Electricity Maps

Forecast accuracy is usually measured with MAE or RMSE. For battery storage, we've found those are the wrong metrics. Our team has been quantifying revenue capture for BESS systems lately, asking whether a more accurate forecast (by standard metrics) actually translates to more revenue for the operator. The answer: nMAE barely correlates with revenue captured. A forecast can look great on paper and still leave value on the table. Spearman correlation works better. It measures whether the forecast gets the order of price points right, which is what tells a battery when to charge and discharge to capture spread. In our modeling, it correlates almost perfectly with revenue capture. The lesson generalizes: the right metric for a forecast depends on the use case. It's the one tied to value created, not the one obvious one. Curious if anyone has seen a similar mismatch in other forecast use cases.

FlexUp benchmarked three forecasts on a 1 MW / 2 MWh BESS trading the German day-ahead market over eight months. A persistence baseline using yesterday's prices as today's forecast. Our day-ahead price forecast. And perfect foresight. The persistence baseline earned €41,302 per MW. Our forecast earned €46,162. Perfect foresight would have earned €50,877. That €4,860 difference is what a better forecast is worth in one of Europe's most liquid day-ahead markets, over eight months, under a realistic model with 85% round-trip efficiency and out-of-the-money bid rejection. It also captured 90.7% of the total possible perfect-foresight margin at a 97.1% fill rate. The remaining 9.3% is the headroom that still separates a strong forecast from a flawless, but fictional one. We published the full methodology and results in a joint research report with FlexUp. Link in the comments.

GridCast v1.4 cut German day-ahead price error from 16.3 to 11.9 €/MWh MAE. Request a backtest against the forecast you trade on today.

Germany cleared €585 on 10 September. GridCast 1.5, now live, forecast €545; v1.4 said €272. Backtest it on the peaks you traded.

GridCast v1.4, our latest forecasting platform for day-ahead prices and fundamentals, is live! It's now been upgraded to better capture the timing of spikes. Read more about it here: https://lnkd.in/emMuS4jj

Still choosing your forecast providers based on MAE? BESS revenue capture doesn't correlate with MAE or RMSE. Spearman rank-correlation is a much better measure, and that's how we optimise our forecasts. Request your backtest today.