Implementing a corporate bond portfolio: lessons learned at the NBP

In this section we present, in a hopefully concise and digestible way, some of the practical lessons we have learned while implementing and ultimately using our corporate bond benchmark index. While we are certainly happy with the changes to our reserve management set-up and overall progress, we nonetheless feel a little bit like the students who reportedly posted this note outside the mathematics reading room at Tromso University:¹¹ From Bernt Oksendal, Stochastic Differential Equations. An Introduction with Applications (New York: Springer).

Hopefully, the discussion below will be helpful to others embarking on the project of building a corporate bond portfolio of their own.

To borrow from Neil Armstrong, going from a pure Treasury bond portfolio to one with a non-trivial allocation to corporate bonds may well be a small step for a finance aficionado but is in many respects a giant leap for the institution and its senior decision-makers. After all, corporate credit is first and foremost about credit. Even if the exposure is only to investment-grade issues, the headline risk that a public investor needs to be mindful of – and the non-zero probability of a sudden default of even highly rated names – means that credit risk management functions will have to take on a considerable burden.

Indeed, in the case of NBP the new approach to managing corporate bonds has resulted in a significant increase in the number of issuers that need to be monitored on an ongoing basis. This raised both investment and operational issues. As the number of people in the credit risk team had not changed and increasing staffing solely due to the changed management style of corporate bonds – which after all remain just a fraction of the reserves portfolio – would prove a tough sell, we had to factor such limitations into our plans. Specifically, when designing the benchmark index we had to think in terms of how many entities we would realistically be able to cover, come rain or shine. This number – or, perhaps more accurately, range – proved an important determinant of the size of our corporate index, alongside more objective criteria stemming from NBP’s investment guidelines and the constraints discussed above.

To be sure, the increase in portfolio size has not only broadened the scope but also the complexity of the entire exercise, forcing our credit team to ponder questions such as: “how do we treat a ‘sukuk’ bond”, “what do we make of Company XYZ moving to the Cayman Islands”, or – most frustratingly – “why is there no info about that ticker on Bloomberg”?

Over and above such largely qualitative issues, changes also had to be made to the quantitative side of the risk assessment process through standardisation of the rating criteria and modifications to the credit limit system. The new limit methodology introduced the concept of a global limit, defining the maximum acceptable risk-weighted exposure to an entity.

Methods used so far limited credit risk at the level of individual transactions. With the expansion in the portfolio of names we had to deal with, and the associated rise in the spectrum of transactions, the lack of a global perspective of investments could result in excessive exposure to individual entities. In other words, without a global limit we could, in theory, be putting on exposures to the same entity through different transactions, limiting the exposure to those transactions as such but at the same time allowing our overall credit exposure to that particular entity to increase.

Simultaneously with the development of the new limit system, we have been developing new analytical tools and IT solutions supporting credit risk assessment. The move towards a more advanced IT environment was partly driven by the desire to have a system that would be fully aligned and consistent with our new methodology, but also by the natural concern to limit the operational risk involved with using less sophisticated solutions. However, for all its superiority, the new system still needs to be operated and supervised, and requires a full assessment of the creditworthiness of each issuer in the portfolio – both at the moment of setting a limit, and later on when monitoring its utilisation.

From the very beginning, our discussions about developing a corporate bond portfolio have always centred around liquidity issues. This is because corporate credit is almost always about credit risk... except when it’s about liquidity. All analyses and market intelligence collected and conducted over the years have consistently shown that the US bond corporate bond market – while perhaps still liquid relative to other asset classes – is certainly not as deep and liquid as the Treasury bond market that we had greatest experience with. However, we were not sure exactly how, and to what extent, the expected lower liquidity would affect our portfolio – which, after all, had been designed in a way as to include mostly highly rated bonds with high amounts outstanding, issued by highly rated, big companies.

As mentioned, our first misadventure with liquidity occurred in the early days of our corporate credit investment journey. Unfortunately, the whole episode taught us more about the philosophy of benchmark construction than the real nature of the problem. Were the wrong criteria for selecting the instruments to blame? Or was it the wrong strategy of a too narrow benchmark? Or perhaps our lack of experience, bad timing, bad luck, or all factors together?

As our knowledge of the market expanded and our activity increased, we have gradually come to realise that although generally speaking liquidity does seem to be an issue, it no longer significantly constrains the management of our portfolio. Of course, the liquidity of corporate bonds is undeniably lower than that of Treasury securities in our US dollar portfolio (see Figure 2.8, which shows the frequency distributions of Bloomberg liquidity scores for both categories of instruments). Still, almost all issues included in the benchmark index are ranked above the median liquidity score in the peer group of government and corporate bonds.²² Based on the Bloomberg liquidity score.

The most visible problem occurs in the case of bonds from the short end of the yield curve. Longer maturities seem to attract better liquidity scores, which is likely related to their greater abundance relative to short-term corporate paper (see Figures 2.9 and 2.10).

A more serious problem relates to the fact that the liquidity of individual corporate bonds is hard to predict and much less stable than that of the corresponding Treasury securities (see Figure 2.11). Indeed, it is not uncommon for liquidity scores of bonds in our portfolio to change by 20%, something that never happens for Treasuries. On the portfolio level, such liquidity disruptions need not present significant difficulties – and, in fact, may constitute potential profit opportunities (see Lesson 4) – but on the level of individual bonds they can lead to transitory frictions for which the front office has to be prepared.

One way to manage liquidity-related problems is through the use of simple derivatives. At the time of writing, we are exploring the introduction of bond futures into our corporate portfolio management strategy, as we have already found that resorting to futures contracts can be a very efficient way of managing duration exposure in our Treasury bond portfolio. We expect that corporate bond futures will help us reduce transaction costs related to managing interest rate exposure in corporates. The next steps along the way will include determining an optimal strategy for managing credit spread exposure, maybe using derivatives as well.

One of the standard motivations for gaining exposure to corporate bonds is to earn the so-called credit risk premium – ie, an excess return net of default losses over comparable duration risk-free bonds. Indeed, corporate bonds typically trade at a positive yield spread over comparable maturity Treasury bonds. The spread – which stands at roughly 130bp for a broad index of US credits – is due to the fact that a corporation, much more so than a sovereign government endowed with taxing power, might default before making good on its debt, leaving investors with recoveries well below par. Accordingly, it is the level and volatility of corporate bond spreads that many investors view as a key determinant for entering or exiting the market. Indeed, according to the results of a questionnaire published in Risk Management Trends 2018, public investors viewed corporate credit less favourably in 2018 due to a general “tightness of credit spreads.”

As a matter of basic financial algebra, a bond’s one-period holding return (net of credit losses) is made up of its yield and capital appreciation/depreciation due to changes in yield levels. Letting MD stand for modified duration, we can express this symbolically in the context of corporate bonds as:

However, since – as stressed above – the corporate bond yield equals the yield to maturity on a duration-matched government bond, YTM, plus a credit spread, s, the above equation actually becomes:

confirming the intuitive insight that when credit spreads are low, then so should be corporate bond returns. However, practical differences between the characteristics of the corporate bond portfolio and the government bond portfolio used as a benchmark to assess the efficiency of credit investments can make such simple textbook decompositions misleading. In particular, one thing we have found was that the spread risk on many occasions tends to be lower than the sensitivity to the risk-free curve. For example, the returns on our corporate benchmark since October 2017 (ie, when index history begins) are explained primarily by the returns on our US Treasury portfolio (see Figure 2.12).

In fact, this seems to be a more universal pattern. Box 2.4 shows the results of a simple exercise in which we have regressed monthly total returns for the Bloomberg Barclays US Investment Grade index on the monthly changes in the average yield on a hypothetical US dollar Treasury index representative of our US dollar portfolio, and monthly changes in the spread between AAA- and BBB-rated corporate bonds. The “betas” on the two spreads represent empirical estimates of interest rate and spread durations, respectively. Over almost three decades, from 1993 to 2018, the spread duration of the corporate bond index was just –1.3, while its interest rate duration is –3.1. In words, a given shift in yields tended to affect corporate bond returns about three times as much as a corresponding change in credit spreads, which suggests that bond returns are likely to be largely driven by yield curve shifts.

However, there is even more nuance to this. Looking more closely at corporate bond returns and their volatility across the maturity spectrum, we have documented a somewhat counterintuitive result that the front end of the credit curve tends to systematically offer the best investment on a risk-adjusted basis (see Table 2.2).

The return of investment-grade corporate bonds in the 1Y–3Y maturity bucket achieves an information ratio above 0.4, and deteriorates to just 0.05 and 0.004 for bonds with maturities beyond 10Y and 25Y, respectively.

Exactly why the short end of the spread curve overperforms so markedly is not clear. Mechanically, this seems to reflect the fact that credit spreads do not increase one-for-one with duration, so that bonds in farther maturity buckets pay disproportionally low spreads, perhaps as a result of some sort of market fragmentation. Be that as it may, we believe there is ample empirical evidence not to think of corporate bonds solely as credit spread products, with lots of room for classic duration-based plays within the broader activist approach to replication.

As should be evident, a constant theme of our internal discussions was how broad the prospective corporate bond benchmark should be. Recall that one of the ideas behind introducing a benchmark (together with the integration of the corporate portfolio within the SAA) has been to provide a framework for consistently measuring the value added to the investment process – so called “alpha”, or excess return – by portfolio managers. As such, a good benchmark should represent as closely as possible the risk and return profile of a given asset class, as well as satisfying three key conditions. First, it should be well defined, in the sense that it should be verifiable and free from ambiguity with regard to its contents. Second, it should be tradeable in the sense that taking positions in the whole index or some of its members should constitute a viable, implementable investment strategy. Finally, the index should be replicable since, without such a provision, portfolio managers might complain (rightfully so) about not being in the position to even generate the benchmark return, not to mention any alpha above it.

Of the three conditions, replicability proved the most contentious, revealing different perspectives of the stakeholders involved in line with their respective roles in the overall investment process. On the one hand, the risk function – being also responsible for strategic asset allocation – was primarily concerned with ensuring a truly representative character of the index. Accordingly, it pushed for as broad a composition as possible within the accepted risk and investment guidelines. On the other hand, portfolio managers were concerned that the broader the index, the less liquid and tradeable its constituents and thus, by extension, the more difficult its replication would be.

Extensive discussions and several iterations of internal memos revealed that at the heart of this “tug of war” was a fundamentally different understanding of replication that, once happily resolved, left both sides feeling they had been right all along. The crux of the matter is that replicating an index does not need to entail taking positions in all constituent bonds. While such passive replication should be at least theoretically possible, it may not be the most effective approach. Indeed, trying to passively replicate a broad index that is truly representative of an asset class – in this case, the investment-grade corporate bonds universe – would entail significant operational costs in monitoring and risk managing individual issuers. However, replication can also be seen as essentially an engineering or optimisation exercise whereby portfolio managers seek to reproduce the main features of the entire benchmark – duration, credit risk profile, etc – by selectively positioning in a much fewer number of bonds, saving transaction and risk management costs. With such an approach, a credit limit would perhaps not even have to be assigned to each individual index member, and portfolio managers would be motivated to engage in more active strategies.

Simulations performed using data on our benchmark index returns and composition over the period January 2018 to September 2018 show that such an active approach to replication is indeed feasible (see Figure 2.13). In particular, a portfolio of 80 carefully selected bonds (out of 220 in the benchmark) is capable of bringing the tracking error to an impressively low number of 0.03%.

In practice, the tracking error of the NBP’s portfolio with respect to the benchmark tends to be considerably greater – around 0.30% – but, to borrow from accounting jargon, it is an error of commission not omission. In other words, portfolio managers have so far purposefully strived not to replicate the index in its entirety, putting on positions leveraged on the insights from Lesson 3 on the overperformance of the front end of the credit curve. Indeed, a decomposition of the tracking error shows that it resulted almost entirely from lower interest rate duration with only a minor contribution from the spread duration effect (see Table 2.3). Put simply, our effective portfolio had significantly lower duration than the benchmark, while preserving similar sectoral composition and credit profile. With tight and generally flat credit spreads, and an environment of rising interest rates, this has produced a healthy excess return of 60bp.

The process of constructing the corporate bond benchmark from scratch and then collaborating with the front office on creating an actual revamped portfolio (accounting now for 2% of the total US dollar portfolio) has forced us to not only consider a myriad technical issues, but also rethink some broader, more philosophical questions, such as: what role should corporate credit ultimately play in our overall strategic asset allocation, and – related to that – how big should our corporate bond portfolio eventually become?

To frame these questions – and our tentative answers – in a proper context, consider once again the historical performance of corporate credit. On the face of it, corporate bonds present themselves as superior investments since they offer a hefty spread over Treasuries. Indeed, as mentioned, the option-adjusted yield spread between the broad corporate bond index and duration-matched US Treasury securities has averaged about 1.30% since the 1980s. Correcting the spread for average default losses, estimated at about 0.20% for investment-grade credits, still leaves a decent 100bp.³³ Moody’s Investor Services, “Corporate Default and Recovery Rates, 1920–2010” (February 28, 2011).

Interestingly, however, the ex ante yield spread tends to markedly exaggerate ex post returns differences. Indeed, in the long period between 1988 and 2018 the excess return of the Bloomberg Barclays Investment Grade index over a hypothetical duration-matched US Treasuries portfolio averaged 0.54% with a standard deviation of 3.80% – ie, just about half of the ex ante yield spread, net of credit losses (see Figure 2.14). Notably, for baskets made up of the highest-rated credits the performance looks considerably less attractive, with an average excess return of a mere 0.08% for Aaa- and 0.35% for A-rated names. This finding seems to be broadly consistent with the empirical literature as academic studies have typically found that investment-grade credit portfolios have generated an average annual spread premium of between 30 and 50 basis points per annum over US Treasury bonds.⁴⁴ See, for example Kwok-Yuen Ng, and Bruce D. Phelps, “Capturing Credit Spread Premium,” Financial Analysts Journal 67.3 (2011): 63–75; David F. Swensen, Pioneering Portfolio Management: An Unconventional Approach to Institutional Investment (New York: Free Press); Antti Ilmanen, Expected Returns: An Investor’s Guide to Harvesting Market Rewards (Chichester: John Wiley & Sons, 2011).

A glass half-full perspective would be that, while somewhat short of the yield spread, the long-run excess return is still noteworthy, especially when contrasted with the typical excess return over the SAA allocation delivered through active management. According to a less sanguine – or perhaps more realistic – view, however, the excess return should be traded off against the cost associated with putting up with three key hurdles associated with holding a position in corporate bonds: (i) credit risk; (ii) inferior liquidity; and (iii) callability. The first two of these were treated above to some extent. In particular, Lesson 1 looked at the non-trivial costs of managing credit risk in terms of both financial and organisational capital, while Lesson 2 complemented the discussion with a deeper look at corporate bond liquidity issues.

Callability refers to the fact that some corporate bonds are issued with a provision that allows the issuer to redeem (call) the bond at a fixed price prior to contractual maturity, exposing bond holders to reinvestment risk. Naturally, much like with fixed rate mortgages, call provisions are most likely to be exercised in an environment of declining yields since borrowers can then refinance their high coupon debt at lower interest rates. Yet, this embedded short option position leaves credit investors with a skewed, or asymmetric, payout profile: if rates rise, they stand to suffer capital depreciation through mark-to-market losses; if rates fall, their bonds are likely to be called, forcing them to reinvest capital less favourably (this effect is often called “negative convexity”). In addition, since rates typically fall in macroeconomic downturns, callability works against bondholders exactly at a time when credit spreads widen and credit risk increases, exerting additional pressure on corporate bond portfolios. Hence, it seems that when it rains, it pours for corporate bond investors.

Clearly, not all corporate bonds have call provisions (roughly 50% in our benchmark index), and some research suggests that call provisions are not exercised optimally. Still, the fact remains that, at least in principle, the historically observed excess return (Figure 2.14) does not come free.

To assess the risk/return profile of corporate bonds somewhat more formally, we have performed a simple optimisation exercise in which we start from a US Treasuries-only portfolio and gradually allow exposures in alternatives: mortgage-backed securities (MBS), inflation linkers, as well as equities and corporate bonds (see Box 2.5). The interesting result is that when exposure to corporate bonds is allowed, the mean variance optimiser puts their weight at 13%. However, when we add equities to the mix, the weight of corporate credit in the portfolio falls to a mere 3% with equities at more than three times that. Indeed, there is almost no meaningful improvement in the Sharpe ratio for Portfolios 2 and 3 – compared to the benchmark UST Portfolio 1 – until an exposure to equities is added.

With all the usual caveats, this analysis seems to suggest that there probably is a limit to how large a corporate bond portfolio one should want to hold, and this limit may be lower than we would like to think. To put it in even more radical terms, while it is difficult to generalise, those who consider the excess return reported above not enough to compensate for all the fuss related with taking on exposure to corporate credit might be better served by investing in equities instead.