latest-news-headlines Market Intelligence /marketintelligence/en/news-insights/latest-news-headlines/fridson-tree-method-vs-comps-to-identify-likely-high-yield-alpha-generators-61359031 content esgSubNav
Log in to other products

Login to Market Intelligence Platform

 /


Looking for more?

Contact Us
In This List

Fridson: Tree method vs comps to identify likely high-yield alpha generators

2021 Leveraged Loan Survey: Defaults edge higher; credit quality a concern

M&A Among Credit Providers to Accelerate Due to Pandemic, Dealmaker Says

ESG, Energy Companies, And Downside Protection For Investors

This is How High-Yield Managers Are Addressing ESG


Fridson: Tree method vs comps to identify likely high-yield alpha generators

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.

Shortcomings of valuation by allegedly comparable issues
The traditional method of relative valuation for individual high-yield bonds relies on “comps.” A bond's yield or spread is compared to yields or spreads on bonds of industry peers. Ideally, the financial ratios of the issue under consideration are in a range closely bounded by those of the comps. In that case, the appropriate yield or spread (we shall focus on the latter) for the bond being evaluated can easily be deduced.

Here is a partial list of the comps method's shortcomings:

* The issuer in question may have no industry peers, at least none operating in a sufficiently narrow category within a broadly defined industry such as “Leisure.”

* The issuer's historical financial ratios may lie far outside the range of the ratios of the few available comps.

* Historical financial ratios capture far less than 100% of the relevant information that determines a bond's spread, leaving out such factors as quality of earnings and quality of management, as well as risks associated with competitive dynamics, potential new government regulation, and environmental hazards.

* The issuer's latest earnings and interest coverage may be heavily influenced by cyclical or one-time factors, undercutting the validity of comparisons with industry peers.

* Financial data for a very small sample size of comps may contain a debilitating amount of statistical noise.

* For an issue that looks cheap by virtue of a spread far out of line with its financial ratios, the comps method offers no insight into the likelihood that the issue's spread will revert to a more appropriate level. Investors consequently get no indication whether they will be able to buy a purportedly rich bond at a more attractive level in the future or whether buying a seemingly cheap bond will lead to a price gain through future spread tightening.

We describe below an alternative approach to issue-level valuation, present empirical evidence regarding its efficacy, and provide current output to facilitate employment of the methodology. Readers who believe his technique can contribute usefully to their investment process are encouraged to provide feedback to marty@fridson.com. If there is substantial demand for it, we can add relevant data for identifying outliers to our regular monthly updating.

MORE FRIDSON: High-yield laggards soar on vaccine news, plus spread analysis

Building a factor tree
A further problem with the comps method is that in effect it assumes that the company being analyzed is competing for capital solely with a handful of companies in the same or related lines of business. In reality, every U.S. company, public and private, in every industry, competes for scarce capital with every other such company. Not to mention cross-border capital flows and competition for capital from a variety of non-corporate users, e.g., government, non-profit enterprises, homebuyers seeking mortgages.

Modeling a capital market as vast as all that, if feasible at all, is beyond our research team's capability. We can, however, improve upon a method that effectively assumes that Yum Brands' cost of debt is determined in toto by its competition for capital with four other restaurant operators. That is what one might infer from the composition of the ICE BofA US High Yield Restaurants Index.

Pitting a single high-yield issuer against all 853 others in the ICE BofA US High Yield Index would be preferable in theory, but counterproductive in practical terms. Such an approach would require adjustments for the numerous factors that influence spreads of individual issues, such as credit quality, maturity, and seniority. Our empirical work on this project led us to conclude that creating a massive multiple regression formula to account for all of these factors was not the most fruitful approach.

We opted instead to create a factor tree, with letter-grade Composite Rating category as the root. At least as an initial step, we limited our universe to the BB portion of the ICE BofA US High Yield Index. Greater dispersion of spreads may limit the reliability of interpretation of output from the lower rating categories. We assigned each bond to a peer group (or leaf node, in tree terminology) based on the following four categories:

* BB alphanumeric Composite Rating — 1, 2, or 3 (note 1)
* Seniority — secured or senior unsecured (note 2)
* Net ratings prospects (NRP) — Positive, Stable, or Negative (based on the majority of rating agency opinions)
* Maturity — 1-5, 6-10, or >10 years (note 3)

Adding more factors would produce unacceptably small peer groups. Note that differences in callability are addressed by using the option-adjusted spread as the dependent variable in the analysis. Choosing the four factors shown above produced a total of 3 x 2 x 3 x 3 = 54 leaf nodes. That resulted in an average of 993 BB issues ÷ 54 = 18 secured or senior unsecured issues per peer group, although of course those issues were not distributed exactly equally among the 54 peer groups.

We chose as our observation dates March 31 and June 30, 2019. The ICE BofA US High Yield Index's option-adjusted spread (OAS) on those two dates was +405 bps and +407 bps, respectively. By selecting observation dates with nearly identical levels of perceived credit risk, we ensured, to the extent possible, that the observed spread changes on individual issues would be a function of either changes in their issuers' fundamentals or drift toward or away from intrinsic value, rather than general market movements.

We judged that the three-month interval between our two observation dates represented an acceptable horizon for high-yield investors seeking to generate alpha by capitalizing on transient mispricing. Clearly, market-wide dislocations may lengthen the period required for realization of a profit from our proposed strategy. Note, too, that the strategy aims at producing relative outperformance. If, for example, our analysis finds that an issue is undervalued but in the subsequent period the high-yield market as a whole declines, a loss smaller than the high-yield indexes will be counted as a successful outcome.

For each peer group we calculated the mean OAS and standard deviation. We considered an issue mispriced if its OAS was outside the range of plus/minus one standard deviation from the group mean. Our objective was to determine whether such outliers had a propensity to return to the fairly valued zone, defined as and OAS within the plus/minus one standard deviation range. If so, investors can reasonably treat an OAS narrower than -1 standard deviation as a short or avoid-for-now signal. Under similar logic, an OAS wider than +1 standard deviation can be considered a buy signal.

We fully recognize that some practitioners will regard this model's use of ratings as a reason to reject it out of hand. Whether or not one agrees with the rating agencies' assessments in every case, however, the ratings are indisputably more comprehensive assessments of credit quality than the handful of financial ratios employed in the comps method (note 4). The agencies' rating committees take into account all of the non-ratio credit factors discussed above. Ultimately, the usefulness of our tree methodology does not hinge on philosophical arguments about the accuracy of ratings but on the results it produces, to which we now turn.

Results
Only peer groups that contained at least one outlier issue (as defined above) on March 31, 2019, are pertinent to the present analysis. Of the 54 leaf nodes, 24, or 44%, satisfied that criterion. Those 24 contained 73 narrow-end outliers (rich) issues on March 31. On our June 30 observation date, 18 were still outliers at the narrow end, producing a repeat rate of 24.7%. At the wide end, there were 88 outliers on March 31, of which 46 repeated, for a 52.3% repeat rate. The difference between 24.7% and 52.3%, given the sample sizes, is statistically significant at a greater than a 99.9% confidence level. We conclude that under reasonably calm credit market conditions (note 5) BB-rated high-yield issues are more likely to be cheap than rich versus their intrinsic value.

SNL Image

These findings tell us that an investor who shorts an outlier, or avoids buying it, in expectation of a more attractive relative valuation within a modest time frame, has a three-out-of-four (100.0% - 24.7% = 75.3%) chance of making a winning trade. Those are appealing odds in a business known for the cliché that the goal is to be right 51% of the time. Portfolio management teams who pride themselves on their analytical prowess should be able to improve on the 75.3% probability of success, assuming their self-assessment is accurate.

Most readers are probably more interested in the other trade, i.e., buying an outlier on the wide side and realizing a gain as it regresses to a fair valuation. Unfortunately, this is essentially just a 50/50 (100.0% - 52.3% = 47.7%) proposition according to our data. Keep in mind, however, that these results are based on all 88 wide-end outliers on March 31, 2019. A subset of them produces a more intriguing result.

Let us zero in on bonds with the striking characteristic of being more than one standard deviation wider than their peer-group means despite positive ratings prospects. We label these “wrong-way” issues. Even investors who habitually denigrate the rating agencies might consider it noteworthy that a bond is trading extremely wide versus its rating peers, normalizing for any differences in seniority and maturity, even though the agencies regard it as an improving credit. Our March 31 sample contains 24 such outliers. Only six of them were still wide-end outliers on June 30. In short, 75% of the wide-end, wrong-way bonds became winners, representing attractive odds, by our lights.

At the narrow end, there were 60 issues with negative ratings prospects. The most ardent rating agency skeptics might sit up and pay attention when a bond is extremely rich versus its rating peer group even though the agencies indicate that it is likely to be rated lower than those peers in the not-very-distant future. This narrow-end, wrong-way subgroup’s success ratio, like the wide-end subgroup’s, was (60 -15) ÷ 60 = 75%.

Caution is warranted in drawing conclusions about the “right-way” issues shown in the table, due to their small sample sizes. These are (a) wide-end outliers with negative or stable net rating prospects or (b) narrow-end outliers with positive or stable net rating prospects. With the sample-size caveat, we note that the clear majority of such issues repeated as outliers on June 30, 2019.

A plausible explanation of the “right-way” results is that many bonds trading extremely cheap relative to their rating peers generally deserve to be trading cheap because the market is anticipating downgrades. Among these are the bonds that the rating agencies are already identifying to investors as possible or likely downgrades. Similarly, many bonds trading extremely rich relative to their rating peers generally deserve to be trading rich because the market expects them to be downgraded. These issues will regress toward their rating-group peers only if the issuers’ fundamentals take an unexpected turn for the better.

The case is much different for bonds that are extremely cheap versus their rating peers despite positive net ratings prospects or extremely rich versus their rating peers despite negative net ratings prospects. Their anomalous spreads reflect, in many if not most cases, factors such as temporary supply/demand imbalances or trading frictions. These issues have a decided propensity to revert toward their peer-group mean spreads within a comparatively short period.

Conclusion
Our tree method of partitioning the BB universe identifies a set of characteristics that makes outlier bonds likely to migrate toward intrinsic value. Investors on either the long or short side have the wind at their backs when they make security choices within the group of bonds bearing these characteristics. If readers express interest in pursuing this approach, we will supply in the future mean spreads and standard deviations for the peer groups described above, enabling them to identify opportunities based on our methodology.

Increased credit availability somewhat eases high-yield overvaluation
The option-adjusted spread on the ICE BofA US High Yield Index tightened to +532 bps on Oct. 31, 2020, from +541 bps on Sept. 30, 2020. That change was dwarfed by the contraction of our Fair Value estimate of the spread to +915 bps from +1,296 bps. The gap between the actual spread and our Fair Value estimate consequently eased to -383 bps from -755 bps, but the high-yield asset class remained extremely overvalued, as depicted in the chart below. For the time being, extraordinary Fed intervention in financial markets in response to the COVID-19 epidemic is dominating the valuation of high-yield bonds as well as various other asset classes.

SNL Image

The latest month's huge decrease in the Fair Value estimate primarily reflected a dramatic improvement in Credit Availability. In the newly released quarterly survey of senior loan officers, the percentage of banks tightening lending standards on businesses exceeded the percentage easing standards by 27.3 percentage points, down from 71.3 percentage points in the previous quarter. In addition, the monthly Industrial Production indicator improved to 1.1% from -0.6% a month earlier and Capacity Utilization rose to 72.8% from 71.5%. Also helping was a rise in the five-year Treasury yield, which is inversely correlated with the spread, to 0.39% from 0.27%.

The default rate, a backward-looking indicator with only a minor impact on the spread, nominally increased to 6.3% from 6.2%. Our dummy variable for quantitative easing remained at 1, indicating that quantitative easing remains in force.

These conclusions are drawn from the updated methodology presented in "Fair Value update and methodology review" (LCD News, Jan. 24, 2018). In brief, we find that 80% of the historical variance in the ICE BofAML US High Yield Index's option-adjusted spread, or OAS, is explained by six variables:

  1. Credit availability, derived from the Federal Reserve’s quarterly survey of senior loan officers.
  2. Capacity utilization.
  3. Industrial production.
  4. Current speculative-grade default rate.
  5. Five-year Treasury yield.
  6. A dummy variable for the period covered by quantitative easing (QE).

Each month we calculate a Fair Value spread based on the levels of these six variables. The extent of high-yield overvaluation or undervaluation is determined by the difference between the actual OAS and the Fair Value number ("estimated"). We define an extreme valuation as a divergence of one standard deviation (124.5 bps) or more from Fair Value. The monthly difference between the actual and estimated OAS is tracked in the chart above.

Research assistance by Lu Jiang and Zhiyuan Mei.

ICE BofA Index System data is used by permission. Copyright © 2020 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.

Notes
1. Composite Ratings are those determined by ICE Indices, LLC. Bloomberg employs a different formula for calculating them.

2. Given that capital structure priorities other than secured and senior unsecured currently account for just 3.6% of the ICE BofA US BB US High Yield Index’s issues, their inclusion would produce peer groups too small to be statistically reliable.

3. For discussion of the non-uniformity of spreads across the maturity spectrum, see "What explains negatively sloped spread curves?".

4. In years past, we have actually heard traders assert that a bond’s “true rating” is the output of a supposedly scientific, purely financial-ratio-based process. This is utter nonsense.

5. Over the period December 1996 to December 2019, the ICE BofA US High Yield Index’s mean monthly OAS was 487 bps. As noted, the March 31, 2019 OAS was +405 bps.