity demand, for example, due to investor redemptions at corporate-bond mutual funds.2 Whether liquidity supply or liquidity demand was a key driver of the liquidity dry-up is important for understanding the COVID crisis and how Federal Reserve policy actions affected investor welfare and dealer profits. There is also debate regarding the effectiveness of Federal Reserve policy actions during the COVID crisis and whether these actions affected dealers' liquidity supply.3
To address these questions, we study the price and quantity of liquidity in the corporate bond market and use these measures to disentangle shifts in liquidity supply and demand. Our measures of the price and quantity of liquidity were developed in Goldberg and Nozawa (2021), which studied the data from 2002 to 2016.
Our main results are the following:
In many asset classes, dealers use their balance sheets to make markets and provide immediacy to investors. When dealers' inventory capacity is strained, assets in dealer-intermediated markets can trade at prices away from their fundamental values. In theories of liquidity with inventory frictions, temporary price deviations-or noise in prices-reflect the expected return for providing liquidity.6 A well-known measure of such price deviations is Treasury noise-deviations of individual Treasury yields from a smooth fitted yield curve. Goldberg and Nozawa (2021) measures noise instead in the corporate bond market.7 Each day, we fit a smooth yield curve using bond-level yields for each large issuer of corporate bonds. Specifically, we fit a Svensson (1994) yield curve to the individual yields on each issuer's bonds. We limit the sample to dollar-denominated publicly offered bonds with fixed coupons and no embedded options other than make-whole call provisions and issuers with at least seven issues outstanding.8
Our noise measure is the average, across all bonds in our sample, of the divergence between a bond's market yield and the yield curve of its issuer. A key advantage of our noise measure is that all issuer-specific information is absorbed by the fitted yield curve and thus the measure is not affected by asymmetric information about issuers. We use this noise measure as our liquidity price.
We begin our discussion of the results by describing the evolution of the noise measure, with a particular focus on the COVID crisis. The top panel of Figure 1 provides a historical perspective covering January 1997 to June 2020. The bottom panel shows the evolution of aggregate noise in the corporate bond market during the first half of 2020, which includes the COVID crisis in financial markets. Historically, the average level of noise is 14 basis points.9 At the start of 2020, corporate bond noise was about 11 basis points. Noise rose only a bit in late February and early March amid steep declines in equity markets. However, noise spiked beginning in mid-March, amid signs of large corporate bond sales by investors (Haddad, Moreira, and Muir (forthcoming)). On March 17, the Federal Reserve announced that it would establish a Primary Dealer Credit Facility (PDCF) to address dealers' funding needs. The PDCF would offer loans to primary dealers with a term of up to 90 days, collateralized by a broad range of investment-grade (IG) debt securities. However, noise increased significantly even after the announcement of the PDCF. The PDCF began operations on March 20. Noise reached a peak of 50 basis points on March 23.
On March 23, the Federal Reserve announced the establishment of the CCFs. The PMCCF would purchase new debt from issuing corporations and the SMCCF would purchase corporate debt from investors in the secondary market. After the CCFs were announced, noise declined quickly, reaching 25 basis points in early April. After the announcement on April 9 that CCF eligibility would include some high-yield (HY) bonds, noise declined further, reaching 17 basis points in early May. There was little change in noise around the start of SMCCF purchases. Since then, noise has declined slightly, on balance, averaging 14 basis points in January 2021.
Note: The price of liquidity shown is the aggregate noise measure, or the root mean square error between individual corporate bond yields and the issuer-level fitted yield curve. Investment-grade and high-yield bonds are included in calculating the noise measure shown here. 'PDCF' corresponds to the announcement of the Primary Dealer Credit Facility on March 17. The PDCF began operations on March 20.
Next, we compare the evolution of noise during the COVID crisis and during previous episodes of fixed income market stress. The spike in noise during the COVID crisis was
Figure 2, top panel, shows the evolution of noise during the COVID crisis for IG bonds and, separately, for HY bonds. Noise for IG bonds peaked on March 20-after the PDCF announcement, on the same day as the PDCF entered operation, and before the original CCF announcement on March 23. IG noise then declined rapidly. In contrast, HY noise continued to rise after the March 23 announcement, peaking on April 9, the day of the announcement describing the extension of the SMCCF to include some HY bonds, after which HY noise declined notably. While many factors might have differentially affected IG and HY liquidity over this period, these different time patterns of IG and HY noise point to the CCF announcement as a key explanation for the improvement in market liquidity in late March and in April.10
Note: The price of liquidity shown is the noise measure for the indicated subset of bonds. The top panel shows noise for investment-grade (IG) and high-yield (HY) bonds. The bottom panel shows noise for bonds in two remaining-maturity categories: less than or equal to 5 years remaining maturity and greater than 5 years remaining maturity.
To provide further evidence regarding the effects of the CCFs, Figure 2, bottom panel, shows the evolution of noise for bonds with less than five years remaining maturity and bonds with more than five years remaining maturity. Eligibility for the SMCCF was restricted to bonds with less than five years remaining maturity.11 Notably, noise for sooner-maturing bonds (bonds with less than five years remaining maturity) initially increased much more rapidly than noise for bonds with longer maturities. However, after the first CCF announcement, noise declined much more rapidly for bonds with lower maturity. Noise for such bonds also declined more over the weeks following the second CCF announcement. These results provide further evidence pointing to the CCF announcements as contributing to improved liquidity during the COVID crisis. Nonetheless, this pattern could also be consistent with, for example, other news or announcements differentially affecting sooner-maturing and later-maturing bonds. Indeed, at times of stress such as the 2007-09 financial crisis, the spike and then recovery of noise was sharper for shorter-maturity bonds than for longer-maturity bonds. Figure 3 shows the liquidity price for securities with less than five years remaining maturity, and for securities with more than five years remaining maturity, during the COVID crisis (left panel) and the Global Financial Crisis (right panel). The liquidity price in Figure 3 is z-score standardized to facilitate comparisons across and within each panel.
Note: This figure shows standardized noise for bonds in two remaining-maturity categories (less than 5 years, and greater than 5 years), during the COVID crisis (left panel) and the Global Financial Crisis (GFC, right panel). Each time series for each sample period is separately z-score standardized.
We complement this analysis of the price of liquidity by examining the quantity of liquidity primary dealers provided using their balance sheets. To absorb the demand imbalances that give rise to noise, dealers take long positions in bonds that investors want to sell and short positions in bonds investors want to buy. Thus, we measure the quantity of liquidity as the sum of dealers' gross long and gross short positions. We calculate this quantity using transaction data from the Trade Reporting and Compliance Engine (TRACE). The procedure to convert transaction data into dealer positions is based on cumulating weekly flows to estimate each dealer's net position in each bond, as described in detail in Goldberg and Nozawa (2021). Our quantity measure is the sum of the absolute value of primary dealer positions in each bond.
The top panel of Figure 4 shows the evolution of dealer gross positions since 2002, when the TRACE data began. The bottom panel shows gross positions during the COVID crisis. Dealer gross positions declined moderately over the course of the COVID crisis, from $43 billion in January to $38 billion at its COVID-crisis trough in May. In late 2020, the quantity of liquidity more than retraced its COVID-crisis decline. The liquidity quantity averaged $46 billion in January 2021. Figure 5 shows the evolution of dealer gross positions with different credit ratings (top panel) and with different remaining times to maturity (bottom panel). The moderate decline in dealer gross positions was driven mostly by IG bonds and bonds with greater than five years remaining maturity.
Note: The quantity of liquidity shown is dealer gross positions, or the sum gross long and gross short positions in corporate bonds. Dealer gross positions are calculated for each bond and each dealer by cumulating transactions and then aggregated across dealers, as in Goldberg and Nozawa (2021). Vertical lines mark the 'Stock market peak' on Feb. 19. The 'PDCF' corresponds to the announcement of the Primary Dealer Credit Facility on March 17. The PDCF began operations on March 20. The 'P/SMCCF announced' vertical line is marked at March 23, the 'P/SMCCF expanded' vertical line is marked at April 9, and the 'SMCCF begins buying' vertical line is marked at May 12.
Note: The quantity of liquidity shown is dealer gross positions, or the sum gross long and gross short positions in corporate bonds. The top panel shows noise for investment-grade (IG) and high-yield (HY) bonds. The bottom panel shows noise for bonds in two remaining-maturity categories: less than or equal to 5 years remaining maturity, and greater than 5 years remaining maturity. Vertical lines mark the 'Stock market peak' on Feb. 19. The 'PDCF' corresponds to the announcement of the Primary Dealer Credit Facility on March 17. The PDCF began operations on March 20. The 'P/SMCCF announced' vertical line is marked at March 23, the 'P/SMCCF expanded' vertical line is marked at April 9, and the 'SMCCF begins buying' vertical line is marked at May 12.
We next use a model of noise and dealer gross positions to infer shifts in liquidity supply and demand using a vector autoregression (VAR) and simple sign restrictions. A supply shift is assumed to lead to opposite-sign changes in price and quantity. A demand shift is assumed to lead to same-sign changes in price and quantity.12 Goldberg and Nozawa (2021) describes this methodology in detail.
We define the liquidity supply index as the quantity of liquidity that (according to the model) dealers would provide if the price of liquidity were 20 basis points. Similarly, we define the liquidity demand index as the quantity of liquidity that investors would demand if the price of liquidity were 20 basis points. An increase in the liquidity supply index thus captures an outward shift in liquidity supply. Similarly, an increase in the liquidity demand index captures an outward shift in liquidity demand.
The usefulness of this decomposition is highlighted by the evidence presented in Goldberg and Nozawa (2021), which shows that liquidity supply, but not liquidity demand, is a priced risk factor. That is, liquidity supply shocks help to explain the cross-section of expected corporate bond returns; investors demand higher returns on bonds that have low returns when liquidity supply is low. In addition, liquidity supply is informative about future aggregate corporate bond returns; expected excess returns on corporate bonds are higher when liquidity supply is low. In contrast, their paper does not find evidence that liquidity demand is a priced risk factor-liquidity demand is not informative about cross-sectional or time-series variation in expected excess returns on corporate bonds.13
Figure 6 shows the evolution of our liquidity supply and demand indexes over the past two decades (top panel) and during the COVID crisis (bottom panel). The liquidity supply index fell from pre-crisis levels near $50 billion to a trough of $29 billion on March 20. Liquidity supply recovered sharply after the CCF announcements and exceeded $40 billion in early May. At the end of 2020, liquidity supply regained its $50 billion pre-crisis level. Liquidity demand rose from pre-crisis levels near $35 billion to a peak of $53 billion on March 20, the same day as the trough of liquidity supply. Liquidity demand declined after the CCF announcements before increasing in late 2020 to levels near $43 billion by year-end.
Note: The liquidity supply index is defined as the quantity of liquidity that (according to the model) dealers would provide if the price of liquidity were 20 basis points. The liquidity demand index is the quantity of liquidity that investors would demand if the price of liquidity were 20 basis points. An increase in the liquidity supply index thus captures an outward shift in liquidity supply. Vertical lines mark the 'Stock market peak' on Feb. 19. The 'PDCF' corresponds to the announcement of the Primary Dealer Credit Facility on March 17. The PDCF began operations on March 20. The 'P/SMCCF announced' vertical line is marked at March 23, the 'P/SMCCF expanded' vertical line is marked at April 9, and the 'SMCCF begins buying' vertical line is marked at May 12.
Although the dry-up of liquidity was thus roughly equally driven by lower liquidity supply and higher liquidity demand, the implications for risk premiums of the fluctuations in liquidity supply and demand are quite different. Applying the estimates in Goldberg and Nozawa (2021), the decline in liquidity supply between February 28 and March 20 increased the expected one-year-ahead excess return on corporate bonds 3.4 percentage points.14 In contrast, increases in liquidity demand are associated with quantitatively small increases in expected excess returns, which are not statistically significant.
This evolution of the liquidity supply and demand indexes differed markedly from the 2007-09 financial crisis. During the 2007-09 crisis, liquidity supply fell even more dramatically and the decline was much more persistent. Liquidity supply declined from a pre-crisis peak of $60 billion in early 2007 to a low of $13 billion in December 2008. Liquidity supply recovered to $40 billion only in March 2010 and to $50 billion only in March 2011. In addition, liquidity demand did not show a systematic response to stress events during the crisis, sometimes rising and sometimes declining or remaining little changed.
This note studied dealers' liquidity supply and investors' liquidity demand in the corporate bond market during the COVID crisis, using Goldberg and Nozawa's (2021) model of market liquidity. The evolution of the liquidity price for bonds with differing eligibilities for the Federal Reserve's corporate credit facilities (CCFs) provides suggestive evidence that the CCF announcements were important causal drivers of the improvement in market liquidity in late March and April. The model attributes the liquidity dry-up during the COVID crisis about equally to lower liquidity supply and higher liquidity demand. Our analysis suggests that roughly one-fourth of the large decline in value of corporate bonds between February 28 and March 20 was due to reduced dealer balance sheet capacity.
Boyarchenko, Nina, Anna Kovner, and Or Shachar (2020). 'It's What You Say and What You Buy: A Holistic Evaluation of the Corporate Credit Facilities,' Staff Reports 935. New York: Federal Reserve Bank of New York, July (revised November 2020).
Brunnermeier, Markus K., and Lasse Heje Pedersen (2009). 'Market Liquidity and Funding Liquidity,' Review of Financial Studies, vol. 22 (6), pp. 2201-38.
Duffie, Darrell (2010). 'Presidential Address: Asset Price Dynamics with Slow‐Moving Capital,' Journal of Finance, vol. 65 (4), pp. 1237-67.
Falato, Antonio, Itay Goldstein, and Ali Hortaçsu (2020). 'Financial Fragility in the COVID-19 Crisis: The Case of Investment Funds in Corporate Bond Markets,' NBER Working Paper Series 27559. Cambridge, Mass.: National Bureau of Economic Research, July (revised May 2021).
Gilchrist, Simon, Bin Wei, Vivian Z. Yue, and Egon Zakrajšek (2020). 'The Fed Takes on Corporate Credit Risk: An Analysis of the Efficacy of the SMCCF,' NBER Working Paper Series 27809. Cambridge, Mass.: National Bureau of Economic Research, September.
Goldberg, Jonathan (2020a). 'Dealer Inventory Constraints during the COVID-19 Pandemic: Evidence from the Treasury Market and Broader Implications,' FEDS Notes. Washington: Board of Governors of the Federal Reserve System, July 17, 2020.
Goldberg, Jonathan (2020b). 'Liquidity Supply by Broker-Dealers and Real Activity,' Journal of Financial Economics, vol. 136 (3), pp. 806-27.
Goldberg, Jonathan, and Yoshio Nozawa (2021). 'Liquidity Supply in the Corporate Bond Market,' Journal of Finance, vol. 76 (2), pp. 755-96.
Gromb, Denis, and Dimitri Vayanos (2018). 'The Dynamics of Financially Constrained Arbitrage,' Journal of Finance, vol. 73 (4), pp. 1713-50.
Grossman, Sanford J., and Merton H. Miller (1988). 'Liquidity and Market Structure,' Journal of Finance, vol. 43 (3), pp. 617-33.
Haddad, Valentin, Alan Moreira, and Tyler Muir (2020). 'When Selling Becomes Viral: Disruptions in Debt Markets in the COVID-19 Crisis and the Fed's Response,' NBER Working Paper Series 27168. Cambridge, Mass.: National Bureau of Economic Research, May.
He, Zhiguo, Stefan Nagel, and Zhaogang Song (2020). 'Treasury Inconvenience Yields during the Covid-19 Crisis,' NBER Working Paper Series 27416. Cambridge, Mass.: National Bureau of Economic Research, June (revised November 2020).
Infante, Sebastian, and Zack Saravay (2020). 'What Drives Treasury Re-use? (PDF)' Finance and Economics Discussion Series 2020-103. Washington: Board of Governors of the Federal Reserve System, October.
Kondor, Péter, and Dmitri Vayanos (2019). 'Liquidity Risk and the Dynamics of Arbitrage Capital,' Journal of Finance, vol. 74, pp. 1139-73.
Kargar, Mahyar, Benjamin R. Lester, David Lindsay, Shuo Liu, Pierre-Olivier Weill, and Diego Zúñiga (2020). 'Corporate Bond Liquidity During the Covid-19 Crisis,' NBER Working Paper Series 27355. Cambridge, Mass.: National Bureau of Economic Research, June (revised May 2021).
Liang, J. Nellie (2020). 'Corporate Bond Market Dysfunction during COVID-19 and Lessons from the Fed's Response (PDF),' Hutchinson Center Working Paper 69. Washington: The Brookings Institution, October.
Nozawa, Yoshio, and Yancheng Qiu (forthcoming). 'Corporate Bond Market Reactions to Quantitative Easing during the COVID-19 Pandemic,' Journal of Banking and Finance.
O'Hara, Maureen, and Xing (Alex) Zhou (forthcoming). 'Anatomy of a Liquidity Crisis: Corporate Bonds in the COVID-19 Crisis,' Journal of Financial Economics.
Sharpe, Steven A., and Alex X. Zhou (2020). 'The Corporate Bond Market Crises and the Government Response,' FEDS Notes. Washington: Board of Governors of the Federal Reserve System, October 7.
Svensson, Lars E.O. (1994). 'Estimating and Interpreting Forward Interest Rates: Sweden 1992-1994,' NBER Working Paper Series 4871. Cambridge, Mass.: National Bureau of Economic Research, September.
Vayanos, Dimitri, and Jiang Wang (2013). 'Market Liquidity-Theory and Empirical Evidence,' in George M. Constantinides, Milton Harris, and Rene M. Stulz, eds., Handbook of the Economics of Finance, Vol. 2B. Amsterdam: Elsevier, pp 1289-1361.
1. We thank Andrea Ajello, Giovanni Favara, Don Kim, and Steve Sharpe for helpful comments. We also thank FINRA for providing transaction data. FINRA screening was limited to whether there is sufficient aggregation such that no particular dealer is identified. The views expressed here are those of the authors and do not necessarily represent the views of the Federal Reserve Board or its staff. Return to text
2. Regarding corporate bond mutual funds, see Falato, Goldstein, and Hortaçsu (2020) and Ma, Xiao, and Zeng (2020). Regarding Federal Reserve interventions, see Boyarchenko, Kovner, and Shachar (2020), Gilchrist et al. (2020), O'Hara and Zhou (2020), Sharpe and Zhou (2020), and Haddad, Moreira, and Muir (forthcoming). An additional question is to what extent the liquidity dry-up and the spike in corporate bond spreads are related (e.g., Nozawa and Qiu (forthcoming)). Return to text
3. Regarding the connections between welfare, dealer profits, and liquidity supply and demand, see Kargar et al. (2020). Liang (2020) reviews the literature and discusses potential reforms. Return to text
4. The differences in eligibility related to credit rating and remaining maturity. Return to text
5. Risk premiums are high when dealers' balance-sheet capacity is low not only because investors value market liquidity (the ability to transact in large quantities at low cost) but, perhaps more importantly, because dealers act as a marginal investor and provider of secured financing to investors. For a quantitative equilibrium model, see Goldberg and Nozawa (2021). See also, among others, Gromb and Vayanos (2018), Kondor and Vayanos (2019), He, Nagel and Song (2020), and Infante and Saravay (2020). Return to text
6. See, for example, Grossman and Miller (1988). Return to text
7. Many common measures of liquidity such as bid-ask spreads and price-impact measures are also affected by dealer inventory constraints but 'contaminated' by other factors such as asymmetric information about issuers. See Vayanos and Wang (2013) for a review of liquidity measures and the factors driving them. This note focuses on liquidity measures from Goldberg and Nozawa (2021) that have been empirically and theoretically connected to dealer inventory constraints. Return to text
8. For issuers with between 7 and 15 issues outstanding, we fit a four-parameter version of the Svensson (1994) yield-curve model, and for issuers with more than 15 issues outstanding, we fit the full 6-parameter Svensson model. Return to text
9. As expected, noise in the corporate bond market is generally much higher than in the Treasury market, for which noise averages 2.3 basis points over the same period (Goldberg (2020a)). Return to text
10. Many factors might have differentially affected IG and HY liquidity over this period, highlighting the need for caution in attributing the evolution of our measures to specific announcements. In particular, the PMCCF and SMCCF were announced concurrently with announcements regarding purchases of Treasury and commercial mortgage-backed securities, the Term Asset-Backed Securities Loan Facility, and other measures; these announcements, while not directly bearing on the corporate bond market, could nonetheless have affected IG and HY liquidity differently. Return to text
11. Eligibility for the PMCCF was restricted to bonds with less than four years remaining maturity. Return to text
12. Goldberg (2020b) performs a similar analysis for the Treasury market, rather than the corporate bond market, finding that liquidity supply shifts have been associated with persistent changes in aggregate liquidity across asset classes and changes in corporate financing conditions. In contrast, liquidity demand shifts have historically been associated with only transitory changes in market liquidity and little, if any, change in corporate financing conditions. Goldberg (2020a) extends this analysis to the COVID crisis. Return to text
13. Goldberg and Nozawa (2021) makes sense of this surprising result using an equilibrium model in which dealers provide liquidity and serve as a marginal investor in the corporate bond market. Negative liquidity supply shocks capture tight dealer inventory constraints, which affect the willingness of dealers both to provide liquidity and to bear aggregate risk not directly related to liquidity (such as aggregate default risk). Indeed, in the quantitative equilibrium model, most of the liquidity supply risk premium arises through the latter channel. The liquidity supply risk premium also reflects that liquidity supply shocks are estimated to have very persistent effects on the price and quantity of liquidity, whereas liquidity demand shocks have fairly transitory effects. Return to text
14. We obtain the estimated increase in the one-year-ahead expected excess return by multiplying (1) the sum of liquidity supply shocks between February 28 and March 20; (2) the percentage change on impact of dealer gross positions, in response to a liquidity supply shock, from Figure 3, panel A of Goldberg and Nozawa (2021); (3) the estimated coefficient on log dealer gross positions for liquidity-supply-dominated periods in table VII, panel C2, 'without crisis' column for 52-week horizon. This estimate is a reasonable one because liquidity supply shocks have quite persistent effects on liquidity supply and the time horizon over which we are summing shocks (less than one month) is relatively short. Fully accounting for the dynamic response of positions implies an estimated increase in the expected one-year-ahead excess return on corporate bonds of 3.6 percentage points. Note that the aggregate return forecasting exercise reported in table VII, panel C2 controls for variables considered informative about expected returns: the term spread, the dividend-price ratio, the variance risk premium, and an option-based skewness measure. Return to text
Chikis, Craig A., and Jonathan Goldberg (2021). 'Dealer Inventory Constraints in the Corporate Bond Market during the COVID Crisis,' FEDS Notes. Washington: Board of Governors of the Federal Reserve System, July 15, 2021, https://doi.org/10.17016/2380-7172.2946.