Home » Research Blog » Commodity trends as predictors of bond returns

Commodity trends as predictors of bond returns

Simple commodity price changes may reflect either supply or demand shocks. However, filtered commodity price trends are plausibly more aligned with demand, economic growth and, ultimately, inflationary pressure. All of these are key factors of fixed income returns. Empirical analysis based on a basket of crude oil prices shows that their common trend is indeed closely associated with empirical proxies for demand and has predictive power for economic output. More importantly for trading strategies, the oil price trend has been able to forecast returns in 20 international bond markets, both in-sample and out-of-sample.

The below post is based on:
Jondeau, Eric, Qunzi Zhang and Xiaoneng Zhu (2019), “Crude Awakening: Oil Prices and Bond Returns”, Swiss Finance Institute, Research Paper Series, N°19-24

The below text consists of quotes from the paper, but headings, emphasis, and cursive text have been added for context and ease of understanding.
The post ties in with the SRSV summary on macro trends, particularly the section on how to use financial market data for tracking macro trends.

In a nutshell

“[Unfiltered] oil price changes fail to predict asset returns because they are too noisy. We construct an oil trend factor that filters out noise and provide evidence that it predicts bond risk premiums well… we find that a substantial portion of the variation in the bond risk premium can be attributed to the oil trend factor…This result holds in developed and emerging markets, both in-sample and out of sample.”

“The oil trend factor is demand-driven. That is, increases in the oil trend factor reflect increases in oil demand and economic output. Therefore, they should negatively predict the bond risk premium.”

“The oil trend factor improves predictions based on current term structure predictors, such as the first three principal components of yields and the Cochrane and Piazzesi factor.”

Why it is the trend that matters

“Oil price changes, a key indicator of business cycle fluctuations, are important for understanding asset prices… By better capturing the process by which agents form expectations for economic conditions, the oil trend factor should help predict future bond returns…The nature of the relationship depends on the nature of the shock. If oil price increases are driven by supply shocks, they may herald economic recessions and should be associated with higher risk premiums. In contrast, demand-driven oil price increases result from strong economic growth and therefore signal relatively lower risk premiums.”

Extreme fluctuations in oil prices…hide their predictive ability. To circumvent this issue, we construct an oil trend factor that synthesizes information from short- and long-term price signals…Two main arguments explain why the oil trend factor predicts bond returns and economic fundamentals better than oil prices do…

  • First, oil prices provide very noisy information about future economic fundamentals. The reason is that oil price movements can be attributed to several sources, including production cost [and] political risk… Moreover, some researchers and policymakers argue that the financialization of the oil market cultivates speculative behavior, which may inject excessive volatility, or simply noise, into oil prices.
  • Second, oil prices affect almost every aspect of the economy, from the cost of electricity and heating to the cost of transportation and production. Although daily oil price changes do not cause inflation concern, steadily increasing oil prices typically signal impending inflation.

On the empirical evidence

“We consider six oil price series, including four oil spot prices (Arab Light, Dubai Fateh, WTI Cushing, and WTI Midland) and two series of futures (NYMEX oil futures and ICE Brent futures) from the Bloomberg database. The oil trend factor is defined as the monthly change in the 6-month moving average of the daily prices.”

“We provide evidence that the oil trend factor is demand-driven following three steps. First, the oil trend factor is positively related to empirical proxies for the oil demand and hardly has any significant relationship with the oil supply. Second, it has significant positive relationships with future economic output variables, such as the industrial production and the real GDP, and a significant negative relationship with future unemployment. Finally, it is driven by aggregate demand as confirmed by the strongly countercyclical dynamics of its prediction of the bond risk premium.

“We conduct an in-sample analysis of the ability of the oil trend factor to predict future excess bond returns. We find substantial predictability for the 20 international bond markets that we consider…The coefficients associated with the oil trend factor are highly significant and negative for most of the developed and emerging countries…For instance, a worldwide oil price shock of one standard deviation (3.83%) predicts a 1.87% decrease in the U.S. annualized excess bond return. The oil trend factor alone impressively explains 10% of the one-month-ahead variation in excess bond returns on average across countries.”

“The current yield curve is known to contain most information useful for predicting future bond returns…We find that the oil trend factor contains substantial information about future bond returns that is not already captured by [yield curve factors].”

“We also conduct an out-of-sample analysis to further assess the robustness of the predictive ability of the oil trend factor… We only rely on information available in the month of forecast formation and use an expanding estimation window…We re-estimate the model every month after updating it with the last month of observations until the end of the period…The out-of-sample ability to predict excess bond returns is strong, statistically significant, and stable over time and across countries. In 11 of the 15 developed countries, the oil trend factor significantly predicts the bond risk premium, with the out-of-sample coefficient of determination (R-squared) ranging from 1.52% to 25.68%.”


Related articles