Data: gold/srw_stocks_residual_predictor_analysis.parquet.
Tests variables that may explain the current SRW September stocks residual: June stocks, Jun-Aug exports, current-crop production, futures spreads, export sales, and balance-sheet candidates.
Best-practice screen: rank simple one-variable residual adjustments by leave-one-out RMSE improvement, not just in-sample R². The sample is short, so the table favors variables that reduce historical errors when each year is withheld from fitting.
Current use: add the residual adjustment to the baseline SRW September prediction from gold/srw_stocks_prediction.parquet. Positive adjustment means the candidate expects stocks above the baseline model.
Tests wheat calendar spreads and wheat-corn spreads through the comparable current-date window before the September stocks report.
Leave-one-out residual predictions for the scatter predictor. Values are million bushels.