ERCOT Battery Storage Backtest
Four years of ERCOT day-ahead prices, a Powerwall-class battery, and a perfect-foresight LP versus a simple sort-based heuristic.
stack: Python, scipy HiGHS, pandas, ERCOT public data, Typst
A backtest of a Powerwall-class residential battery across five ERCOT load zones from 2022 through 2025. Compares a perfect-foresight linear program (scipy HiGHS) against a sort-based dispatch heuristic on the day-ahead market.
Latest
The heuristic captures 93 to 97 percent of LP revenue on DAM across all five zones and all four years. The 3 to 7 percent gap is the entire economic value of perfect price forecasting on this market, which is much smaller than I expected. The real edge lives in real-time and ancillary services, not day-ahead. Companion essay at /essays/2026-ercot-battery-backtest.
Status
Shipped. The honest finding is that on ERCOT DAM, a battery operator does not need a forecasting model; they need an execution model.