Simple OLS: Y(stocks) = f(X), where X = prev quarter stocks − exports (non-production quarters) or X = prev stocks + production − exports (production quarter: Sep for wheat, Dec for corn). Highlighted = forward projection.
Data: gold/hrw_stocks_prediction.parquet,
gold/hrs_stocks_prediction.parquet,
gold/srw_stocks_prediction.parquet,
gold/corn_stocks_prediction.parquet,
gold/crop_production.parquet.
Current-year production uses yield prediction models by default
(winter wheat yield predictions for HRW/SRW, spring wheat yield predictions for HRS, corn yield prediction history for corn).
X = prev quarter stocks − exports (non-Sep) or June stocks + production − exports (Sep). ● = predicted (no actual yet). Dashed line = OLS regression on historicals.
Y = residual (actual stocks − predicted stocks from simple OLS above). X = NC + VA winter wheat production, deviation from linear trend. Tests whether southeastern production explains the misses.
Regional buildup = sum of per-region OLS predictions. Total OLS = single regression fitted on summed historical totals.
LOO RMSE (M bu) and R² for simple OLS (Y ~ X), by region and quarter.