Seasonal Trading in Commodity Futures: Evidence from Regression and Singular Spectrum Signals

Battle Count: 350

Highly relevant to quantitative trading as it directly tests whether seasonal signals in commodity futures generate robust out-of-sample trading profits under realistic implementation constraints (transaction costs, liquidity screening, delivery avoidance). The paper provides a transparent empirical framework for comparing seasonality models and demonstrates that signal detection does not equal tradable alpha. Findings are directly actionable for commodity trading desks and systematic strategy developers.

Moderate complexity. DVR is straightforward (calendar dummy regression with Newey-West inference). SSA requires trajectory matrix construction, SVD, diagonal averaging, and recurrent forecasting. RLSSA adds alternating L1 SVD with convergence checks. The MEB bootstrap with 500 paths, paired Sharpe tests, and Holm adjustment adds statistical complexity. Portfolio construction (Top-1 ranking, delivery-avoidance contract selection, transaction costs) is straightforward but requires careful implementation. Volatility normalisation via EWMA adds another layer.

Code and derived monthly panels are available in a public GitHub repository. Complete 500-run artifacts are available from the authors. The underlying market histories are attributed to TradingCharts. Fixed random seed (42) is used for MEB. All parameters (L=36, r=12, lambda=0.97, cost=0.00086) are explicitly stated. However, raw data access depends on the original data provider's conditions.