Uncertainty diffusion across commodity markets

ABSTRACT

While numerous studies investigate volatility transmission across commodity markets, particularly oil and agricultural markets, uncertainty diffusion across commodity markets remains absent from the literature. This circumstance is primarily related to the lack of appropriate measures of commodity price uncertainty, which differs from volatility. This study focuses on measuring commodity price uncertainty and how it is transferred from one commodity market to another. Our contributions are twofold. (i) We construct an aggregate predictability-based measure of uncertainty for each group of commodity markets and different maturities, and (ii) we analyse uncertainty diffusion across different commodity markets using a vector autoregressive model. Our findings clearly demonstrate a bi-causal uncertainty transfer between agriculture, energy, and industrial markets, excluding precious metals markets. Additionally, the industrial commodity market is assumed to be the transmission channel of commodity uncertainty spread, given its close link with global economic activity. Notably, we validate the efficacy of using industrial uncertainty as a proxy for macroeconomic uncertainty. Finally, our confirmation of precious metals’ insensitivity to other markets’ shocks reinforces its nature as a safe haven.

KEYWORDS:

• Commodity uncertainty
• vector autoregressive model
• macroeconomic uncertainty

JEL CLASSIFICATION:

• Q02
• C32
• E32

Acknowledgments

We would like to thank Fabien Rondeau, Louise Narbonne, Thibaut Arpinon, Martina Dattilo and Roberto Brunetti for the careful reading of the manuscript. We are also grateful to anonymous reviewers for the insightful and constructive suggestions to greatly improve the quality of the manuscript. Finally, we acknowledge the contributions of participants at international conferences, workshops and seminars that helped divulging the paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Commodity financialization is defined as the increase in the volume of transactions on financial instruments, such as “Futures” contracts on commodity markets. For more details about see Silvennoinen and Thorp (2013), Chari and Christiano (2017).

² The aggregate measure of uncertainty is computed using weights provided by World Bank data.

³ Bloom (2014) and Baker et al. (2014) provide large reviews of literature regarding uncertainty shocks.

⁴ The VIX is a stock market-based option-implied volatility that can be decomposed into two components, a proxy of the risk aversion and expected stock market volatility. VIX represents the option-implied expected volatility of the S&P500 index with a horizon of 30 calendar days (22 trading days). It is an implied or “risk-neutral” volatility, as opposed to actual or physical expected volatility. If we consider a discrete state economy, physical volatility will intuitively use actual state probabilities to arrive at the physical expected volatility, whereas risk-neutral volatility would make use of probabilities adjusted for the pricing of risk.

⁵ Several Studies of uncertainty shocks use either VAR models or Dynamic Stochastic General Equilibrium (DSGE) models (Nicholas et al. 2018; Leduc and Liu 2016).

⁶ Baker, Bloom, and Davis (2016) construct an economic policy-related uncertainty measure based on newspaper analysis and its impact on financial and macroeconomic variables.

⁷ World Bank Commodity Markets: https://www.worldbank.org/en/research/commodity-marketsCommodity-prices-pinksheets. The data are seasonally adjusted using a deterministic log-additive decomposition method with values close to those obtained with seasonal-trend decomposition using LOESS.

⁸ In this study, we use one-month horizon uncertainty. For interested readers, an uncertainty for each market and different forecast horizon $h$ is available upon request.

⁹ ${\epsilon }_0={\epsilon }_1=.\dots .={\epsilon }_{-q+1}=0$. The roots of the characteristic polynomial associated with the coefficients of the MA process, $\psi =({\psi }_1,.,{\psi }_q{)}^{\prime }$ are outside the unit circle, which ensures the stability of the model.

¹⁰ See Chan and Jeliazkov (2009) and Chan and Hsiao (2014) The MATLAB code used to estimate the MASV model is freely available from Joshua Chan’s website. For each variable, we obtain 20,000 loops or draws from the posterior distribution using the Gibbs sampler after a burn-in period of 1000.

¹¹ Markets weights are available on the World Bank Commodity Markets website: https://www.worldbank.org/en/research/commodity-marketsCommodity-pricespinksheets.

¹² GARCH stands for Generalized Autoregressive Conditional Heteroskedasticity model. For each group of commodity markets, we compute a GARCH (1,1) volatility measure of commodity price returns. In turn, the VIX represents implied volatility.

¹³ Table 4 (in appendix A) displays the descriptive statistics of uncertainty variables.

¹⁴ We are thankful towards an anonymous referee also highlighting as an insightful point implications for commodity of cryptocurrency markets.

¹⁵ The lag length criteria retained is two, following standard information criteria. We use different lag specifications to test the robustness of our results. The qualitative results of our study are not affected by the choice of the lag specification.

¹⁶ The test reveals that our uncertainty variables are all stationary at a 5% significance level. Moreover, structural break dates identified by the test are agriculture (08/2012), industry (05/2010), energy (06/2009, 08/2014), precious metals (09/2011).

¹⁷ Despite aluminium and iron have considerable weights, we insist on copper and its prominent role in the industry sector.

¹⁸ The reverse causality of uncertainty diffusion from agricultural to energy markets is confirmed in Figure 4.

¹⁹ Tables on variance decomposition of endogenous variables are available upon request.

²⁰ This includes: real output and income, employment and hours, real retail, manufacturing and trade sales, consumer spending, housing starts, inventory sales ratios, orders and unfilled orders, compensation and labour costs, capacity utilization indices, price indices, bond and stock market indices, and foreign exchange rate measures.

²¹ A brief overview of the technical construction of macroeconomic uncertainty is presented in Appendix A.

²² The implied volatility measure (VIX) is available on Bloomberg or Yahoo Finance website. Macroeconomic uncertainty is free to access on Ludvigson homepage: https://www.sydneyludvigson.com/data-and-appendixesmacroeconomic-uncertainty.

²³ Regarding industry and precious metals historical decomposition graphs, more details are available upon request.