We’re using cookies, but you can turn them off in Privacy Settings. Otherwise, you are agreeing to our use of cookies. Learn more in our Privacy Policy.

This is a summary of “Boosting the Equity Momentum Factor in Credit,” by Hendrik Kaufmann, Philip Messow, and Jonas Vogt, published in the Fourth Quarter 2021 issue of the Financial Analysts Journal.


Overview

Listen to an audio version of this summary.

Investors can double alpha in the credit markets by using simple equity momentum strategies enhanced by applying machine learning with boosted regression trees.

What’s the Investment Issue?

Equity momentum can be combined with machine learning and boosted regression trees to enhance investor alpha in the corporate bond market. This approach builds on simple momentum models that take advantage of the fact that the bond market digests news less quickly than does the stock market.


How Do the Authors Tackle the Issue?

The authors seek to enhance an equity momentum factor strategy by using machine learning with boosted regression trees.

Features are added to the simple equity momentum factor model:

A) Measures of liquidity (liquidity cost score/daily volume) in the bond/equity markets, respectively

B) Size of the company (market cap/market value) in the equity/bond markets, respectively

To develop the model, the authors collect monthly data from the Intercontinental Exchange for the Global Corporate Investment Grade (IG) and Global High Yield (HY) Indexes covering the period January 2002 through February 2021. These data include information on credit spreads, credit ratings, time to maturity, total returns, excess returns over Treasuries, and market sector.

The final sample of bonds includes only those with the following attributes:

• They can be mapped to a listed company.

• They are US dollar denominated.

• They are from the financial, utility, or industrials sector.

• They do not include cash, government/agency, or securitized bonds.

• For companies with more than one bond, a single market value–weighted proxy, including key characteristics such as credit spread, is developed.

The authors then allow the machine learning algorithms to develop a model in the so-called formation period using historical data. Within this learning process, the algorithm splits decision tree nodes based on a feature to improve performance.

Using the information from the formation period, bonds are collected into portfolios; the fifth quintile has the highest factor exposure, and those in the first have the least factor exposure.

The model is then tested against fresh data with holding periods of one and 12 months.


What Are the Findings?

Both IG and HY bonds in the fifth quintile produce high and stable excess returns that enhance results compared with simpler models. Specifically, the strategy leads to annual alphas of up to 4.4% in IG and up to 14.3% in HY after accounting for transactions costs. Similarly, the information ratios go as high as 1.9 and 2.8 for IG and HY, respectively.

“[These results indicate] that simpler momentum factors do not exploit all relevant information in the data,” the authors state.

For both categories of bond, IG and HY, the size and liquidity features rank alongside the variables of the highest importance, the study shows. The authors find the importance of equity returns is related to their recency, with less recent data generally being less relevant for future fixed-income returns.

“It is highly likely that this strategy can be integrated within a multi-factor strategy where the overall signal can benefit from diversification effects between the different factors,” assert the authors.


What Are the Implications for Investors and Investment Managers?

Bond market investors can enhance their simple equity momentum factor models by using machine learning with boosted momentum trees that include bond/equity size and liquidity features.

About the Author

Simon Constable

Simon Constable is an Edinburgh-based journalist.