Predicting output well is only half the value — the other half is bidding it correctly. This POC shows how bidding on a diversified, cross-site portfolio basis (instead of site-by-site) captures revenue that siloed OEM/site bidding leaves on the table, on the same 6-site, 3-OEM portfolio as the previous POC.
$0
Estimated annual revenue uplift from portfolio-optimized bidding vs. current (naive) practice
vs. naive point-forecast bidding–
vs. siloed per-site bidding–
forecast-error diversification–
01 · Why bidding is not just forecasting
NEWSVENDOR-OPTIMAL BID FRACTILE
Balancing markets settle asymmetrically: unsold surplus is bought back at a discount, but shortfall is bought back at a premium. That asymmetry means the revenue-maximizing bid is not the point forecast — it's a specific quantile of the output distribution.
Surplus sold back at a discount to day-ahead–
Shortfall bought back at a premium to day-ahead–
Given this market's settlement structure, the revenue-optimal bid sits at the – percentile of the forecast distribution — i.e. it should be bid below the point forecast, trading a little expected volume for materially lower shortfall-penalty risk.
02 · The portfolio effect
FORECAST-ERROR STD BY SITE VS. PORTFOLIO AGGREGATE
Each site's forecast error is only partly correlated with the others — different regions, different OEM forecast models. Pooled across the fleet, errors partially cancel: the portfolio's relative uncertainty is well below the sum of the individual sites'.
Three approaches to the same portfolio: bid the point forecast (naive), let each site hedge on its own uncertainty (siloed), or hedge once on the portfolio's true, lower aggregate uncertainty (portfolio-optimized).