There’s a widely held proposition that mean reversion has been, and can be, a powerful and reliable driver of alpha generation. Having recently re-tested the existence of mean reversion in corporate bond markets, we would agree with this assertion. But only up to a point. 

A mean reversion strategy involves buying low and selling high, with trading signals informed by historic trading data. In a bond fund, such a strategy might help identify when to buy a cheap bond or sell an expensive bond in the belief that pricing will return closer to the mean. However, due to the payoff structure of corporate bonds, we believe it’s more important to avoid issuers that are downgraded from investment grade, or those that default. Fund performance can be adversely affected if one or more issuers fail to revert and subsequently face difficulties.

We've seen a rise in quantitative or rules-based trading strategies across credit markets. Many are passive, while others are more active or labelled ‘smart beta’. Some newer strategies utilise artificial intelligence or techniques with little human influence or interaction. Despite their sophistication, many of these mirror mean reversion strategies or rely on historic trading data.

At abrdn, our recent analysis showed that, statistically, mean reversion exists. It’s particularly pronounced for European investment grade bonds, but also evident in markets such as US municipal bonds. Our back-testing confirmed that mean reversion can potentially add significant alpha. It also showed that it can go wrong – sometimes spectacularly so. This failure can counteract good performance in the rest of the portfolio.

How does mean reversion look in practice?

Take Chart 1. It details the relative performance of euro-denominated bonds from satellite operators Eutelsat (ETLFP) and SES (SESGFP). Both bonds have low coupons and 2028 maturities. In 2022, Fitch rated the bonds BBB. So, we have bonds with the same credit quality, operating in the same industry, with the same maturity and similar trading liquidity. Their credit spreads (the darker lines in our chart) were even the same at the end of June 2022.  


Chart 1: Does mean reversion always work?

The dashed lines (right-hand axis) show each bond’s Z-scores. Investors can use Z-scores to help identify when a bond is ‘cheap’ or ‘expensive’ versus its trading history. In other words, when they are prime mean reversion candidates.

A Z-score higher than two or three usually means a bond is cheap. The highest Z-scores for our bonds were in the summer of 2022. Eutelsat’s Z-score exceeded seven in late-July 2022. SES’s highest score (nearly four) was in early-August 2022. Both investment-grade bonds should have been a high-conviction buy.

Key determinants of whether mean reversion works include a) the length of trading history used to calculate the Z-score, and b) the period over which one would expect the mean reversion to occur.

Our analysis produced estimates for both (a) and (b). For illustration, let’s assume mean reversion happens most frequently after six months. We’ve marked six-month periods on our chart (the arrows) following the highest Z-scores (indicating a buy).

As we can see, there was mean reversion for SES, with spreads tightening over 100 basis points (prices go up as credit spreads fall, all else being equal). However, there was no reversion for Eutelsat. In fact, spreads were a little higher after six months.

Let’s assume it takes longer than six months for mean reversion to occur and look at the nine months post-signal. SES spreads still tighten, albeit by just over 50 basis points (a loss in performance versus six months). By contrast, Eutelsat spreads revert, tightening over 100 basis points. The take-home message? Following a set of prescribed rules or trading guidelines can result in markedly different outcomes.

Look to the future

The later dates in the chart also show that re-testing or re-setting for mean reversion might not work. Eutelsat’s Z-score was again above two at the beginning of 2023. However, spreads materially widened over the year (albeit with some volatility). Investors would have significantly underperformed if they’d applied the learnings from 2022 (that is, a holding period of over nine months for Eutelsat bonds is better than six months or fewer). By contrast, SES continued to demonstrate some mean reversion, with spreads tightening after bouts of volatility.

This is just one example. There are many more that rely on mean reversion and don’t work. This emphasises the importance of skilled credit analysts who, while considering trading levels and history, effectively integrate relative value assessments with a forward-looking, fundamental credit evaluation.

Now, back to our satellite operators. At one point, the market (and Fitch) prescribed the same credit risk to Eutelsat and SES (that is, the same spreads and rating). However, a forward-looking fundamental assessment could have flagged significant risks about Eutelsat’s operational performance and subsequent cashflow generation relative to its indebtedness. Throughout 2023, the market and rating agencies caught up and reflected the deterioration in credit quality. This caused material spread widening and ratings downgrades from investment grade into high yield.

Final thoughts…

There are limitations to corporate bond trading strategies, including mean reversion. Recent changes in the market environment, such as the fallout from Covid-19 and years of ultra-low interest rates, mean it’s not ideal to just rely on trading history or assumptions based on normalisation.

Credit analysts must consider a variety of data and variables to provide accurate investment recommendations. At abrdn, we don’t rely solely on mean reversion or rules-based investment approaches. Instead, we prefer meeting with issuers, understanding what’s changing, and identifying the triggers that will influence future investment performance – not the past.