The authors provide a comprehensive comparison of commonly used bond liquidity measures for US corporate bonds. They consider measures that use intraday (high-frequency) data and measures (called liquidity “proxies”) that use daily (low-frequency) data. All high-frequency measures seem to measure transaction costs equally well. Three of the low-frequency proxy measures, which have fewer computational burdens and data, are shown to track the high-frequency measures particularly well.

Because bond trading occurs over the counter rather than through a central exchange, high-frequency intraday quote data are unavailable. This circumstance has led researchers to design their own liquidity metrics based on transaction data or to use low-frequency liquidity proxies developed for equity markets. Such proxies may or may not translate well in the decentralized bond market. Rather than develop yet another potential liquidity measure, the authors offer independent advice to practitioners and academics on the relative quality of high-frequency measures and on the reliability of any of the low-frequency liquidity proxies.

The authors analyze bond transactions over an eight-year period to implement and compare six high-frequency intraday methodologies for gauging transaction costs, price impact, and price dispersion. The high degree of correlation among the six measures indicates that they deliver similar results when measuring liquidity. For that reason, the authors deem them appropriate benchmarks for evaluating low-frequency liquidity proxies. They consider 13 low-frequency liquidity proxies that need only daily information for their calculations. These proxies range from commonly used measures of transaction costs to measures hitherto applied only in equity markets. The authors evaluate the efficacy of the low-frequency proxies by testing their time-series and cross-sectional correlations with the high-frequency liquidity measures.

Of all the proxies examined, the authors consider three superior. The low-frequency liquidity measures developed for equities by Corwin and Schultz (Journal of Finance 2012), Hasbrouck (Journal of Finance 2009), and Roll (Journal of Finance 1984) provide high time-series and cross-sectional correlations with the high-frequency liquidity measures and hold up well in numerous robustness tests.

Traders and securities operations managers will find useful data with respect to transaction costs, as will portfolio managers and analysts concerned with valuation and implementation. The authors’ findings will also be of interest to regulators.

The authors’ observation period runs from 1 October 2004 to 30 September 2012. For the intraday (i.e., high-frequency) measures, the authors rely on the recently introduced enhanced Trade Reporting and Compliance Engine (TRACE), which includes uncapped transaction volumes along with information indicating whether the trade is a buy, a sell, or an interdealer transaction. The bonds included are active for at least a year during the observation period and trade for at least 75% of the trading days in their lifespan. Bonds in default are included only up to three months before the default date to prevent abnormal trading behavior from skewing the results. Finally, the authors exclude erroneous entries and extreme outliers. Because some of the authors’ high-frequency metrics depend on a bond’s fair market value, they use Markit Group’s composite price calculations. For the daily (i.e., low-frequency) liquidity proxies, the authors use daily end-of-day high and low prices, trading volumes, and bid–ask quotes from Bloomberg.

They construct both high- and low-frequency liquidity measures, which are based on very different sets of information. For the former, the authors use regression analyses to estimate transaction costs and consider relative differences between average customer buy-and-sell transaction prices. For low-frequency liquidity proxies, they construct 21 measures, of which 8 are transaction cost measures, 10 are price impact measures, and 3 fall into the “other liquidity measures” category.

The authors first note that the high-frequency measures are highly consistent with each other. They then consider which low-frequency liquidity proxies best track the high-frequency measures. They find that although most proxies can detect transaction cost variations on both a time-series basis and a cross-sectional basis, the proxies calculated with data from Bloomberg and TRACE, rather than with quote data, prove best at measuring transaction costs. Specifically, the Corwin and Schultz, Hasbrouck, and Roll proxies perform best.

Measuring fixed-income liquidity is no easy task given the opacity of markets and the lack of a centrally organized exchange. Add to that the lack of any consensus on the best approach to measuring liquidity in these markets, and you arrive at a conundrum. The authors shed important light on this gray area by offering a robust comparison of the merits and demerits of all the measures, considering both high- and low-frequency approaches. Adapting a similar method to global corporate bond markets could be an interesting exercise but would face such challenges as data availability, markets of varying depth, different pricing conventions, and data transparency.