How to maximize the value of merchant batteries in the spot market?
In a volatile spot market, the option value of a merchant battery is worth more than its intrinsic value for durations up to four hours. But monetizing this value requires asset-backed trading and a risk framework suitable for algorithmic trading.
Some developers of energy storage solutions are planning to offer a significant portion of their new capacity on the spot market to provide additional flexibility. Instead of relying solely on long-term contracts with creditworthy counterparties, some merchant capacity can be deployed profitably now with the ability to enter long-term agreements later.
However, managing merchant energy storage requires elaborate trading activities to monetize its extrinsic value, often exceeding its intrinsic value. Indeed, a storage asset is nothing more than a series of time spread (e.g., buy and sell-back, sell and buy-back) options where their premium may be more valuable in a volatile price environment than their inherent time spreads.
Trading around merchant lithium-ion batteries using only the spot market in California or Texas can significantly increase their value for up to four hours of duration. Beyond four hours, California and Texas do not yet offer enough actual time spread opportunities to justify their cost.
What are the different ways to manage a merchant battery?
A familiar approach to managing merchant energy storage is to fill it when the energy price is low and empty it when the price is high. The strategy is relatively straightforward to implement when the essential sources of uncertainty are generation outages and load variability. However, intermittent renewable energy creates more uncertainty about when to buy and sell, which markets to use between day-ahead and real-time, and what to offer between energy and ancillary services. The following exhibit shows the impact of renewable intermittency on day-ahead and actual residual load (i.e., load less renewable energy).
Therefore, the objective is to make the most of the flexibility embedded in energy storage to monetize the discrepancies created by renewable intermittency. In particular, lithium-ion batteries have favorable operating characteristics (e.g., high round-trip efficiency, fast charging and discharging rates) to capture the associated value comprehensively.
We can think of four activities to monetize merchant batteries in spot markets operated by a Regional Transmission Operator (RTO) in the U.S.:
- Make purchases and sales in the real-time market
- Make purchases and sales in the day-ahead market
- Make speculative bets in the day-ahead market against the real-time market — we call these bets “flip-flops”
- Offer ancillary services
The last three activities involve future commitments that create risk exposure if there is a possibility of not meeting them. We mitigate the risk by requiring the necessary capacity against these commitments (e.g., a speculative day-ahead sale will require the corresponding energy in storage). We describe these activities in more detail for CAISO (California) and ERCOT (Texas) in the next exhibit.
For simplification, we associate the first activity, i.e., real-time only, to the intrinsic value of a storage asset and the other three to its trading value as we must make decisions that carry market risks.
In Europe, they have a continuous intraday market that allows trading hourly, half-hourly, and 15-minute contracts up to 24 (or more) hours before delivery. This market creates even more trading opportunities for merchant batteries as we can sell and buy back the same product multiple times in very uncertain residual load scenarios.
Why California and Texas?
California (i.e., CAISO) and Texas (i.e., ERCOT) both have a significant share of their installed capacity coming from solar and wind power, with good potential to increase this renewable capacity. CAISO and ERCOT have about 25% and 40% of their capacity from intermittent renewables, accounting for about 20% and 30% of their generation.
Both markets use a nodal approach to establish the energy price in day-ahead and real-time. Such an approach accurately represents transmission congestion, thus penalizing sizeable renewable projects in congested areas where supply exceeds demand. For example, suppose renewable generation forces the reduction in thermal generation. It can lead to very low or even negative prices as some thermal power plants may want to avoid a shutdown. Therefore, it is advantageous to have a merchant battery in these areas to capture the value associated with congestion.
For this reason, we have selected three nodes backed by large renewable projects in relatively congested areas to assess the value of merchant batteries, namely:
- In the Mojave Desert, next to a large photovoltaic power plant (“CAISO SP15 Solar”)
- In Texas, near the border with New Mexico, next to a large photovoltaic plant (“ERCOT West Solar”)
- In Texas, near the border with Mexico, next to a large wind plant (“ERCOT South Wind”)
We assessed 50 MW batteries with durations of up to ten hours based on actual prices on these nodes between April 2021 and March 2022.
The following exhibit shows our three cases.
What is the current value of merchant batteries in CAISO and ERCOT?
There are two types of value, the maximum one and the one we can achieve with proper risk controls. The former is relatively easy to determine, while the latter depends on the approach and risk tolerance. In our case, we determine the maximum value through our backtesting model and use our automated asset-backed trading model for dispatching the storage asset. We provide a more detailed description of our approach in the following exhibit.
This article shares the maximum value because it does not depend on the approach used to dispatch the asset, which can vary significantly between solutions. Usually, the performance of a trading model is highly dependent on its ability to capture the outliers. The greater the risk tolerance, the more likely the trading model will capture the outliers, but with a potentially much more volatile P&L. If you contemplate implementing such a trading model, Warren Powell has an excellent framework for sequential decisions that goes beyond the popular Q-learning algorithm in reinforcement learning. And, if you have the choice to start from scratch, you might want to consider Julia as the programming language that has performance like C but with the convenience of Python.
The trading activities make merchant lithium-ion batteries at our selected nodes quite profitable for up to four hours of duration. Beyond four hours, these chemical batteries are currently too expensive. At this time, ERCOT offers more value to merchant batteries than CAISO. The next two exhibits provide the value related to the different ways of monetizing a merchant battery and for different durations.
Finally, we have two observations worth mentioning about capturing the value. First, trading activities reduce battery usage as we use some of its capacity to back-up day-ahead commitments. For instance, regarding ERCOT South Wind, real-time time spread opportunities involve 1.7 cycles per day for a 4-hour battery, while trading activities bring it down to 0.9 cycles per day. We present in the following exhibit the charge and discharge profiles related to real-time only but limited to 1 cycle per day to fulfill the warranty and when we include all trading activities.
Second, as we said before, a good trading model must have the ability to capture the outliers. However, it is interesting to observe that we must anticipate more outliers to realize the intrinsic value than to obtain the extrinsic value. In other words, realizing the full value is not necessarily more complicated from an algorithm perspective than capturing the intrinsic value. The last exhibit illustrates the daily margin distribution for ERCOT South Wind, where we see the outliers associated with the different ways of monetizing a merchant battery.
