Modelling other impacts to intraday pricing

There is more to the difference between Day-Ahead and intraday prices than just forecast errors: human behaviour also impacts prices in the intraday continuous market.

We make several assumptions in building the intraday price model that require mitigation and improve the backtest.

Assumption 1: actual wind and solar generation, demand, and plant outages are known when intraday prices are set

Intraday trades occur over the counter (OTC) on a pay-as-bid basis. Each half-hourly settlement period can involve thousands of individual transaction prices. This makes reporting a single price for each settlement period challenging. The most common approach is to use the volume-weighted average price of all intraday trades for each half-hourly period - often called RPD HH. This is the price we forecast.

The method of using day-ahead forecasts for our day-ahead prices and then outturn forecasts for intraday prices does not directly calculate RPD HH. The resulting price is closer to the intraday closing price (i.e., the price at gate closure) than to RPD HH. In reality, RPD HH includes trades made hours before delivery when there is still uncertainty around renewable output, demand, and outages.

That said, most volume is traded in the last 90 minutes, when renewable and demand forecasts are unlikely to change much.

To capture this additional uncertainty and improve the backtest of observed price differences, we add noise to our intraday forecast. We base the magnitude of this noise on historical differences between RPD HH prices and intraday closing prices, modelling it using a Markov process to reflect the same persistence observed in actual price discrepancies. This approach shifts our forecast from a closing-price estimate to a more realistic RPD HH forecast.

Assumption 2: all trading behaviour is logical and rational

It's not just forecast uncertainty that can skew intraday prices - human behaviour can also play a role. In particular, "herding" is where traders (or algorithms) make decisions based more on the actions or signals of other participants than on their analysis. This can cause amplification of price movements.

Other human behaviours - like mistakes - can also cause fluctuations in prices.

To capture this behaviour, we increase the difference between Day-Ahead and intraday prices by 30%. For example, if we initially calculate the intraday price for a given settlement period to be £10/MWh higher than the Day-Ahead price, this is increased to £13/MWh.