Factors in Quant Credit | Man Institute (2024)

1. Introduction

Traditionally, discretionary managers in credit have sought to identify mis-pricings through fundamental analysis. Credit tends to be dominated by large, buy-and-hold investors and the availability and speed of information dissemination has historically been slower than in equities. With the advent of electronic trading, significant availability of data and the subsequent rise of quant credit, all this is changing. The meeting of a slow-moving, inefficient market with quant strategies that exist to exploit such inefficiencies has led to a raft of opportunities for those with the platforms to exploit them.

Style factors emerged from the investigation of market anomalies – strategies that appeared to generate consistently above-market returns. All factors seek to capture that anomaly. So, in the equity markets, Value looks for firms whose stocks are undervalued relative to some consistent financial metric – price-to-book (‘P/B’) or price-to-earnings (‘P/E’); Momentum looks for firms whose shares have been rising in value and seeks to ride that wave; Size attempts to exploit the fact that smaller firms have historically tended to outperform larger companies. When it comes to quant credit, style factors are somewhat different, as we shall see.

2. Overview of Key Factors

We now set out the most common style factors employed in quant credit and how they specifically address the nuances of the corporate bond market.

2.1. Value

One of the principal fundamentally driven style factors, Value looks for bonds that are undervalued relative to other bonds judged by a specific financial metric. At its simplest, Value selects cheaper (higher spread) bonds over more expensive (lower spread) for the same underlying credit rating. We could also calculate a theoretical fair spread level by estimating the probability of default and recovery rates and establishing whether the market is correctly rewarding us for the risk of loss.

We use a variant of the second of these approaches as an example in Figure 1, highlighting several of the biases inherent in many relative value approaches. There is a spectrum of signal refinement in quant approaches that can span from more raw factors with biases or tilts left embedded, to more highly refined factors which aim to largely remove risk exposures focusing on more idiosyncratic dislocations. Of course, with raw factors comes simplicity of construction and often ease of interpretability: you know exactly what you are getting. We find that with more highly refined approaches, removing unintended biases or risk exposures often improves consistency and reduces volatility of what can be often quite volatile raw factor returns.

Below we start by looking at OAS per unit of leverage, measured as net debt to EBITDA. A raw relative value score like that can produce excess return but can also come with several embedded biases. We find that looking at overall factor tilts as well as group biases can help refine the signal, improving efficacy.

Figure 1. Signal Transformation Schematic

Source: Man Group. For illustrative purposes only.

It has been noted by several studies that Value has performed better as an investment strategy in credit markets than it has in equities. This appears particularly to be the case over the past decade. While the concept of buying cheap can be applied across asset classes, we find Value as a factor requires entirely new model specifications compared with equities. Cheap equity does not imply cheap bond and vice versa. It’s worth noting also that Value stands out as being particularly bond-specific, that is to say that within the same capital structure, even after accounting for liquidity, subordination, duration, etc., mispricing can remain.

2.2. Momentum

Again, when we come to Momentum, there is a basic definition that cleaves closely to the version employed by quant equity. This seeks to follow price trends for bonds, looking at price movements over various horizons and investing in those securities that have outperformed over the specified time horizon. It is worth highlighting that we find greater overlap between equity and credit Momentum than in other factors. We generally find trends to be consistent across a capital structure and indeed across broader groups/ecosystems as well; these trends therefore tend to be less issue-specific and more at an issuer-level.

As noted above, however, the bond markets are significantly more price-inefficient than equities due to the buy-and-hold approach of many of the largest players. This means that looking at bond prices alone is deemed insufficiently robust as far as signal construction goes, particularly when it comes to investment grade credit. Research by Khang and King (2004) and Gebhardt, Hvidkjaer and Swaminathan (2005) indicated that bonds had a tendency to revert to mean pricing levels over time, while further analysis by Jostova et al. (2013) indicated minimal effects from following Momentum strategies in investment grade, but a more meaningful impact in the high yield market. This superior performance of Momentum strategies in high yield is confirmed in research by Pospisil and Zhang (2010).

2.3. Other Factors: Size, Carry, Quality, Low-Risk and Informed Investor

There are a number of other style factors employed in quant credit that are worth mentioning here. It should be noted that many of these intersect with Value and/or Momentum, or with each other, such that factors can be thought of as overlapping Venn diagrams rather than conceptually discrete. As the quant credit investment world grows and matures, we expect these style factors to become more clearly defined and to have a greater degree of statistical corroboration.

Size privileges smaller companies based on the market value of their outstanding bonds. There is little academic analysis of Size as a factor, with only Houweling and Van Zundert (2017) suggesting that there is a positive Size premium in corporate bonds, while Alquist, Israel and Moskowitz (2018) argued that the Size effect was either immaterial or marginally negative for bonds. We have seen little evidence to suggest that Size is a meaningful factor in quant credit, especially after transaction costs; we certainly do recognise that it can be a driver of volatility (i.e. it is a clear risk factor). There can be fairly meaningful market-relative returns both positive and negative to a Size factor.

Like Value, Carry is a factor that will be familiar to discretionary market participants. Koijen et al. (2018) analysed Carry in a variety of asset classes, including corporate bonds. They define Carry as the “return of an asset assuming that prices stay the same”. Notwithstanding the higher risk implied by higher yields, Koijen et al. find that the Carry factor has, on average, positive returns, with Sharpe Ratios between 0.4 and 0.5. Over the long run, Carry appears to offer a significant premium, but with incredibly high volatility of returns. This lack of consistency means that Carry as a signal can have both fantastic and very poor years in terms of benchmark relative performance.

Low-Risk strategies (which are often expressed as a sub-set of the Quality factor) exploit the observation that bonds that exhibit lower volatility tend to perform better on a risk-adjusted basis over time. This reflects the fact that – unlike equities – bonds tend to experience more volatility on the downside than on the upside: the tightening of a bond on the upside towards risk free rates is typically asymmetrically smaller than the jump to recovery in situations of distress. Low-Risk factor signals generically target shorter-dated, higher-rated bonds.

The Informed Investor style factor looks to exploit information about investor activity to drive investment decisions, working along a ‘wisdom of crowds’ thesis. This harnesses data points such as short interest and call/put positioning to build a picture of investor activity across markets. While this factor is hampered by the slow-moving nature of the traditional bond-buying investor base, there is some evidence (Illmanen, 2011) to show that it delivers superior returns, particularly when employed in combination with other style factors.

This later point segues nicely into the conclusion of this paper on style factors. In quant equity, it has become clear that no single style factor is sufficient to address the different faces of the markets. As such, investors are increasingly using multi-factor models that seek to achieve two distinct goals: to deliver strong returns in a variety of different market conditions and to deliver excess returns in a manner that is uncorrelated with both the corporate bond market indices and with equities. Figure 2, taken from Henke et al. (2020) shows pairwise correlations across various style factors. The low or negative nature of these correlations means that a multi-factor portfolio substantially increases the risk-adjusted return while reducing volatility.

Figure 2. Pairwise Correlations Across Style Factors

Source: ‘Factor Investing in Credit’ Henke et al; as of 2020.

The most sophisticated investors will create their own proprietary models that adapt traditional factor models and deploy them in varying combinations based on the market environment.

3. Conclusion

We will conclude this paper with a few thoughts about complexity. We recently participated in a survey of quant credit managers by a Dutch pension fund. They broke managers in the quant credit space down into those who were more sophisticated in terms of both model construction and data usage, and those whose engagement is more simplistic. It’s a striking observation. On the one hand, simple models tend to experience faster alpha decay (as we have seen in the equity markets). On the other hand, more sophisticated models may expose themselves to overfitting and a loss in transparency.

Our belief is that the quant credit space will necessarily follow quant equity by increasing in complexity and sophistication and that simplistic models will find that they are overtaken by those firms with the modelling skills and data analysis expertise that allow them to execute with greater precision and speed. Overfitting is always a risk in quant strategies but can be mitigated by best practice in research and analysis of model inputs and outputs.