Commodity futures returns are correlated across many different raw materials and products. Research has identified various types of factors behind this commonality: [i] macroeconomic changes, [ii] financial market trends, and [iii] shifts in general uncertainty. A new paper proposes to estimate the strength and time horizon of these influences through mixed-frequency vector autoregression. Mixed-frequency Granger causality tests can assess the interaction of monthly, weekly, and daily data without aggregating to the lowest common frequency and losing information. An empirical analysis for 37 commodity futures from all major sectors, based on mixed-frequency Granger causality tests, suggests that macroeconomic changes are the dominant common driver of monthly commodity returns, while financial market variables exercise commanding influence at a daily frequency.
This post ties in with this site’s summary of quantitative methods for macro information efficiency.
Commodity future returns and common drivers
“It is well established that prices of various ‘unrelated’ commodities tend to co-move…Existing [academic] literature attributes the presence of commonalities in commodity prices to three major factors…
- Commodity prices react to changes in macroeconomic fundamentals that shift aggregate supply and demand or its expectations. These fundamentals include real economic activity or exchange rates, for instance…
- Co-movement can be attributed to the financialization of commodity markets. With financialization, commodities became new investable asset class in the eyes of financial investors [pointing to] growing importance of financial variables in explaining commodity returns…
- Various uncertainty measures have been found to affect commodity prices. Effects originating from uncertainty affect commodity markets both through fundamental and financial channels.”
“[In empirical analyses explained below] we find that most commodity futures returns are driven mainly by changes in real economic activity on a monthly basis, whereas financial variables affect price movements on a daily level. Many futures returns are also influenced by uncertainty, both in the short- and long-term, depending on the cause of uncertainty.”
A method for identifying common drivers
“The majority of the existing literature on the drivers of commodity prices and their co-movement uses temporally aggregated monthly, quarterly, or even annual returns for empirical analyses as most macroeconomic indicators are available at a monthly or lower frequency. This aggregation results in the loss of valuable information inherent in higher-frequency data and can distort empirical findings.”
“We build our testing framework on the recently proposed mixed-frequency Granger causality test based on mixed-frequency Vector Autoregression (VAR). In contrast to traditional VAR models that can be estimated only from variables that share the same data frequency, mixed-frequency-VARs combine information of multiple time series at distinct sampling frequencies. Hence, when analyzing the nexus of low-frequency and high-frequency processes, data from the high-frequency process does not need to be temporally aggregated to the common lower frequency.”
“Instead of losing valuable information…through temporal aggregation, [the method] proposes to form a ‘stacked’ vector that contains all available observations of both the low and the high-frequency variable…Mixed-frequency Granger causality tests have higher asymptotic power…Temporal aggregation is likely to lead to either spurious causality or spurious non-causality.”
“While most other MIDAS (mixed data sampling) models are used for predicting low-frequency time series with high-frequency information, mixed-frequency VARs also allow to study the opposite direction, e.g., testing for Granger causality from low-frequency macroeconomic indicators to high-frequency commodity futures returns.”
“The null hypothesis [of the causality test] states that the low-frequency variable does not Granger-cause the high-frequency variable at horizon h, if the h-step ahead forecast of the high-frequency variable, based on available information remains the same whether or whether not past information about the low-frequency variable is utilized. Simply put, [the hypothesis to be rejected to detect causality is] that the prediction of the high-frequency variable cannot be improved by looking at past values of the low-frequency variable.”
Empirical analysis and findings
“We collect daily settlement prices of 37 front-month commodity futures traded at various exchanges via Bloomberg from January 1998 to December 2019. Motivated from theory and empirical findings of previous studies, we select 21 fundamental, financial, and uncertainty variables, that potentially affect the futures prices of many commodities. While some are readily available at daily frequency, many of the potential drivers we present below are only available on a monthly basis.”
“Our data set covers a wide range of agricultural and energy commodities as well as industrial and precious metals…[The table below] presents summary statistics for daily log returns.”
“We also construct equally-weighted commodity portfolios and extract a common factor in the returns defined as the first principal component of standardized log returns of all 32 commodities, for which data is available over the full-time period. [The figure below] depicts how the returns of different types of commodities and the common return factor evolve throughout our sample…Considering the full sample, the common factor grasps 20% of total commodity return variation, whereas its correlation is highest with Brent (0.73) and lowest to Feeder Cattle (0.04).”
“Our set of fundamental variables primarily encompasses indicators of global real economic activity. Shocks to real activity indicate changes in the aggregate demand for commodities as real physical assets. Recent work by suggests that aggregate economic activity primarily drives spot price fluctuations in the cross-section of commodities. Popular indicators of global economic activity that we include in our study are the Baltic Dry Index (BDI) as published by the London Baltic Exchange, Global Crude Steel production (STEEL) published by the World Steel Association, World Industrial Production (WIP), and the Global Economic Conditions (GECON).”
“As financial variables we use the S&P 500 (SPX) and the Cboe Volatility Index (VIX), that reflects the market’s expectations of near-term SPX volatility, to capture the stock market’s risk and return…We also include the Investor Sentiment Index (ISENT).”
“The following measures form our set of uncertainty variables…Indices of Macro (MUNC), Financial (FUNC) as well as Real Uncertainty (RUNC)…We also incorporate the daily US Economic Policy Uncertainty (EPU) and the monthly Global Economic Policy Uncertainty (GEPU) indices.”
“[The table below] provides summary statistics for all variables in each category distinguished by their data availability, which is either daily (HF) or monthly (LF).”
“To avoid parameter proliferation, we transform daily futures prices to weekly log-returns, when analyzing effects from monthly-available driver variables. For daily-available driver variables, we use non-aggregated daily commodity returns. Thereby, we split our analysis into high-frequency (daily) and low-frequency (monthly) drivers to discard as few information as possible.”
“For the full sample [the figure below] depicts rejections of the null of non-causality from potential drivers to commodity futures up to the 10% level.”
“In line with expectations, monthly measures of global economic activity, which are most represented by the low-frequency fundamental first principal component, appear to be ‘common’ drivers of a large subset of commodity futures…We can observe the most pronounced Granger causalities for energies, industrial metals, and precious metals except gold…Agricultural commodities are only partly driven by fundamental variables, whereas we find more Granger causalities by looking at single drivers instead of principal components, particularly from GECON to the majority of grains.”
“[Daily] financial variables…Granger-cause daily returns of the majority of softs, energies, industrial metals, precious metals, and soybean-related grains…We find almost no Granger causalities based on low-frequency financial indicators…Our results suggest that financial variables are important drivers of daily commodity returns, presumably reflecting the influence of financial markets on commodity futures markets during their financialization.”
“Results for uncertainty variables depend on the group of commodities and the uncertainty measure. Many agriculturals and industrial metals are Granger-caused by the first principal component of monthly uncertainty indicators.”
“With an out-of-sample trading study, we demonstrate that mixed-frequency models improve the economic value of directional return predictions over traditional models estimated from monthly data only…Our findings emphasize the importance of using mixed-frequency techniques to uncover relationships between monthly-published macroeconomic variables and commodity prices.”
