Investing with Style in Liquid Private Debt

Abstract

This paper extends the analysis of systematic investment approaches to broadly syndicated leveraged loans. We find that exposures linked to (short-term) momentum and valuation styles (and a combination thereof) are well-compensated: monthly rebalanced long-only portfolios of high value and momentum loans generate Sharpe and information ratios well above one and economically and statistically significant alphas. Factor portfolio performance deteriorates but remains significant over longer investment horizons. An important implication of our research is that active credit managers employing loan trading strategies that are momentum- and value-neutral do not make use of a viable source of additional return.

Keywords:

• factor investing
• leveraged loans
• momentum
• value

PL Credits: 2.0:

Disclosure statement

No potential conflict of interest was reported by the author(s).

Acknowledgements

The views expressed in this article are the authors’ and do not necessarily reflect the views of DWS International GmbH (“DWS”).

Notes

1 CLO is the name of an asset-backed security set up to hold and manage pools of leveraged loans (almost exclusively term loans B, C, etc.), and to a lesser extent high-yield bonds. These investments are financed through the issuance of several debt and (one or two) equity tranches that have rights to the collateral and payment stream, in descending order. CLOs are issued by a special purpose vehicle/entity (SPV/SPE).

2 To enable a fair comparison between active factors and passive benchmark, we construct a customized investable market index that only includes loans that are accessible to an investor in the secondary market.

3 Due to the absence of a unique ID number, the loan level match across the four data sources is time-consuming and requires a significant amount of hand-matching. For example, loans are identified in DealScan by a “FacilityID”, and by a completely different “lxid” in IHS Markit. A detailed description of how we approached the matching task can be obtained from the authors upon request.

4 Information (e.g., trade date and price) on CLO trades is collected from CLO trustee reports available in CLO-i and the trading behavior of loan participation funds (Lipper style code “LP”) is inferred from monthly holding reports (i.e., 13F reports) of funds that file with the SEC. Fund trades are assumed to equal changes in loan par amounts between subsequent holding reports, adjusted for repayments, refinancings, and restructurings. Data source for fund holding reports is CRSP.

5 For the average loan in our sample in an average month with at least one CLO trade, we observe 3.9 CLO buys and 4.2 CLO sells (the median number is two, respectively).

6 Almost all loans in our sample are institutional term loans B that typically repay 1% of par annually over their life and the remaining notional at maturity. Therefore, at a monthly frequency, the principal repayment return should be of second-order importance. Supporting this claim, Beyhaghi and Ehsani (2017) report that the 0.38% average monthly total return for their loan series consists of 0.53% interest return, 0.01% principal repayment return, and −0.16% price return.

7 In sharp contrast to bonds, most loans are callable any time during their life. Because of this callability feature, prices of loans normally do not rise beyond par. Hence, the ability of the borrower of a loan to repay the principal prior to maturity places a cap on the investment’s upside potential.

8 For example, with 1-month rebalancing (and equal weights), the month t + 1 return $r_{t,t+1}^I$ on the index portfolio, formed in month t, is calculated as:$r_{t,t+1}^I=\frac{1}{N_t}{\sum }_{i=1}^{N_t}{r}_{t,t+1}^i,$ where $r_{t,t+1}^i$ denotes the t + 1 return of loan i. Note that the monthly set of investable loans, $N_t,$ is dynamic. With a 12-months holding period, the month t + 1 index return is the simple average of the month t + 1 returns from the twelve index portfolios formed from month t-11 to t.

9 The S&P/LSTA Leveraged Loan 100 Index (LLI100) is a daily index for the U.S. market that consists of 100 loans (mostly term loans, both amortizing and institutional) and intends to mirror the market weighted performance of the largest institutional leveraged loans to reflect the most-liquid side of the market. The index is published by S&P’s Leveraged Commentary & Data (LCD) unit, dates back to 2002, and the pricing source are average bid quotes from the LSTA/LPC mark-to-market service.

10 To calculate credit excess returns for the LLI100, we obtain aggregate (equal weighted) time series of secondary market STMs and average bid quotes of index constituents from LCD. Because these time series exclude the prices and STMs of defaulted loans, our estimated CERs for the LLI100 likely overestimate the actual returns accessible by investors.

11 Besides the fact that value and momentum are probably the two most common styles across asset classes, our choice of these two styles is motivated by two additional considerations: data availability and overfitting. Our data is probably not rich enough to allow for a rigorous assessment of other styles like betting-against-beta or quality. Recall that we do not have much information on loan issuer characteristics. Second, we wanted to avoid the impression that we cherry-picked the results by fishing for the styles that worked best in-sample.

12 As robustness check, we alternatively calculate price returns from log changes in monthly average bid quotes, or from month-end changes in mid or bid quotes. All results remain similar.

13 In robustness checks, we employ alternative volatility proxies. We estimate historic volatility from the previous 60 days of raw price returns, requiring at least 20 daily returns, and we utilize a market model to calculate idiosyncratic volatility. We use price returns of the LLI100 as the market index. Because the results for the different volatility measures are similar, we stick with the simpler measure (absolute price return) which is also less data demanding.

14 The value and momentum measures VAL_R and MOM_t-1,t are weakly negatively correlated ($\rho =-0.0012,$ p-value = 0.88) in the full sample.

15 We focus on equal weighted portfolios in the body of the paper. The “Value-Weighted Portfolios” section and online supplementary material Tables IA.4 and IA.5 provide performance statistics for value weighted portfolios, where each loan is weighted by its market value in month t. The market value of a loan is the product of the outstanding balance and market price (monthly average bid quote), ignoring any accrued income. The value weighted portfolio tests are meant to further highlight the economic significance of style investing in loans and to help mitigate concerns that the equal weighted results are entirely driven by the smallest loans in the sample.

16 We thank Michael Wolf for sharing his R code to perform the test. We choose the optimal block size according to Algorithm 3.1 in the Ledoit and Wolf (2008) paper and we set the number of bootstrap resamples to 1000.

17 Their defining and differentiating characteristics (first lien, amortizing notional, floating coupon) likely reduce the volatility of loans compared with corporate bonds with similar maturity and rating. Consistent with such a risk-reducing effect, Beyhaghi and Ehsani (2017) found that returns on loans are less volatile than the returns on speculative grade bonds. This might partly explain our relatively high Sharpe ratios (for comparison, Houweling and van Zundert 2017, report long-only value and momentum factor Sharpe ratios for high yield bonds of below 0.5). Alternatively, quote staleness might artificially depress credit excess return volatilities of loans. We discuss this concern in Section “Quote staleness”

18 The five asset classes are represented by the same total return indexes used in the construction of the efficient frontiers in Figure 1.

19 To illustrate the risk-return tradeoff between the two factor portfolios, Figure IA.2 in the online supplementary materials displays the (in-sample) efficient frontier from a top-down approach that mixes the factor portfolios, and not the loan characteristics that are used to construct the portfolios in the first place. This allows for a better understanding of how factor combinations outperform each factor individually. For example, the maximum Sharpe portfolio has a 97% weight on the momentum style.

20 Note that our dataset is free of survivorship bias: whenever a loan exits the sample (because of calls, repayments, or defaults), the price returns are based on the loan’s final quotes. As we require just two consecutive months of daily price quotes for each loan that passes the liquidity filter, the number of observations in Table 5 drops for longer return horizons.

21 The other return standard deviations are: 2.30% (1-month), 4.40% (3-months), 9.38% (9-months), and 10.80% (1-year).

22 Table IA.3 in the online supplementary materials reports results from Fama and MacBeth regressions with only the controls as predictors, leaving out the investment style proxies. Coefficients for LN_SIZE, QHS, and the number of CLO trades are generally insignificant. This strengthens the conclusion that momentum and value do not pick up a predictive ability of these other loan characteristics.

23 As further comparison numbers, Keßler and Mählmann (2022) construct a liquidity index out of a sub-sample of IHS Markit loans assumed to be widely traded (see their Table 1, Panel B). This index depicts lower quoted liquidity (higher spreads) than our trade sample, with mean and median half-spreads of 62 bp and 49 bp, respectively. In addition, S&P’s Leveraged Loan Commentary & Data (LCD) unit reports average dealer bid and ask quotes for all 15 constituents in their U.S. “flow-name composite”. This composite is a regularly updated sampling of loans widely traded in the U.S. secondary market, per LCD’s discussion with dealers and investors in the market. Over the period from May 2002 to July 2020, the 15 most liquid loans possess half-spreads of about 25 bp on average (median: 21 bp), somewhat less than what we found for our sample.

24 Even if loan strategy Sharpe ratios are not comparable to traded return-based Sharpe ratios from other asset classes (e.g., high-yield bonds), any within loan class comparison would only be affected by a systematic association between styles and staleness. To the degree that HIGH portfolios are more liquid, and, hence, their returns are less stale, the staleness argument works against our finding that HIGH portfolios outperform LOW ones.

Additional information

Notes on contributors

Thomas Mählmann

Thomas Mählmann is a professor of finance and Chair of Banking and Finance at the Catholic University of Eichstaett-Ingolstadt, Ingolstadt, Germany.

Galina Sukonnik

Galina Sukonnik, CFA, is a portfolio manager at the Multi Asset and Solutions Group, DWS International GmbH, Frankfurt am Main, Germany.