A PGGM research project on a range of asset managers’ credit factor programmes shows the breadth of approaches and confirms the potential use in multiple ways for investors. However, a first real live stress test event of the programmes is required to confirm the robustness

Key points

  • Credit factor investing is in its early stages, but models and data are improving
  • The recent volatility and illiquid market conditions offer a first live stress test
  • Credit factor investing has a large quantitative and data intensive foundation, but its success depends as much on the qualitative and judgmental decision making of the portfolio manager and trader
  • Asset managers need longer live track records to show how they actually implement models 
  • The approaches of asset managers differ from one another therefore diversification benefits for end investors arise

Investing based on factors or risk premia, common within equities, has started expanding to other investment categories in recent years. Factor investing in corporate credits has also picked up significantly and assets managed in these programmes are starting to grow from a still relatively modest base. The drivers of the increased interest are more research and better data, as well as the increasing range of managers implementing factor strategies. For these reasons, and also because PGGM already has extensive experience in equity factor investing, we started researching the merits of expanding this to credits. We approached it from different angles, as credit factors can be used in multiple ways within the allocation and investment process. 

This article covers the ‘broadest’ part of our research project and established a consistent and comprehensive overview of the different approaches currently in the market. For this purpose and in addition to our review of academic research and conversations with managers, we sent a comprehensive survey to a dozen investment managers with multifactor credit programmes. 

The survey provided us with a large number of observations, leading to a range of conclusions. It also raised new questions and follow-up research topics. It underscored our expectations regarding implementation issues in less-liquid credit markets and revealed the fat tails and skewed distributions of many factor models. These issues can make or break the factor strategy and provide lessons on if and how to incorporate credit factor investing in portfolios and investment processes. The current corona crisis is the first real live stress test for credit factor investing and therefore acts as a watershed moment. 

This article describes the main observations from our survey. It also reflects upon the questions allocators and investment managers should ask and answer before starting to allocate to and or integrate credit factor models in their investment processes. 

Models and implementation considerations
Factors analysed and used for credits are often derived from (or are similar to) equity factors in their name and rationale. This includes traditional factors like (relative) value, quality, size, momentum and risk. Like in equities, the reasons why these factors or risk characteristics can explain cross-sectional returns often relate to behavioural factors, market and benchmark inefficiencies and/or preferential habitat and market segmentation effects. However, the calculation for individual credits, issues and issuers, is quite different from equities. For example, for ‘value’ the credit spread substitutes equity ‘value’ measures like dividends or earnings, while for price momentum sometimes both bond and equity momentum are used. 

Another difference is that a company has often just one ‘class’ of stock outstanding, but dozens of different bonds that also can be issued by different parts of the company with their own credit fundamentals. This significantly increases complexity of obtaining the required data as well as the modelling. 

The granularity of the different models used is diverse. Most managers build a relatively simple, robust, well-explainable model to be used across both the European and US corporate bond markets that can be applied to investment grade (IG) and high yield (HY). In other words, applying the ‘KISS’ principle: ‘keep it simple, stupid’. However, several managers have built more extensive and data-intensive models that are more adaptive to different markets and periods. 

While this can improve the return/risk profile, it increases the risks of overfitting. The different approaches of the managers can be mapped upon a continuum of pure ‘risk-premium’ factor investing towards a systematised, alpha-focused, credit investment process. In the latter, machine learning and big data are also starting to be researched and applied to credits. Regardless the model approach, investors allocating to these programmes should thoroughly research how much the model is ‘optimised’ to historical data.

While managers have different approaches towards data series used, their basic investment philosophies are still quite similar to traditional investing in credit. The models prefer good ‘quality’ companies, based upon credit fundamentals, which are relatively attractively priced. Subsequently, the portfolio construction and transaction models make sure the portfolios are sufficiently diversified and not tilted to illiquid bonds. Therefore, we also do not observe, on average, negative correlations of most factor models’ excess returns compared to traditional investment processes, nor that portfolio characteristics are totally different. It will be interesting to see how factor models perform versus the ‘traditional’ investors in 2020.

All managers build multi-factor portfolios to create a more robust return pattern, as returns of the different factors differ over time. There are significant diversification benefits derived from combining factors, like (relative)-value, momentum and quality. Therefore, these factors are present in almost all programmes. Diversification is just one driver when selecting factors to make it into the final multi-factor model. Implementation issues are starting to play a larger role. For instance, relative high trading costs and/or difficulties in sourcing bonds imply that factors like size (or illiquidity) are often omitted from the model, or play only a small role when picking individual bonds after the model portfolio is created. Combining factors – that is, the weights of the factors, is often a simple and straightforward process not driven by optimisation. For robustness over different credit markets, the weights are often fixed and equal over the factors used. The optimisation occurs afterwards, ie, when selecting the issues and issuers with the highest multi-factor scores to fit within the actual portfolio risk constraints while taking into account transaction costs. 

Portfolio and return characteristics: fat tails and the need for diversification
Most credit factor models have modest constraints for sector and credit quality deviations from the benchmark. Only a few have very strict relative risk constraints. Based upon portfolio characteristics and time series analyses, we observe that while diverse for each manager, on average, factor models don’t have a natural credit-beta tilt versus an index. This contrasts empirical results for ‘traditional’ fundamental qualitative managers that in IG have, on average, a higher credit beta than the benchmark, while active HY managers, on average, play defence – in other words, have lower credit beta. 

We have analysed in depth the time series of returns of the backtests over almost two decades. Longer historical data is not available for almost all programmes. While it still seems quite an extensive period, it remains a relatively short period. It covers less than three credit cycles, includes a period where parts of the European corporate bond market were less developed, and also the breadth and data quality were less than at the end of the research period. An additional caveat is that the two decades, depending upon the manager, contain both an in-sample and out-of-sample period of the model. 

In general, we observed that the level of relative returns and risks from the backtests are in line with observed results for traditional IG managers. That might seem a relatively poor result, as these are backtests that can be impacted by overfitting. However, as stated, traditional IG managers often ‘create alpha’ from overweighting beta. In addition, the backtests include modelled transaction costs and managers often claim their actual trading costs are lower. For HY the average relative return/risk ratios are positive, but lower than IG. This can be explained by higher transaction costs and idiosyncratic risks – that is, less exposure to ‘general’ risk factors. In other words, in HY the merits of credit factor investing are less obvious than in IG. 

“The corona crisis is an unfortunate first ‘live’ stress test of almost all multifactor programmes”

An important observation from the times series was the existence of sizable fat tails and sometimes quite negatives skews. So even though average excess returns can be quite favourable, a gradual build-up of positive returns can be whipped out in just a few months. This result has multiple implications for allocators to these credit factor programmes, but also for asset managers that use and/or manage these models. For asset managers it means more and longer historical data is needed to draw more reliable and firm conclusions and/or that their models must be adjusted. 

It also requires a longer live track record to see how the model behaves in real time and how portfolio managers and traders actually respond or not when faced with such a large negative outlier. These outliers often occur during market stress, where the model may suggest a significant change in portfolio composition, but that this cannot be executed at reasonable prices. 

The current unfolding corona crisis is therefore an unfortunate first ‘live’ stress test of almost all multifactor programmes. Most index providers even postponed the usual adjustments in their credit indices at the end of March 2020 due to bond market illiquidity. For allocators to these programmes fat tails also imply that diversification across multiple managers is required even more in addition to the reasons listed below. 

We also analysed modelled versus actual returns for the few managers able to provide these. Over the brief ‘live’ periods, before corona, we observed that, at least on a monthly basis, the returns can differ and sometimes substantially. This is due to the nature of the bond market – that is, less liquid, frequent maturities, coupons reinvesting and so on. This impacts portfolio construction throughout the month while the back-tested model is often only rebalanced on a monthly basis. 

The deviations underscore the need for cautious use of longer-term modelled returns when selecting a programme. Managing the daily tradeoffs of market conditions in the less liquid credit markets with the recommended model portfolio is central to the success of the programme. It again underscores that although credit factor investing has a large quantitative and data-intensive foundation, its success depends as much on the qualitative and judgmental decision making of the portfolio manager and trader. It makes the due diligence and selection of managers an even more interesting challenge. 

Other interesting observations are derived from the correlations of the excess returns. The correlations inform how truly different the outcomes of the various factor models are, but also the stability of factor returns between different regions (EU/US) and markets (HY/IG). Without mentioning individual managers, we observed ‘clusters’ in the correlation matrices. Several managers had high correlations, in the order of 0.5 or more, between them, as well as between their models in different credit markets. This result can imply that even though the models can be quite different, they seem to generate similar returns. This is not unexpected as often similar factors like value and momentum are used. Subsequently their different models pick up the ‘exposure’ to these basic factors. Therefore, the average of all managers’ correlations is likewise also positive, although relatively small. 

For several managers, we observed from the survey that, while they use similar basic factors, their factor model approaches were quite distinct, including the weighting schemes and data used. This resulted in sometimes negative correlations with other managers. Their models seem to pick up different ‘factor loadings’. This underscores the potential of diversification, but also raises interesting questions for additional research. Overall, the return analyses still leaves an important question not firmly answered: do these factors have a positive ‘risk’ premium net of trading costs that is persistent and unlikely to be ‘arbitraged’ or fade away over time, or are the factors ‘risk characteristics’ that don’t provide sustainable excess returns? In other words, just an explanatory risk variable that creates both cross sectional and time series volatility. And how much of the return can be considered as ‘alpha’? 

Will 2020 be the year credit factor investing matures?
In terms of research conducted, credit factor investing is catching up with equity factor investing. However, it is just at an initial stage of actual adoption by investors and has several implementation caveats, including availability and quality of data, as well as higher transactions costs. This makes credit factor investing complex compared to equities. Credit factor investing puts a higher burden on combining the modelling and data efforts with human judgement in balancing model portfolios with actual market conditions. With better data, additional research and more managers entering the market, combined with the results so far, credit factors are expected to fit more purposes in the future. 

The corona crisis will be a watershed event for the use of credit factor models. The models themselves, but more importantly, the implementation of these models in illiquid and volatile bond markets are ‘stressed’ to the limit. The behaviour of the live programmes during their first live stress test can provide answers to important questions that again arose from our observations of the available research as well as our survey and data analysis:

● How have modelled returns behaved, but more importantly, how have the portfolio managers and traders managed their live programmes and what were the actual outcomes? 
● Did diversification properties of individual factors within multifactor strategies hold? ● Did diversification among managers hold – that is, did the crisis avoid driving correlations to one?  
● How did multi-factor programmes stand against traditional managers during and after the ‘storm’?
● Did the crisis reveal the true nature or source of the returns of factor programmes; is it a well-known theoretical explanation for risk premiums or a new ‘source’ of unmodelled ‘event’ risk?

These questions, and their answers, will impact the adoption rate. It should help allocators, selectors and other users of credit factor models on issues like allocations, the breath of due diligence required to assess programs, the need for diversification and the usefulness of factor models for  risk and performance management. The answers will reveal the ‘true’ potential of credit factor investing. Therefore, 2020 could be the year in which a young adolescent matures. 

Mark Geene is senior investment consultant at PGGM

To read the latest IPE Digital Edition click here

Factor investing: Crisis factor