Backtesting Equity Signals to a Fixed Income Universe

The expanding importance and liquidity of fixed-income markets has coincided with more interest in systematic strategies for fixed income investing. But most of the proposed strategies are based on pure fixed income signals. The reason? It’s difficult for researchers to map a corporate debt security to its credit parent over time. The credit parent, a corporation’s structural obligor, or the entity responsible for servicing the debt, can change over time due to corporate actions. Because of this difficulty, investigations of equity signals as predictors of corporate fixed-income performance are still comparatively rare.

FactSet, however, has been mapping corporate debt to the parent equity for many years. A recent enhancement in our quant solutions adds history to credit-parent mapping, making it easy to link corporate debt to historical parent equity signals.

The purpose of the article is to walk you through this by backtesting monthly equity signals on a universe of US high-yield corporate bonds over a 16-year period starting in December 2006. We used Alpha Testing, our workstation solution for signal research. Here is our universe specification: Corporate issues, time to maturity over one year, fixed coupons, USD denomination, minimum amount outstanding of $150M, Moody’s rating of Baa3 or lower, and available FactSet fixed income analytics.

Performance is measured as excess return over Treasury, weighted by outstanding market value. We approximate excess return as the change in a bond’s option-adjusted spread multiplied by the bond’s spread duration. We sourced this from FactSet fixed-income analytics, which offer hundreds of fixed income analytics covering over 15 years of history. Per the Alpha Testing methodology, performance is measured over the forward period to answer a key question: How do current security characteristics predict the next period's returns?

We tested the ability of five equity signals to predict bond performance in our high-yield universe.

• Value factor, Sales to Price

• Size factor, Market Value

• Solvency factor, Merton Default Probability

• Market factor, Short-term Return Volatility

• Sentiment factor, Analyst Estimate Revisions for EPS

The factor definitions draw from our Quant Factor Library dataset, which stores over 2,000 precalculated factors for the universe of traded equity securities—along with an extensive history of daily point-in-time snapshots. To facilitate comparability and to avoid sector bias, all factor data is normalized by z-score within a bond’s issuer sector group.

Equity factor data is sourced from a bond's credit parent. Equity data is only available for public companies, while debt is also issued by non-public companies. As a result, we can't source equity signals for our entire universe. The percentage of bonds with equity coverage hovers around 60% in our universe and analysis period.

Performance and Distribution Effect of Public vs Private Company Debt

A first question might be whether the selection of bonds with equity data can, by itself, affect performance or skew the sector distribution of the original universe.

Source: FactSet

For performance, we see a distinct difference. On average, bonds with public credit parents outperform those with private ones. While the cumulative performance difference is notable, it can almost entirely be attributed to the Great Recession of 2008 – 2009. Following that time, both groups' excess performance is an order of magnitude below the standard deviation of returns and therefore is not significant by any measure.

Source: FactSet

Source: FactSet

The sector distribution is not skewed significantly. In the above heatmap, size represents the portion of the sector in the original universe, and color represents the allocation difference from selecting public company debt. In relative terms, a focus on public companies overweights Consumer Durables and underweights Finance the most.

The significance of equity signals can only be determined within the group of securities for which equity data is available. Performance comparisons should be relative to this group. For this reason, we chose the aggregate of public company debt as our benchmark.

Performance of Individual Equity Signals

All equity factors experienced a regime change during the Great Recession. For a clearer signal, we reported metrics for the period from December 31, 2009, to April 28, 2023.

Numerous metrics can gauge the strength and significance of a signal. In the following chart, we plotted quintile spreads as well as the significances of the information coefficient and a market alpha. Quintile spreads represent the strategy of forming five groups of securities by factor value, then holding the top group and shorting the bottom group. As a metric, it looks at the extremes of the factor value distribution.

The information coefficient is a broader metric that measures the correlation between factor values and forward returns. We plotted the t-statistic of the correlation and loosely considered a value greater than 2 to be significant.

We derived market alpha from a time-series regression of the quintile spread on the benchmark return over 161 monthly dates. The portion of the quintile spread that is not explained by the market is the alpha. Again, we show the value of the t-statistic.

Source: FactSet

As our chart above shows, not all of our equity signals are significant predictors of returns in our universe.

• Sales to price and market value are not significant by any measure.

• Merton default probability and short-term return volatility are significant by all measures.

• Analyst sentiment is a good predictor at the value extremes but not significant across the full spectrum of factor values.

In the next chart, we show hit rates of the top and bottom quintiles during periods of positive and negative benchmark performance, respectively.

The circle represents the correlation between quintile spreads and market performance.

Source: FactSet

Almost all signals are highly correlated with market performance, predicting security performance with high reliability in up markets but poorly in down-markets. With a correlation close to zero, Sentiment is the only signal that isn’t highly correlated with market performance. It still has a positive hit rate in up markets, and it maintains a positive hit rate in down markets, suggesting this factor may offer some downside protection in contrast to the other signals.

Signal persistence can be important when transactions costs are significant because it indicates how frequently a strategy would need to rebalance to capture the signal effect. Breaking a three-month horizon into one-month periods, we can see that our solvency and volatility signals retain their information-coefficient significance over the entire period. The significance of the analyst sentiment signal declines over this period, as do the signal’s spread returns, which we don't show here.

Source: FactSet

Another way to look at potential transactions costs is to consider the stability of factor values over time. In the following chart, we show the average percentage turnover when we rebalance a portfolio to the first quintile of each factor each period. To clarify the metric, a complete replacement of all securities would correspond to a turnover of 200%.

Our solvency factor has the lowest turnover of the significant factors, while analyst sentiment has the highest.

Equivalently, a rank correlation of factor values over time shows the same pattern of stability. If a bond's parent equity has high short-term return volatility today, it is probable the volatility will remain high over the next period as well.

Source: FactSet

Besides the risk of high transactions costs with a signal strategy is the obvious risk of negative performance. Maximum drawdown represents the lowest point of a portfolio's performance before recovering to the original level. For context, we charted the benchmark return on the left axis. Recalling the market correlation of our solvency and volatility factors, it is not surprising their drawdowns coincide with market downturns. Analyst sentiment, which was uncorrelated with the market, is remarkably stable over the entire period.

Source: FactSet

To understand the unusual riskiness of the default probability signal, it's worth noting how the signal's quintiles were constructed in this analysis. Bonds were placed in the top quintile if their parent equity had the highest probability of default. Since the creditworthiness of public companies is easily visible, we could attribute the overall outperformance of this quintile to a risk premium on the parent equity's shaky balance sheet.

It is therefore unsurprising that market downturns strongly affected these companies and their debt. Indeed, the following chart illustrates that, with the exception of the Great Recession period, negative spread performance of the solvency signal is primarily driven by a negative performance of the top quintile (blue)—the companies with lowest solvency.

Source: FactSet

Performance of a Combined Equity Signal MFR

Combining multiple signals can often improve signal strength. We examine an equal-weighted multi-factor-rank (MFR) of all equity signals that we have considered so far.

Over the period from 2009 to 2023, the combination MFR maintains a high hit rate for its top quintile in up markets.

Source: FactSet

Average quintile returns are more consistent than those of the components. The MFR’s quintile ranks predict performance across all values, not only in the extremes.

Source: FactSet

The information coefficient of the combined MFR, for which we use a pooled calculation each year, is also more consistent over time.

Source: FactSet

Over time, a strategy based on the top quintile of the combined rank outperforms the benchmark significantly. A strategy based on the quintile spread avoids the extreme risks we saw for the component signals for solvency and volatility.

The following chart shows cumulative performance on a log scale on the right axis, and it plots maximum drawdown on an absolute scale on the left axis.

Combination MFR – Quintile Returns and Spread Drawdown

Source: FactSet

In conclusion, the US high-yield debt market is clearly reactive to parent-equity characteristics. Using factor definitions based on quant factor library data and associating them with our fixed income universe through credit-parent mappings, we have chosen signals to represent common styles and have found several to be highly predictive for future bond returns.

Further analysis could determine whether these signals can be exploited profitably after taking account of transactions costs. In addition, analysis would certainly identify signals that can take advantage of multi-asset dependencies with even greater efficiency.

For additional discussion of this topic, view our recent webcast, Factor Research In Fixed Income Markets. 

The authors of this FactSet article are Peter Sachs and Ravinder Dosanjh. 

Peter Sachs is a Principal Product Manager of Quant Solutions at FactSet. In this role, he is responsible for the development of workstation quant solutions, especially the factor backtesting application Alpha Testing. Prior to FactSet, he was part of the investment strategy team at Banc of America Securities. Peter earned a Master of Finance from the Goethe University in Frankfurt, Germany.

Ravinder Dosanjh is a Specialist in Risk and Quantitative Analytics at FactSet. In this role, he is one of FactSet’s experts for Portfolio Risk and Quantitative Analytics and has spent the last four years specializing in workflows and solutions for portfolio and quantitative analytics including, but not limited to, factor research, portfolio construction, optimization, performance, and risk and factor attribution. Starting as a Consultant in 2012, he spent three years working with some of FactSet’s largest buy-side clients across the UK. He then joined the Analytics team in 2015 covering the same region, before working with clients across EMEA from 2019. Mr. Dosanjh is a CFA charterholder since 2017, and he earned a Bachelor of Science in Economics from the University of Birmingham.

This blog post is for informational purposes only. The information contained in this blog post is not legal, tax, or investment advice. FactSet does not endorse or recommend any investments and assumes no liability for any consequence relating directly or indirectly to any action or inaction taken based on the information contained in this article.