Re-price flexible power at every market gate

To revise a market quote, an operator needs a forecast from the latest realized price. The forecast must return the remaining daily curve before the gate closes. A day-ahead schedule cannot incorporate information that arrives during the delivery day.

The case compares a previous-day schedule with rolling eomer forecasts. Each forecast uses price history and returns quantiles without a separate covariate feed.

To measure forecast quality, the backtest scores the same delivery periods at commitment lengths from 24 hours to 30 minutes. It reports mean absolute error, directional accuracy, spike recall, spread capture, and interval coverage.

To translate the signal into value, the case selects the eight cheapest and eight most expensive forecast half-hours for each day. This calculation assumes a price-taking position and excludes asset limits, efficiency losses, and power constraints.

At the 30-minute horizon, eomer records 87% directional accuracy and captures 81% of perfect-timing spread, compared with 69% and 42% for the previous-day baseline. Across 194 delivery days, the case reports SGD 33,000 of additional spread value per MW.

The value estimate describes signal quality rather than a full dispatch backtest. A deployment must add asset constraints, transaction terms, and settlement rules before it can estimate realised asset revenue.