The Negawatt Market: Trading Power You Can't Store

Every second of every day, the power grid has to make exactly as much electricity as everyone is using. Not close. Exactly.

There is nowhere to put spare electricity. No warehouse, no back room, no "we'll send it Tuesday." If the world starts using more than the grid is making, the whole grid slows down. Make too much, and it speeds up. Let it drift too far and machines start switching themselves off so they don't fry.

Meanwhile, electricity gets bought and sold ahead of time, in 15-minute chunks, sometimes a day and a half early, based on the weather forecast.

So one side of this story needs perfect balance right now, this exact second. The other side is people making promises about the future using a weather app. Everything else -- the auctions, the alphabet soup of acronyms, the $800 price at 7am -- exists to glue those two things together.

This page is a simulation, built to teach how balancing markets actually work rather than to copy any real place. You run a small green-energy firm in a made-up zone called Positron: a few wind farms, a few solar farms, a battery, a book of customers who buy your power, and a shed full of computers mining crypto. The firm is invented -- but the market it plugs into is real, basically how most of Europe runs today.

Simulation

The knobs set up the world and your firm's choices. One day shows a single day up close -- hover over any 15-minute block and the cards below explain what happened in plain words. The month races three different strategies against the exact same weather. Pro screen shows the same month the way a real trading desk's wall of screens would.

The grid is one giant spinning wheel

Start with why any of this exists.

Picture the whole grid as one enormous spinning wheel that every power plant turns together, all in sync. The speed of that wheel is called the frequency, and it is a live scoreboard of the balance between power made and power used.

Use more than you make, and the wheel slows down. Make more than you use, and it speeds up. Nobody chooses this. It is just physics.

So somebody has to stand there every second, ready to shove more power in or pull some out on a moment's notice. That somebody is the system operator (the industry name is the TSO). Its tools are layers of backup power, sorted mostly by how fast they can jump in:

Forget the names for now. The shape is what matters: the fast backups are expensive and small, the slow ones are cheap and deep, and the operator is always trying to hand the problem down to the cheaper, slower ones.

The sim's "system needle" strip shows the one thing this whole machine watches: how far out of balance the zone is, minute by minute. One detail that bites later -- the official needle everyone can see runs about 30 minutes behind what is really happening.

Three chances to be right

Trading bumps into physics as a sequence -- three chances to get your energy right, each one closer to the moment the power actually flows.

1. Day-ahead. At noon, an auction sells off every 15-minute block of tomorrow. You bid based on your forecasts, the auction hands back one price per block, and whatever you win becomes a promise to the operator: this is exactly what I will deliver.
2. Intraday. From that evening right up until just before each block, a nonstop market lets you fix your promise as your forecasts get better. A known mistake, a known price, sorted.
3. Imbalance settlement. Whatever gap is still left between what you promised and what you actually did gets settled afterward, at the imbalance price -- a number you only find out later, and one that can hurt.

A beginner sees three separate markets. Look again and it is one thing quietly running out of time: every hour that ticks by, you have fewer ways left to fix a mistake, and the price of fixing it gets harder to guess.

And "energy trading" is a sneaky phrase, because it makes you picture the wrong thing. These three are not even the same kind of market.

The day-ahead auction is not a stock exchange. Everyone hands in a secret list of "I will buy or sell this much at this price," a computer finds where all those lists cross, and everyone in a block pays one shared price.

The intraday market is more like a stock exchange -- but a thin, lonely one. Every 15-minute block of every zone is its own little product that expires fast, so instead of one busy market you get hundreds of nearly-empty ones.

The balancing layer is not traders at all. It is a single buyer -- the operator -- walking down a price-sorted list of offers because physics left it no choice. The closer you get to real time, the less it looks like finance and the more it looks like engineering.

Bids are curves, not bets

Nobody who actually trades this wastes effort guessing tomorrow's day-ahead price. Not because it is impossible -- because they do not need to.

A day-ahead bid is not "I want 5 MW." It is a curve: at this price I want this much, at that price I want that much. Your own economics -- the price where running, buying, or storing is actually worth it to you -- ride inside the bid. The auction runs your rules against everyone else's, and whatever price pops out, you are automatically happy with what you got.

Your crypto miners make this real. Mining turns power into computing worth about $120 for every MWh. So should the miners run at 7 tomorrow morning? Wrong question. You hand in a curve -- buy power whenever it is $120 or cheaper -- and the auction answers it for you, hour by hour.

A "block" bidder who says "I want exactly 5 MW" has to guess the price or lose money. A curve bidder only has to know their own costs. That is why, in practice, all the price-guessing effort moves somewhere else -- toward the one price you cannot write a curve against, because by the time it exists, the auction is long finished. That price is next.

Where a shortage becomes a price

When the moment of delivery arrives and the zone is short 80 MW -- a forecast missed, a cloud rolled in early, a plant tripped offline -- the operator has to buy 80 MW right now. It buys from the stack. Think of a discount store that piles its cheapest power right by the entrance and tucks the pricey, exclusive stuff deep at the back. The operator walks in, grabs from the front first, and only pushes further into the shop when it needs more.

It keeps grabbing until the 80 MW is covered, then pays the price on the deepest shelf it had to reach -- for the whole basket. That price becomes the imbalance price -- and everyone's leftover mistakes settle at it, not just the operator's.

The needle is just how deep into the shop the shortage pushes the operator. Drag it yourself:

Two things matter here, and everything later leans on them.

The buyer here has no choice. Back in the day-ahead auction, the buyers were people with options -- they would walk away if the price was bad. Here, the zone needs 80 MW at any price, because the alternative is the grid losing its balance. There is no walking out empty-handed, and no "too expensive" -- the needle pays whatever the shelf it reaches is charging.

So why is the store sorted this way at all? Because the same electricity costs wildly different amounts to make, depending on who makes it. A hydro dam or an already-running nuclear plant is cheap -- the money was spent building it, and squeezing out one more unit costs almost nothing. A gas plant has to buy fuel for every single unit, so it is dearer. An emergency generator, or a factory you pay to switch itself off, is dearer still. Line every source up by what it costs to run right now, cheapest first, and you have built the store: cheap-to-run power by the door, expensive-to-run power at the back.

If you could store electricity, the expensive back of the shop would barely matter -- you would fill a giant tank with cheap nuclear power overnight and pour it out at breakfast. But you cannot keep electrons in a warehouse (batteries hold a little, but that is its own story). So at the exact second everyone wants power at once, the cheap sources are already flat out, and the only thing that can jump in this instant is the pricey stuff at the back. That is why no country runs on 100% cheap nuclear even though it looks cheaper on paper: some hours, you simply have to burn something costly to keep the wheel spinning.

Now watch the shape bite. A normal little wobble never leaves the cheap shelves by the door, so the price stays boring -- a few dollars above normal. But a 300 MW plant tripping marches the operator all the way to the back, and nobody rigged anything: it is just what that last, most desperate source costs to run. The same machine prints $600 while the normal price sits at $100. Everybody pays that deepest price, which is why the spikes are sudden cliffs, not gentle slopes.

It works upside-down too. When the zone has too much -- too much wind, a blazing sunny noon -- the operator pays plants to turn down. Now the shop runs in reverse: instead of you paying to take power, it pays you to make the extra go away, and the price can go negative. Sit with that for a second: there are 15-minute windows where spare electricity is worth less than nothing.

Why would anyone pay to give power away? Because for a lot of plants, stopping costs more than paying does. A nuclear plant or a heat plant cannot just flick off and on cheaply, and a subsidized wind farm that gets paid per unit it makes stays profitable even at a small negative price. When there is a glut, the extra power physically has to go somewhere -- so the market pays whoever can soak it up. Real European markets now have hundreds of negative-price hours a year, with a floor around -$500 (which is where this sim's floor comes from).

You are always trading

Now the rule that turns every wrong forecast into a trade, whether you wanted to trade or not.

Your firm promised a schedule yesterday. Reality delivered something a bit different. That difference is your imbalance, and it gets settled at the imbalance price, both directions, under a rule called single pricing: short? you buy the missing bit at the imbalance price. Long? you sell the extra bit at the imbalance price. One price, no fine, no forms. Just a price.

Sounds fair -- until you notice which way the luck runs. Most beginners assume this: "if I make too much, no big deal -- extra power just gets sold, minor thing." The sim will spend a whole month beating that idea out of you, in two steps.

1. Your mistakes are everyone's mistakes. Your solar farms and your rivals' solar farms sit under the same sky. When your output surprises you by running high, the whole zone's does too -- everyone is long because of the same sunshine that made you long.
2. The price comes from the zone's state, not yours. Whole zone long → needle sits at the cheap front of the shop → imbalance price at its floor, maybe negative.

Put those together and it is brutal: you have the most extra to sell at the exact moment extra is worth the least, and the biggest shortfall to cover at the exact moment power is priciest.

And it is far worse than plain bad luck. If your forecast errors were random -- a bit high one day, a bit low the next -- they would cancel out over a month and cost you almost nothing. But your errors are not random. The size of your mistake and the price you pay to fix it are driven by the same weather, so they swell together, on the same side, every single time. Random errors add up and cancel each other; these two just keep multiplying -- a big error times a bad price, over and over -- and the losses compound instead of washing out.

That is why the sim's ledger files this under imbalance · unavoidable: whatever error you are left with leans this same wrong way, because the moment your weather surprises you, it surprises the whole zone right along with you.

There is a flip side, and a later strategy is built on it. Single pricing cuts both ways: your surplus settles at that same imbalance price too. So on the rare occasion you have extra power exactly when the whole zone is short, the price is spiking -- and single pricing sells your extra straight into that spike. Being off-schedule by the same amount, but in the opposite direction to everyone else, flips the sign: instead of paying the spike, you pocket it. That is the one time a forecast error makes you money, and later you will try to make it happen on purpose.

Repair, carry, or lean

So forecasts keep updating all night and all morning. The evening run says your wind guess was 20 MW too high. Now you are holding a known mistake. There are three schools of thought about what to do, and the sim runs all three on the same weather and the same firm.

Do nothing (the industry term is set-and-forget) submits the day-ahead promise and walks away. Ignore every update, let every mistake ride all the way to settlement. This is not a meme strategy -- for a tiny retail-only firm whose mistakes are rounding errors, it is the smart, lazy choice. But for a weather-driven firm it means carrying known, everyone-shares-them mistakes straight into the one market that punishes exactly that. Its month has the widest swings and the ugliest worst case.

Fix everything (also called the full hedge) trades every mistake away in the intraday market the second it appears. You get certainty -- for a price. By the time you know about the cloud bank, so does everyone, so the intraday price has already slid toward the bad imbalance price, and you pay the gap on every trade. Fixing does not dodge the cost of a bad forecast. It just swaps a scary unknown bill for a smaller known one. Very much like buying insurance on weather and production. But if you can buy insurance, someone smarter can sell, it is a full circle which leads us to...

Fix, then bet (the jargon for the bet is a lean) is where the strategy settles once you have been burned enough times. Fix what you know, bet on what you believe. Trade away the part of the mistake you are sure about -- then, where your model has a hunch about which way the whole zone will lean, deliberately keep a small position pointed the helpful way, sized to how confident you are. That is you being wrong the opposite way to everyone else on purpose, to collect the reward single pricing pays for it. Actually, now it sounds exactly liek weather insurance, but in real time.

The month tab shows the verdict as three overlapping bell curves. Flip our weather forecast between poor and good and watch the PnL.

With a good forecast, the bet lifts the whole curve to the right, past zero. With a bad one, the bet melts into random guessing and the edge vanishes. A bet is only as good as the forecast under it. The gap shows how much you get paid being good at forecasting, in dollars per day.

The flexibility desk

Everything so far cast your firm in one role -- the balance responsible party (BRP), the one that promises a schedule and pays whenever it misses. Mostly on the receiving end of the market.

The store, though, is built out of offers, and anyone who can control their own power can sell into it. That is the other role the simulation hands you: the balancing service provider (BSP), paid to stock a shelf in the store and cover the gap when the operator wants the product stocked behind the back of the shelf.

The sim lets you play both at once for one reason -- to show how much option value, and how much extra profit, a plain BRP unlocks the moment it owns a battery or compute farm it can dial on demand.

The miners are your compute farm load you can switch off in seconds -- and switching off a 5 MW load is exactly the same, to the grid, as switching on a 5 MW generator. The trade even has a name for it -- a negawatt (yes, I have the same reaction when writing this).

The math is pure opportunity cost: suppose mining earns about $120/MWh, then the miners offer to power down for $120. Whenever the stack's price beats that, mining pauses and the balancing market pays better than the mining would have. (Watch the event feed during a shortage: Negawatt's up ⚡.) And in the deep-negative windows, the same shed gets paid to gobble up more power.

The battery is another way of absorbing the load, so the sim gives you both a size slider and a job switch. In earn in the market mode it stocks a shelf in the shop, collecting payment each time the operator reaches deep enough to buy it, which gets huge during outages, small but steady on calm days. In cover our mistakes mode it quietly soaks up your firm's own errors, shrinking the leftover-gap line instead of earning visible income. The same battery cannot do both jobs at once.

Mess with it for a while and a few things become clear.

1. Bigger pays, but not double-for-double. In the default world, your first few MW each earn about $2,500 a month. By the time you are at 80 MW, an extra megawatt earns a few hundred. The zone's shortages are only so deep -- pile on more battery and the extra just stands around waiting for a disaster big enough to need it, so a big capex in batteries is NOT an autowin strategy.
2. Past the slider, the sim would start lying to you. The sim pretends you never move prices. A giant battery would flatten the very spikes it feeds on -- the "infinite battery, infinite money" dream dies the moment you become the back of the shop. That is why the slider stops where it does.
3. Set the job to cover our mistakes under Fix, then bet and watch the money vanish. The battery cannot tell a mistake from a bet -- it dutifully cancels out your bet. And because single pricing sometimes rewards honest mistakes, even the sloppy strategies gain less from self-insurance than from just selling that flexibility to the market. That is not irrational, and it is why real firms in single-price zones run their batteries in the market instead of hiding them behind their own books.

The risk on this desk -- the BSP side -- is the mirror image of the BRP side you played until now. A balancing-book position has bottomless downside -- a few bad tail hours can wipe out a month.

A flexibility asset's downside is capped: the worst case is not a giant bill, it is a reliability strike -- you fail to deliver when you are called on (the battery runs empty halfway through an outage) and the operator writes you up. Three strikes and you are kicked out of the reserve market for the rest of the month. The sim tracks this. Strikes are rare equipment faults that only bite during real activations, so they show up as bad luck in busy months, and after the third one your battery income flatlines.

Tails, not averages

Set breakdowns to zero and the weather to calm, and this market looks boring -- spreads of a few dollars, an unprofitable and unexciting business. This is the world a lazy "let's look at a typical day" analysis would see, and it is a lie by leaving things out.

When outages becomes a frequent thing, a single tripped plant now takes the electrons 300 MW deep into the liquidity pool and prints hours of huge spreads.

The month's profit stops being a sum of days and becomes a sum of events: twenty-six forgettable days plus three that decide whether the month is green or red. This makes electron trading a 'tail-end oriented' business.

Check the charts, strategies' averages sit within a few thousand dollars of each other, while their worst cases are tens of thousands apart. In this market you do not manage the average. You manage the worst case.

The flexibility strip shows the same fact from the other side: the desk that owns the batteries and miners gets those exact same events as income spikes.

How profitable is all of this?

Great question. After all, the grid has kept the lights on for 100 years, long before anyone built a stock exchange for electricity, and it managed fine. So why, suddenly, are we paying millions for quant traders and order-book plumbing to do a job the grid did quietly for a century?

Start with why it was quiet for a century. The old grid was a few dozen big power stations an operator could basically phone. Supply was controllable -- need more, call a coal plant and turn a dial. Demand was predictable -- people boil kettles at roughly the same times every day.

Balancing was a small in-house chore, a handful of knobs the operator owned outright. There was nothing to trade, because there was nothing to argue about.

Then we bolted the weather onto the grid. Wind and solar cannot be dialed up -- the sky decides, not a dispatcher -- and instead of a few dozen controllable plants there are now thousands of weather-driven ones, mostly owned by people the operator has never met.

The gaps got bigger, hit more often, and all leaned the same way at once (same sky, same surprise). You cannot fix that by phoning everyone. The job stopped being an engineering chore and became a coordination problem across thousands of strangers.

A market is simply the cheapest way to coordinate that mess. It works out who can adjust for the least money -- the store, sorted cheapest-first -- without any central planner needing to know everyone's costs.

It pays the ones who help and charges the ones who caused the gap. The order book and the quants show up the instant real money starts flowing through that coordination. Nobody got greedy, we just made the balancing problem so painful to work with that we crowdsourced to strangers.

So who actually gets rich? Almost nobody, and that is the part the "energy trading is the hot new frontier" pitch leaves out. For a normal weather-driven firm, the balancing market is a cost you shrink, not a profit you make -- it is the unavoidable-imbalance bleed you already watched in the ledger. You do not hire quants to strike gold. You hire them so you bleed a little less than the firm across town.

And the money that does get made is mostly a transfer, not fresh wealth ¹. It flows from the firms that forecast badly or cannot flex, into the ones that forecast well or own a battery. The quants' salaries come straight out of that transfer. It is an arms race: everyone spends on forecasting and flexibility to avoid being the sucker funding everyone else -- and once everyone spends, most of the edge cancels, and the spending itself becomes the price of staying in the game.

The operator is not cleaning up either. It still earns the boring, regulated way -- a set cut of the power delivered, roughly how your water utility earns -- and it runs the balancing market not to profit but to buy the emergency flexibility it is legally forced to keep on hand, as cheaply as competition allows. The market is its shopping cart, not its cash register. The alternative -- the operator building and owning a mountain of idle backup itself, or letting the frequency wander until things trip -- costs far, far more. ²

So, is it profitable? System-wide, it is the opposite of a gold mine: it is the cheapest way anyone has found to keep a weather-powered grid reliable and consistent.

We spend millions on quants and screens because going green created a billion-dollar coordination problem, and an auction is the least-bad tool we have for it.

Firm by firm, it is mostly defensive -- a handful of battery owners and sharp forecasters come out ahead, paid for by everyone else, and most players would switch the whole thing off if they could. They cannot, because the problem is physical and real, and somebody has to be paid to close the gap every fifteen minutes, forever.

Why models rot here

Seeing all these prices, there is a tempting shortcut: skip the physics, feed a few years of old imbalance prices into a model, predict the next ones. The sim's poor forecast is exactly that shortcut (it literally guesses tomorrow from yesterday's prices), and its real problem is not being wrong on a quiet Tuesday. It is that the thing making these prices is a machine whose parts get swapped out while you are still studying it.

Balancing markets are young and get rebuilt constantly: settlement windows shrink, zones merge their backup pools, a fleet of new batteries moves into the stack and shaves every spike. Each change rewrites the link between weather and price.

A model that only learned old prices inherits none of how the machine works and all of its fragility. The other way -- forecast the physical stuff (wind, sun, breakdowns), then let the market rules turn that into a price, exactly how this sim is built -- ages gracefully, because each of its parts matches a real part of the machine.

How is pricing set in practice?

You might be picturing a trading floor full of people shouting. There isn't one.

The day-ahead price is one computer run a day: at noon a program takes everyone's secret curves, finds where they cross, and stamps one price on each block of tomorrow. The intraday market is a matching engine pairing buyers and sellers around the clock. The imbalance price is arithmetic the operator does after the fact -- read the deepest shelf it had to buy from, print the number. Humans write the curves and pick the leans; the clearing itself is a program that runs whether anyone is watching or not.

Which answers the next worry: yes, this runs every fifteen minutes, all day, forever, because power is used every fifteen minutes, all day, forever. But that is the whole reason you bid a curve instead of manning a desk. You write your rule once -- buy under $120, sell my spare above $200 -- and the machine runs it at 3am while you sleep. The market never rests; the trader mostly does. Babysitting it by hand is the exact job the curve was invented to delete.

Then comes the question of using price as its own prediction parameters. Can you even feed price into your decisions, though, when a price is really just physics wearing a dollar sign?

Weather makes the shortage; the shortage makes the price. Predicting tomorrow's price from yesterday's prices is chasing the shadow instead of the thing that casts it. Electric markets is very different from equity markets. Electric is a consumable and ephemeral commodity, so prices shouldn't have memory. Businesses are almost identical in the short term, so prices can have memory. ³

A price is not a thermometer, though -- it does not just sit there being read. The moment everyone can see it and react, it changes the very thing it is measuring. A scarcity price is meant to fix the scarcity by telling everyone what to do: expensive means "come online, or stop using."

But when a thousand miners and batteries all obey it at once, the shortage flips to a glut, the price craters, everyone backs off, and the shortage comes straight back. It is a map app routing every driver down the one empty side street, which stays empty only until the app fills it. The signal built to smooth the swing becomes the reason for the swing.

This is not a bug someone forgot to fix -- it is a law with a name. Economists have called it the cobweb for a century: high crop prices, so everyone plants, so a glut, so low prices, so everyone quits, so a shortage, and around again. George Soros would call it reflexivity. Control engineers would call it a loop that rings.

The page already handed you the accelerant -- the official needle runs about thirty minutes late, so everyone is steering by looking at where the road was half a mile back. React hard to stale information and you overshoot, every time, together.

So is a real-time market for this even possible? Yes -- but only because it is not perfectly efficient. The things that look like friction are the brakes: the gate that slams shut before delivery, so you cannot keep reacting; the plant that physically cannot switch on the instant the price twitches; the imbalance penalty that punishes anyone who overreacts. Those frictions are the damping that keeps the feedback loop from exploding. ⁴

What the simulation ignores

At the end of the day, this page is just a nerd's toy. We are just showing the wires inside the zone and its traffic jams. We skipped:

• The fine detail of how intraday trades actually execute (they run at a modeled price).
• The fastest backup layer's spin up time.
• Price differences between zones.
• Your firm's own market impact (we assume you are too small to move anything, anywhere).
• The fine print of how plants can be switched on and off (some can't).
• The real numbers -- the shape of the stack, the size of the errors, the $120 mining value -- is invented, picked to sit in believable ranges.

The one faithful part is the shape of the cascade. The intention is to let people understand the business, profit opportunity, and cost incurred to civilization if not solved.

The comic version

Low cortisol link. How Electricity and Compute Get Traded.

Glossary