Breaking down trading in Wholesale Power Markets
Note: All views expressed in this article are my own, and sourced from publicly available material (all of which is linked below)
Power Trading is one of the most interesting and niche areas of financial and commodities trading, and in my opinion, presents some of the most complex applied mathematics problems in the industry. It’s especially relevant today as data-centers, unpredictable climate conditions and growing renewable footprints change the operations of global power grids. But what exactly does it mean to trade power? The explanation will follow this structure. Feel free to skip ahead.
- A brief history of power markets
- How does the Market Function and Price Power?
- Trading
A Brief history of Power Markets
This section is intended to motivate why power markets exist and why they are structured as a competitive wholesale tradeable market in some parts of the U.S.
Vertical Integration
Up to the 1990s, power markets were structured according to a vertical model. Utility companies handled generation (the burning of fuel to produce electrical energy), transmission (the movement of electrons through large transmission lines), and distribution (the supply of the electrons to the end user via local lines). This is analogous to the Seven Sisters in the oil industry controlling the discovery, extraction, refining, transport and distribution to the retain pumps. This so-called ‘natural monopoly’ existed in the power industry due to the enormous fixed-costs associated with transmission and generation; the competitive development of transmission lines and distribution networks was economically inefficient, so utilities and similar corporations were given monopolistic control of power generation and transmission over a specific region. Regulators compensated the utilities according to a cost-of-service model where they could earn an approved return on their investment and recover operating costs. An obvious flaw of this approach is that utilities/producers may not have much of an incentive to invest in research to improve efficiency in generation.
Deregulation
Advancements in generation technology (smaller and more efficient gas generators like combined-cycle gas turbines) reduced the economies of scale associated with monopolistic generation. Power markets underwent a phase of de-regulation in the late 90s where generation was structured as a competitive market-based process while transmission remained regulated. Under the de-regulated wholesale market structure, producers / utilities compete under an auction framework to supply power to the grid, while a central non-profit operator known as the Independent System Operator / Regional Transmission Operator (ISO/RTO) schedules and maintains the transmission processes. Cleared generators generally receive the market-clearing price at their respective locations, rather than their individual offer prices. The difference between the market price and a generator’s marginal production cost forms their margin, providing an incentive to improve operating efficiency and reduce costs. This creates a strong economic incentive; because all cleared generators earn the market price regardless of how low their own costs are, producers are motivated to increase fuel efficiency and lower operating costs to maximize their margin. A more technical explanation is often given in Power System Economics textbooks (e.g., Kirschen & Štrbac) where it can be shown that the deregulated model theoretically maximizes the total-utility or socal-welfare in the market, but I will skip this argument in this write-up.
Some regions in North America, particularly the Southeast (Georgia, Florida) still operate under the vertically integrated model. However, roughly 2/3 of U.S electricity demand operates under the ISO/RTO model.
How does the Market Work and Price Power?
How does an ISO actually determine the price for power? This process is quite complex so I will present a high-level description with some simplifications. The full in-depth process is described in ISO business practice manuals which can be found on their respective websites.
Locational Marginal Pricing
Before describing the mathematical process behind pricing, an important concept to understand is that at any given time, the price of power varies spatially. Different points on the power grid are priced differently. An interesting reality of any commodity is that prices can vary significantly across different axes. For example, a barrel of North Sea Brent is fundamentally a different commodity from a barrel of Iranian heavy crude, because the chemical properties create different characteristics for refining. Similarly, a MWh of power delivered during mild-temperatures in the New York countryside is fundamentally a different commodity than a MWh of power delivered in Manhattan during a cold-snap. Thus, the ISO runs a market-clearing algorithm that outputs a different price for each node on the power grid rather than an individual price across the board.
How the ISO prices power
Consider the simplified power grid above. We have eight nodes, or pricing points, each of which will have a specific LMP. We have three generators/producers producing at specific nodes, and we have four nodes where there exists some level of demand. A real power grid is of course, vastly more complex and requires a more sophisticated representation. Some basic terminology. Let G be the set of generators / producers in the grid. We will use i to index the nodes, and x to index individual transmission lines connecting the various nodes.
- Weather / Load / Outages
First the ISO must determine an approximate level of demand that production must account for. They may approach this by forecasting demand based on weather, and looking at any planned maintenance or outages that would affect the ability to meet this forecasted demand.
2. Security Constrained Unit Committment (SCUC)
The ISO then runs a mathematical optimization algorithm to determine which generators g ∈ G should be on or off to meet the forecasted demand in a forward looking period. In other words, for each hour of the the next 24 to 48 hours, which set of generators G’ ⊆ G should be activated at a specific time t ∈ T. This is a very difficult problem because there is a tremendous number of potential combinations of generators that could be on in any given hour. Further more, the choice on which generators are on at time t impact which generators can be on at time t + 1 because of ramp-up/ramp-down times, combined with many other constriants. This problem is solved with a multi-period Mixed-Integer Linear Program. An example formulation is shown below, but note that the exact formulation will vary from ISO to ISO, and can be found in the respective business practice manuals.
Let us break down the above optimization formula. The decision variables P ≥ 0, u ∈ {0,1}, v ∈ {0,1}, w ∈ {0,1}, and r ≥ 0 represent generator power output (MW), commitment status, startup status, shutdown status, and spinning reserve capacity (MW), respectively. The optimization minimizes a total offered-cost function that can incorporate energy production costs, no-load costs, startup costs, shutdown costs, and reserve procurement costs, subject to physical and operational constraints (see the BPMs for exact formulations). Conceptually, the optimization produces time-indexed schedules for generator commitment, startup and shutdown decisions, energy dispatch, and reserve procurement, subject to network, reliability, and generator operating constraints. This optimization is solved with respect to several constraints that ensure physical feasibility, grid stability etc. For example, one of the most important requirements is that supply and demand are balanced. Generators also have physical constraints (it may not possible for a gas-turbine generator to go from 0 MWh in one hour and immediately ramp to a certain number of MWh in the next hour). Transmission lines can only carry a certain amount of energy before failing. Some illustrative constraints are written below.
The list of constraints above is not exhaustive, view BPM for full list. Conceptually, this optimization outputs a matrix of size |G| x |T| which fixes the binary variables on which generators are active, when generators start up, shut down, how much they produce, etc.
3. Security Constrained Economic Dispatch (SCED)
SCUC determines a cost-effective, physically feasible commitment and dispatch schedule across multiple periods. SCED optimizes dispatch subject to operational constraints and applicable unit commitments. For those familiar with mathematical optimization, an interesting property of linear programming is that the dual variables associated with constraints can be interpreted as marginal values. In an appropriately formulated economic dispatch LP, the marginal cost of serving an additional MW of demand at a particular node corresponds to its Locational Marginal Price (LMP). Because SCUC involves binary commitment decisions and a nonconvex feasible region, its dual interpretation is more complicated; nodal marginal prices are generally obtained from an associated pricing or dispatch optimization rather than directly from the mixed-integer formulation. An example SCED formulation is written below.
4. Locational Marginal Pricing
LMP = System wide cost of energy + congestion component + loss
The ohmic loss component is easiest to understand. As electricity is transmitted on a transmission line, some amount of energy is lost via heat due to electrical resistance. The system wide cost of energy can be conceptually thought of as the cost of power if there was unlimited transmission capacity between any two points. Mathematically, it is the dual of the energy balance constraint of the SCED linear program. However, transmission capacity is absolutely not unlimited and the physical constraints associated with the topology of the grid creates severe price separation between points on the grid. For a power trader, this is the most interesting component of the LMP.
Get Prasun Banerjee’s stories in your inbox
Join Medium for free to get updates from this writer.
5. Market Payoffs
The cost coefficients in the optimization formulation above are among its most important components. These coefficients are derived from offer curves submitted by generators to the ISO. For example, consider a generator capable of producing 250 MW of power. It might submit a piecewise offer curve specifying that it is willing to produce the first 100 MW at $30/MWh, the next 100 MW at $50/MWh, and the final 50 MW at $100/MWh.
Whether these offers clear depends on the market-clearing optimization, including prevailing demand and transmission constraints. Suppose the LMP at the generator’s node clears at $60/MWh, and the generator is dispatched to produce 150 MW for one hour. Its energy-market revenue would be 150 × 60 = $9,000. If we assume that the submitted offer curve accurately represents the generator’s marginal production costs, the total variable cost would be 100(30) + 50(50) = $5,500, resulting in a gross operating margin of $3,500, before accounting for startup costs, fixed costs, and other settlement adjustments.
6. The Real-Time Market
The processes described above are central to the Day-Ahead Market (DAM), in which the ISO clears an hourly schedule for the following operating day. As actual conditions evolve, the ISO continuously reassesses system requirements and dispatches generation through the Real-Time Market (RTM). In PJM, real-time LMPs are calculated at five-minute intervals. The difference between the day-ahead and real-time LMP at a given location and delivery interval is commonly referred to as the Day-Ahead/Real-Time (DART) spread.
Trading
So where do financial speculators / traders fit into this landscape of wholesale power markets?
Virtual Trading
Virtual trading is a involves taking a paper position in the Day-Ahead market which the ISO financially settles in the Real-Time Market. In other words, the financial trader speculates on the difference in the Day-Ahead/Real-Time spread. There are two types of virtual trades; INC bids (going short DART or virtual supplying bids), and DEC bids (going long DART or virtual demand bids). Virtual trades are submitted directly into the ISOs market optimization algorithm described above. Suppose a trader submits a 10 MW DEC at $50/MWh for one hour. The day-ahead LMP clears at $40/MWh, so the DEC clears. Real-time LMP subsequently averages $70/MWh for the corresponding settlement period. There are also other types of virtual contracts available, depending on the specific ISO.
PnL = 10(70–40) = $300
Derivatives Trading
Various derivatives trade on the Intercontinental Exchance (ICE) such as futures, options as well as various OTC swap contracts. For example, ICE lists a PJM Western Hub Day-Ahead Peak Fixed Price futures contract. This contract is a cash-settled future that pays out the weighted mean of price of power during hours 8–23, for each trading day (excluding weekends and NERC holidays) of the contract month. Western Hub means that the price being averaged day-to-day is the average of the LMPs across all the nodes within the Western Hub of the PJM ISO.
Financial Transmission Rights
We previously discussed the congestion component of the LMP creating tremendous price separation between different nodes in a power-grid. ISOs hold an auction where they distribute a financial contract known as an FTR (Financial Transmission Right). These FTRs function either as options or obligations, which exposes the holder to the magnitude of electrical congestion along a certain path. FTRs are financial instruments that settle against the difference in the day-ahead congestion components of LMP between specified source and sink locations. An FTR obligation can generate either positive or negative settlement value depending on the direction of congestion, while an FTR option generally provides a nonnegative target allocation. Actual payments may also depend on the market’s FTR funding rules .While these contracts were initially created to provide utilities / power producers with a hedge to protect them from congestion, FTRs have become very attractive investments for financial speculators active in the power space, including hedge fund commodity desks and proprietary trading firms. Congestion modeling is particularly difficult because it requires an understanding of the physics behind power flow, mathematical optimization, portfolio management concepts to manage the risk of holding various paths across the grid, etc.
Players in the Space
Brief description of different types of firms that trade power, and why they do so. Note that this list is neither exclusive nor exhaustive. There are firms that may combine multiple different types of the following businesses (a commodity merchant that trades financial power might also own physical assets to produce power), and there are business not listed that also might be interested in financial power trading (like a big tech firm using power futures as a way to hedge against power price volatility to support their AI/datacenter business lines).
- Utilities and Load-Serving Entities: These are firms that are responsible for procuring power and serving the power to customers. As a result, they are exposed to the price of wholesale power and can use financial trading as a way to hedge this exposure.
- Independent Power Producers: These are the producers submitting their bid/offer curves to the ISO to produce power. Their business is built around generation dispatch, asset optimization, power sales, etc. They are also exposed to the price of power and might use derivatives to hedge against a fall in the price of power.
- Hedge Funds / Proprietary Traders / Merchants: Hedge funds / Proprietary traders speculate and take views on power, investing in and trading power for a profit. Recently, many of these formerly financial speculators have been aggressively building out physical footprints and expanding business lines to physical assets as well, effectively functioning more like merchants. Merchants are firms that are involved both with physical and financial trading.
- Investment Banks: Many investment banks are involved with Power / Commodities trading. A trading desk at an Investment Bank might focus on structured transactions, OTC market making (such as with swaps), financing, derivatives trading, etc.
Many of the aforementioned sequences of pricing and trading provide a rich ground for fundamental and quantitative research. This is a lesser-known area of trading that is becoming increasingly important with weather volatility and skyrocketing demand for energy. This article is intended to be a very crude overview of the landscape of power trading. There are many components I did not mention such as ancillary services, mitigations, curtailments, the landscape of renewables, hydropower, batteries, etc. I’ve included links to useful further reading below.
Sources + Further Reading
- https://www.ferc.gov/electric-power-markets
- https://www.ferc.gov/wholesale-electricity-markets-overview-and-guide
- https://learn.pjm.com/pjmfiles/directory/manuals/m11/index.html
- https://www.pjm.com/-/media/DotCom/documents/manuals/m12.pdf
- https://www.pjm.com/markets-and-operations/energy/real-time/historical-bid-data/unit-bid.aspx?p=1
- https://learn.pjm.com/three-priorities/buying-and-selling-energy/lmp.aspx
- https://www.pjm.com/-/media/DotCom/markets-ops/energy/real-time/reserve-shortage-pricing-paper.pdf
- https://learn.pjm.com/three-priorities/buying-and-selling-energy/energy-markets.aspx
- https://www.pjm.com/markets-and-operations/energy
- https://www.pjm.com/-/media/DotCom/documents/manuals/m06.pdf
- https://www.ice.com/products/6590366/PJM
- https://en.wikipedia.org/wiki/Independent_power_producer
- https://en.wikipedia.org/wiki/Electric_utility
