Tactical Asset Allocation: Can You Time Bond Markets?

Book: Systematic Fixed Income: An Investor’s Guide Author: Scott A. Richardson, Ph.D. Publisher: John Wiley & Sons, 2022 ISBN: 9781119900139

Chapter 2 made the strategic case for owning bonds. Chapter 3 asks a harder question: can you do better by adjusting your bond exposure over time? Richardson builds timing models for both the term premium and credit premium, tests them on decades of data, and delivers a refreshingly honest verdict.

Timing the term premium

Long-term US government bonds earned an average excess return of about 1.8% with 5.1% volatility over the past century. The question is simple: can you predict when that premium will be higher or lower than average?

Richardson uses three signals, each targeting a different aspect of expected returns.

Carry is measured as the term spread, the difference between long-term and short-term government bond yields. When the yield curve is steep, you earn more from holding long bonds. This is the most straightforward signal. It captures the return you get just from the passage of time if nothing else changes. A steeper curve means more carry, so you want more bond exposure.

Momentum is measured as the trailing 12-month average of government bond excess returns. The idea is simple: if bonds have been doing well recently, there is some tendency for that to continue. This could reflect slow-moving macroeconomic trends or gradual shifts in market expectations about monetary policy.

Value is the trickiest one. It is measured as the real bond yield, which is the nominal yield minus a survey-based long-term inflation forecast. When real yields are high relative to history, bonds are “cheap” and you expect yields to fall back toward normal. When real yields are low, bonds are “expensive” and you expect yields to rise.

Each raw signal gets transformed into a trading position. Richardson caps extreme values, benchmarks against expanding historical windows, and scales by volatility. The result is a signal between negative one and positive one that tells you how much to over or underweight your bond exposure.

So how did these signals perform over the 1920 to 2020 period? Here is the honest truth. The Sharpe ratios are positive but small. Carry delivered 0.31, momentum 0.15, and value just 0.07. Combining all three gave a Sharpe of 0.29.

For context, the passive buy-and-hold term premium had a Sharpe of 0.35 over the same period. The timing signals add something on top, but the added value is modest relative to just showing up and holding bonds.

Richardson titles the underlying research paper “Sin a Little,” borrowed from Asness, Ilmanen, and Maloney. The message: there is some evidence you can time the term premium, but do not bet the farm on it. The return from timing is small compared to the return from just being invested.

The correlations between signals are helpful for diversification. Value and momentum have a modest negative correlation of negative 0.17. Value and carry are basically uncorrelated at 0.01. Momentum and carry are more positively correlated at 0.48, partly because momentum includes the realization of carry in recent returns.

Timing the credit premium

The framework for timing credit is similar but with one key difference. Corporate bond returns have two components: rates and spreads. Since we already covered rates timing with government bonds, the credit timing model focuses entirely on spread returns. Think of it as working with interest-rate hedged corporate bond portfolios.

Expected credit excess returns break down into initial spread (carry) plus expected changes in spreads. Same basic structure as term premium timing, just applied to credit markets.

The three signals mirror the term premium approach. Carry is measured by the credit spread itself. Momentum is the trailing 12-month average of corporate bond excess returns. Value is more interesting. Richardson builds it from a regression model that explains credit spreads using three fundamental variables: average market leverage, average profitability, and average equity volatility. The regression residual, the part of spreads not explained by current fundamentals, is the value signal. If spreads are wider than fundamentals justify, credit is cheap.

The value model makes intuitive sense. Higher leverage means more default risk, so wider spreads. Higher profitability means stronger cash flows, so tighter spreads. Higher equity volatility means more uncertainty, so wider spreads. This connects directly to structural credit models like Merton’s framework, where the credit claim is essentially a short put option on the firm’s assets. As the firm gets riskier and the “distance to default” shrinks, credit and equity start behaving more similarly.

Results differ dramatically between investment-grade and high-yield markets. For US IG corporate bonds over 1989 to 2020, timing signals barely worked. The combined value-momentum-carry model had a Sharpe of just 0.13. Individual signals were even weaker.

But for US high-yield bonds over 1994 to 2020, the story was better. The combined model delivered a Sharpe of 0.48, with value and momentum each contributing meaningfully (Sharpe ratios of 0.23 and 0.24 individually).

Why the difference? The IG market simply has less credit premium to time. Spreads are tighter, less variable, and there is less signal to extract. The HY market has more credit risk, wider spreads, and more variability, giving timing models something to work with.

The Merton model connection

Richardson includes a discussion that ties credit and equity markets together through option pricing theory. Think of a company’s debt as a risk-free bond minus a put option on the firm’s assets. The equity is a call option on the same assets.

When a company is very healthy (asset value far exceeds debt), the credit claim is not very sensitive to changes in asset value. The put is deep out of the money. But as the company gets riskier and asset value approaches the debt level, the credit claim becomes very sensitive. The put moves toward being at the money.

This means timing signals for credit markets should incorporate equity market information, especially for riskier companies. Equity index returns contain information about changing expectations for corporate fundamentals that directly affect credit spreads.

Data mining: the elephant in the room

Richardson spends serious time warning about data mining risks. Timing models have inherently low breadth. You only have one history of one asset to test against. With term premium, you get a century of data. With credit premium, less than three decades.

There are two sneaky forms of data mining. First, the many small choices in model construction. How do you measure inflation expectations? What look-back period for momentum? Rolling or expanding windows? Each choice seems minor, but they compound. Research has shown that different implementations of the same basic strategy can produce Sharpe ratios ranging from negative 0.10 to 0.78.

Second, there is memory bias. If you have been investing for years, you remember past mistakes. When you “improve” your model based on hindsight, you are implicitly data mining even if you are not running thousands of regressions.

The remedy is consistency. Use the same data sources, the same transformation procedures, and the same model structure across different markets and time periods. Build your model with one data provider and test returns using a different one.

The bottom line

Can you time bond markets? Yes, a little. Value, momentum, and carry signals have modest predictive power for both term premium and credit premium. But the returns from timing are small relative to just being invested. The adage from the research is “sin a little”: take small tactical bets on top of your strategic allocation.

The signals work better in high-yield than investment-grade. They work better when combined than individually. And they require serious attention to implementation costs, because the gross returns from timing are not large enough to survive sloppy execution.

Most importantly, timing decisions in bonds should not be made in isolation. If your model says to underweight bonds, where does that capital go? In a world where every asset class looks expensive, the answer is not obvious.

Previous post: Strategic Asset Allocation: Bonds in Portfolios

Next post: Active Fixed Income Managers: Performance