Agricultural Futures Time Series Analysis
ARIMA, GARCH, and multivariate GARCH modeling of daily CBOT wheat, corn, and soybean futures (2000–2024) to characterize volatility clustering and cross-commodity correlation.
TECH STACK
Deep Dive
Time series analysis of daily wheat, corn, and soybean futures prices from the Chicago Board of Trade, January 2000 through December 2024 (~6,300 trading days). Spectral analysis and STL decomposition confirmed none of the three series exhibit significant seasonality, consistent with futures prices tracking macroeconomic conditions rather than production cycles.
Fit per-commodity ARIMA models on price differences, then ARFIMA+GARCH(1,1) models with skewed Student-t errors on daily returns to capture volatility persistence. A multivariate DCC-GARCH model quantified time-varying correlation across all three commodities (wheat–corn strongest). Extended each ARIMA model with ARIMAX terms using the most correlated peer commodity as an exogenous regressor.
Cross-validated every model class against actual Oct–Dec 2024 prices before drawing conclusions: the ARIMA models' residuals failed independence/normality checks, and the GARCH forecasts produced negative out-of-sample R² for all three commodities. Rather than force a forecast from an unreliable model, the project's conclusion was that univariate/multivariate GARCH alone isn't sufficient here — a useful negative result on model selection discipline.