A Cross-Balancing-Authority Demand-Forecast-Error Innovation Predicts Near-Term Net-Load Ramp Stress in MISO

Steep swings in net load — electricity demand net of wind and solar generation — are the operational pressure point of a renewables-heavy grid, because they drive the fast conventional dispatch that keeps the system balanced. Anticipating when the next net-load ramp will be unusually severe is therefore directly useful for scheduling and reserve decisions (subject to real-time caveats). We show that a single, price-free scalar built from one balancing authority's day-ahead demand forecast anticipates near-term net-load ramp stress in a neighboring balancing authority. The predictor, |S1|, is the absolute value of the AR(1-3)-stripped innovation in the PJM day-ahead demand-forecast error, standardized using only past data (prequentially). Hours in the top decile of |S1| are 2.65-2.90x more likely than baseline to precede a top-decile MISO net-load ramp-stress event over the next one to three hours, in each of three annual cohorts (2022, 2023, 2024; matched-null p < 1e-4). The dependence rises monotonically in trend across |S1| deciles (Spearman 0.94-0.98), improves cross-fit calibrated probabilities (positive Brier skill), and survives episode leave-one-out in every cohort. Four ensemble weather-dispersion features (temperature, wind, cloud, solar), tested under the same battery, add no forecasting information beyond |S1|. The result was reproduced by two independent blind implementations, agreeing to within ~2% on every statistic with identical verdicts (R1). Skill is measured against the unconditional base rate only; we do not benchmark |S1| against competing ramp forecasters. We frame the finding as forecasting on archived data and delimit — without claiming — the separate question of real-time tradeability. Code and the reproduction engine are released openly.