In our most recent update to Signal, we’ve made two key changes to our methodology. The first is the inclusion of the Balancing Mechanism into the future battery revenue stack (read more here). The second (and the subject of this article) is a revision to our wholesale methodology.
The key difference in our new Signal forecast methodology is our newly developed hourly forward curve. Before getting into the details of how we build this, check out our explainer on the forward market for electricity if you’re not already familiar with this topic.
Why are we building an hourly curve?
The value of battery energy storage is fundamentally driven by exploiting short-duration spreads in the price of electricity. This means charging when prices are low and selling when they are high, often within the same 24h period.
The average duration of the GB battery fleet is around 1.2h, meaning batteries charge and discharge within hours, not days. So, for Signal to more accurately forecast battery energy storage revenues, we need to understand how prices change on an hourly basis.
Unfortunately, the forward curves available in the market are specified in terms of month-long contracts. These aren’t suitable for our Signal forecast. As such, we must build an hourly curve, consistent with the monthly data available on the forward market. This is how we do it...
Building an hourly forward power curve
1. Calculate off-peak futures prices
Two futures products are available to trade on Intercontinental Exchange (ICE): baseload and peakload. Since the delivery hours of these contracts overlap, we can derive the future value of an off-peak product, even though it is not directly traded on the exchange.
Off-peak prices are calculated as the weighted difference between baseload and peakload prices, where the weighting is determined by the number of peak-load hours each month. Figure 1 (below) shows the base and peakload futures curves taken from the market and derived off-peak prices.
2. Resampling and smoothing
The futures curves shown in figure 1 (above) are then resampled to hourly granularity. This, however, results in a problem: discontinuous changes in prices between successive months (see dashed series in figure 2 below).
To rectify this, we apply a smoothing algorithm to the resampled curves to avoid large jumps at the beginning and end of each month. Importantly, this must preserve the average value of the curve for each month to ensure the new, smoothed forward curves are consistent with the original, market forward curves. The results of this process are shown in figure 2 (below).
3. Characterizing daily price shape
Next, we need to add a daily shape to the smoothed curves. To do this, we characterize daily price profiles via ‘hourly price scalars’. Hourly price scalars are defined as the ratio of historical day-ahead prices to the monthly peak or off-peak prices. Median scalars are then used to characterize an indicative price profile across a 24h period and normalized to ensure they average to 1 across the day (more on this later).
This process is repeated across different types of days -namely summer and winter, weekdays and weekends. Figure 3 (below) shows the resulting daily price profiles for the different types of days modeled. Prices (and resultant scalars) are taken from the Nordpool day-ahead hourly auctions.
4. Adding daily shape to the curve
To apply historical daily profiles, we multiply the daily price profiles (shown above in figure 3) by the smoothed hourly forward curves (shown above in figure 2). Since the daily profiles have an average value of one, the resultant hourly curve (see Figure 4 below) is consistent with market futures curves.
5. Including extreme prices
The above curve (figure 4) assumes a consistent daily price shape for all similar days (e.g., winter weekdays). In reality, however, certain system conditions see prices move away from their average behavior resulting in very high or very low prices. These are the days when batteries stand to make the most money - therefore, we model them separately.
Using a historical view of wind generation, residual demand, available dispatchable generation, and the resulting system margin, we train a machine learning model to predict the system conditions that result in extreme pricing events. Then, using forecasted values for demand, renewable generation, and available capacity, we identify when these extreme prices are most likely to occur. The days where we predict extreme events are shown in Figure 5 (below), with green indicating high prices and red indicating low.
For future days identified as ‘extreme pricing days’ by our model, we apply a daily profile shape derived from historical extreme price days instead of using the average profiles from figure 3 (above).
6. Et voilà!
Bringing this together, we arrive at the hourly forward price curve, shown in figure 6 (below).
- The above curve shows forward electricity prices at an hourly granularity.
- Each day exhibits a price profile in line with historically observed prices at a given time of year.
- Extreme prices are included in line with our future expectations of system conditions.
- The curve is consistent with the forward market - averaging across all hours in a given month gives the baseload price we see in the forward market. The same is also true for peak and off-peak hours.
From this hourly curve, we calculate and include in our Signal forecast a forward view of battery revenues by modeling optimal dispatch against varying wholesale prices. These resulting revenues represent the current forward market value of battery energy storage.