Introduction

Underpinning so much of the modern economy, the electrical grid is the backbone of the world’s infrastructure.  The often under-appreciated system will become even more strained over the coming years as population growth and the increased digitization of society demands more environmental resources. Meeting these demands will require a multitude of power generation technologies, but a large part of it is expected to be met with renewable sources as the technology matures and policy makers set carbon reduction targets. The challenge of increased renewable penetration is the inherent volatility and unpredictability of wind and solar conditions even a few hours out. An entire system built on renewable resources faces enormous challenges to provide power instantaneously and reliably. To begin tackling these challenges, we must first understand the grid’s underlying systems of supply, demand, and price formation in domestic power production.

By the Way…

Unless otherwise denoted, all plots and tables shown on this site are created by me in R using publicly available data. Images borrowed from other websites are sourced with a link below them.

Market Structure Overview

The U.S. power grid is subdivided into several independently operated regions, each with its own particular market structures and regulations.  Many regions have found success with market “deregulation,” as state governments have opened up the wholesale power markets to competition between power producers and utility-scale buyers (I quote “deregulated” because the energy market rules are still highly regulated by the state and federal governments). The oldest and most mature regional operator is PJM, which was founded in 2001 and covers much of the demand of the Mid-Atlantic.  More recently, in mid 2011, Texas formed a deregulated market called ERCOT, covering about 90% of the populous state’s electricity demand. I will use ERCOT data to illustrate the price dynamics of domestic power production; the same framework exists across the country’s major grid systems.

Energy Price Formation

One hallmark of such market designs is the Locational Marginal Price, or LMP, which sets price at a specific point, or node, in the region.  These LMPs can vary across the system when transmission capacity is congested, and regional generators need to be incentivized to start up or shut down.  For this analysis I will be using ERCOT’s North Hub LMPs, which is an average of a large basket of nodes and is less prone to extreme congestion.

Another essential component of these markets is transparency and ease of access to data, as they depend on encouraging competition to keep prices efficient.  Therefore, ERCOT publically publishes useful information on prices, supply, and demand each day. Here is an overview of the average LMP for onpeak hours (6AM - 10PM each business day) over time in ERCOT:

ERCOT North Hub Real-Time Daily LMPs, June 2011 - July 2018

The units here, $/MWh, are a standard measure of power price, representing the wholesale cost of 1 megawatt of generation running for 1 hour.  1 MW can power about 1000 homes, so a typical price settlement of $40/MWh means it cost about $1 to run a home for a day (40/1000*24). This is the wholesale cost of energy, and does not include the utilities’ transmission costs. The extreme spikes seen above show how power prices in ERCOT can be highly volatile, sometimes shooting up to 20 times the price of a median day.  This is a highly non-normal distribution, with a very long tail. To better see how the distribution of LMPs has changed over time, we can group each daily price by year in a box-and-whisker plot (or “boxplot” in R terms). The inner box shows the 25%-75% range, with a line for the median, while the whiskers show the 5%-95% range.  Outliers are dots outside the whiskers. I’ve also included the number of points in each box at the bottom (~261 business days per year). The top of the plot is truncated at the overall 95%ile, with another label to indicate how many points are above that level.

ERCOT North Hub Real-Time Daily LMPs by Year

Summary Statistics

Min: 11.21, Median: 27.76, Mean: 34.36, Max: 704.12


ERCOT had a tough time regulating prices when the market was new, with the hot summer of 2011 causing extreme volatility right out of the gate.  Even though the market only existed for two thirds of the year, 2011 had the lion’s share of the points above the 95%ile level. However, as designed, these high prices incentivized new generation build out, and prices stabilized over the next year.

Supply and Demand

This high level of skew between the mean and the median mirrors the highly non-linear cost curve of generating each additional MW of power.  Below is an example of a hypothetical dispatch curve produced by the federal Energy Information Administration as ERCOT was preparing to launch ahead of summer 2011.

Using these submitted bids, the system dispatcher will call on units to start up or shut down as the demand moves up and down the x axis.  Some fuel types will bid very low, like nuclear units because they take days to start up and shut down, or renewables because their marginal costs are nearly $0.  The rest of the stack is made up of a diverse mix of units with unique operating characteristics, typically with a trade off between cost and flexibility (eg slow baseload coal vs fast peaker gas units).  At the top of the stack are highly uneconomical units that should only run in emergencies, such as diesel generators.

The sum of all generator bids is the total available capacity.  However, for a given day, not all units will be available, as some go offline for maintenance, refueling, or upgrades.  ERCOT publishes a daily report detailing the total commitment of all generation, and we can see that availability is highly seasonal.

ERCOT Daily Available Generation Grouped By Month

Summary Statistics:

  Min.     1st Qu. Median  Mean 3rd Qu. Max.

 39,220   47,895 54,461   56,027 64,622 78,323

This pattern reflects the average demand level of the time of year, as units try to go into maintenance during periods they are less likely to be dispatched.

ERCOT Average Supply and Demand By Month


The difference between the available generation and the peak load each day is the surplus capacity, or in other words, how close the demand is away from the top of the stack.  If surplus reaches 0 MW, load has exceeded available capacity, and blackouts occur.

This concludes the introduction to the US power markets. To see how these supply and demand dynamics have been affected by increased wind generation in Texas, continue on to the next article.