Extended Abstract
The purpose of this study is to determine whether the volatility of major financial markets provides additional predictive value for agricultural commodity volatility forecasts. Despite the growing body of research documenting the increasing financialization of commodity markets and cross-asset volatility spillovers, the agricultural realized volatility forecasting literature has mostly stayed limited to self-persistence specifications. This leaves the potential role of broader financial-market information unexplored, especially during periods of heightened systemic stress.
To fill this gap, we calculate daily realized volatility measures using the iPath Bloomberg Agriculture Subindex (JJA) and five-minute intraday data for corn, soybean, and wheat futures. Additionally, we incorporate cross-market intraday information from 45 assets in seven different sectors (agriculture, livestock, energy, metals, foreign exchange, bonds, and equity indexes) within the Heterogeneous Autoregressive (HAR) framework. The sample covers the period from September 2014 to June 2023, which includes the COVID-19 pandemic and the Russo-Ukrainian War, two significant structural disruptions in global asset markets. To assess both in-sample fit and out-of-sample forecasting performance, we compare single-asset and sectoral HAR-X specifications with regularized HAR-ML models (LASSO, Ridge, and Elastic Net) and eight forecast combination schemes.
Our findings indicate that agricultural commodity volatility is embedded in a broader financial system rather than evolving independently. The baseline HAR model is consistently dominated by in-sample shrinkage-based specifications, suggesting that agricultural volatility dynamics absorb multi-market information that is diffused across assets, requiring dimensionality reduction for efficient extraction. Out-of-sample, regression-based forecast combination schemes perform best overall. Specifically, during the crisis episodes, the least absolute deviations combination emerges as the most robust approach, particularly for macro-financial sectors.
Beyond method choice, different commodities have significantly different sources of predictive content, and such heterogeneity can be economically informative. Forecasts for corn are mainly influenced by equity market volatility, as stock market volatility contains forward-looking information about the business cycle and future demand conditions. Thus, it may help predict corn market developments because corn demand is closely tied to cyclical feed and ethanol use. Due to the export-intensive nature of global soybean trade and its susceptibility to currency swings, foreign exchange volatility predominates in soybean forecasts. The most reliable indicator of wheat and the JJA index is bond market volatility, and it is also important for corn, highlighting the importance of financing costs and cost-of-carry mechanisms in storable commodities. Diebold-Mariano and the model confidence set tests support these findings.
The results may impact both market players and policymakers. For instance, risk and portfolio managers who aim to reduce agricultural exposure should include both macro-financial volatility indicators and commodity-specific persistence measures in forecasting and hedging models. Given the importance of financial market volatility for forecasting performance, organizations responsible for food security should consider macro-financial volatility indicators (especially bond markets) in their commodity price monitoring and early-warning systems, since monetary tightening may amplify agricultural price volatility with further effects on inflation and food prices.
JEL codes: C53; C58; G15; G17; Q14