Causal Greedy Pair Scheduling for Residential Battery Storage under Day-Ahead Electricity Prices

Residential batteries operated against day-ahead electricity prices require schedules that simultaneously respect battery physics and enforce charge-before-discharge causality. Naive price-ranking heuristics can violate this ordering in real time, converting apparently profitable arbitrage plans into partially infeasible execution. This study presents a deterministic causal greedy pair-matching heuristic that eliminates this failure mode by construction and evaluates it on a full-year measured dataset of 2025 Estonian household demand and spot-price data with mixed hourly and 15-minute resolution. After chronology repair, daylight-saving correction, and linear imputation of 165 sparse missing-price rows, the cleaned dataset covers exactly 8760 hours and serves as a common input for all simulations. Under the canonical 24-hour non-overlapping planning horizon with no grid export, a 10 kWh battery reduces the ex-VAT spot-energy bill from €1095.91 to €648.53, saving €447.38 per year (40.82%), while a 30 kWh battery saves €629.91 per year. The marginal value of added capacity declines sharply: the first kilowatt-hour delivers €79.04 per year and the thirtieth only €2.67 per year, confirming strong diminishing returns. Across the four tested planning horizons, the 24-hour window performs best for small-to-mid capacities, while the 48-hour window marginally surpasses it only above 25 kWh. Against an exact matched-constraint mixed-integer linear program benchmark, the heuristic recovers 95.65% of optimal value at 10 kWh and approximately 97.95% at 20–30 kWh, with an annual gap below €21 in the mid-to-large capacity range. A descriptive event-study around the February 2025 Baltic synchronization shows no immediate reduction in intraday price spread or heuristic savings. The results demonstrate that a causally valid, computationally lightweight scheduler can capture the large majority of available arbitrage value on real household data without requiring optimization solvers.