How does the model compare to reality, when we look at historic day-ahead prices?

We run the model over 2021 & 2022 and compare our half-hourly price to N2EX day-ahead hourly prices. We replace some of the inputs to the model with historic data to get an assessment of 'the model' (as opposed to the inputs).

Backtesting fundamentals model: price

We compare a data point every half hour (our forecast) to a data point every hour (the N2EX day-ahead price). We take that hourly data point for and make two per hour so we can compare every forecasted data point.

Using historical data for some of the inputs

We normalize the model as much as possible - so give it real historic data and see what the model generates. This allows us to compare our model to historic data and hope to achieve something vaguely close.

  • Using daily SAP gas price (this means the shapes track)
  • Actual wind data
  • Plant availability (via MEL data) for CCGTs and OCGTs
  • We use historical demand data

For 2021:

For 2022:

  • We underpredict absolute prices by an average of £30/MWh and intraday spreads by a similar amount.
  • Both 2021 and 2022 saw far more intraday volatility than our model predicts. There are a few reasons for this:
    • The Russian invasion of Ukraine - and linked record high gas prices - added a significant risk premium to market operators.
    • The gas and electricity markets were in contango (future markets were worth more than spot markets) - and peak power especially was over-valued.
    • In periods of tightness, particularly over £500/MWh, a scarcity premium emerges in power markets.
    • Considering these factors, a huge risk premium was priced into the market: this is not captured in our fundamentals model.
  • We don't model CCGT ramp rates in our fundamentals model. This also explains some of the shortfalls in the intraday price volatility.