Our latest white paper explains why wholesale electricity prices in the U.S. have become more volatile in recent years. The paper looks at the four macro trends causing greater volatility and explains why they aren’t going away any time soon. But there’s a fifth dynamic at play in ERCOT (Electric Reliability Council of Texas) that requires more explanation.
ERCOT is an energy-only market. There’s no capacity market, which means generators only get paid for the electricity or ancillary services they sell in real-time or short-term markets. They do not get paid to build future capacity or maintain reserve capacity. This market structure, which is different from other U.S. wholesale electricity markets, inherently leads to greater swings in prices.
Let’s look at why that is.
In capacity markets like PJM, CAISO, or MISO, customers pay extra fees to ensure enough generation capacity is available, even if it’s not used. Similarly, those markets require utilities to maintain reserve margins (extra capacity) to handle peak demand. The capacity payments and regulatory mechanisms act as a floor that keeps prices from going below a certain level.
In ERCOT, however, generators aren’t paid to develop new capacity or maintain extra capacity, so those costs don’t get passed on to consumers. As a result, ERCOT’s lowest wholesale prices (often $0/MWh or negative) are extra low in comparison to other markets when demand is stable.
But when it isn’t…
Because there are no guaranteed payments to keep extra capacity available (besides ancillary services which are very low) ERCOT relies on high prices during scarcity to incentivize the development of new generation. The price cap is $5,000/MWh (compared to around $1,000/MWh in other markets). This allows for extreme electricity prices during shortages.
Also, because there are low reserve margins in ERCOT compared to other ISOs, shortages happen more regularly and more easily depending on the season. Demand can surge during a heatwave or cold snap, causing shortages. Or, increasingly, when the sun starts to set and solar generation falls off. Or, a power-plant outage can cause a shortage, if there’s little reserve margin.
Since generators rely entirely on energy-market revenues, they need to make most of their money during high-price events. This puts additional upward pressure on pricing. It also can drive some generators out of the market if prices stay low for too long, which further reduces supply and increases the chance of future price spikes.
In short, ERCOT’s market structure acts as a magnifier for the macro trends that are driving greater volatility across all the wholesale markets in North America.
For more detail on each of these macrotrends and how they impact different market participants, be sure to check out the full white paper.
ERCOT is anything but boring, and it’s getting more interesting with every passing year, because Texas is one of the fastest growing states in terms of both population and economy. The demand for electricity is expected to increase sharply, as data centers and manufacturing facilities continue to locate there to take advantage of business-friendly policies, the relatively low cost of living and the abundance of low-cost solar and wind power.
But with the price volatility in ERCOT, it takes a lot of moxie to participate in the market as a power producer, energy trader and/or utility. Where there’s volatility, there’s opportunity. But the right risk mitigation strategy and tools are imperative, which is why we launched our price forecasting service in ERCOT first. That is where it’s needed the most.
Amperon’s price models are built on historic generation data, weather forecasts and static, time-based features (e.g., hour of the day, holidays, etc.). The data is sourced from public databases, third-party vendors, and proprietary sources, all of which record the nuances of ERCOT data and price behavior.
Generator and grid data: We use publicly available data from the EIA and ERCOT to understand the fleet of generators available and physically model their marginal costs to operate.
Amperon’s Net Demand Forecasts: Amperon’s net demand forecast is an input feature to understand the total demand minus renewable generation.
Historic Prices Data and Behaviors: We use historical data of market prices for natural gas since they are generally the price setter in ERCOT, as well as the historic location marginal prices in ERCOT’s thirteen hubs and zones. The historical price behaviors are quite different in ERCOT than elsewhere because of the price jumps that happen from being in a capacity market.
Weather Data: Amperon uses weather input from multiple vendors with hyper-granular weather points that update hourly. This helps pick up the sudden changes and nuances for a state that commonly says, “If you don’t like the weather, wait five minutes.”
In addition to machine learning algorithms, we also incorporate fundamentals-based dispatch or generation-stack modeling into our price forecasts. This conventional approach considers how power generators meet demand at varying costs. Also known as a merit order stack, this approach helps predict electricity prices by ranking generation resources based on their marginal costs. Generation units with the lowest marginal costs (typically renewables like wind or solar) get dispatched first, followed by more expensive resources (typically natural gas or coal) as demand increases.
The price of electricity in this model is set by the marginal cost of the last unit of generation needed to meet demand. Our unique approach to dispatch modeling incorporates dynamic adjustments of the gen-stack to scale, based on recent history.
Amperon runs all of this data through a set of models designed to infer day-ahead locational marginal prices (LMPs). These forecasts employ a variety of machine learning (ML) and statistical techniques to capture complex data relationships. Each model contributes unique advantages to enhance forecasting accuracy while minimizing the risk of overfitting. At the same time, we evaluate each model’s historical errors and their error correlations. By applying individualized weighting to each model, we dynamically blend them to generate a final price prediction every hour that optimizes accuracy while accounting for a diversity of errors.
Amperon’s unique gen-stack model grounds our price forecasts in the proven fundamentals that reflect the physical attributes of generators on the grid. The simple and transparent methodology is verifiable with straightforward economics concepts, while also taking advantage of the compounded machine learning that goes into our net demand forecasts.
