GridCast Webinar #1: How much better have our forecasts become?

The team behind GridCast walks through the rebuilt forecasting engine, five model versions from v1.1 to v1.5, and how we backtest without hindsight.

In this webinar, three of the people behind GridCast explain how we rebuilt the forecasting engine and what each model release has changed for the day-ahead price forecast.

What we cover

  • Why we rebuilt the forecasting engine, and why GridCast is an engine rather than a single model

  • How we backtest without hindsight

  • What changed in each model version, from v1.1 to v1.5

  • Why MAE alone doesn’t tell you what a forecast earns, shown with a German battery case

  • Two questions to ask any forecast vendor

  • What’s next, from v1.6 to probabilistic forecasts

  • Live Q&A

Timestamps

00:00 - Welcome and agenda
03:09 - October 29: one spike, three releases
04:06 - From a paused engine to monthly releases
05:42 - Why we rebuilt the forecasting engine
09:48 - GridCast is an engine, not a model
12:33 - Built to be relied on
14:47 - Every improvement compounds
15:57 - Backtests without hindsight
20:24 - v1.1: residual demand
21:52 - v1.2: cross-border spreads
22:52 - v1.3: the right model for each horizon
25:07 - v1.4: listening to the market
27:18 - v1.5: following the market into the spike
28:29 - Back to October 29
30:04 - MAE doesn’t make money: a German battery case
33:21 - Spikes: still the hardest hours
34:30 - Ask any forecast vendor two questions
35:52 - What’s next: v1.6
38:22 - Probabilistic forecasts and the roadmap
42:07 - Request a backtest
43:16 - Live Q&A

Speakers
Julien Lavalley, Product Lead
Marcus Garsdal, Senior MLOps Engineer
Yunus Ergemen, Senior Data Scientist

Want to see how GridCast’s day-ahead price forecast performs on your market? Request a backtest