Module 00 · Origins

Before Markets: The Regulated Century

For seventy years, a utility was a legal monopoly earning a guaranteed return on everything it built — until an accidental experiment, three federal orders, and one spectacular California disaster created the markets we have now.

All you need to bring is one physical fact: the grid stores almost nothing, so electricity must be generated the instant it's consumed. Everything strange about these markets follows from that — and any physics you need along the way gets explained where it appears.

The deal: how a utility made money, 1920–1990

Before there were markets, there was a deal. Samuel Insull — Thomas Edison's former secretary, who built Chicago's electric empire — championed it himself around 1900: give each utility an exclusive franchise territory and an obligation to serve everyone in it, and in exchange let state governments set its prices. Wisconsin and New York created the first modern state utility commissions in 1907, and nearly every state followed. Lawyers call this the regulatory compact.

The economics rested on one idea: electricity was a natural monopoly. Building two competing sets of wires down the same street is pure waste, so one company should own generation, transmission, and distribution together — the vertically integrated utility — and a regulator should stand in for the missing competition.

Here is how the money worked. Every few years the utility filed a rate case with its state public utility commission. The commission added up the utility's operating expenses, then took the value of everything the utility had prudently built — plants, poles, wires, called the rate base — and multiplied it by an allowed rate of return, say 10%. Expenses plus that return equaled the revenue requirement, divided by expected sales to set the price per kilowatt-hour. This is cost-of-service, rate-of-return regulation: profit was a regulated markup on invested capital, not a reward for efficiency.

Notice the incentive: the more you build, the more you earn. Economists Harvey Averch and Leroy Johnson formalized this in 1962 — a systematic bias toward over-building, or "gold-plating." For decades nobody minded, because ever-bigger plants kept driving costs down; the real price of electricity fell for most of the mid-century. Then the 1970s arrived.

The cracks — and PURPA's accidental experiment

The 1970s broke the compact's happy math. Oil shocks sent fuel costs soaring, inflation drove up interest rates, and the industry's giant bet on nuclear power became a graveyard of cost overruns — in 1983 the Washington Public Power Supply System defaulted on $2.25 billion in bonds for abandoned reactors, then the largest municipal bond default in US history. Under cost-of-service rules, customers ate much of this. "Rate shock" entered the vocabulary, and economists asked a heretical question: the wires are a natural monopoly, but why should generation be? A power plant is just a factory. Factories compete.

The proof arrived almost by accident. The Public Utility Regulatory Policies Act (PURPA), signed November 9, 1978, was an energy-conservation law, not a deregulation law. It forced utilities to buy power from qualifying facilities — cogenerators and small renewable producers — at the utility's avoided cost: whatever it would have spent to generate or buy that power itself.

The response stunned everyone. Non-utility generating capacity grew from 17.4 GW in 1979 to 56.8 GW by the end of 1992 — by 1991, qualifying facilities alone had brought roughly 32,000 MW online, nearly triple early predictions. PURPA proved the load-bearing fact of everything that followed: companies that were not utilities could finance, build, and reliably operate power plants. Once demonstrated, the monopoly on generation was intellectually dead.

Opening the wires: EPAct 1992, Orders 888/889, and Order 2000

Independent generators still needed to reach buyers — over transmission lines their competitors owned. Congress moved first: the Energy Policy Act of 1992 created exempt wholesale generators (merchant plants freed from Depression-era holding-company restrictions) and let FERC order a utility to transmit a generator's power — but only case by case. File, litigate, wait. Too slow to build a market on.

So FERC went wholesale. On April 24, 1996 it issued Order No. 888, requiring every utility that owns interstate transmission to file an open access transmission tariff — a standing menu of transmission service offered to all comers on the same rates and terms the utility gives itself. Its companion, Order No. 889, built the plumbing: every transmission provider had to post available capacity and prices on a public electronic system (OASIS), with standards of conduct walling off its own traders from inside information.

But a utility still ran its own dispatch desk, and discrimination was hard to police. FERC's answer, Order No. 2000 (December 20, 1999), encouraged — voluntarily, not by mandate — utilities to hand grid operations to independent Regional Transmission Organizations. That invitation built the RTO map you'll study for the rest of this course.

Say "Restructuring," Not "Deregulation"

These markets are not unregulated — arguably the opposite. Restructuring replaced cost-of-service regulation with a different, heavier apparatus: FERC-approved tariffs thousands of pages long, independent market monitors screening every offer, price caps, and mitigation rules. What changed is what gets regulated: the rules of the auction, rather than each company's allowed profit. "Deregulation" is the word headlines use; almost nothing was deregulated.

California: the crash test

Separate two revolutions people constantly conflate. Wholesale competition — generators competing to sell into the grid — is federal, FERC's domain. Retail choice — you picking your electricity supplier — is a state decision. A state can have either, both, or neither.

California chose both, at maximum speed. AB 1890, signed September 23, 1996 after passing the legislature unanimously, created a wholesale spot market alongside a new ISO, opened retail choice, and — the fatal clause — cut residential rates 10% and froze them while utilities recovered past costs. Regulators pushed the utilities to sell off most of their gas plants and buy power through the spot market, largely unhedged by long-term contracts. Trading began March 31, 1998.

For two years it worked; wholesale power averaged about $30/MWh in 1999. Then in summer 2000 everything tightened at once: drought slashed Pacific Northwest hydro imports, natural gas prices spiked, demand had grown with almost no new in-state plants — and suppliers, Enron most infamously, learned to game the market's flaws by withholding capacity and creating phantom congestion. How much was genuine scarcity versus manipulation is still debated; FERC later found extensive manipulation, but the drought and gas spike were real.

The squeeze was mechanical. Wholesale costs jumped from $7.4 billion in 1999 to $27.1 billion in 2000, while PG&E and Southern California Edison could only charge frozen retail rates — buying high, selling low, they hemorrhaged an estimated $12–14 billion. Localized rolling blackouts had already hit the San Francisco Bay Area on June 14, 2000; the first statewide rolling blackouts came January 17–18, 2001, as the ISO declared a Stage 3 emergency and the governor declared a state of emergency. PG&E filed for Chapter 11 on April 6, 2001. Only in June 2001, when FERC imposed West-wide price mitigation, did prices collapse. Total estimated damage: $40–45 billion.

Field Note · Paid to Relieve Congestion That Never Existed

The most damning documents of the California crisis were written by Enron's own lawyers. In memos dated December 6 and 8, 2000, they described a strategy traders called Death Star: scheduling power flows that made the grid look congested, then collecting payments from California's ISO for "relieving" it — in the memo's own words, getting paid "without actually moving any energy or relieving any congestion." The memos, with sibling schemes Fat Boy, Ricochet (buy price-capped California power, sell it out of state, re-import it uncapped), and Get Shorty, sat in files until FERC forced them public in May 2002. Enron settled with three West Coast states for $1.52 billion in 2005 — by then, mostly a claim against a bankrupt shell. Remember these when you meet the market monitors in Module 06: this is why they exist.

The freeze and the patchwork

The political effect was a nationwide flash-freeze. Around 2000, roughly half the states were implementing or seriously studying restructuring. Afterward, seven — California, Nevada, Arizona, New Mexico, Arkansas, Montana, and Virginia — suspended or repealed their programs. States that had already jumped (Texas opened retail choice in January 2002; Pennsylvania, Illinois, New York, and most of the Northeast) stayed the course. States that hadn't jumped never did.

The Southeast is the clearest holdout: Southern Company, Duke Energy, and TVA had relatively low rates, cooperative regulators, and healthy vertically integrated businesses — no upside in surrendering generation to markets, and after California, no politician would make them. The Southeast still runs largely on the 1935 model: monopoly utilities, cost-of-service rates, power traded bilaterally between neighbors.

The non-California West kept its vertically integrated utilities too, but its trading is no longer purely bilateral: most Western utilities joined CAISO's real-time Western Energy Imbalance Market after 2014, CAISO's Extended Day-Ahead Market went live with PacifiCorp on May 1, 2026, and SPP's rival Markets+ targets a 2027 launch. So today's RTO map is largely a fossil record of a fight that stopped in 2001 — though the West has begun redrawing its corner of it. Seven ISO/RTOs now operate the grid for about two-thirds of US electricity demand. Everything ahead — day-ahead auctions, LMP, capacity markets — happens inside that jagged boundary.

Checkpoint · Module 00

1. Under classic cost-of-service regulation, what was the most reliable way for a utility to grow its profits?

Cut fuel and labor costs below what the regulator projected Add more capital assets — plants, poles, and wires — to its rate base Sell more kilowatt-hours at prices set by supply and demand

Profit was the allowed rate of return multiplied by the rate base, so earnings grew with invested capital, not efficiency. Averch and Johnson showed in 1962 that this creates a systematic bias toward over-building — "gold-plating."

2. What did PURPA (1978) end up proving that made full wholesale competition thinkable?

That companies other than utilities could finance, build, and reliably run power plants That renewable energy was already cheaper than coal and nuclear power That residential customers were eager to switch electricity suppliers

PURPA was a conservation law, but its purchase mandate triggered a boom — non-utility capacity grew from 17.4 GW in 1979 to 56.8 GW by 1992 — demonstrating that generation didn't have to be a monopoly function.

3. What was the core financial mechanism that nearly destroyed PG&E and Southern California Edison in 2000–01?

An absolute shortage of generating capacity meant there was no power to buy at any price Enron's fraud singlehandedly caused the crisis, and the utilities were its direct victims They had to buy power at exploding wholesale spot prices while retail rates stayed frozen by law

AB 1890 froze retail rates while pushing utilities to buy through the spot market largely unhedged; when wholesale costs jumped from $7.4B (1999) to $27.1B (2000), they bought high and sold low. Manipulation amplified the crisis, but the buy-high/sell-low design is what bankrupted PG&E.

Module 01 · The Institutions

The Seven Operators

Seven nonprofit control rooms dispatch two-thirds of America's electricity — meet the operators, the three jobs they all share, the things they never do, and the one state that built a legal moat around its grid.

One job description, three jobs

An ISO (independent system operator) or RTO (regional transmission organization) is a nonprofit referee. It owns no power plants, no wires, no substations — those belong to the utilities and independent generators in its footprint. What it owns is control: the legal authority to operate everyone else's transmission network as a single machine and to decide, minute by minute, whose generation runs. Every one of the seven does three things:

Reliability operations. A 24/7 control room monitors flows, keeps frequency at 60 Hz, and re-dispatches generators every five minutes using security-constrained economic dispatch — running the cheapest available generation first (the "merit order"), while respecting every transmission line's limits.
Market administration. Concretely: in CAISO, by 10 a.m. every day, generators submit offers and utilities submit demand bids for each hour of tomorrow. The ISO's software chews through thousands of nodes and constraints and by about 1 p.m. publishes the schedule: which units run which hours, at what price. A real-time market then trues up every 5 minutes. The ISO also runs settlements — billing and paying every participant.
Transmission planning. RTOs study where the grid will break ten years out and order members to fix it. Real money: MISO's board approved a $21.8 billion portfolio of new long-range transmission in a single vote in December 2024.

What an ISO Is Not

It owns nothing, sells nothing, and profits from nothing. An ISO/RTO doesn't own plants or wires, doesn't build anything (utilities and developers build what the plan calls for), doesn't set the retail rates on your bill (state commissions do), and doesn't sell power to consumers. It is a revenue-neutral counterparty: every dollar collected from buyers is passed to sellers, except a small administrative fee — well under $1/MWh — that funds its operations. When prices spike, the ISO earns nothing extra. Keep this in hand for every module that follows: the ISO runs the auction; it is never a bidder.

Meet the seven

Operator	Footprint	Record peak	Personality
PJM	13 states + DC, ~65M people	165,563 MW (2006)*	The bellwether — oldest (1927 pool), biggest, most litigated
MISO	15 states + Manitoba, ~45M	127,125 MW (2011)	The sprawler — Canada to the Gulf
ERCOT	~90% of Texas, 27M+ customers	85,508 MW (2023)	The island — see below
SPP	Plains; 17 states since Apr 2026	56,184 MW (2023)	The wind machine
CAISO	~80% of California, ~32M	52,061 MW (2022)	The lab — duck curve, 10+ GW of batteries
NYISO	New York, ~20M	33,956 MW (2013)	The specialist — one giant load pocket
ISO-NE	Six NE states, ~15M	28,130 MW (2006)	The worrier — winter gas anxiety

*A record that very likely fell the week this course was written: July 1, 2026 hit 161,859 MW and July 2 ran higher still (~162,700 MW instantaneous despite ~6,000 MW of activated demand response) — PJM says the July 2 load likely surpassed the 2006 record, official figure pending its ~60-day accounting. Data centers are the driver; Module 06 picks this up.

Flavor worth keeping: SPP's instantaneous renewable penetration topped 90% of its load (90.2%, March 29, 2022) — a first for any North American RTO, with wind alone supplying a record 88.5% in the same event. CAISO is the only one whose board is appointed by a governor rather than independent — a fact with big consequences in Module 06. MISO was FERC's first fully approved RTO (December 2001); PJM got provisional status in 2001 and full RTO status in December 2002; SPP followed in 2004, ISO-NE in 2005.

ISO or RTO? Mostly a history lesson

You'll hear both terms, often interchangeably, and honestly: the distinction is more archaeology than substance. "ISO" came first — FERC Order 888 (1996) blessed the independent system operator as a way to guarantee open access: a neutral operator with no financial stake in any generator. CAISO switched on March 31, 1998; NYISO followed in 1999. Then FERC decided voluntary ISOs weren't spreading fast enough and Order 2000 (1999) defined a beefier creature, the RTO, with required characteristics — regional scope, operational authority, planning responsibility.

Today PJM, MISO, SPP, and ISO-NE are formally RTOs; CAISO and NYISO carry the older ISO label; ERCOT is an ISO under Texas law. Do they behave differently? Not meaningfully. All seven run control rooms, administer day-ahead and real-time markets with locational pricing, and plan transmission. "RTO" is the label FERC's paperwork prefers, "ISO" is the label history left behind, and everyone — including FERC — lumps them together as "ISOs/RTOs."

The Texas exception

ERCOT is genuinely different, and the difference is physical before it is legal. The Texas Interconnection is one of North America's three synchronous grids, and it does not synchronously connect to the other two. Its only links are DC ties: two to the Eastern Interconnection totaling 820 MW, plus small ties to Mexico of roughly 400 MW. Against an 85,000+ MW peak, that's a rounding error — when Texas runs short, almost no one can help.

The legal consequence: the Federal Power Act gives FERC jurisdiction over wholesale sales and transmission in interstate commerce. Because ERCOT's power stays inside Texas, its wholesale market is regulated by the Public Utility Commission of Texas, not FERC. But nuance this correctly — ERCOT is not lawless space. FERC and NERC still enforce mandatory reliability standards there (through the Texas Reliability Entity), and the DC ties themselves are FERC-jurisdictional. "Outside FERC" means outside FERC's rate jurisdiction, not outside federal reliability rules.

ERCOT's market personality follows from its independence: the purest energy-only market in the country — no capacity market at all, scarcity prices allowed to scream. The offer cap sat at $9,000/MWh during Winter Storm Uri in February 2021 and was lowered to $5,000/MWh in 2022. Modules 04 and 05 return to both.

Field Note · The Midnight Connection

Texas's grid independence survived an inside job. On the night of May 4, 1976, West Texas Utilities closed a 69-kilovolt switch in Vernon, Texas, and sent power north to Altus, Oklahoma — deliberately placing the entire Texas grid in interstate commerce to drag it under federal jurisdiction. Within hours, other Texas utilities began physically disconnecting rather than be pulled in too. Years of litigation later, the tie was ordered cut and Texas stayed sovereign. Fifty years on, ERCOT still touches the rest of America only through about 1.2 GW of DC converter stations — a legal moat made of power electronics.

The other third — and the West's slow-motion merger

About two-thirds of US load is served inside ISO/RTO regions. Who flies the plane for the rest? The same three jobs get done — just inside vertically integrated utilities. Southern Company, Duke Energy, and TVA each run their own control rooms, dispatch their own fleets, and plan their own wires under state (or federal, for TVA) oversight. Trade between neighbors happens through bilateral contracts: utility A calls utility B and buys 200 MW for tomorrow afternoon at a negotiated price. Since 2021 southeastern utilities have shared an automated 15-minute trading platform called SEEM — a marketplace, but nothing like an RTO.

The West, though, is regionalizing in slow motion — as a competition. CAISO's Western Energy Imbalance Market (launched with PacifiCorp in 2014) lets outside utilities buy into CAISO's 5-minute dispatch; it now spans 22 balancing authorities covering about 80% of Western Interconnection load, with $8.6 billion in cumulative benefits. On May 1, 2026, CAISO and PacifiCorp switched on the Extended Day-Ahead Market (EDAM). SPP is chasing the same customers from the other direction: its Markets+ won FERC approval in January 2025, targets an October 2027 launch, and has landed the Bonneville Power Administration. And on April 1, 2026, SPP extended its full RTO across the seam — the first grid operator spanning two interconnections. Which market map wins the West is the decade's live question; Module 06 finishes the story.

Checkpoint · Module 01

1. Why is ERCOT largely outside FERC's jurisdiction over wholesale power rates?

The Texas legislature passed a law exempting Texas utilities from all federal energy regulation Its grid doesn't synchronously cross state lines, so its power isn't sold in "interstate commerce" under the Federal Power Act ERCOT is a private company rather than a government-chartered organization, putting it beyond FERC's reach

FERC's rate authority covers wholesale sales in interstate commerce, and ERCOT's electrons stay inside Texas (its only external links are asynchronous DC ties). Note the limit: FERC and NERC still enforce mandatory reliability standards in Texas — no state law could opt out of those.

2. Which of these is actually part of an RTO's job?

Owning and maintaining the high-voltage transmission lines in its footprint Setting the retail electricity rates that households pay Running the control room and re-dispatching generators every five minutes

RTOs operate the grid, administer wholesale markets, and plan transmission — but they own nothing and build nothing; the wires belong to member utilities, and retail rates are set by state commissions. Think referee, not player.

3. What best describes the difference between an "ISO" and an "RTO"?

RTOs run wholesale markets while ISOs only handle reliability operations They're largely historical labels from two different FERC orders — day to day, they do the same three jobs ISOs operate within one state while RTOs must span multiple states

"ISO" comes from Order 888 (1996) and "RTO" from the stricter checklist in Order 2000 (1999), but CAISO and NYISO — formally ISOs — run control rooms, markets, and planning exactly like the certified RTOs. The single-state pattern (NYISO, CAISO, ERCOT) is a coincidence, not the definition.

Module 02 · The Auction

A Day in the Day-Ahead Market

By 11:00 a.m., billions of dollars of tomorrow's electricity has been bid; by 1:30 p.m., a room-sized optimization has decided which power plants wake up — before a single electron flows.

First, the layer under the auction

A correction to a misconception this course must not teach: most wholesale power is not bought in the ISO's spot market. Utilities and big buyers hedge the bulk of their needs years or months ahead through bilateral contracts and power purchase agreements (PPAs) — private deals struck at trading hubs, at negotiated prices. What the ISO's markets do is settle everything against a common, transparent price: they are the balancing and pricing layer on top of a mostly-hedged world, with the ISO as a revenue-neutral central counterparty. When you read that "95 percent of energy settles at day-ahead prices," it means positions are scheduled and priced through the day-ahead market — not that everyone shows up unhedged to buy tomorrow's power at auction. Hold that frame; now let's see how the auction itself works.

Why tomorrow's electricity is bought today

The reason a day-ahead market exists is thermodynamic before it is economic. A large steam unit is a building-sized kettle: boiler walls and turbine rotors must be heated slowly and evenly or the metal cracks. From a cold start, a big steam unit needs roughly 12–24 hours to synchronize to the grid. A combined-cycle gas plant needs a few hours; only a simple-cycle combustion turbine can go from off to full output in 10–30 minutes. If the operator waited until demand showed up to decide which plants to run, the cheap, slow units could never participate. Someone has to decide today which machines will be warm tomorrow.

So every US ISO runs the same ritual: each morning, an auction for every hour of the following day — the day-ahead market — alongside a real-time market that trues up differences every five minutes. The pairing is called a two-settlement system, and the key move is that the day-ahead auction is financially binding, not physically binding. Sell 100 MWh for hour 17 and you don't strictly have to generate it — you have to deliver it or buy it back at whatever the real-time price turns out to be. That rule lets the day-ahead market function as a genuine forward market. It works so well that roughly 95% of wholesale energy transactions settle at day-ahead prices — a share that varies somewhat by ISO and how it's measured — leaving real time as the balancing market for the leftovers.

The morning: everything in by 11:00 a.m.

Take PJM, the largest US market, serving 67 million people. Its day-ahead window closes at 11:00 a.m. the day before the operating day — a deadline FERC approved extending from 10:30 a.m. in December 2018, giving generators (gas-fired units especially) an extra half hour to watch morning natural gas prices and firm up fuel costs before locking in offers. (NYISO runs far earlier: bids due at 5:00 a.m.) By the deadline, three groups have filed electronically:

Generators submit a three-part offer per unit: an incremental energy curve ("$28/MWh for my first 200 MW, rising to $45 for my last 50"), a startup cost (the one-time bill for firing up — often tens of thousands of dollars), and a no-load cost (the hourly cost of simply being synchronized and spinning). Attached is the unit's physics: min/max output, ramp rate, minimum run time, minimum downtime.
Load-serving entities — utilities and retailers buying for consumers — submit demand bids: mostly fixed ("buy 5,000 MWh in hour 19 at any price"), some price-sensitive ("only if under $80").
Virtual traders — banks, hedge funds, anyone with credit and a login — submit purely financial positions: an INC sells day-ahead to buy back in real time; a DEC does the reverse. No power plant required. Why allow this? Arbitrage: if day-ahead prices run predictably higher than real-time, INC traders sell into the gap until it closes. Virtual bidding is what keeps day-ahead prices honest forecasts of real-time prices — price convergence. It's not parasitic; it's the market's error-correction mechanism. PJM has allowed it since its two-settlement market launched June 1, 2000.

11:01 a.m.: the machines take over

The moment the window shuts, the ISO's computers confront one of the largest optimization problems solved anywhere on a daily basis, in two stages. First, SCUC — security-constrained unit commitment — answers the yes/no question: which units should be ON in which hours? It's a mixed-integer program: every unit in every hour gets a binary on/off variable, because you cannot run 40 percent of a boiler. SCUC minimizes total cost — energy offers plus startup and no-load costs — across all 24 hours simultaneously, honoring every ramp rate, every line limit, and the N−1 contingency rule: the plan must keep the grid safe even if any single line, transformer, or generator fails without warning.

Second, SCED — security-constrained economic dispatch — takes the on/off decisions as given and answers the continuous question: exactly how many MW from each committed unit? SCED's shadow prices become the locational marginal prices — a distinct hourly price at each of PJM's 10,000-plus pricing nodes.

By about 1:30 p.m., PJM posts the answer: 24 hourly LMPs for every node and a schedule for every unit — all financially binding. Clear 300 MW in hour 15 at $42 and you have sold 300 MWh at $42, period. Produce nothing and you buy 300 MWh back at the real-time price. A generator that clears day-ahead and then trips offline during a $500/MWh scarcity hour learns exactly how binding "financially binding" is.

Pay-as-clear: why the cheapest plant gets the most expensive price

Here's the feature that startles newcomers: everyone who clears is paid the same price — the offer of the marginal unit, the last one needed. A nuclear plant that offered $5/MWh gets $45 if a gas peaker set the clearing price. Isn't that a giveaway?

Economists overwhelmingly say no, and the argument is about incentives. Under uniform pricing, your offer determines only whether you run, not what you're paid — so your best strategy is to offer your true marginal cost. The auction harvests honest cost information, and the merit order it builds is the real one. Under pay-as-bid, that honesty collapses: a $5 nuclear operator would never bid $5 and accept $5. Everyone would bid their guess of the clearing price, converting the market into a forecasting contest. Prices wouldn't fall — and dispatch would scramble whenever a genuinely cheap plant guessed badly and priced itself out.

Try it yourself:

Auction Console · Clear the Day-Ahead Market

DEMAND BID (HOUR 18)550 MW

$—

Clearing Price /MWh

—

Marginal Unit

$—

Total Payment /hr

$—

Wind Farm Profit /hr

Drag demand and watch the marginal unit set the price for everyone. Every cleared unit is paid the clearing price — the wind farm offering $0 profits most when an expensive peaker is on the margin.

What do real numbers look like? In normal conditions, prices mostly land between $20 and $60/MWh — PJM's 2024 load-weighted average was $33.74/MWh, up from $31.08 in 2023. But the tails are wild: during Winter Storm Uri, ERCOT held prices at its then-cap of $9,000/MWh for about four days. And the scale is the final point: PJM's settlement system moved $51.74 billion in 2024 — all of it decided each afternoon, by a computer, for a tomorrow that hasn't happened yet.

Field Note · The Math Upgrade Worth $60 Million a Year

Until 2004, the day-ahead commitment problem was literally too hard to solve; ISOs used an approximation called Lagrangian relaxation — clever, fast, provably not optimal. That year PJM became the first ISO to clear its market with true mixed-integer programming. Per a 2011 FERC staff report, the switch cut production costs by an estimated $60–100 million per year — money saved not by building anything, but by choosing a smarter set of ON/OFF decisions among the same plants. FERC still convenes an annual software conference where researchers compete to shave minutes off solve times, because in this market, a better algorithm is a power plant you never had to build.

Checkpoint · Module 02

1. A hedge fund with no power plants submits an INC offering 100 MW at $38/MWh for hour 17. It clears day-ahead at $41; the real-time price lands at $33. What happens?

The fund is penalized for failing to deliver physical power it promised The fund earns $8/MWh on 100 MWh: it sold at $41 day-ahead and bought back at $33 in real time — no electrons required The trade is voided, because only companies that own generation may sell day-ahead

Day-ahead positions are financial: any shortfall is simply bought back at the real-time price. Profitable arbitrage like this is the point — it pushes day-ahead and real-time prices toward convergence, making day-ahead prices better forecasts.

2. After the 11:00 a.m. close, the ISO runs two giant optimizations. What is the division of labor between SCUC and SCED?

SCUC decides which units will be ON or OFF each hour, weighing startup and no-load costs; SCED then sets each committed unit's output and produces the prices SCUC calculates the locational marginal prices; SCED performs the reliability and contingency checks SCUC schedules transmission maintenance; SCED decides which generators to commit

SCUC is the mixed-integer "which machines turn on" problem — integer because you can't run 40 percent of a boiler — while SCED is the continuous "how many MW from each" problem whose shadow prices become the LMPs. Both enforce transmission limits and N−1 contingencies.

3. A nuclear unit offers energy at $5/MWh and clears in an hour where the marginal gas unit sets the price at $45/MWh. The nuclear unit is paid…

$5/MWh — you receive what you bid $25/MWh — an average of the accepted offers $45/MWh — every cleared seller receives the marginal unit's price

Day-ahead markets are uniform-price (pay-as-clear) auctions: your offer decides whether you run, not what you're paid, so the profit-maximizing strategy is to offer true marginal cost. Pay-as-bid sounds cheaper but would turn offers into guesses at the clearing price — degrading dispatch without lowering prices.

Module 03 · Operations Meets Money

Real Time: The Five-Minute Market

By sunrise the day-ahead plan is already wrong. Every five minutes, software redraws the dispatch and reprices the entire grid — and the gap between plan and reality settles in cold cash.

The plan meets the day

Yesterday afternoon, the day-ahead market produced a beautiful hour-by-hour plan. Then the actual day arrives and starts shredding it. The cold front moves through two hours early, so the heating load ramps at 6 a.m. instead of 8. The solar forecast assumed thin clouds; the clouds are thick, and 1,500 MW of expected output isn't there. At 9:14 a.m., a coal unit that promised 600 MW trips offline on a boiler-tube leak — no warning, no apology.

This is what the real-time market exists to do: it is the residual market that trues up every deviation between what was scheduled and what actually flows. It is not a backup market or a penalty box. It is the only market that dispatches actual electrons. The day-ahead market bought a plan; real time buys reality.

And critically, real time is not run by traders. It is run by the control room. The same software that decides which generators change output — right now, to keep the system balanced — also sets the price. In real time, the market and grid operations are not two things that talk to each other. They are one computation. Hold that thought; it's the key to this module.

Five minutes, four seconds

The heartbeat of real time is SCED. Every five minutes, the ISO's computers ingest the current state of the grid — actual load, wind and solar output, line loadings, unit status — solve the merit-order problem respecting every transmission limit, and emit two things simultaneously: a base point (target output) telegraphed to every generator for the next five minutes, and a fresh set of real-time LMPs at every node. That's 288 dispatch runs, and 288 price sets, every day. The price is the shadow of the dispatch solution.

But load wobbles second-to-second, and five minutes is an eternity for frequency. Between SCED runs, AGC — automatic generation control — takes over: in ERCOT, regulation signals go out every four seconds, nudging designated units up or down to hold 60 Hz. Think of SCED as steering and AGC as the constant micro-corrections on the wheel.

Two modern refinements matter. FERC Order 825 (2016) forced RTOs to settle at the five-minute interval instead of hourly averages — before that, a generator sprinting for a five-minute price spike got paid a blended hourly price, which blunted the incentive to actually follow dispatch. And on December 5, 2025, ERCOT launched RTC+B (Real-Time Co-optimization plus Batteries), its biggest market change since the nodal market launched in 2010: SCED now procures energy and reserves in one optimization every five minutes.

Two settlements, one worked example

Here is the machinery connecting the two markets. Everything you scheduled day-ahead settles at the day-ahead price; every deviation settles at the real-time price.

The buyer. A load-serving entity bought 100 MWh day-ahead at $40/MWh: a $4,000 bill, locked. The day runs hot and its customers actually use 105 MWh. Real-time price this hour: $200/MWh. The LSE pays for the 5 MWh deviation at real time — $1,000 — total $5,000. Nobody repriced the first 100 MWh. The day-ahead purchase hedged 95 percent of the volume — and because the unhedged 5 MWh repriced at $200, that last 5 percent of volume drove 20 percent of the bill. Volatility lives entirely in the deviation.

The seller. A generator sold 200 MWh day-ahead at $40 ($8,000, locked) but a feedwater pump limits it to 190 MWh. It is short 10 MWh against its schedule, so it buys back 10 MWh at the $200 real-time price: minus $2,000, netting $6,000. Underdelivering during a spike stings — exactly as it should.

Now flip the real-time price to $25. The same generator could choose to back down 10 MWh, buy them back for $250, and keep $7,750 — profitable if its own fuel cost exceeds $25/MWh. This is why the day-ahead award is a financial position, not a physical straitjacket. If someone else can produce the energy cheaper in real time, the market lets them — and pays the original seller to step aside.

Volatility, scarcity, and getting made whole

Because real-time prices reprice every five minutes with zero storage slack, they swing like nothing else in commodities. In CAISO's solar-soaked SP15 zone, prices cleared below zero for roughly 1,180 hours in 2024 — about 13% of the year — against a bid floor of −$150/MWh: the duck curve's belly, priced. The same design produces the opposite extreme: on August 17, 2023, ERCOT real-time prices hit about $4,750/MWh at 3:45 p.m. — versus $75 at the same time a day earlier — pressing against the $5,000 cap. Elsewhere, FERC Order 831 caps offers at $1,000/MWh, or up to $2,000 if verified against actual costs.

Those few hundred scarcity hours a year are where peakers and batteries earn their living. A gas peaker running 100 hours a year at $30 prices is a charity; at $4,750 it pays its mortgage. A battery charging through negative midday prices and discharging into the evening ramp runs the same play twice a day.

One loose end: sometimes the control room orders a unit to run for reliability — voltage support, a local constraint — when the LMP is below the unit's offered costs. Forcing a resource to lose money on command would end the market, so the ISO writes a make-whole payment (also called uplift) topping the unit up to its offer, smeared across participants. It's deliberately kept small because it's invisible to the price signal: PJM's energy uplift in 2024 was $269.3 million — about 0.5% of total billing. The rest flowed through transparent market prices — day-ahead and five-minute real-time — computed by the same software family that steers the grid. Dispatch is the market; the market is dispatch.

When it breaks: Uri, from the control room to the kitchen table

Winter Storm Uri (February 2021) is this module's stress test, and two details complete the story most tellings skip.

Who actually turns off your power. When ERCOT ran out of generation, it did not flip neighborhood switches. The ISO declares emergency levels and orders each transmission and distribution utility to shed a quota of megawatts; the utilities' operators physically rotate feeder outages. "The ISO turned off my power" is half true — ERCOT commanded the shed; your local wires company executed it. ISOs direct; utilities touch the equipment.

Who actually got the bill. Most Texans on fixed-rate retail plans never saw wholesale prices — their retailers ate the exposure (several went bankrupt). But customers of Griddy, a retailer that passed raw wholesale prices straight through, lived at the $9,000/MWh cap for days: some received bills around $16,000 for a single week. Griddy filed for bankruptcy within weeks, and Texas banned wholesale-indexed residential plans. It is the clearest lesson in the course on the difference between wholesale and retail: your bill is normally a heavily hedged, regulated abstraction of the spot market — until someone wires you directly to it.

Field Note · The Most Expensive Days in Market History

During Uri, Texas regulators overrode the very system this module describes: with plants freezing offline, the PUCT ordered prices administratively pinned at the $9,000/MWh cap — and held them there roughly four days. The independent market monitor, Potomac Economics, later found prices stayed at the cap about 32 hours longer than necessary after emergency load-shedding ended, overpricing real-time energy by some $16 billion — though because most load was hedged or self-supplied, the monitor put the net settlement impact at roughly $3.2 billion. Regulators refused to reprice; the Texas Supreme Court upheld the orders in June 2024. The five-minute market's superpower — every price instantly becomes real money — is also its blast radius.

Checkpoint · Module 03

1. A retailer bought 100 MWh day-ahead at $40/MWh, but its customers actually used 105 MWh. The real-time price is $200/MWh. What does it pay?

105 MWh × $200 — actual usage repriced at real time 100 MWh × $40 plus 5 MWh × $200 — only the deviation settles at real time 105 MWh × $40 — the day-ahead price applies to actual usage

Under the two-settlement system, the day-ahead purchase is locked at $40 and only the 5 MWh deviation settles at real time ($4,000 + $1,000 = $5,000). This is why a day-ahead position works as a hedge against real-time volatility.

2. A generator sold 200 MWh day-ahead at $50/MWh but only produces 190 MWh. The real-time price is $30/MWh. What happens?

It forfeits its day-ahead revenue and is paid only for 190 MWh at the real-time price It pays a fixed non-performance penalty set by the ISO's tariff It keeps its full day-ahead revenue but buys back the missing 10 MWh at $30

The day-ahead award is a financial position: the generator keeps the $10,000 it locked in and settles the 10 MWh shortfall at real time (−$300), netting $9,700. There's no penalty beyond the real-time price itself — which is punishment enough when prices spike.

3. Where do real-time LMPs actually come from?

The same SCED optimization that sends generators their dispatch instructions every five minutes Continuous bidding among traders on an exchange, like a stock market An hourly average of day-ahead prices adjusted for actual load

SCED solves one problem every five minutes and outputs both the base points telling each unit what to produce and the LMPs at every node — the price is a byproduct of the dispatch solution. In real time, dispatch literally is the market outcome.

Module 04 · The Price Signal

Anatomy of an LMP

Every five minutes, every node on the grid gets its own price — this module dissects one, from the cheap-megawatt baseline to congestion spikes, negative prices, and $9,000 emergencies.

One price, three ingredients

A locational marginal price answers one question with brutal precision: what would it cost to deliver one more megawatt-hour to this exact spot on the grid, right now? PJM invented the modern version, switching it on in April 1998, and today computes it at more than 10,000 pricing nodes. Every LMP decomposes into three parts:

LMP = system energy + congestion + marginal losses

The system energy component is the familiar merit-order price — the marginal unit's offer if the grid were a single copper plate with no wires to worry about. Identical at every node, usually the biggest piece, typically $20–60/MWh.

The congestion component is what makes the price locational. When a line hits its limit, the software must dispatch more expensive generation on the constrained side, and the extra cost lands on the nodes downstream. Zero most of the time at most nodes — then it swings violently, tens or hundreds of dollars, positive or negative, when a constraint binds.

The marginal loss component reflects that pushing power down a wire heats the wire; delivering a marginal MWh to a distant node might require generating 1.05 MWh. Typically a few percent of the total — low single digits per MWh, positive or negative, depending on how far the node sits from generation and how heavily the system is loaded.

One big shared number, one violent local number, one small local number. The rest of this module is about the violent one.

Two nodes, one wire: watch a price split

The canonical example. Node A is Windville: 2,000 MW of generation offering $20/MWh. Node B is Load City: big demand, plus one local gas unit offering $80/MWh. A single line connects them, rated at 500 MW.

While the line has headroom, one more MWh anywhere comes from a $20 Windville turbine: both nodes price at $20, one market, one price. The instant demand pushes flow to the 500 MW limit, re-ask the LMP question. One more MWh in Windville? Still a $20 turbine. One more MWh in Load City? It cannot cross the full wire — it must come from the $80 local unit. The city snaps to $80.

Notice the shape: prices do not drift apart as the line fills from 60% to 99%. At 499 MW of flow, both nodes read $20. At 500, the constraint binds and a $60 congestion component appears out of nowhere. Constraints are cliffs, not ramps. Drive it yourself:

Congestion Console · Two Nodes, One Wire

LOAD CITY DEMAND400 MW

$20

LMP · Windville

$20

LMP · Load City

400/500

Line Flow MW

Congestion Rent /hr

Push demand past the 500 MW line limit and watch Load City's price snap from $20 to $80 — while Windville never moves.

This is why traders call an RTO's live LMP contour map an X-ray of the grid. You cannot see a binding constraint from the highway, but on the map it is unmistakable: a sharp color boundary — green $20s on one side, red $80s on the other — tracing the exact overloaded line. Congestion, invisible in the physical world, fluoresces in the price data.

Congestion rent, and the hedge market built on top

Follow the money in the bound case. The ISO charges Load City's consumers $80 per MWh but pays Windville's generators only $20 for the 500 MW crossing the line. That $60 spread × 500 MW leaves $30,000/hour in the ISO's account: congestion rent. It is not the ISO's profit — ISOs are nonprofits and must pass it through.

Where it goes: to holders of financial transmission rights (FTRs in PJM; TCCs in NYISO; CRRs in CAISO and ERCOT) — purely financial contracts defined by a source, a sink, and a MW quantity, paying (sink LMP − source LMP) × MW each hour. Hold a 100 MW Windville→Load City FTR during our binding hour and you collect $6,000 — exactly offsetting the congestion charge you'd pay to move 100 MW of your own power along that path. That's the point: FTRs let a utility that bought cheap remote generation lock in delivery cost despite unpredictable congestion. RTOs allocate the underlying value to the customers who paid for the transmission system, then auction the rights to anyone — including pure financial traders.

The sums are serious: PJM's market monitor put total congestion at $1.07 billion in 2023 and $1.75 billion in 2024 — a 64% jump in one year, driven by rising load and a transmission build-out that hasn't kept pace.

Field Note · The Trader Who Broke the Hedge Market

FTRs are financial instruments, and financial instruments attract financial disasters. In 2018, a two-person firm called GreenHat Energy assembled the largest FTR portfolio in PJM — with almost no collateral, under credit rules that assumed FTR positions were self-securing. When the portfolio went bad, GreenHat defaulted, and PJM's rules socialized the loss: roughly $179 million spread across every member of the market. The debacle triggered a FERC investigation, a rewrite of RTO credit policy, and a standing lesson: every market layer you build on the grid — even the hedging layer — needs its own risk management.

Nodes, zones, and hubs: reading a price screen

Three flavors of price, one market — essential literacy for watching this world:

Nodes — the raw, physical granularity. Generators are typically paid at their individual node, so a plant behind a constraint feels its congestion personally.
Zones — utility-sized aggregations. Load usually settles at a zonal average price, smoothing nodal noise across a delivery territory.
Hubs — stable, liquid averages of many nodes (PJM's Western Hub, ERCOT's North Hub) designed for trading: when someone quotes "PJM power at $42," they mean a hub, and most bilateral contracts and futures settle against one.

Generators paid nodal, load settled zonal, contracts quoted at hubs. Now you can read any price screen in the industry.

Below zero: when generators pay to keep running

LMP maps regularly show negative prices. A node at −$15/MWh means a generator must pay $15 to inject a megawatt-hour there. Why would anyone? Three real reasons, none irrational:

Inflexibility. A nuclear or big coal unit can't ramp off for a three-hour price dip; shutdown-and-restart takes days and costs hundreds of thousands of dollars. Paying $15/MWh for a few hours beats cycling. Many thermal units also sit at minimum-generation levels because they're needed tonight for tomorrow's peak.
Tax credits. The federal production tax credit pays wind (and now solar) owners per MWh generated — $27.50/MWh at the 2023 rate, rising to $30/MWh in 2024, for projects meeting the Inflation Reduction Act's prevailing-wage and apprenticeship requirements. A wind farm earning ~$30 in credits still nets money at any price above roughly −$30. Its true marginal cost is negative, and its offers say so.
Congestion + sunshine. Trap lots of zero-fuel-cost generation behind a constrained line — West Texas wind, California solar — and the local price collapses first and furthest. This is the duck curve's belly, priced: in CAISO's solar-saturated SP15 zone, prices went negative in roughly 1,180 hours of 2024 — 13% of all hours — up from ~530 hours in 2023. When even negative prices can't shed enough supply, the ISO curtails solar — deliberately wasting free energy because the wires can't move it. A negative LMP is the grid shouting: stop sending electrons here, or build a battery.

The ceiling: scarcity pricing and the $9,000 lesson

If negative prices are the floor, the ceiling is not the most expensive generator's offer. When reserves run short, RTOs deliberately push prices above any unit's cost using administrative scarcity adders, on the theory that the last megawatts before blackouts are worth what customers would pay to avoid losing power — the "value of lost load."

ERCOT built the purest version: the Operating Reserve Demand Curve (ORDC), live June 2014 — as spare reserves shrink, a rising adder attaches to every LMP, small at comfortable levels, then exponential. Since ERCOT has no capacity market, these scarcity spikes are how its generators earn their fixed costs. Then came the stress test: during Uri, prices sat at the $9,000/MWh cap for roughly four straight days. The aftermath rewrote the rules: on December 2, 2021 the PUCT cut the cap to $5,000/MWh effective January 1, 2022; two weeks later its Phase I market-redesign order raised the ORDC's minimum contingency level from 2,000 to 3,000 MW, also effective January 1, 2022; multi-step price floors followed in November 2023. (In December 2025, ERCOT's real-time co-optimization overhaul retired the ORDC entirely, moving the same job inside the dispatch engine.)

Other RTOs run tamer versions: under FERC Order 831 (2016), offers are capped at $1,000/MWh — cost-verified offers up to a $2,000 hard cap — with reserve-shortage adders on top. Texas remains the high-wire act; everyone else works with a net.

Checkpoint · Module 04

1. In the two-node example, Load City's demand grows until the 500 MW line from cheap Windville ($20/MWh) is exactly full, forcing the city's $80/MWh gas unit to run. What do the two LMPs read?

Both nodes rise to $80/MWh, since the gas unit is now the system's marginal generator Load City jumps to $80/MWh while Windville stays at $20/MWh Both nodes settle near $50/MWh, splitting the difference across the constrained line

Each node prices the cost of serving one more MWh at that exact spot: an extra MWh in Windville still comes from a $20 turbine, but an extra MWh in the city can't cross the full line and must come from the $80 local unit. Prices separate instantly when the constraint binds — no blending, no averaging.

2. When that line binds, the ISO charges the city's load $80/MWh but pays Windville only $20/MWh, collecting a $60 spread on 500 MW. Where does that $30,000/hour of congestion rent go?

The ISO keeps it as operating revenue — that's how RTOs fund themselves It funds payouts to holders of financial transmission rights, whose allocations and auction proceeds largely flow back to load It is rebated to the Windville generators whose cheap power was blocked by the constraint

ISOs are nonprofits and can't pocket the spread; congestion rent funds FTR/CRR payouts, and the rights are allocated or auctioned with proceeds going mainly to the customers who paid for the transmission system. Blocked generators get nothing extra — their remedy is buying an FTR along that path.

3. A West Texas wind farm keeps generating even when its nodal price is −$15/MWh. Why is that rational?

The federal production tax credit pays roughly $27–30 per MWh generated, so the farm still nets money Wind turbines cannot be curtailed quickly enough to respond to five-minute price signals RTO rules require renewable output to be dispatched before fossil generation regardless of price

The PTC pays per MWh actually generated, so a wind farm's effective marginal cost is around −$27 to −$30/MWh — it profits at any price above that. Modern turbines can curtail in seconds and RTOs dispatch wind economically like any other resource; it's tax policy, not physics or priority rules, that pushes prices below zero.

Module 05 · The Other Products

Beyond Energy: Reserves, Regulation & Capacity

Energy is only one product on the grid's shelf. This module tours the auctions that buy headroom, split-second accuracy, and promises to exist years from now — and the multibillion-dollar fight over how to fund the fleet.

The shelf behind the counter

When the day-ahead market clears, the software isn't just buying energy. It's simultaneously buying ancillary services — commitments of capability rather than output — in the same optimization. The main items, fastest to slowest:

Regulation: capacity that follows the AGC signal — beamed every 2 seconds in PJM — absorbing the second-by-second wobble that keeps frequency at 60 Hz.
Spinning (synchronized) reserve: headroom on units already online, deliverable within 10 minutes if a generator trips.
Non-spinning reserve: offline fast-start units producing within 10–30 minutes, backfilling once an emergency stretches on.
Ramping products: the newest shelf item — maneuvering room against forecast error. CAISO's Flexible Ramping Product launched in 2016; ERCOT's ECRS in June 2023. Neither buys energy; both buy optionality. Both exist because of the duck curve.

The elegant part: because one optimization clears everything — co-optimization — the prices are internally consistent. A megawatt held as reserve is a megawatt not sold as energy, so reserves are priced at opportunity cost. Concretely: a gas plant with $40/MWh costs running into a $60 LMP earns $20/MWh of margin; ask it to hold 50 MW back as reserve and the reserve price must cover that $20, or the plant is being punished for helping. When capacity is plentiful, reserve prices sit near zero. When the system tightens, reserve scarcity drags energy prices up with it — reserves and energy are two claims on the same iron.

The need for speed: how batteries ate the regulation market

Regulation is the grid's smallest market — PJM procures a few hundred MW against a peak north of 150,000 — but the most demanding, and it's where batteries first proved themselves. In October 2011, FERC's Order 755 found it unjust that a lumbering steam plant and a lightning-fast flywheel earned the same regulation payment, and required "pay-for-performance": compensation tied to mileage (how much a resource actually moves) and accuracy. PJM split its AGC signal in two: RegA, slow, for conventional plants; RegD, fast and roughly energy-neutral over short windows — so a battery can follow it all day without running out of charge.

Batteries were built for this: millisecond response, near-perfect accuracy scores, no minimum run times. PJM's battery fleet went from zero in 2005 to 280+ MW by 2017 — about 41% of its regulation procurement. The catch: a few hundred megawatts of need saturates fast. Prices thinned, and the storage gold rush moved to Texas — where batteries now arbitrage energy prices and stack ancillary products at grid scale.

Demand gets a seat too

Not everything on the shelf is a generator. Demand response — factories, big-box stores, and aggregated households paid to not consume during tight hours — participates in these markets as a resource. FERC's Order 745 (2011) required demand response in energy markets to be paid the full LMP, like a generator; the Supreme Court upheld it in FERC v. EPSA (2016), the jurisdictional twin of the Hughes v. Talen case you'll meet in Module 06. Demand response is a multi-thousand-megawatt resource in PJM's capacity market — during the July 2026 heat wave, ~6,000 MW of activated demand response is part of why the lights stayed on. When you hear "virtual power plant," this is the market door it walks through.

Missing money: when the energy market can't pay the mortgage

Now the harder question: do energy and reserve revenues, by themselves, pay for enough power plants to exist? In theory, yes — a peaker running 40 hours a year can recover its mortgage if those hours price near the value of lost load. In practice, regulators flinch: most markets cap offers around $1,000–2,000/MWh, far below blackout-avoidance value, and operators call on emergency measures before prices spike, suppressing the very signal investors need. Economists Peter Cramton and Steven Stoft named the resulting gap "missing money." If scarcity hours are capped and rare, energy margins fall short of the cost of keeping plants alive — so plants retire, and new ones don't get financed, until reliability erodes.

The industry produced two rival answers — plus a third path. Pay for capacity explicitly (PJM, ISO-NE, NYISO): run an auction where resources sell a promise — I will exist and show up during emergencies — and collect a per-megawatt-day fee whether or not they ever run. Remove the flinch (ERCOT): keep one market, raise the cap sky-high, let scarcity do the financing. And state-mandated resource adequacy (California): the CPUC forces every load-serving entity to sign bilateral capacity contracts covering forecast peak plus a reserve margin — since 2025 demonstrated hour-by-hour for the worst day of every month ("slice-of-day," resource adequacy rebuilt for the duck curve).

Answer one: PJM's capacity machine, from CONE to Elliott

PJM's Reliability Pricing Model, running since 2007, holds a Base Residual Auction nominally three years before each June–May delivery year. Sellers offer capacity; the buyer side is an administrative demand curve anchored to Net CONE — the annualized cost of new entry for a reference plant, minus its expected energy and reserve earnings. Clear the auction and you collect a fee for every megawatt-day, doing nothing but existing.

Except existing isn't enough anymore. After roughly 22% of PJM's fleet failed in the January 2014 polar vortex, PJM adopted Capacity Performance: perform during emergencies or pay penalties at a rate of Net CONE × 365 ÷ 30 expected emergency hours, up to an annual stop-loss of 1.5 × Net CONE × 365 per committed megawatt. Note the cap is tied to Net CONE, not to what the resource earned in the auction — with clearing prices far below Net CONE, a unit's maximum penalty could be several times its annual capacity revenue. The theory met reality on December 23–24, 2022, when Winter Storm Elliott knocked out more than 40,000 MW — roughly 46,000 MW at peak, close to a quarter of the fleet's installed capacity, 70% of it gas. PJM assessed $1.8 billion in non-performance charges; a FERC-approved settlement later trimmed it to about $1.25 billion. That's why the stop-loss detail matters: the penalties were the point.

Elliott also exposed inflated capacity ratings, accelerating the shift to ELCC (effective load carrying capability) accreditation: your capacity value is your measured contribution to avoiding lost load, not your nameplate. For the 2026/27 auction: fixed-tilt solar 8%, tracking solar 11%, onshore wind 41%, 4-hour storage 50%, gas combined cycle 74%, nuclear 95%.

Then prices detonated — the "missing money," found. The 2025/26 auction cleared at $269.92/MW-day, up from $28.92 — the region's capacity bill jumping from ~$2.2 billion to ~$14.7 billion. The 2026/27 and 2027/28 auctions slammed into FERC-approved caps ($329.17 and $333.44/MW-day) under a two-year price collar approved in April 2025 after Pennsylvania's governor complained; PJM's own simulations say they'd otherwise have cleared near $389 and $530. Driver: data-center-led demand growth colliding with retirements. The 2027/28 auction was the first in which the entire RTO fell short of its reliability requirement. Module 06 finishes this story.

Answer two: ERCOT's high-wire act

ERCOT runs the purest energy-only experiment on Earth: no capacity market, scarcity prices as the sole signal telling investors to build. Then came Uri: generation froze, 4.5 million customers lost power for days, and 246 people died by the state's official count. Prices were ordered to the $9,000 cap and held there roughly four days; the market monitor concluded they stayed about 32 hours too long, overpricing real-time energy by some $16 billion (net settlement impact roughly $3.2 billion, since most load was hedged). Texas reformed without converting: cap to $5,000, mandatory inspected weatherization, earlier scarcity pricing, new reserve products. The legislature flirted with a capacity-market-lite — the Performance Credit Mechanism, adopted in concept in January 2023, capped by statute at $1 billion a year — then the PUCT shelved it in December 2024, voting unanimously to redirect its effort into real-time co-optimization instead. The energy-only faith survived its worst week — barely, and not unchanged.

Field Note · One Auction, Seven Times the Money

On July 30, 2024, PJM posted capacity auction results that made governors call their lawyers: the RTO-wide price jumped from $28.92 to $269.92 per MW-day — more than ninefold — and the region's total capacity bill from about $2.2 billion to $14.7 billion. Nothing physical changed overnight; the auction simply repriced scarcity as data-center demand collided with retirements and stricter ELCC accreditation. Capacity — the product nobody sees on their bill — abruptly became one of its biggest line items. Whether that's the market working (a genuine scarcity signal, arriving in time) or failing (a windfall to existing plants that can't build new ones fast enough) is the sharpest fight in the industry right now.

Checkpoint · Module 05

1. A gas plant with a $40/MWh marginal cost is running while the LMP is $60/MWh. The ISO wants it to hold 50 MW back as spinning reserve. What must the reserve clearing price cover for this to be a fair deal?

The fuel the plant burns while keeping that capacity spinning in reserve The $20/MWh profit the plant gives up by not selling those 50 MW as energy The full $60/MWh energy price, since co-optimization forces reserves and energy to clear at the same price

Co-optimized markets price reserves at opportunity cost: capacity held in reserve forfeits the energy margin (LMP minus marginal cost), so the reserve price must make the generator whole. That's why reserve prices hover near zero in slack hours and spike during scarcity.

2. A developer finishes a 100 MW fixed-tilt solar farm in PJM. Under the ELCC ratings used for the 2026/27 capacity auction, roughly how much capacity can it sell?

About 8 MW — solar output barely overlaps PJM's riskiest hours, which increasingly fall on winter evenings About 25 MW, matching its annual capacity factor 100 MW — capacity is sold at nameplate, and performance penalties handle any shortfall

Marginal ELCC asks how much a resource actually reduces loss-of-load risk; PJM rated fixed-tilt solar at 8% for 2026/27 because risk hours cluster when the sun is low or down. Capacity factor measures average energy over the year — a different question from reliability value.

3. ERCOT has no capacity market. What is its substitute mechanism for paying generators to exist?

State-mandated bilateral resource adequacy contracts, like California's program Very high scarcity prices — an offer cap of $9,000/MWh before Uri, $5,000/MWh since 2022 — so rare shortage hours finance new plants Direct FERC-regulated capacity payments to Texas generators

ERCOT is "energy-only": investment is funded by occasional extreme prices, smoothed by scarcity-pricing adders. The FERC option also gets the institutions backwards — ERCOT sits largely outside FERC's wholesale-rate jurisdiction, which is partly why Texas can run this experiment at all.

Module 06 · Power Over Power

Governance, Watchdogs & the Road Ahead

Every market rule you've learned was approved in a beige building in Washington, is policed by economists with no badge, and is now being rewritten under the fastest load growth America has seen in thirty years.

Three referees, one grid

Start with the seam that runs through everything. The Federal Power Act of 1935 split electricity in two: FERC governs wholesale sales and interstate transmission; state public utility commissions govern retail rates, most generation siting, and utilities' long-range build-out plans. The Supreme Court polices the line: in Hughes v. Talen (2016) it unanimously struck down a Maryland program because it effectively set a wholesale price — FERC's turf. (Its companion that same year, FERC v. EPSA, upheld FERC's reach in the other direction, blessing demand-response payments in wholesale markets. The 2016 term drew the border in both directions.)

Concretely: when PJM wants to change how its capacity auction works, it files the tariff change with FERC under Section 205 of the Federal Power Act on 60 days' notice. FERC can accept it, reject it, or suspend it and set it for hearing — and if FERC does nothing, the filing takes effect by operation of law. Every rule in this course — LMP formulas, auction parameters, penalty structures — lives in one of these tariffs, thousands of pages long, amended constantly. Under Section 206, FERC or a complainant can force a change by proving the existing rule unjust. The exception you know: ERCOT, where the Public Utility Commission of Texas plays both roles.

The third referee is NERC — born after the 1965 Northeast blackout, but toothless for decades: its reliability standards were voluntary. What changed that was the August 14, 2003 blackout — a sagging Ohio transmission line, a failed alarm system, and a cascade that put ~50 million people in the dark from Detroit to New York City. The US–Canada task force report traced violations of those voluntary standards, Congress responded with the Energy Policy Act of 2005, and NERC's standards became mandatory and enforceable in June 2007, with penalties up to $1 million per violation per day. So: FERC approves the market rules, states control retail and siting, NERC writes the physics-of-reliability rulebook — and everyone in this course answers to at least two of them.

Who runs the room — and who watches it

RTOs are nonprofits with independent boards, but the rules get drafted below the board, in stakeholder committees. PJM's Members Committee votes in five sectors — generation owners, transmission owners, other suppliers, electric distributors, end-use customers — with weighted thresholds so no single interest can steamroll the rest. Critics call the machinery both slow (major reforms take years of working groups) and captured (incumbents with paid staff outlast consumer advocates at every meeting). Defenders answer that the people who bear the costs should write the rules. Governance is also a live fault line out West: CAISO's board is appointed by California's governor — precisely why other states hesitated to join it. More below.

Then the watchdogs: Independent Market Monitors. Every ISO/RTO has one — Monitoring Analytics for PJM; Potomac Economics for MISO, NYISO, ERCOT, and ISO-NE. And they don't just observe; here's how the policing actually works:

Structural tests. Screens like the pivotal supplier test ask: is this generator's capacity needed to meet demand right now — could it name its price? A supplier that fails is subject to mitigation before it ever misbehaves.
Conduct-and-impact screens. Offers wildly above a unit's reference cost (conduct) that would materially move prices (impact) get automatically mitigated — reset to a cost-based level by the software itself, mid-auction.
Referrals. Suspected manipulation goes to FERC's Office of Enforcement, which can fine. The famous scalp: JP Morgan's trading unit paid $410 million in 2013 to settle FERC charges of manipulative bidding strategies in CAISO and MISO — schemes designed to extract make-whole payments, a direct descendant of the Enron playbook from Module 00.
The megaphone. Annual public State of the Market reports grade their own RTO's competitiveness and propose fixes the RTO often resists. IMMs can't fine anyone. Their power is analysis, made public — recall Uri's $16 billion letter.

The modernization docket

Five landmark FERC orders define the modern reform era. Learn them as a set:

Order 1000 (2011) — regional transmission planning and cost allocation: transmission must be planned regionally, with costs allocated to those who benefit — the legal foundation under every RTO planning process (and every cost-allocation fight since).
Order 841 (2018) — storage gets a seat: every RTO must let storage as small as 100 kW buy, sell, and set prices in all its markets. States sued; the D.C. Circuit upheld FERC in 2020 — a landmark for federal authority.
Order 2222 (2020) — the VPP order: the same logic for aggregations of distributed resources — rooftop solar, home batteries, EVs, thermostats bundled into a 100 kW-or-larger virtual power plant that bids like a generator. Implementation has been grindingly slow: PJM's full DER-aggregation participation, originally set for February 2026, was pushed to February 2028 — an order on paper taking the better part of a decade to become a market you can trade in.
Order 2023 (2023) — fixing the queue: scrapped first-come-first-served interconnection studies for first-ready-first-served cluster studies, with higher deposits, site-control requirements, and withdrawal penalties to flush out speculation.
Order 1920 (2024) — planning for the grid we'll need: transmission providers must plan on a 20-year horizon, across multiple scenarios, refreshed every five years, with cost allocation settled up front.

Together: wires planned regionally and long-term, storage in, DERs in, queue unclogged.

The squeeze: a 2,300 GW queue meets the return of load growth

Now the collision that defines the late 2020s. At the end of 2024, Lawrence Berkeley National Lab counted roughly 2,290 GW of active capacity waiting in US interconnection queues — about 10,300 projects, nearly twice the entire installed US fleet: 956 GW solar, 890 GW storage, 271 GW wind, with gas jumping 72% in a year to 136 GW. History is sobering: of projects that entered queues from 2000–2019, only about 19% (roughly 13% of capacity) ever reached commercial operation, and the typical wait now runs four and a half to five years.

Meanwhile, after two flat decades, demand is roaring back — AI data centers, electrified heating and transport, new factories. NERC's latest long-term assessment forecasts US-plus-Canada summer peak demand growing 224 GW in ten years, the fastest growth since its tracking began in 1995, with data centers the dominant driver. Add accelerating retirements and you get the resource adequacy squeeze: demand rising into a fleet that can't connect fast enough. You watched PJM's capacity auctions price it in Module 05; you watched PJM's all-time load record likely fall in Module 01. Same story, three data streams.

The flashpoint is co-located load: in March 2024 Amazon paid $650 million for a data-center campus at Talen's Susquehanna nuclear plant, planning up to 960 MW served behind the meter. FERC rejected the expanded interconnection deal that November — opponents claimed up to $140 million a year in costs shifted to other ratepayers — the parties restructured it as a conventional front-of-meter contract in June 2025, and in December 2025 FERC declared PJM's tariff unjust for lacking co-location rules at all, ordering new ones in 2026. Who pays for the grid when the biggest customers try to step halfway off it — that is the fight.

The next map — and how to watch this world

The last frontier is the West, where two rival day-ahead markets are racing to organize the non-RTO half of the map. CAISO's EDAM went live May 1, 2026 with PacifiCorp; Portland General Electric follows in October 2026, with LADWP, PNM, BANC, and Turlock slated for 2027. SPP's Markets+ won FERC approval January 16, 2025 and targets an October 2027 go-live for its first cohort — anchored by the Bonneville Power Administration, which chose Markets+ citing governance independent of any single state. California answered: AB 825 (the "Pathways" bill), signed September 2025, opens a path to move governance of CAISO's regional markets to a new independent body — blunting the it's-run-by-Sacramento objection. Within a few years, nearly every US megawatt-hour will clear through some organized day-ahead market. The map you memorized in Module 01 is not finished being drawn.

How to keep watching after this course:

FERC dockets — search eLibrary (elibrary.ferc.gov); tariff filings are "ER-" dockets. Every controversy in this module lives in a docket you can read for free.
IMM State of the Market reports — Monitoring Analytics (PJM) and Potomac Economics (MISO, NYISO, ERCOT, ISO-NE): the sharpest free analysis of how these markets actually perform.
Live dashboards — PJM Data Miner, CAISO Today's Outlook, ERCOT's grid dashboard, or gridstatus.io for all of them at once. Real LMPs, live — you can now read every number on the screen.
Annual rhythm — LBNL's Queued Up (interconnection queues, each spring), NERC's Long-Term Reliability Assessment (each winter), and daily trade press: RTO Insider, Utility Dive.

Field Note · The Watchdog That Cried $16 Billion

During Winter Storm Uri, ERCOT held wholesale prices at the $9,000/MWh cap for 32 extra hours after emergency load-shedding ended. Potomac Economics — ERCOT's own Independent Market Monitor — publicly called it a $16 billion pricing error, likely the largest ever flagged in a US power market, and urged regulators to reprice it. The Public Utility Commission of Texas refused, and the Texas Supreme Court upheld the refusal. The episode is the purest demonstration of what an IMM is: enormous analytical authority, zero enforcement power — and a megaphone.

Checkpoint · Module 06

1. PJM wants to change the rules of its capacity auction. Before the change can take effect, who reviews it?

FERC, via a tariff filing under Section 205 of the Federal Power Act The public utility commissions of all thirteen PJM states, by majority vote NERC, since capacity auctions are a reliability function

RTO market rules live in FERC-jurisdictional tariffs, filed under Section 205 on 60 days' notice — FERC can accept, reject, or suspend and hear them (and silence means the filing takes effect by law). State PUCs govern retail rates and siting; NERC writes physical reliability standards.

2. Your startup bundles 5,000 home batteries into a 40 MW virtual power plant and wants to bid it into MISO's markets. Which FERC order requires MISO to let you participate?

Order 841, the storage participation order Order 2222, the DER-aggregation order Order 1920, the long-term transmission planning order

Order 2222 (2020) requires RTOs to admit aggregations of distributed resources of 100 kW or more as wholesale participants. Order 841 (2018) covers individual grid-scale storage; Order 1920 (2024) governs transmission planning, not market participation.

3. Roughly how much capacity was actively waiting in US interconnection queues at the end of 2024?

About 230 GW — around a fifth of the installed US fleet About 23 GW — roughly one year's worth of new construction About 2,300 GW — nearly double the entire installed US fleet

LBNL counted ~2,290 GW active across ~10,300 projects, roughly twice the existing fleet, with solar, storage, and wind about nine-tenths of it. Historically only about one in five queued projects ever reaches operation — which is why Order 2023's queue reforms matter so much.

Course Complete

Certificate Pending

Pass all seven module checkpoints to earn your (entirely unofficial, deeply prestigious) Market Operations certification.

Your homework, if you want it

Watch a market clear. Open gridstatus.io or PJM's Data Miner around 1:30 p.m. Eastern and look at tomorrow's day-ahead LMPs the moment they post. You now know the machine that produced every number.
Find a binding constraint. Pull up any RTO's live LMP contour map and look for a sharp color boundary — then remember Module 04: you're looking at an overloaded wire, fluorescing in price data.
Read one State of the Market report. Monitoring Analytics' PJM report or Potomac's ERCOT report — start with the executive summary. You'll recognize every concept.