To effectively operate a merchant energy storage project, you need to forecast energy prices. While storage projects also rely on earning capacity payments and selling ancillary services, to compete, you need to know when to buy low and sell high in the energy markets. This is commonly known as energy arbitrage, and with greater competition it is an increasingly important share of revenue. We walk through approaches to estimating energy arbitrage revenues at different locations in California, and why both location and predictability matter for finding the best opportunities.
Summary Contents
Wholesale energy prices are hard to predict.
First and foremost, prices vary considerably. Unlike in other commodities markets, supply and demand need to be balanced instantaneously. This creates large price swings on a daily and seasonal basis because demand during a hot summer afternoon looks different than the cooler evening later that day, and especially different than a mild day in the fall.
Second, prices vary locationally. In the US, wholesale electricity markets are nodal, meaning every generator and substation has a unique (or multiple unique) pricing point(s). The old real estate adage applies here: location matters. Price points closer to power demand tend to be higher, as meeting that demand requires more grid infrastructure. Besides the cost of building that infrastructure (paid separately from the wholesale prices), the physical elements can be overloaded leading to localized price increases. Think of this like surge pricing for Uber/Lyft.
Both of these features makes energy prices volatile and difficult to predict, however you slice it: within a day (and even individual hours), over the year, and across thousands of locations in a single market.
Nodal value by location, Southern California
Notes: see description of perfect foresight below.
Given the rapid pace of storage deployments, particularly in CAISO and ERCOT, market participants have turned to energy markets to justify long term investment value. The amount of operating storage capacity is expected to exceed the ancillary service requirements of the ISOs. With that increased competition over “finite” ancillary opportunities, storage developers, investors, and asset owners/IPPs are increasingly focused on the revenue opportunities associated with energy arbitrage.
To aid with this, Tyba provides modeling software that helps companies figure out how to predict and execute on energy arbitrage strategies, in conjunction with capacity, ancillary service, and bilateral commitments. We focus on addressing the core modeling and integration challenges related to energy market operations:
For the rest of this blog, we focus on the impact of price forecasts on arbitrage revenue. In future posts, we will investigate the impact of optimization and bidding.
A common starting point to understand the revenue potential for a project is to simulate performance and revenue under a “perfect” foresight case. This represents how the resource would operate if you had advance knowledge of market prices and could bid/clear/operate to capture that. If using historical prices, this is the maximum revenue the resource could have generated if operational. If using forward curves and assuming you believe the price trajectory and volatility of those curves, this is the forward looking maximum revenue potential for the project.
This is a great starting point – but has shortcomings including that having perfect knowledge of prices at all the times is unrealistic (or we’d all be the Princess of Monaco). So what discount should be applied given real world uncertainty?
There are various modeling approaches companies use to determine what percent of the perfect case (PoP) is realistic for the “non-perfect” case. These methods range in complexity from using yesterday’s actual prices to set today’s plan to leveraging data-driven models, such as AI/ML methods or regressions, to forecast prices. Whether trying to defend asset value to financiers, align on the right price to acquire an asset, or determine if a project will meet your return threshold – it is critical to develop a point of view on what is realistic and to support that point of view in a defensible way.
Without further ado, we will dive into the case study next.
Variability of perfect and non-perfect DART revenue
Note: This chart tracks how percent of perfect foresight varies across nodes. Vertical separation represents variability between nodes.
For this case study, we selected a node (Node A) in the SCE region of CAISO, with the goal of answering the following questions.
Case study assumptions
See footnotes, here.