The Forecast Was Wrong — But the System Risk Was Bigger | Week 34, 2026

Generation forecasting showed persistent weakness throughout Week 34. The weekly mean absolute error reached 4.69 GW, while Tuesday recorded the worst daily MAE at 6.02 GW. The strongest single forecast miss occurred on Saturday, 22 August at 22:00 UTC, when forecast generation exceeded actual output by 29.29 GW.

The errors were not directionally balanced. Mean signed error was -3.59 GW, showing a systematic over-forecast bias. Actual generation exceeded forecast for only 33.2 hours, compared with 134.8 hours when forecast generation exceeded realised output. All ten of the largest forecast failures were over-forecast events.

Forecast performance also varied significantly by time of day. 18:00 UTC recorded the highest average hourly forecast error at 8.77 GW and the strongest hourly bias at -8.77 GW, identifying the evening period as an important concentration of forecast risk.

Operationally, the week also contained substantial physical-versus-scheduled deviations. Maximum schedule deviation reached 31.56 GW, while the maximum border-balancing gap reached 2.68 GW. However, the signed correlation between forecast error and schedule deviation was only -0.03, and the magnitude correlation was just +0.12.

The EUnix Intelligence Platform consequently classified the story as critical, with forecast quality scoring 94.99/100, schedule deviation 88.78/100, and border balancing 33.48/100. The overall story priority reached 85.58.

Week 34 was not simply a week of occasional forecasting mistakes. The stronger signal was the directional persistence of those mistakes. Forecast generation exceeded realised generation during most of the reporting period, producing a sustained negative signed bias rather than a symmetrical distribution of positive and negative errors.

That distinction matters operationally. Random errors can partly offset each other over time, but persistent directional bias can repeatedly leave market participants and system operators positioned against realised conditions. The -3.59 GW mean signed error therefore deserves as much attention as the headline 29.29 GW maximum miss.

The concentration of error around the evening period adds another dimension. With 18:00 UTC showing the highest average hourly MAE, forecast uncertainty was particularly elevated during a period when system conditions can already be changing rapidly.

At the same time, the schedule-deviation analysis provides an important caution. Although the week contained both major forecast misses and substantial schedule deviations, their interval-level relationship was weak. The data therefore identify concurrent operational risks, not evidence that forecast failures directly caused the observed schedule deviations.

The resulting picture is one of layered uncertainty: imperfect generation forecasts, persistent directional bias, significant schedule deviations and measurable border-balancing exposure occurring within the same operating week.

The supplied Week 34 data do not include balancing prices, imbalance settlement prices, reserve activation revenues or direct trading P&L. A monetary cost or revenue figure therefore cannot be calculated from these charts alone. The commercial relevance instead comes from exposure. Forecast errors increase the risk that scheduled market positions diverge from realised physical output, potentially increasing the need for intraday corrections, balancing actions or portfolio flexibility. The concentration of the largest forecast failures is particularly relevant. The ten largest misses averaged 19.00 GW, and every one was an over-forecast event. For portfolios exposed to forecast-dependent scheduling, such persistent directional misses could be more consequential than isolated symmetric forecast noise. Storage, demand response and diversified generation portfolios may therefore carry option value during periods of elevated forecast uncertainty because they provide mechanisms to adjust physical positions when realised conditions depart from expectations. However, the actual financial value of that flexibility cannot be inferred from Week 34 without corresponding intraday prices, imbalance prices, reserve prices and asset-specific operating constraints.

The first indicator to watch after Week 34 is whether the negative forecast bias persists. Continued over-forecasting would suggest that the forecasting issue is systematic rather than a one-week collection of isolated deviations.

The second is the evening risk window. If forecast MAE continues to concentrate around 18:00 UTC, that period should receive particular attention in forecast updates, intraday position management and flexibility scheduling.

Schedule deviations should also be monitored independently from forecast errors. Week 34 demonstrated that large deviations can coexist with major forecast misses without showing strong statistical co-movement.

A particularly important signal would be simultaneous deterioration in forecast accuracy, schedule adherence and border-balancing exposure. Such convergence would increase the operational relevance of the balancing-risk story considerably.

For flexible portfolios, the practical opportunity lies in being able to respond when forecasts change or realised production diverges from scheduled positions—not in assuming that every forecast miss will automatically translate into balancing activation or revenue.