Utilities receiving their first ISO/RTO settlement statement often discover that imbalance energy charges — the cost of deviating from scheduled positions in real time — constitute a larger share of total energy costs than anyone anticipated. Understanding how settlement works is a prerequisite for reducing those charges through better forecasting.
How Locational Marginal Pricing Creates Imbalance Exposure
ISO/RTO markets operate in two timeframes: the day-ahead market, where energy is bought and sold at day-ahead LMPs (Locational Marginal Prices) based on day-ahead load forecasts, and the real-time market, where deviations from day-ahead schedules are settled at real-time LMPs determined by actual system conditions during each 5-minute dispatch interval.
A load-serving entity (LSE) that schedules 400 MW in the day-ahead market and consumes 430 MW in real time must purchase the additional 30 MW at real-time LMP. If the real-time LMP is close to the day-ahead LMP — as it is during most operating hours — the imbalance cost is modest. During scarcity conditions (high demand, constrained transmission, generator outages), real-time LMP can spike 10x–100x the day-ahead price, and the same 30 MW imbalance becomes a significant line item.
The asymmetry is important: forecast errors that result in over-scheduling (purchasing more energy day-ahead than you consume) typically don't produce severe costs, because selling back surplus energy in the real-time market at depressed real-time prices is less damaging than buying shortage energy at high real-time prices. Systematic under-forecasting — consistently purchasing less energy than you consume — generates the worst imbalance settlement exposure.
The 5-Minute Dispatch Interval and Its Settlement Implications
FERC Order 764 (2012) required ISOs and RTOs to implement 15-minute settlement intervals for energy imbalances, replacing the previous 1-hour settlements. Several RTOs subsequently moved to 5-minute settlement to align settlement intervals with dispatch intervals. In 5-minute markets (CAISO, NYISO, ERCOT), every 5-minute period produces a separate settlement calculation based on the deviation from schedule and the 5-minute real-time LMP.
The consequence of 5-minute settlement is that aggregate daily MAPE doesn't capture settlement cost exposure. What matters is error in the 5-minute intervals that coincide with high-LMP periods. A model with 2.5% aggregate MAPE that consistently under-forecasts during afternoon peak hours — the periods with highest real-time LMP volatility — produces higher settlement costs than a model with 3.5% aggregate MAPE that distributes errors uniformly across hours.
This is the argument for evaluating forecasting systems not by aggregate accuracy metrics but by settlement-weighted error: MAPE weighted by hourly average LMP rather than unweighted. A forecasting system that achieves 2% MAPE during off-peak hours and 5% MAPE during peak hours performs worse on settlement-weighted error than one that achieves 3% MAPE uniformly, even if the aggregate MAPE appears similar.
Ancillary Services and the Secondary Settlement Layer
ISO/RTO settlement includes ancillary services costs that are often less visible than energy charges but contribute substantially to total balancing costs. Regulation service (spinning reserves committed to automated frequency regulation) and operating reserves are procured by the ISO/RTO and charged to load-serving entities based on their capacity obligation. For utilities with poor load forecasting, two ancillary service cost drivers are particularly significant:
Regulation capacity obligation: ISOs size their regulation reserve procurement based on system uncertainty, which includes uncertainty contributed by each LSE's load forecast error. LSEs with systematically high forecast errors may face higher allocation of regulation service costs in RTOs that use performance-based allocation mechanisms. In CAISO and MISO, forecast error is explicitly factored into the capacity obligation for spinning and non-spinning reserves.
Uplift charges: When an ISO/RTO commits generators out-of-merit-order to maintain reliability (typically during high-demand or constrained-transmission conditions), the resulting uplift costs are allocated to LSEs. An LSE that contributes to the reliability event through persistent forecast error during high-demand periods may face higher uplift allocation under certain distribution methodologies.
Translating MAPE to Settlement Dollars
The relationship between forecast accuracy improvement and settlement cost reduction is not a fixed ratio — it depends on the LMP volatility profile of the specific ISO/RTO and delivery point. However, benchmark analysis across five Mid-Atlantic and South-Central utilities provides useful reference points:
For a 500 MW peak load utility in PJM with 3% day-ahead MAPE and typical PJM LMP volatility (2022 annual average real-time LMP of $44.80/MWh with 95th percentile of $180/MWh): reducing day-ahead MAPE from 3% to 1.5% reduces expected annual imbalance energy costs by approximately $380,000–$520,000, depending on the correlation between forecast errors and high-LMP periods. Moving from day-ahead forecasting to 15-minute-ahead forecasting that reduces real-time imbalance accumulation contributes an additional $200,000–$400,000 in reduced charges, for a combined annual benefit of $580,000–$920,000 at current LMP levels.
These numbers scale roughly linearly with peak load and with LMP volatility. Utilities in markets with higher LMP volatility (ERCOT during summer scarcity events, NYISO during winter cold snaps) see proportionally higher savings from accuracy improvements.
How Demand Response Changes the Settlement Math
An effective demand-response program changes the settlement optimization problem. Rather than always minimizing deviations from schedule, a utility with dispatchable DR assets can choose to under-schedule slightly in the day-ahead market and fulfill the position with lower-cost DR dispatch in real time when real-time LMPs are high — effectively arbitraging the day-ahead/real-time price spread.
This strategy requires two capabilities: a sufficiently accurate real-time forecast to predict when real-time LMPs will be high enough to justify DR dispatch (rather than purchasing energy in the real-time market), and a dispatch engine that can execute the DR call with sufficient lead time for the curtailment to materialize during the high-LMP period. The 15–30 minute response delay for HVAC-based DR assets means the dispatch decision must be made 20–35 minutes before the anticipated high-LMP period begins — which is only possible with a short-interval load forecast, not with real-time telemetry alone.
Meter Data and Settlement Dispute Resolution
Settlement disputes — disagreements with the ISO/RTO about the measured deviation that generated charges — require detailed interval-level metering data. Utilities that lack automated interval data management systems often find themselves unable to reconstruct the metering basis for disputed settlement calculations, paying charges they might have successfully contested with adequate documentation.
A forecasting system that logs both forecast values and actual SCADA measurements at the interval level provides the audit trail needed for settlement dispute resolution. The log should include: forecast MW by interval, actual SCADA-measured MW by interval, the deviation (actual minus forecast), and the real-time LMP for each interval. With this data, a utility can reconstruct the settlement calculation independently and identify discrepancies with ISO/RTO billing. As we explore in our article on NERC BAL-001 and BAL-002 compliance monitoring, interval-level logging serves dual purposes: operational performance monitoring and regulatory/commercial documentation.