Scarcity and the Missing Money

Somewhere in every large power system there is a generator that runs for perhaps forty hours a year. It is old, it is inefficient, it starts quickly, and for the other eight thousand seven hundred hours it earns nothing. It exists for the evening when a cold front stalls, the wind drops, the sun has set, and every other machine on the system is already running flat out. That machine is a peaking plant, or a peaker: usually a simple cycle gas turbine (similar to an aircraft jet engine) of the kind in Chapter 4, built to start in minutes and to sit idle the rest of the year.

The question this chapter is about sounds like an accounting detail and turns out to be the deepest argument in electricity markets. Who pays for that plant, and does an energy-only market, one that pays a generator for the electricity it produces and for nothing else, pay enough to make anyone build the next one?

Why the money goes missing

Under the clearing rule from Chapter 8, a generator earns the difference between the market price and its own marginal cost. For the peaking plant that difference is zero in almost every hour, because it is the most expensive machine on the system and is therefore either not running at all or running as the marginal unit at its own cost. Its entire return has to come from a few hours a year when the price rises far above the cost of the most expensive thing running.

In theory that works. Peak-load pricing says a plant needed rarely should recover its fixed costs through rare and very high prices, and the arithmetic does add up. In practice something is usually in the way. High enough prices to pay for peaker plants are often not available.Offer caps, the ceiling a grid operator sets on what a generator may bid, exist because a market with genuinely unbounded prices during a shortage invites both abuse and ruin. Grid operators also intervene before prices reach extremes, dispatching reserves out of merit to keep the lights on and suppressing the price as a side effect.

A second problem is nobody finances a power plant against revenue that arrives in a handful of hours which might not occur at all in a mild decade. That is a shortage of long-term contracts rather than a shortage of revenue.

So, the price level often is limited (missing money to peaker plants) and there is limited financing available for building a power plant that operates a few hours every few years (a missing market for peaker plants).

Putting a number on the lights going out

If a market is going to price scarcity, somebody has to say what scarcity is worth. That number is the value of lost load, the cost to customers of not being served, and it is the least comfortable figure in this book, because it is an attempt to price an involuntary outage. It is estimated from studies of what interruptions actually cost households and businesses, varying by season, time of day and duration. ERCOT in Texas places a value on this scarcity of 5,000 dollars per megawatt-hour. What actually gets added to the price in any given interval is a fraction of that number, and is called the adder.

The reason the number matters is that everything above it is anchored to it. Texas builds this in explicitly through an operating reserve demand curve, the most thoroughgoing attempt anywhere to make an energy-only market price scarcity on purpose rather than by accident.

The idea is that reserves have value even when they are not used, because holding them is what stops a shortfall becoming an outage. So the grid operator computes, from the reserve margin left in real time, the probability that load will have to be curtailed, multiplies that probability by the value of lost load, and adds the result to the energy price. When reserves are plentiful the adder is negligible. As reserves tighten the probability climbs and the adder climbs with it, so the price rises toward the value of lost load before anything has actually failed.

What it looks like when it fires

In February 2021, Winter Storm Uri gave the mechanism the most severe test any electricity market has had, and afterwards everything that could be contested was.

ERCOT prices went to the cap, which at the time was 9,000 dollars per megawatt-hour, and stayed there while a large share of the state was without power. The independent market monitor, Potomac Economics, later put total energy costs for the event at roughly 46 billion dollars. It also concluded that prices should have come off the cap at 11:55pm on 17 February, and that by holding the adder high enough to keep them there for a further 32 hours, through the whole of the 18th and into the 19th, roughly 16 billion dollars of additional charges were created after the conditions justifying them had passed.

The Texas Public Utility Commission declined to reprice, reasoning that unwinding two days of settled trades would do more damage than the error. Litigation ran for three years. A court of appeals held that the commission had exceeded its authority and had eliminated competition by fixing the price at the system-wide offer cap. Then on 14 June 2024, in Public Utility Commission of Texas v. Luminant Energy, the Supreme Court of Texas unanimously reversed that decision and held the orders lawful. The 16 billion dollars stayed.

Scarcity pricing worked, in the sense that it did exactly what it was designed to do, which is transfer very large sums to whoever was generating. Retail customers on floating tariffs discovered that a market design had been quietly delegated to them, and some received bills for a single week larger than a normal year. And the boundary between the market functioning and the grid operator making an administrative error turned out to be a question for a supreme court.

The standard nobody defends and nobody replaces

Most of North America plans to aone-day-in-ten-years standard, generally implemented as aloss-of-load expectation of 0.1 events per year. Planned reliability is therefore not perfect reliability. The system is built on the explicit expectation that it will fail to serve everyone roughly once a decade.

The standard is old, and it was designed for a system whose risk was a summer afternoon and whose generators failed independently of one another. That describes neither a winter storm which freezes nat gas wellheads and power plants at the same time, nor a system where a still evening reduces output across an entire region at once. NERC, the body that writes North American reliability standards, and others are moving toward measures that count hours and unserved energy rather than events, precisely because a metric counting events treats a two-hour shortfall and a four-day one as the same thing.

Why the argument never ends

In Chapter 11, prices fall below zero for more than a thousand hours a year because there is more zero-cost generation than the system can use. In this chapter, the same system may not be paying enough to keep a machine available for the forty hours it desperately needs. Both are becoming more pronounced, because what is being added to these systems produces energy without producing availability.

That is the whole disagreement. One camp holds that the answer is to let scarcity prices go high enough, for long enough, that the peaking plant finances itself, and that every intervention to soften them is what created the missing money in the first place. The other holds that no political system will tolerate prices at those levels once they have been experienced, that Uri is the proof, and that it is more sensible to buy availability explicitly through a separate market than to hope it emerges from a volatile one.

Neither camp has won. Texas and Australia took the first view of letting prices rally to a very high offer cap. PJM and much of Europe took the second of buying availability through a separate market. That second choice is a market design described in Chapter 16.