Why Prices Are Locational
Energy plus congestion plus losses, and why the same megawatt-hour is worth different amounts a few miles apart.
Words the industry uses are marked like this at the sentence that defines them, so you can tell a term you will hear on a desk from a phrase this book happens to be using.
Two substations forty miles apart, connected by wires, in the same market, under the same grid operator, at the same instant. One clears at 32 dollars a megawatt-hour. The other clears at 480. Nothing has broken, nobody is being cheated, and both numbers are correct.
A substation is where high-voltage lines hand power down to the local network, so it is a point at which the system meets real demand. This is the property that separates electricity from every other commodity in this series, and it follows from Chapter 3. A market that treated the grid as a single point, where any generator could serve any load, would be assuming what engineers call a copper plate. In electrical engineering and power markets, a copper plate is a theoretical assumption that a grid has infinite transmission capacity and effectively zero electrical resistance. The model treats the entire geographic transmission network as a single solid piece of highly conductive material.
In reality, the grid is a finite network with lines that carry a limited amount of power, and when one of those limits is reached, the cheap generation on one side of it stops being available to the demand on the other.
Three components
The price at a location, its locational marginal price, is the cost of serving one more megawatt-hour of demand at that specific point, given every constraint the system is currently respecting. It decomposes into three parts.
Table 10-1: What a locational price is made of
| Component | What it reflects | Varies by location? |
|---|---|---|
| Energy | The system-wide marginal cost of production, the number Chapter 8 described | No. One value for the whole system |
| Congestion | The cost of transmission limits forcing a more expensive dispatch | Yes, and it is the large term |
| Losses | Energy dissipated as heat getting power to this point | Yes, but usually small |
Losses are intuitive. Moving power along a conductor wastes some of it, so serving a megawatt-hour far from generation requires slightly more than a megawatt-hour to be produced, and distant locations price a little higher for that reason alone.
Congestion is where the interesting behaviour is. When a transmission limit is reached, the dispatch algorithm can no longer choose the cheapest generator. It has to back down cheap generation on the constrained side and start something more expensive on the other side. The extra cost of that substitution, per additional megawatt of demand at a given point, is the congestion component of that point's price.
The shadow price, and why a constraint has a value
Every limiting constraint in the dispatch has a shadow price: the amount the total cost of serving the system would fall if that limit were relaxed by one megawatt. A line running at its limit through a tight evening might have a shadow price of several hundred dollars, meaning one more megawatt of capacity on that specific line would save the system that much in that interval.
Shadow prices are the mechanism by which a physical limit becomes a financial quantity, and they produce results that look wrong until the cause is clear.
A congestion component can be negative, so that a node prices below the system energy component, which happens where adding demand relieves a constraint rather than worsening it. A price can exceed the offer of every generator running, because serving another megawatt at a constrained location may require redispatching, meaning moving several machines up and down at once. And prices at two nodes on the same short line can differ enormously while nothing physically distinguishes the locations except which side of a limit they sit on.
Illustrative values for one interval, in dollars per megawatt-hour, chosen to show the range rather than to represent a particular market. The total is printed on each bar because the components stack across zero and are hard to add by eye.
Nodes, zones and hubs
A nodal market computes that whole price, energy plus congestion plus losses, at every point where the network model has a bus. A bus is an electrical junction, and the bar is a literal one: a strip of copper or aluminium inside a substation that lines, transformers, generators and loads all bolt onto. A single substation can hold several. The name is short for omnibus, Latin for the benefit of all, which is also where the vehicle gets its name, and it is exactly right for a piece of metal whose job is that everything connects to it. PJM clears pricing on roughly 11,000 of these geographic nodal points, recomputed every five minutes in real time.
Eleven thousand prices can all be settled, but they cannot all be traded individually.
Generators settle at their own nodes. That is where they inject, and where their effect on a constraint is real, so a plant sitting behind a limit should feel the price behind that limit rather than an average that hides it.
Load (demand) usually settles at a zonal average, a load-weighted average of the nodes inside a defined area. The nodal price under a house is perfectly knowable. It is averaged away because a household cannot respond to it, and should not carry the risk of which substation nodes it happens to sit near.
Trading concentrates at hubs, and this is where eleven thousand nodal prices genuinely become unusable. A hub is a fixed basket of nodes, chosen and weighted in advance, that gives a liquid reference point, in the same way that a crude benchmark exists so that cargoes with no market of their own can price against something.
A zone and a hub are both averages of nodes and they are built for opposite purposes. A zone follows a utility's territory and settles the load actually inside it, so its nodal membership is whatever the geography says. Mainland Europe clears about 40 bidding zones for the entire continent. A hub is selected for hedging and trading purposes: PJM's Western Hub is a fixed basket of fewer than a hundred nodal buses, well under one per cent of the network, picked because the price there is stable and liquid rather than because anyone in particular consumes power at those points.
Nodal pricing is how American organised markets run: PJM, MISO, NYISO, ISO-NE, SPP and CAISO all price this way. Texas is a useful case to look at, because it ran the zonal pricing design before moving its market to nodal pricing. Until December 2010 ERCOT priced the entire state through four congestion zones, Houston, North, South and West, which meant a constraint inside a zone was invisible to the price and had to be handled by paying units out of merit and spreading the cost across everybody. It switched to nodal pricing on 1 December 2010, four years later than first scheduled. Chapter 15 takes up that switch.
Congestion rent, and where it goes
When a constraint becomes actively limiting, the grid operator collects more from load than it pays to generation, because load is paying the higher constrained price while generation behind the constraint is receiving the lower one. That surplus is congestion rent, and it can be very large.
The grid operator does not keep it. In a well-formed nodal market the rent fundsfinancial transmission rights, instruments that pay their holder the difference in price between two points. A generator whose output is stuck behind a constraint can buy the right between its node and a hub, and be made whole for the discount it suffers. The congestion rent collected from the physical market is what pays those claims, which is why the two exist as a matched pair. Chapter 24 takes up the instruments.
Those rights are not held only as insurance. The grid operator auctions them in long-term, annual and monthly rounds. They also change hands afterwards, and American trading firms staff a seat for them. A congestion trader's whole book is the difference between two locations, which makes this the one place in the business where the output of an optimisation algorithm is itself the traded product. Chapter 22 describes the seat and Chapter 24 the instrument.
The risk in that congestion trader's book is not small. In June 2018 a firm called GreenHat Energy defaulted on a PJM portfolio of more than 800 million megawatt-hours of these congestion rights, and roughly 179 million dollars of losses fell on the other members, 992 of the 1,054 then in PJM. What undid the position was transmission being built: upgrades changed the congestion patterns the firm had modelled from history. A congestion position is a bet that a network keeps the shape it has, and the whole purpose of a congestion price is to fund the upgrades that change it.
Basis is the risk that ruins contracts
The practical consequence for anyone signing a long-term contract is that the hub price and the node price are different numbers, and their difference due to energy, losses and congestion, is neither stable nor predictable.
As an example of this, say a solar developer signs a power purchase agreement, a long-term contract to sell its output at an agreed price, referencing a hub. The project settles at its own node. If new generation is built nearby, or a line is derated (meaning it cannot serve the load it had previously done), or load patterns shift, the node can drift persistently below the hub. The developer is then delivering energy worth less than the contract assumes and has to pay the difference. That exposure is called basis risk, and it has damaged more renewable projects than construction cost overruns, and it exists because Chapter 3 is true: the wires have limits.