A New Way to Set Issuer Limits in Corporate Credit Portfolios

14 Oct 2014

AT A GLANCE

  • Credit ratings are not always a reliable indicator of credit risk
  • Credit spreads have been shown to predict spread volatility across credit classes
  • Credit spreads can be used in place of credit ratings to determine maximum individual issuer limits
  • This approach to position sizing can help improve the efficiency of portfolio management and protect portfolio performance in the event of a corporate blow-up

Fixed income portfolio managers constantly grapple with the problem of setting appropriate issuer limits in corporate credit portfolios. However, in spite of its fundamental importance in portfolio management, most approaches to setting issuer limits are surprisingly crude. More often than not, limits are based simply on credit ratings, with a maximum exposure defined for each rating class: for example, AAA: 5%, AA: 4%, A: 2%, BBB: 1%. But credit ratings can be inaccurate, and can also be slow to reflect changes in credit quality. The most obvious examples include the AAA ratings of Enron prior to 2001, WorldCom before 2007, and most European sovereigns prior to 2010. In this article, we describe an alternative approach to setting upper limits to the exposure to single issuers that is based on issuer spreads. It is rooted in the widely observed fact that many measures of spread risk, such as spread volatility and the distribution of spread changes, are proportional to absolute spread levels.

BACKGROUND: A SUMMARY OF SPREAD BEHAVIOR

It was shown as far back as 1999 that the volatility of US investment-grade corporate bond spreads is proportional to the spreads’ absolute level. In the years since, this observation has been shown to hold in Europe as well as the US, and across industries, rating classes and maturities.

A useful rule of thumb emerges from these studies: over long horizons, the annualised volatility of a corporate bond spread is approximately 30% of its current level. For example, if a spread is 200 basis points (bps), its spread volatility will be about 60 bps per year. If the spread tightens to 100 bps, its spread volatility falls to about 30 bps per year, while if it widens to 400 bps, its spread volatility doubles to about 120 bps per year.

This observation has important implications for investors, and has in the past been used as an input in fixed income risk management software and analytic tools, and also as a measure to identify bonds that are expensive or cheap relative to their true probability of default.

DON’T FORGET TAIL RISK

But looking at spread volatility alone is not enough, in our view. We also need to take into account tail risk when setting issuer limits in credit portfolios – and credit tails can be very long.
With this in mind, we examined spread behaviour in two very different corporate bond indices: the Barclays US Corporate Index, which consists of US corporate bonds (looking at data from May 1993 to July 2013); and the JP Morgan CEMBI Index, which is made up of dollardenominated emerging market corporate bonds (data from December 2001 to July 2013).

Our aim was to analyse tail risk in credit portfolios. We calculated the following for each index:

  • The spread return (the percentage change in spread over a month) of each issuer in the index in each month
  • The volatility of the spread returns over the periods analysed
  • The range of monthly spread returns for all issuers in the indices trimmed at two percentiles: first the 1st and 99th, and then the 5th and 95th

The expected shortfall of monthly spread returns and spread changes at the same percentiles (1st/99th and 5th/95th). Our main findings were as follows.

  • Despite the two indices differing widely in their construction and in the types of bonds they contain, the monthly spread returns of their constituents exhibit very similar patterns of behaviour. This suggests that our findings are applicable across credit types.
  • Over the entire period, the volatility of spread returns for both indices is around 12% per month. When the sample is trimmed by 1% on each side, this figure falls to 10%, and when it is trimmed by 5%, it falls to 7%.
  • The corresponding one-month expected shortfall levels at the 99% and 95% confidence limits (that is, the average spread return in the best and worst 1% and 5% of cases) are 45% and 26% respectively.
Download to read more