Measuring Bond-Level Liquidity

Abstract

Market liquidity is important for investors, portfolio managers, and policy makers because it affects decision making and portfolio performance. Yet, for all its importance, liquidity is difficult to measure. Although various approaches to measuring systemic, market-wide liquidity have evolved over time, there have been no reliable security-level metrics. The problem is particularly acute for bonds, many of which trade thinly and mostly over the counter. This article describes a bond-level liquidity measure, Liquidity Cost Score (LCS), that fills this void in the fixed-income investors' toolbox. LCS is defined as the cost of a standard, institutional-size round-trip transaction. It is expressed as a percentage of a bond's price and can be aggregated across bonds in a portfolio, as well as compared over time. LCS provides investors with a rigorous and consistent metric and facilitates academic research of market liquidity.

T ransaction cost considerations enter the decision-making process of all participants in fixed-income secondary markets. The timing and size of individual trades are influenced by the trade-off between the cost of trading and the opportunity cost of not trading. Yet, although its importance is recognized, liquidity is not easy to measure.

Academics and policymakers are interested in aggregate market liquidity and use a variety of variables to study or monitor it. Investors, however, are concerned about the liquidity of their portfolios and individual holdings. How does one measure the liquidity of a bond? There has been some success in measuring liquidity in transparent and active equity markets; but in bond markets, it's a challenge because of the sheer number of instruments and the infrequent trading in most of them. Moreover, unlike stocks, which largely trade in exchange-based markets, bonds still trade mostly over the counter. Although some markets (notably USD credit) have regulations requiring that trades be reported, transaction data is scarce in others.

In 2009, Barclays created a bond-level liquidity measure to fill a gap in the bond investor's toolbox. Because of the dearth of transaction data, this measure relies on simultaneous two-way quotes from Barclays' traders who make markets in a significant number of bonds. These are automatically collated, parsed, and saved as part of the Barclays index validation process. On any given day, many more bonds are quoted than traded. bid-ask spreads themselves do not incorporate the market impact of large trades, which is often of interest, but many traders and investors find bid-ask spreads to be sufficiently positively correlated with market-impact costs.

Barclays' bond-level liquidity measure,Liquidity Cost Score (LCS), is defined as the cost of an institutional-size, round-trip transaction; therefore, a lower LCS signifies better liquidity. LCS is expressed as a percentage of the bond's price and can be aggregated across bonds and compared over time. Portfolio managers can use this measure to quantify the liquidity of their holdings and compare them to a benchmark. A consistent quantitative metric facilitates rigorous studies of market liquidity and other market phenomena.

LCS METHODOLOGY

Traders post bid and ask quotes in two different ways: as yield spreads over Treasuries or as bid and ask prices. The former arespread quotes (typical for USD Investment Grade Credit), and the latter areprice quotes (most USD High Yield and non-USD bond markets). As a result, LCS is computed in one of two, conceptually identical, ways:

1

[Figure omitted. See PDF]

2

[Figure omitted. See PDF]

As an example, in March 2016, the ALCOA 5.4% of 4/5/2021, a Ba-rated bond issued in 2011, was quoted with a bid price of 99.89 and an ask price of 101.01. Its LCS is calculated as (101.01 - 99.89) / 99.89 = 1.121, which means it would cost a manager 1.12% of the bond's value to execute an immediate round-trip transaction for a "normal-size" amount.

Every month, thousands of simultaneous bid-ask quotes are collected and matched to the Committee on Uniform Security Identification Procedures (CUSIPs). For every bond, LCS corresponding to each quote is computed and, at the end of the month, averaged into the bond's monthly LCS value.

Any particular bond's bid-ask spread is unlikely to be the "effective" market--that is, the highest bid and lowest offer across all broker/dealers. A trader's quotes for a particular bond are often influenced by his inventory or outlook. A long position may be quoted with a tighter spread to entice a bid, andvice versa . Investors, however, choose their counterparty and shop for best execution, so LCS may overstate "best-execution" cost. However, Barclays' material presence in fixed-income markets ensures that its quotes are not too far from market levels. Moreover, idiosyncratic circumstances are unlikely to persist throughout a whole month, and monthly averaging largely eliminates their influence. Nevertheless, LCS is a conservative measure of transaction costs.

The quality of trader quotes is a key factor and may be uneven across bonds. Actively traded issues are likely to be quoted both at executable levels and uniformly among broker/dealers. The LCS methodology uses the termbenchmark to describe such bonds. Benchmarks are closely monitored high-profile securities with good two-way flow. A trader is unlikely to quote such a bond carelessly because it would signal inattention to the market or weak market-making capability. The benchmark status is determined by two criteria, "on the run" and "high volume". To be on the run, a bond has to meet several conditions--for example, to be a large and recent issue and to have a maturity close to one of the main issuance points (2-, 5-, 10-, and 30-year). However, if a bond has extremely high trading volume (high-volume criterion), these conditions are waived.

Sometimes, trader quotes areindications , as opposed to live, transactable, two-way markets. The LCS methodology determines whether a quote is realistic or an indication. In the latter case, the model widens the bid-ask spread to ensure it is not too narrow compared with the "true" market. bid-ask spreads are never tightened, in the spirit of making LCS a conservative measure.

Last but not least, a bond may have no two-way trader quotes at all. Its bid-ask market must be estimated.¹ In the next section, we look at how this is done.

LCS MODEL FOR NONQUOTED BONDS

In the absence of a trader-quoted market for a bond, the model estimates what investors would likely have to pay to trade this bond. The model works as follows:

Monthly cross-sectional regression analysis is used to estimate a statistical relationship between quoted bonds' attributes and their observed LCS. It is assumed that the same relationship holds for nonquoted bonds, and their LCS is calculated accordingly. Then, LCS is adjusted upward, because a bond without a single trader quote in a month is likely to be less liquid than a quoted bond with similar attributes.² These models vary across markets. Attributes important for, say, EUR covered bonds may not matter, or indeed even exist, in the USD credit market. So although the key set is usually the same, the econometric models reflect specific properties of each market. We will now use the large and diverse USD IG corporate market to illustrate the LCS modeling approach.

Investors will find most attributes used in the LCS model intuitive. Recent and large issues are cheaper to trade than seasoned and small ones, so bond age and issue size matter. High-risk securities (i.e., bonds with wide spreads) tend to be costlier to trade than low-risk ones. A trader taking a position in a high-risk bond will quote wider bid-ask spreads, so some measure of credit risk must be among the model variables.

The decision on whether to use a certain attribute is based on its relationship with observed LCS. The "heat map" in Exhibit 1 segments the universe of trader-quoted corporates by age and issue size while controlling for maturity (hence, four tables), with darker backgrounds for higher LCS. Two clear gradients emerge: LCS increases for higher age and lower issue size.

Exhibit 1

Average LCS by Issue Size and Age, USD IG Corporates, March 2016

[Figure omitted. See PDF]

Next, we look at the historical relationship between observed LCS and credit spread (OAS). The strength and stability of the relationship (Exhibit 2) is striking. Clearly, credit spread has to be one of the model variables.

Exhibit 2

LCS vs. OAS, USD IG Corporates, Trader-Quoted Benchmarks, January 2007-March 2016

Source: Barclays Research.

[Figure omitted. See PDF]

When liquidity is defined as the cost of trading, one needs to examine how trading volume affects it. Intuitively, one might assume a negative relationship between LCS and volume; indeed, this was true during the credit crisis, when LCS and volume moved--or rather jumped--in opposite directions. However, in the recent, more normal environment, there has been no meaningful relationship between volume and LCS. The scatter plot of LCS versus volume for trader-quoted corporates in Exhibit 3 shows no discernible relationship between the two. Yet the LCS model controls for volume because it can become important during periods of market turbulence.

Exhibit 3

LCS vs. TRACE Trading Volume, USD IG Corporates, Trader-Quoted Benchmarks, March 2016

Source: Barclays Research .

[Figure omitted. See PDF]

Two proxies of market risk are VIX and Treasury over Eurodollar (TED) spread. Yet they are highly correlated with OAS (Exhibit 4), so including them in the model is redundant.

Exhibit 4

LCS vs. VIX, USD IG Corporates, Trader-Quoted Benchmarks, January 2007-March 2016

Source: Barclays Research.

[Figure omitted. See PDF]

Some of the more important variables in the model are Option-Adjusted Spread Duration (OASD) × OAS, LCS benchmark status, industry sector and quality, age, issue size, trading volume, and price distance from par. The coefficients of these factors are almost always statistically significant. TheR² of the regression usually ranges between 60% and 80%.

Liquidity in different markets is driven by factors specific to that market. For example, the country of issuer does not matter for USD corporates but is important for Pan-Euro corporates, so models for different markets are implemented individually.³

PROPERTIES OF LCS

Investors are concerned not only with a bond's current LCS, but with its variability. Portfolio managers want to know what the LCS of their portfolio is likely to be in the near future, when they may have to transact. "Current liquidity conditions" should reflect not only the aggregate LCS level, but also intramarket dispersion. The stability of relative LCS rankings within a market is also important: How likely is a bond in a particular LCS quintile to be there next month? Is it possible to compare the liquidity of different markets?

Relationship between LCS Level and LCS Volatility

High spread levels are usually accompanied by high absolute short-term spread volatility. Spread is also a major driver of transaction costs. When market conditions deteriorate, traders demand more compensation for holding inventory. Hence, one might expect the short-term volatility of LCS to be proportional to its level.

This relationship is investigated on a sample of about 450,000 trader quotes for bonds in the USD IG Credit and high-yield (HY) indexes. Separately for IG and HY, each bond's one-month ΔLCS is measured, forming a sample of LCS-ΔLCS pairs. Bonds are then sorted into 10 buckets, based on their beginning-of-the-month LCS. For every such bucket, average absolute value, standard deviation, and distribution of ΔLCS are computed.

Exhibit 5 shows that during the same stressful period, the relationship is both strong and linear, except for the sparsely populated high-LCS buckets. A regression testing this relationship yields highly significant coefficients for both IG and HY.

Exhibit 5

LCS Level and Liquidity Risk in Turbulent Markets, USD IG Credit, December 2007-December 2009

Source: Barclays Research.

[Figure omitted. See PDF]

In addition to the systemic relationship between current liquidity conditions and short-term liquidity uncertainty, portfolio managers are concerned about liquidity tail risk--that is, being stuck with bonds that are costly to trade. To examine this, we sort the bonds in each LCS bucket into percentiles based on their absolute one-month ΔLCS and then measure each LCS bucket's 95th and 99th percentile of one-month ΔLCS. For both IG and HY, high-LCS buckets show higher liquidity tail risk.

Cross-Sectional Distribution and Persistence of LCS

In broad markets, such as USD credit, there is a wide dispersion of LCS, which parallels equally wide distributions of factors that influence LCS--for example, issue size, age, and spread. The LCS distribution contains valuable information about market conditions. LCS dispersion can change significantly. Exhibit 6 shows cross-sectional distributions for the pre-crisis month of July 2007--widely considered a time of very good market liquidity--and the turbulent November 2008. The properties of the two distributions could not be more different. The March 2016 distribution lies between the two extremes.

Exhibit 6

Historical Cross-Sectional Frequency Distribution of LCS, USD IG Credit

Source: Barclays Research.

[Figure omitted. See PDF]

Another important question concerns the stability of LCS for groups of similar securities. For any particular bond, LCS may change significantly but, on average, bonds in a particular place on the liquidity spectrum should remain there a month later. In other words, howpersistent is LCS? How quickly do bonds migrate along the liquidity scale? This question is answered by dividing the Barclays USD IG Credit Index into LCS quintiles and measuring transition rates among different quintiles. Exhibit 7 shows the transition rates for March 2016. In normal liquidity regimes, bonds tend to stay in their last-month's quintile. Not surprisingly, it is particularly true for the most- and least-liquid quintiles (85% and 90%, respectively). At the peak of the credit and liquidity crisis of 2008, the corresponding rates were 63% and 71%.

Exhibit 7

LCS Quintiles:Transition Rates and Average LCS, USD IG Credit, March 2016 (%)

[Figure omitted. See PDF]

Cross-Market Liquidity Comparisons

Portfolio managers are also interested in comparing liquidity across different markets. Exhibit 8 shows historical time series for USD and EUR corporates. For most of the time period, USD LCS was significantly higher than EUR LCS. Is the EUR corporate market more liquid? Although this may be true in an absolute sense, this is not how portfolio managers look at it. They want to know in which market it would be cheaper to trade thesame bond. Comparing market-level LCS says little about relative liquidity because the markets have different bond characteristics, which invalidates the comparison. For example, if one market largely comprises short-duration, newly issued low-spread bonds, while the other consists mainly of long-duration, seasoned high-spread bonds, a lower aggregate LCS of the first market would be expected.

Exhibit 8

Historical LCS of USD and EUR Credit, May 2010-March 2016 (%)

Source: Barclays Research.

[Figure omitted. See PDF]

The USD and EUR corporate markets differ in several attributes important to LCS. As Exhibit 9 shows, USD corporates have longer spread duration and wider spreads. Both contribute to higher LCS. Their product, used in the LCS model, is more than 1.5 times that of EUR corporates. In addition, bonds in the USD market tend to be older and have a smaller issue size. Every one of these attributes is an important determinant of LCS.

Exhibit 9

Market Attributes Driving Corporate LCS:USD vs. EUR, March 2016

[Figure omitted. See PDF]

To evaluate the relative liquidity of these markets, one needs to account for differences in bond attributes. This is achieved by regressing the LCS of the most liquid, trader-quoted USD and EUR corporates on their age, issue size, OASD × OAS, trading volume, and a dummy variable that indicates whether the bond is USD or EUR. If the coefficient on this variable is statistically significant, its sign will show whether the USD market has worse (positive) or better (negative) liquidity.

Exhibit 10 shows the regression results for March 2016. The USD dummy coefficient is positive and significant, indicating relatively less liquidity in the USD corporate market, all other things equal. However, although the average LCS difference between USD and EUR LCS is 0.332 (Exhibit 9), more than a third of it can be explained by market attributes. The true measure of relative liquidity--that is, the LCS difference between identical bonds in the two markets--is 0.201.

Exhibit 10

USD and EUR Corporates:Cross-Sectional Regression Results, March 2016

[Figure omitted. See PDF]

LCS APPLICATIONS FOR PORTFOLIO MANAGEMENT

Measuring Bonds' Relative Liquidity

LCS is an absolute measure that fluctuates with overall market liquidity, so a time series of a bond's LCS does not show where the bond has stood against its peers over time. Another liquidity measure, derived from LCS, isTrade Efficiency Score (TES) . TES is a bond-level liquidity rank ranging from 1 (best) to 10 (worst); it helps to quickly judge a bond's liquidity relative to similar bonds, both currently and over time.

LCS captures the cost of trading but does not directly measure trading flow. Many corporate bonds trade infrequently, so LCS may not reflect the difficulty of implementing large or numerous trades. When comparing similar bonds, traders interested in immediate execution may prefer a bond with a high current trading volume to a bond with the same LCS but lower volume.

TES blends LCS and trading volume into a single relative score that reflects both the cost and flow. Within one market, bond-level TESs are comparable over time and among bonds, and they come close to representing how traders think about liquidity--that is, in terms of both transaction costs and market impact. As a relative measure, TES can serve as a liquidity filter in portfolio construction. It also helps with backtesting investment strategies. Using only low-TES bonds in a backtest shows how realistic the strategy is in practice, and how achievable are its promised returns.

To compute TES, we assign each bond in a particular market to an OASD-adjusted LCS quintile and to a monthly trading-volume decile. (LCS is a product of the bid-ask spread and OASD, so the duration adjustment is necessary for relative-liquidity comparison of bonds with different durations.) Then, these two quantiles are added, and the sum (ranging from 2 to 15) is mapped to a TES ranking from 1 to 10 (Exhibit 11).

Exhibit 11

Trade Efficiency Scores, Barclays USD IG Corporate Index (ex. 144A), March 2016

[Figure omitted. See PDF]

The TES buckets differ in the number of bonds and market value. The TES1 bucket comprises approximately 15% of the corporate market by number of bonds and 32% by market value, whereas the TES3 bucket accounts for 7% of bonds and 7% of market value.

The attributes of bonds in different TES buckets vary substantially and predictably. By construction, low-TES buckets have bonds with low LCS and high trading volume. As Exhibit 11 shows, the average LCS for TES1 is less than a third of that for TES10. Its average monthly trading volume is $392 million per bond compared with $1 million for TES10. Low-TES buckets tend to be populated by large, recent issues. Average issue size decreases dramatically in higher-TES buckets, while average age increases.

Liquidity and Market Efficiency

Market efficiency is an important topic for both academics and investors. The main characteristic of efficiency is howquickly asset prices reflect available information; insufficient liquidity is among the reasons why they may not. In a liquid market, prices adjust rapidly to news and changes in portfolio preferences. With many potential buyers and sellers constantly inquiring, quoting, and trading, prices (and, hence, excess returns⁴ ) quickly reflect an equilibrium of many viewpoints. However, if a market has limited quoting and trading activity, the propagation and evaluation of news is slower. Hence, one way to assess efficiency is to check forprice inertia when past returns help explain current-period returns.

To investigate informational efficiency of the USD IG corporate market, we partition the index into liquidity strata based on TES. One would expect more liquid segments to display less price inertia. The comparison of price inertia in various TES buckets can reveal whether low-TES buckets are indeed more efficient than high-TES ones.

Price inertia is measured by regressing current-month excess returns (ER) on previous-month excess returns.⁵

3

[Figure omitted. See PDF]

For a market with no price inertia, the estimated regression coefficient on the lagged return term would be statistically insignificant. For example, the one-month lag coefficient is not significant for the Treasury Index and SPX, so their previous-month returns do not help explain current-month returns--which is consistent with the common view that these are very liquid markets with prices and returns quickly adjusting to new equilibrium levels. This is not so for the Corporate Index. The lag coefficient (0.34) is statistically significant, and 11% of the variation in the current-month returns is explained by the previous-month excess returns (Exhibit 12).

Exhibit 12

Estimated Autoregression Coefficients by TES Bucket, Monthly Returns, February 2007-April 2015

[Figure omitted. See PDF]

What could explain this pattern in the Corporate Index? Unlike Treasury bonds and stocks, many corporate bonds trade rarely or not at all in a particular month. Hence, changes in investors' views affect infrequently-traded bonds gradually. Eventually, the news does become fully reflected in their prices, but the delayed adjustment causes lagged returns to be positively correlated with current-period returns.

How uniform is price inertia within the corporate market? This analysis is repeated for each TES bucket. In low-TES buckets, lagged excess returns should have little explanatory power (i.e., statistically insignificant coefficients and lowR² ); however, in high-TES buckets, one would expect significant coefficients that explain a meaningful percentage of the bucket's excess return volatility. Exhibit 12 presents the autoregressive model output by TES bucket, which shows a large variation in price inertia in the corporate market.

For the most liquid bucket, TES1, the lagged ER coefficient is statistically zero, and the regressionR² is close to zero. TES1 bonds are relatively cheap to trade and have relatively high trading volumes, so it is not surprising that new information is quickly and fully reflected in their prices and, hence, excess returns.

Beyond TES1, the picture quickly changes. For TES2, the coefficient for the lagged ER term is positive (0.28) and statistically significant: TheR² is 7%. Moving from TES1 to TES2 produces reduction in liquidity, confirmed by their LCS and volume (Exhibit 11). As TES increases, both the lagged ER-term coefficient and regressionR² rise. For TES9 and TES10, the lag coefficient reaches 0.50, with anR² of 24%.

To summarize, market efficiency varies significantly within the corporate market and is determined largely, if not entirely, by liquidity. Also, the results suggest that TES and, hence, LCS, do a good job of partitioning the market by liquidity.

Credit Spread Decomposition

Credit spreads compensate credit investors for the possibility of bond default. However, many studies have shown that spreads of credit bonds are generally much wider than is justified by their subsequent default and recovery experience. One of the explanations for this "excess" spread is expected liquidity cost. LCS can help illustrate this by allowing the decomposition of a bond's spread into expected default loss, expected liquidity cost, and "risk premium" components.

The approach is to regress bond-level credit spreads (OAS) on liquidity cost (LCS) and expected default cost (issuers' market-quoted five-year credit default swaps [CDS]). The intercept term represents a market-level risk premium common to all bonds. Exhibit 13 shows the results of this decomposition from January 2007 through March 2016.

Exhibit 13

Risk Premium, Default, and Liquidity Components of the USD IG Credit OAS, January 2007-March 2016

Source: Barclays Research.

[Figure omitted. See PDF]

Portfolio managers look at spread decomposition to gain certain insights. For example, although the OAS levels in September 2007 and April 2010 were similar, the components of OAS were very different. The OAS in September 2007 consisted mainly of a market-wide risk premium, whereas default cost was the main contributor to OAS in April 2010. During 2008 and early 2009, the risk premium and liquidity cost were the largest components of average OAS. For buy-and-hold investors, who may not need to sell in the foreseeable future, the unusually high risk premium and liquidity components of OAS may have presented an opportunity to add credit exposure.

Liquidity-Adjusted Tail Risk

In times of market upheavals that trigger massive portfolio liquidations, portfolio managers find it difficult to realize the mark-to-market value of their holdings. As a result, actual losses may far exceed the estimates of traditional value-at-risk (VaR) models based on published bid prices. To correct for this, one needs to recognize that losses in tail events are exacerbated by high transaction costs. LCS allows investors to modify their tail-risk models accordingly. One method of adjusting VaR models is to model a bond's midprice, rather than its bid price, and lower it according to the bond's LCS to arrive at a bid price more realistic in adverse market conditions.

To illustrate, we apply both a traditional VaR model and the "LCS VaR" model to three structurally similar portfolios of USD corporate bonds--in the stressful environment of November 2008 and in March 2011 when markets returned to normal (see Exhibit 14). The first two portfolios differ in risk (OAS) and liquidity (LCS). The spread of the "Illiquid High-OAS" portfolio is about twice as high as that of the "Liquid Low-OAS" portfolio. Its LCS is higher as well.

Exhibit 14

Traditional vs. Transaction-Costs-Adjusted 99% VaR, Test Portfolio Losses (% market value)

[Figure omitted. See PDF]

Exhibit 15

Excess Returns of the Liquid Tracking Portfolio vs. the USD IG Credit Index, January 2009-March 2016

Source: Barclays Research.

[Figure omitted. See PDF]

To control for spread, we created a third portfolio, "Illiquid Low-OAS," with an OAS similar to that of the "Liquid Low-OAS" portfolio, but with higher LCS and its tail risk modeled for November 2008. This experiment asks two questions: In extreme scenarios, how much more does it cost to liquidate illiquid bonds compared with equally risky but more liquid bonds? And, do transaction costs affect tail risk in difficult times more than in calm periods?

Exhibit 14 shows that in November 2008, both models predict a bigger loss for the Illiquid High-OAS portfolio than for either the Illiquid or Liquid Low-OAS portfolios. In addition, the LCS VaR model predicts significantly larger losses for all three portfolios compared to the traditional VaR model. Notably, the traditional VaR model, based on spreads only, calculates identical losses for the Liquid Low-OAS and the Illiquid Low-OAS portfolios. The LCS VaR model, however, predicts a 0.6% bigger loss for the Illiquid Low-OAS portfolio (-8.7% vs. -8.1%) due to poorer liquidity. In March 2011, as the markets are returning to normal, the difference between the tail-risk estimates of the two VaR models is smaller, for both the Liquid Low-OAS and Illiquid High-OAS portfolio. LCS data for USD credit are available back to 2007, so investors have liquidity data for the 2008-2009 crisis experience to adjust their tail-risk models.

Benchmark Replication

Portfolio managers often look for ways to obtain credit index beta exposure. In broad markets with thousands of securities, this is not a trivial task. Although good tracking is the main objective, realistic implementation at a reasonable cost is critical. Total return swaps on broad credit indexes are often unavailable or expensive. Credit derivatives tend to track cash indexes poorly. Yet index replication with cash bonds is not easy because it may be hard to decide which bonds are sufficiently liquid. A tracking portfolio needs to be rebalanced on a regular basis, so liquidity is important.

Having a quantitative measure of bond liquidity helps a portfolio manager to objectively select a universe of liquid bonds from which to construct a tracking portfolio. Then, the manager can apply a transparent, rules-based methodology that relies on stratified sampling. An example of such liquid proxy is a 50-bond portfolio that tracks the Barclays USD Credit Index (6,465 bonds as of March 2016). To construct it, we divide the index into five sectors (basic, consumer, financial, technology, and other), and five duration categories (0-3, 3-5, 5-7, 7-10, and 10+). We form "the eligible universe" by selecting, for each of the resulting 25 buckets, the top 20% most liquid index bonds according to their LCS rank, and adding to this set the top LCS quintile of bonds by duration category. Finally, we select 50 bonds by stratified sampling to match the contribution to OASD × OAS, as well as the market value percentage, of the index in each of the 25 buckets.

In March 2016, the LCS of the tracking portfolio was 0.478 versus 0.896 for the credit index. Since January 2009, it has tracked, out of sample, the index with an average monthly tracking error of just -0.5 bps and a monthly tracking error volatility of less than a quarter of the index's excess return volatility over the same period (30 bps versus 130 bps).

Dealer Inventories and Market Liquidity

Beginning in early 2008, corporate bond dealer inventories collapsed after a steady multiyear buildup. Higher capital requirements, diminished risk appetite, and new legislation have kept inventories low. The dramatic decline in inventories was concurrent with an equally dramatic spread widening.

This decline in dealer inventories gave rise to all kinds of speculation about liquidity consequences. The prevailing opinion is that constrained inventories cause a material reduction in liquidity. Indeed, Exhibit 16 shows a concurrent jump in LCS. However, most views have lacked empirical support. LCS helps bring some hard evidence to the discussion.

Exhibit 16

U.S. Dealer Corporate Bond Inventories vs. the Nonfinancial USD IG Corporate LCS, January 2007-December 2012

Source: Federal Reserve Bank of New York, Barclays.

[Figure omitted. See PDF]

However, regression analysis is needed to control for changes in other market factors besides changes in inventories: market risk (OAS), trading volume, and dealer distress (TED Spread). Even after accounting for these other factors that could impact LCS, Exhibit 17 reveals a significant relationship between LCS and dealer inventories in the immediate post-crisis period, confirming and quantifying the negative effect of the reduced inventories on transaction costs.⁶ More precisely, a $10 billion decline in dealer inventories is associated with a 5.2 bps increase in LCS. Assuming 100% annual credit portfolio turnover, this deterioration in liquidity corresponds to a portfolio performance drag of approximately 5.2 bps a year.

Exhibit 17

LCS vs. Dealer Inventories, Regression Results, July 2009-April 2012

[Figure omitted. See PDF]

CONCLUSION

Liquidity Cost Score (LCS) is a bond-level metric that provides an objective, quantitative way to measure individual bonds' liquidity. LCS can be aggregated to the portfolio level and compared over time. A few LCS applications described in this article illustrate how bond portfolio managers can use LCS. Finally, LCS provides valuable and relevant data for academics and policy makers studying and monitoring liquidity in bond markets.

To order reprints of this article, please contact Dewey Palmieri at [email protected] or 212-224-3675.

Footnote

ENDNOTES

The authors would like to thank Siddhartha Dastidar, Ariel Edelstein, and Lokesh Munirajulu for their contribution to LCS research and development.

¹ Such bonds are a minority in most markets. For example, in March 2016, 69% of bonds in the USD IG Corporate Index were quoted, of which 24% were benchmarks.

² The LCS for a bond without quotes in a particular month may not immediately undergo the full adjustment. The model has a smoothing algorithm that takes into account whether the bond was quoted in preceding months.

³ In March 2016, Barclays published LCS for about 18,700 fixed-income securities with the total market value exceeding $47.6 trillion. The LCS asset class coverage includes the following markets: USD and Pan-Euro investment-grade and high-yield credit; USD, Pan-Euro and Asia-Pacific nominal and inflation-linked treasuries and government-related debt; global covered bonds; USD emerging markets; and USD mortgage-backed securities.

⁴ Corporate excess return is total return less the return of a duration-matched Treasury portfolio

⁵ All corporate bonds are trader-priced at the end of each month.

⁶ We include quoted bonds only, to eliminate model dependency.