Exploring the way liquidity risk affects bond spreads, the authors find that illiquidity contributes to bond spreads nonlinearly. Illiquidity also affects different classes of bonds heterogeneously. Although the liquidity effect varies over time, it is highly correlated with different bonds.
How Is This Research Useful to Practitioners?
Although investors generally agree that the liquidity risk premium becomes larger with less liquid bonds, the contribution of corporate bond liquidity to yield spreads has not been fully investigated. The authors find nonlinear effects of illiquidity on yield spreads.
First, increasing illiquidity does not increase yield spreads at a constant rate, because their relationship is nonlinear. The marginal impact of illiquidity on bond spreads increases at an increasing rate when illiquidity is at a medium or high level. Practitioners should be more alert to liquidity risk when illiquidity intensifies.
Second, bonds with different liquidity risk sensitivities behave differently when liquidity changes. For bonds that are highly sensitive to liquidity risk, the marginal effect of liquidity risk on spreads becomes larger and larger. For bonds in the middle or lower quartile, the liquidity effect on spreads is nearly neutral or slightly negative. Overall, highly sensitive bonds react to liquidity changes more drastically at extreme levels of illiquidity.
Third, using a clustering technique, the authors categorize bonds into various groups on the basis of heterogeneous liquidity effects. They find that bonds with a longer time to maturity and a higher ranking, as well as more frequently traded bonds, are less sensitive to liquidity risk. But bonds with the opposite characteristics are more sensitive to liquidity changes and show higher levels of nonlinearity.
Finally, bonds in all groups exhibit similar timing and direction of changes in liquidity measures over time. During financial crises, liquidity contributes different premiums among the groups. During the 2008 financial crisis, the spread of the most sensitive group soared more than one percentage point, while the spread of the least sensitive group increased around 10 bps owing to liquidity shock.
How Did the Authors Conduct This Research?
The authors use an empirical approach by examining the trade data for US corporate bonds from TRACE (Trade Reporting and Compliance Engine), bond and issuer characteristics from CUSIP (Committee on Uniform Securities Identification Procedures), and fundamentals from Compustat. The sample covers the period 1 October 2004 to 31 December 2012 and comprises 5,729 bonds from 1,419 issuers.
The authors use three measures to proxy bond liquidity—bid–ask spread, market spread, and duration between days with trading activity—to capture both idiosyncratic and market illiquidity.
In the first stage, regression models are estimated using liquidity as the independent variable and bond spread as the dependent variable. The authors find that the liquidity effect is significant at the 10% level for about two-thirds of the bonds and significant at the 1% level for about half the bonds. AnF-test is used to assess linearity.
In the second stage, the authors categorize bonds with significant liquidity effects into various groups on the basis of their similarities in reacting to liquidity changes. Using a clustering technique called Ward’s algorithm, they identify seven groups. The two largest groups are characterized by lower spreads and a longer time to maturity; they are less sensitive to liquidity risk and show little nonlinearity. Smaller groups with larger spreads and a shorter time to maturity are more sensitive to liquidity risk and show stronger nonlinearity, especially when the liquidity measure is above the medium threshold.
Finally, the authors examine the time variation of liquidity effects, showing a connection between time, daily liquidity measures, and estimated liquidity effects.
The authors’ approach is novel in that it captures nonlinear effects while avoiding both subjective divisions and such time-varying variables as credit ratings. Using a clustering algorithm, the authors uncover the latent group structure that governs liquidity effects, which may help practitioners identify risky bonds through more objective fundamentals.
Abstractor’s Viewpoint
The impact of liquidity risk should never be underestimated, especially during financial crises. The authors shed light on the management of liquidity risk for bonds.
It is generally believed that holding illiquid assets is compensated by the liquidity risk premium. But the authors find that about one-third of bonds have no significant liquidity effect at the 10% level, with the ratio increasing to over half at the 1% level. In other words, around half of investors are not compensated at all for this risk-taking behavior. Even for the bonds with liquidity effects, the effects for the majority of them are very mild (close to zero) most of the time. The reason is unclear and merits further investigation.
Using a statistical technique, the authors identify a few groups of bonds that are highly sensitive to liquidity effects, which practitioners should pay more attention to. The economic rationale behind the groupings is unclear. Further research is needed to help fixed-income analysts identify the liquidity risk factor.
However, group differences are observed in credit-risk-related measures—including spread, debt ratio, and quality rank of issuer—suggesting that credit risk and liquidity risk are correlated. To avoid liquidity risk, the rule of thumb is to choose bonds with lower credit risk.