This commentary is written by Martin Fridson, a high-yield market veteran who is chief investment officer of Lehmann Livian Fridson Advisors LLC as well as a contributing analyst to S&P Global Market Intelligence.
Two of our recent pieces dealt with the determinants of variance in spreads on high-yield bonds. “What's the best financial ratio for HY analysis?” produced the surprising result that of eight standard credit ratios we tested, the one that best explained spreads was perhaps the least heralded, as well as being based on widely disparaged GAAP earnings and book equity. Net Income - Dividends / Total Debt + Total Equity by itself explained more than half the spread variance in our test sample. "How maturity and issue size affect high-yield spreads" found that maturity and face amount outstanding (a proxy for liquidity) had correlations with option-adjusted spread, or OAS, that did not rise to the level of "weak." Contrary to prevailing wisdom, calculating amount outstanding per issuer rather than per issue reduced that variable's OAS correlation.
A thorough inquiry into the question countless credit analysts devote their working lives to ("How can we determine whether Bond X is trading too tight, too narrow, or just right?") ought to give at least a nod to Modern Portfolio Theory, or MPT. Yes, one renowned market strategist dismissed MPT early on with a barnyard epithet and other practitioners derided its equating of risk with volatility. The theory's founder, Harry Markowitz, did win the Nobel Prize in economics, though, and his innovation has spawned a vast body of investment research. Furthermore, the case study included at the end of this piece demonstrates that MPT has even had an impact on corporate strategy. Ignoring the possibility of a connection between corporate bond risk premiums and risk — as MPT defines it — could be considered neglectful if not oblivious.
MORE FRIDSON: Anatomy of a high-yield spread reversal
Indeed, anyone who rejects out of hand the possible relevance of volatility to corporate bond spreads must confront the chart below. For each major rating subdivision of the corporate universe, the chart plots mean OAS for the period 1997-2020 versus standard deviation of monthly returns. Visual inspection confirms a strong connection between the two.
More formally, the correlation (R) of 99.7% means that at the aggregate level (based on ratings), volatility single-handedly explains 99.4% of the variance in spreads (R-squared). Only the most devoted (a polite way of saying "disturbingly obsessive") analyst would expend nontrivial effort in seeking to explain the remainder. Our work is not done, however. We have shown that the possibility of a link between volatility and spreads cannot be brushed off, but it remains to be determined whether the correlation is as strong at the individual issue level as it is at the aggregate level.
It also bears noting that volatility of returns is a function of other factors that could alternatively be used as explanatory variables to create a chart like the preceding one. The lowest-rated, most leveraged issuers have smaller cushions between their debt and their underlying equity value than the highest-rated ones. A decline in the value of a bottom-rated issuer's equity will put it closer to outright elimination of the equity cushion (that is to say, default) than in the case of a commensurate decline in the value of a top-rated company's equity. We should therefore expect the bottom-rated company's bonds to experience the greater swings in default risk premium and price as the economy alternately boosts and punishes the earnings of companies generally. It follows, Q.E.D., that one should be able to construct a graph of aggregate spreads about as persuasive as the one above by scaling the horizontal axis not with volatility but with a financial ratio such as Total Debt / Total Debt + Market Value of Equity.
Our above-referenced Aug. 10 piece, though, showed the limitations of correlating financial ratios with spreads at the individual issue level. Those findings underscore the importance of proceeding from the aggregate spreads to issue-specific spreads to determine how well volatility serves as an explanatory variable in the latter context. Spoiler alert: We do not find a correlation as high as 99.7%.
In principle, standard returns of monthly returns on individual high-yield bonds can be plotted against their mean spreads, just as we did with the aggregate rating data in the preceding chart. The practical impediment is that neither Bloomberg nor ICE Data Indices LLC reports issue-specific monthly returns. They do provide the raw data from which one could create a database of monthly returns, but we elected not to invest the hundreds of hours of labor necessary to create a graph directly comparable to the one displayed above. We opted instead to use data that Bloomberg does supply in readily manipulable form, namely, price histories.
Another constraint is that very few if any high-yield bonds (possibly a negligible number of fallen angels) were outstanding on Dec. 31, 1996. Even those that have been outstanding for five years may have been much less or much more volatile in the earlier years than more recently, as their credit quality may have changed materially in the interim. We made the assumption that if the market does care about an issue's volatility, it is probably concerned about how volatile the issue has been in relatively recent times, which we decided to define as Aug. 26, 2020, to our observation date, Aug. 26, 2021.
We further diverged from the procedure reflected in the preceding chart by measuring OAS on the observation date, i.e., the end of the period, rather than plotting the mean OAS for the period on the vertical axis. This choice was driven by the desire to link up our findings with investment decisions facing high-yield portfolio managers. They must decide whether to buy, hold, or sell an issue on the date that they review the credit, so the issue's mean spread over the preceding year is not directly relevant to them.
Substituting price volatility for total return volatility represented a departure from the tenets of Modern Portfolio Theory. We therefore had to ask whether our revised procedure constituted a valid test of the proposition that corporate bond premiums. The chart below answers that question in the affirmative.
In creating this graph we focused on the three broad speculative-grade rating categories. We normalized for the categories' disparate maturity mixes by selecting the 5-to-8-year maturities in each case. As proposed for our individual-issue-spread analysis, we defined volatility as the standard deviation of the Aug. 26, 2020, prices divided by the mean price for that period. We measured OAS on Aug. 26, 2021. As the chart's inset indicates, the correlation between volatility and spread was 92.9%, meaning that the percentage of the variance in the spread explained by volatility (R-squared) was 86.3%. In short, the relationship proposed by MPT remained quite powerful at the aggregate level, despite the adjustments necessitated by data limitations. Accordingly, we went ahead with our idea of testing the modified process at the individual security level.
A key objective in creating our test sample was to minimize the impact of potential determinants of spread other than volatility and credit factors that differentiated the issuers them other like-rated companies. We therefore restricted our sample to senior unsecured bonds rated BB3 (the subcategory of ICE BofA BB U.S. High Yield Index containing the most issues). To remove maturity as a factor, we included only bonds maturity in 2027, roughly corresponding to the index's 7.36-year average maturity on our observation date. In further pursuit of homogeneity, we eliminated a small number of financials to leave only industrials (there were no utilities in our initial sample). We also excluded Energy on the grounds that on our observation date Energy bonds were systematically wider than non-Energy bonds. The ICE BofA U.S. High Yield Energy Index was 108 basis points wider than the ICE BofA US High Yield Index on Aug. 26, 2021, even though Energy's average rating was BB3 versus B1 for the all-industry index. The December 1996-December 2020 monthly mean OAS differential was -14 bps.
One of a single issuer's two bonds maturing in 2027 was eliminated from our sample to avoid duplication. We removed the one with an Aug. 26, 2021, price of 113.25, the highest the preliminary sample. Another issue did not make the final cut due to unavailability of Bloomberg pricing data. (We elected not to collect all of the bond's 250 or so daily prices of the past year one by one from the ICE BofA U.S. High Yield Index.) Two other issues entered the index more recently than Aug. 26, 2020, so their price histories of less than a full year plus a day would not be comparable to those of the other issues.
The remaining bonds constituted a test sample of 28. That was a bit shy of the rule-of-thumb of 30 required to make up a valid scientific sample (note 2). In light of the details of our findings, we do not consider that shortcoming fatal. Within the test sample, our volatility measured ranged from 0.34 to 4.58 and OAS ranged from +168 bps to +530 bps.
Results and interpretation
The correlation between price volatility and OAS in our final sample was 36.4%. That qualifies as a weak correlation. At the individual bond level, volatility and spread were much less tightly connected than at the aggregate ratings level. The issues lined up much less neatly than the rating categories in the two charts above. Some of the widest-OAS bonds had some of the lowest volatilities and vice versa.
On the other hand, the correlation with spreads was stronger than our previous research found for maturity, face amount outstanding, and most standard financial ratios. Accordingly, our price-based volatility measure merits consideration as an explanatory variable in constructing a multi-factor model aimed at identifying misvalued high-yield bonds.
Case study: How MPT drove business innovation
During my years at Merrill Lynch (1989-2002) I attended periodic management gatherings at which senior executives made presentations on the company's business strategy. The CEO might announce, for example, that Merrill Lynch was reducing the number of divisions to foster a more cohesive, integrated interface with clients and customers. Another year, the number of divisions would increase. A cynic could have supposed that these changes had more to do with divvying up power than with operating more efficiently and effectively. The management consultants presumably made out well, in any case.
One year came a breath of fresh air. COO Herb Allison presented an analysis that genuinely pointed to a productive, new direction for the company. He displayed a table that ranked several prominent stocks from highest to lowest by price-earnings ratio. Merrill Lynch stood near the bottom. What accounted for the superior valuations of the companies at the top of the chart?
Allison explained that the E in their P/Es was less volatile than Merrill Lynch's. For example, consumers purchased Johnson & Johnson's personal care products on a steady basis, regardless of recession or expansion. In contrast, when the Dow plummeted, Merrill Lynch's retail customers' interest in the stock market waned and with it the company's commission volume.
To achieve a higher P/E, Allison concluded, Merrill Lynch needed to shift its business more toward segments with steady revenues. Wealth management fees, for example, were tied to assets under management, a stabler number than the New York Stock Exchange trading volume. Allison's analysis was perhaps less revelatory to the MPT-conscious research directors in the room than to many other managers, but it illuminated a promising path forward that all could comprehend and buy into.
When the managers reconvened a couple of years later, Allison presented an updated version of his chart. Lo and behold, Merrill Lynch had moved up a few slots, thanks to a shift in business mix that had reduced its earnings volatility. It was truly a case of MPT advancing from theory into practice in Fortune 500 business management.
Research assistance by Manuj Parekh and Weiyi Zhang.
ICE BofA Index System data is used by permission. Copyright © 2021 ICE Data Services. The use of the above in no way implies that ICE Data Services or any of its affiliates endorses the views or interpretation or the use of such information or acts as any endorsement of Lehmann Livian Fridson Advisors LLC's use of such information. The information is provided "as is" and none of ICE Data Services or any of its affiliates warrants the accuracy or completeness of the information.
1. The AAA category's refusal to line up correctly is presumably a function of statistical noise owing to the fact that it contains bonds of just 19 issuers, of which 14 are universities, foundations, or arts institutions. By comparison, 845 issuers are represented in the BBB subindex of the ICE BofA U.S. Corporate Index. A rule-of-thumb puts the minimum number of members to constitute a valid scientific sample at 30. Removing the AAA sector from the analysis increases the correlation with spread from 99.7% to 99.8%.