Exchanges & Market Structure

:::section{mins=5} What a CEX Actually Does :::

:::lead You buy crypto with a bank card, and a minute later it shows up in your exchange balance. In that minute the exchange did several different jobs at once, and none of them involved a blockchain transaction in your name. In 2022 FTX showed what that arrangement costs when it fails, with roughly $8 billion of customer money gone. :::

:::p{html} CEX (Centralized Exchange) is a company that matches buyers and sellers through its own trading system. When you deposit BTC, the exchange credits your account after enough network confirmations. From that point, trades inside the venue don't need a separate Bitcoin transaction for every buy and sell. What changes is a row in the exchange's internal database. The coins themselves sit in wallets whose private keys are controlled by the exchange or its custodian, and the on-chain world only sees movement again when you request a withdrawal. If the mechanics of keys are hazy, 3.1 Wallets covers why whoever holds the key controls the asset. :::

:::p Why does this matter? Because while the asset sits on a CEX, you depend not only on the blockchain but on the company itself. It controls the keys, the withdrawal rules, identity checks, and access to your account. The FTX collapse was the extreme version of that dependence. In 2024 the U.S. Department of Justice described the case as Sam Bankman-Fried's misappropriation of billions of dollars in customer funds, and he was sentenced to 25 years in prison. :::

:::p{html} The account itself comes with paperwork. KYC (Know Your Customer) is usually required on large regulated venues for full access, especially for moving fiat in and out. The exact documents and the exact set of features depend on your country and on which legal entity serves your account, a detail we return to in §2. The broader rules around identification are the subject of 7.5 KYC/AML. :::

:::callout An all-inclusive resort with a wristband. At check-in you hand over money at the front desk and get a plastic band on your wrist. Inside, everything works with a tap. The bar, the spa, the excursion desk, no card, no cash, the speakers hum and the pool smells of chlorine. It feels like money, but it's a number in the resort's computer. The kitchen and the cleaning crew settle up between themselves in that same computer, and nothing leaves the building. Real money only moves again when you check out and the desk pays out what's left. If the resort goes bankrupt overnight, the band on your wrist still beeps at the bar, but the number behind it belongs to the bankruptcy lawyers now. :::

:::keypoints{label="KEY POINTS"}

• A CEX balance reflects your claim on the exchange inside its own books, not coins at an address under your control. The blockchain doesn't know your account exists, which is exactly why internal trades are instant and free of gas.
• A hundred trades inside a CEX leave not a single extra transaction on the blockchain; the network sees only your deposits and withdrawals. That's why a block explorer can't reconstruct your trading history: it lives only in the exchange's database, and leaves with the exchange if the exchange goes.
• CEX risk is wider than hacks. Management fraud, liquidity gaps, withdrawal freezes, and legal restrictions have each cost users real money, and neither a strong password nor 2FA protects against any of them — those guard the account, not the company.
• Experienced users treat an exchange as a workshop rather than a vault. What's actively traded stays on the venue, the rest lives in self-custody, and the account sits behind authenticator codes rather than SMS. :::

:::details{badge=Advanced title="Where the exchange's revenue comes from"} :::p{html} The visible layer: trading fees :::

:::p{html} Many CEXes run a maker/taker model, where the fee depends on whether an order adds liquidity to the book or removes it; the mechanics live in §5. As of 2026, base spot fees on the large venues mostly sit in the 0.05–0.1% range per trade and fall with 30-day volume, and on the top VIP tiers the maker rate approaches zero, which means the largest flows effectively trade for free in exchange for supplying liquidity. Schedules change often enough that the fee page of the specific exchange beats any memorized number. :::

:::p{html} Everything above the fees :::

Margin lending, derivatives, fiat on- and off-ramps, staking products, custody, listing services, launch platforms, and stablecoin economics each add revenue lines. For the industry at large, an exchange is less a marketplace than a financial supermarket built around one, and the trading screen works as the storefront that brings customers to the higher-margin aisles.

:::p{html} What a public company shows :::

Coinbase, as a listed company, publishes the anatomy every quarter, and Q2 2026 makes the supermarket thesis measurable. Total revenue was $1.2 billion, with transaction revenue at $599 million and subscription and services at $555 million, a record 48% of net revenue against 29% in Q4 2024. Bitcoin-related transactions provided only about 12% of total revenue, average USDC held in Coinbase products reached $20 billion, and prediction markets crossed $100 million in annualized revenue. All of that happened in a quarter when industry spot volumes fell roughly 25%, while Coinbase's share of crypto trading hit a record 10.3%. A business once described as a cash register attached to the BTC chart now demonstrably isn't one. :::

:::further{label="Further Reading"}

• DOJ: Samuel Bankman-Fried Sentenced to 25 Years in Prison | justice.gov
• Coinbase Q2 2026 Results (Form 8-K) | sec.gov :::

:::quiz{id=q-5-4-s1 label="Check yourself" correct=1 explanation="Trades inside a CEX update the internal database, not the blockchain. The chain only gets involved when you deposit or withdraw. That's why internal trades are instant and gas-free, and also why your balance there is a claim against the company rather than coins at an address you control."} You bought 0.5 BTC on a CEX and sold it two hours later at a profit. How many Bitcoin transactions did this trading produce on-chain?

• Two, the network recorded one transaction for the buy and one for the sell
• Most likely none, both trades changed rows in the exchange's internal ledger
• One, because buy and sell get batched into a single transaction
• It depends on how the BTC price moved between the buy and the sell :::

:::section{mins=8} How to Judge an Exchange :::

:::lead In Q2 2026 the top ten spot CEXes, where the coins themselves change hands, traded $1.95 trillion, and Binance alone took 38.7% of it. Impressive, but volume answers only one question, where trading is easy. Whether your money survives the company's bad quarter is a different question, and the industry's graveyard is full of exchanges that had plenty of volume right up to the end. :::

:::p Start with the size of the graveyard. Mt. Gox announced in 2014 that about 850,000 BTC were missing, and creditors waited over a decade for partial repayments. QuadrigaCX collapsed in 2019 owing clients around C$190 million after its founder died, and investigators later found most of the money had been lost in trading well before his death. Celsius froze withdrawals in 2022 owing users about $4.7 billion. FTX added roughly $8 billion the same year. Different decades, different failure modes, one common thread. Users judged these companies by how busy they looked. :::

:::p{html}

:::p So what does predict survival better than volume? Four things are checkable from the outside: :::

:::p{html}

• The legal entity. One brand often operates through different companies under different rules in different countries, so the question is which company signs the agreement with you and what license it holds in your jurisdiction. Regulatory history counts here too. In November 2023 Binance paid a $4.3 billion settlement to U.S. authorities over anti-money-laundering and sanctions violations, and its founder CZ served a four-month sentence.
• Custody and transparency. What the venue discloses about reserves and liabilities, whether addresses are verifiable, whether independent financial reporting exists. We take this layer apart below.
• Incident history. The interesting part is the reaction. Bybit lost about $1.5 billion to a hack in February 2025, covered client balances from its own funds, and its market share recovered within months, which tells you more than an unbroken record would.
• Execution quality. Spread and book depth on your pairs rather than the advertised daily volume, a topic for §6.

:::p Aggregators try to compress this into one number. In May 2026 CoinGecko rebuilt its Trust Score around exactly this logic. Liquidity carries half the weight, and the rest goes to cybersecurity, regulation, incident history, and Proof of Reserves. A useful filter, but a relative grade on a curve, not a guarantee. :::

:::p It also helps to picture how an exchange physically stores coins. The standard architecture splits funds into a hot wallet, a small share connected to the internet to process withdrawals, and cold storage, where the bulk sits behind multisig or MPC (multi-party computation), the key split across several people and machines. Some venues hand custody to third-party custodians. The Bybit hack from the incident list above exposed the weak point of this scheme, the attack hit not the cold wallet itself but the signing procedure for a transfer out of it, and the signers approved a doctored transaction. Cold storage protects keys from theft, yet every transfer out of the cold remains a moment of exposure. :::

:::p Now the transparency layer. Proof of Reserves (PoR) lets an exchange show that certain assets existed under its control at a snapshot moment, and some schemes let each user verify that their balance was included in the liability set. That is far better than a bare claim that the money exists. It is also far less than an audit. The U.S. audit regulator PCAOB warned back in 2023 that such reports may omit liabilities, claims on the assets, borrowed coins, and anything that happened after the snapshot date. A correct PoR still doesn't prove that tomorrow's withdrawals will clear. :::

:::p{html} The word "insurance" hides a similar trap. Coinbase, for example, explains that crypto assets get no FDIC protection at all. For U.S. clients, pass-through FDIC coverage may apply to part of the dollar balance if it actually sits as cash in eligible bank accounts and the program's conditions hold. Insurance always insures a specific list of events for a specific asset class, and the fine print is where that list lives. If an exchange's marketing leans on safety words this hard, the habits from 4.1 Scams apply to glossy institutions too. :::

:::callout Judging a winery by how many bottles it ships. The tasting room is packed, trucks leave the yard every morning, the brand is on every restaurant list. None of that tells you what's in the cellar. Down there, among cool stone and the smell of oak, the questions change. How many barrels are actually here, which of them are already pledged to the bank, which are sold forward two vintages ahead. A tour of the cellar with a flashlight, that's Proof of Reserves, and it beats taking the owner's word. But the flashlight doesn't show the loan contracts in the office upstairs, and it doesn't show whether the barrels will still be here in spring. Shipping volume measures popularity. Solvency lives in documents nobody shows at the tasting. :::

:::keypoints{label="KEY POINTS"}

• Volume is a metric of the matching engine; solvency is a metric of the custody machine. An exchange can print record turnover while already hollow inside, because internal trades are ledger rows and need no real assets to exist.
• The legal entity decides more than compliance formalities. Courts, arbitration, and applicable law follow the company's registration, so a dispute with an offshore entity plays out under offshore rules — and the entity's name sits in the Terms of Service, one minute away.
• A single PoR is weaker than a series. Borrowing coins to pass one snapshot is cheap; borrowing them for every monthly publication gets expensive, so the credibility of reserves grows with regularity rather than with the size of any one report.
• No solvency signal comes with a guarantee, and all of them can fail at once. Concentrating an entire deposit on one venue stacks those errors into a single point of failure, while spread across several venues, any one collapse shrinks from a catastrophe to a nuisance. :::

:::details{badge=Advanced title="ZK proofs of reserves: what the math covers and what it can't"} :::p{html} First wave: the Merkle tree :::

:::p{html} The post-FTX construction hashed every user balance into a leaf, built the leaves into a Merkle tree, and published the root together with a list of exchange addresses. Any user could recompute their own leaf and the path to the root, proving their balance was included in the liability set, the same inclusion-proof mechanics that 1.2 introduced for blocks. Two flaws pushed the design further. The tree leaked aggregate information, and, worse, nothing in plain Merkle math stops an operator from quietly counting some accounts with negative balances, shrinking the declared liabilities. :::

:::p{html} Second wave: zero-knowledge :::

Zero-knowledge proofs patch both holes at once. As of 2026, Binance publishes zk-SNARK-based proofs on top of its Merkle tree and OKX publishes zk-STARK-based ones, both on a roughly monthly cadence. The math proves statements like "the sum of all user balances equals X, and no account was counted with a negative balance" without revealing a single individual balance. Regularity matters as much as the technology, since borrowing coins to dress up one snapshot is cheap, while borrowing them for every monthly publication becomes an expensive habit.

:::p{html} What the math can't see :::

A proof cannot look outside its own inputs. If a material liability never entered the model, a bank loan, an obligation of a sister company, a court claim, the proof stays perfectly valid and perfectly silent about it. This is the substance of the U.S. audit regulator PCAOB's 2023 advisory, such reports may omit liabilities, third-party claims on the assets, borrowed coins, and everything after the snapshot date. ZK upgrades the verifiability of what's declared. Whether the declaration covers the whole company is still an accounting question, which is why PoR, published liabilities, independent financial statements, and rules on segregating client assets together tell more than any one of them alone. :::

:::further{label="Further Reading"}

• CoinGecko: Trust Score Methodology | coingecko.com
• PCAOB Investor Advisory: Proof of Reserve Reports | pcaobus.org :::

:::quiz{id=q-5-4-s2 label="Check yourself" correct=3 explanation="PoR is a snapshot of declared assets and, in the better schemes, of the liability set. It says nothing about undeclared liabilities, pledges on the assets, borrowed coins, or anything after the snapshot date. PCAOB's 2023 advisory makes exactly this point. Solvency needs liabilities and rights over assets, not just an asset photo."} An exchange publishes a fresh Proof of Reserves showing assets that cover all user balances. What does this NOT prove?

• That the listed assets existed under the exchange's control at the snapshot
• That your balance was included in the liability set when user checks exist
• That the sum of user balances matches the declared total
• That the company is solvent and that withdrawals will still work tomorrow :::

:::section{mins=5} The Order Book: Where Price Comes From :::

:::lead The best buyer on a market is offering $99.95, the best seller wants $100.05. Ten cents apart, no trade, and yet your screen confidently shows "the price". Where does that number come from when nobody is trading at it? The order book answers this, and the answer explains half of everything else in this card. :::

:::p The order book stores limit orders from both sides. A limit order is a request with a price written in advance, ready to buy at $99.95 or ready to sell at $100.05. It doesn't demand an immediate trade; it joins the book and waits until someone accepts its price. Bids are prices at which people stand ready to buy, asks are prices at which they stand ready to sell. The best bid sits on top of the buy side, the best ask on the bottom of the sell side, and the gap between them is the spread. In our example the spread is $0.10, and as long as nobody crosses it, the book just sits there, humming with intentions. :::

:::p{html} A trade happens when someone stops waiting. A new order crosses the other side's price, and most often it's a market buy order. A market order carries no price of its own; it asks for execution right now at the best available terms, which is why it starts consuming the cheapest asks. The full system of order types waits in §5. If the first price level holds too little volume, the remainder walks deeper into the book and fills at progressively worse prices. This is why "how much is available near the top" matters as much as the top price itself, a theme §6 develops into slippage. :::

:::p{html} That leaves the number an exchange displays in large type above the chart as the current price. Usually it's the last trade, an index averaged across several venues, or the mid price, the midpoint between the best bid and best ask. Convenient reference points, all of them, but none is an objective "fair value" of the asset. On a DEX the whole mechanism gets replaced by an AMM formula, a story told in 5.5 DEX & AMM. :::

:::p{html}

:::callout An auction house on a quiet afternoon, between hammer strikes. The auctioneer's ledger holds absentee bids for the painting, one collector left $18,000, another $19,500, a third $19,800. The seller, meanwhile, set a reserve of $21,000, below which the lot doesn't sell. Everything is written down, nothing is happening. The room smells of dust and varnish, the clerk turns pages. Then a buyer in the third row raises a paddle at $21,000, the reserve is crossed, the hammer falls, and only now is there a price. The ledger of standing offers is the order book. The gap between the highest written bid and the seller's reserve is the spread. The hammer is a trade, and it only strikes when someone agrees to the other side's terms. :::

:::keypoints{label="KEY POINTS"}

• The impatient side makes the price. Whoever crosses the spread prints the last trade, so a price chart is a tick-by-tick history of someone's urgency, not a curve the market draws by itself.
• On most spot CEXes, inside one price level time decides. Orders queue first come, first served, so the market can touch your price, fill those ahead of you, and leave — the right price alone doesn't guarantee your turn ever comes.
• On an illiquid pair the "last price" can be minutes old and one small trade away from anywhere. Alerts, stops, and liquidation marks tied to a single venue's last trade inherit that fragility in full.
• How many levels of the book an order will consume is visible before it's sent. That number decides the final price of the trade, and it's the one thing a price chart cannot show. :::

:::details{badge=Advanced title="Level 1, Level 2, and what order flow does and doesn't predict"} :::p{html} Data levels :::

Level 1 shows only the best bid and best ask with their sizes, which is what most retail widgets display. Level 2 opens more of the book and reveals depth, where the real information lives. Professional feeds add the trade tape and order-by-order updates, so an algorithm sees not just the state of the book but every change to it, additions, cancellations, and executions with timestamps.

:::p{html} What flow predicts :::

Microstructure studies do find that aggressive flow imbalance correlates with the next price move, on horizons of seconds. That is the edge high-frequency strategies actually harvest, and it decays almost immediately, because acting on the signal consumes it. Nothing in this literature supports reading a book imbalance as a forecast for the next hour, and stretching a seconds-scale signal to a day-scale decision is how the spread gets fed.

:::p{html} Why the visible book lies :::

:::p{html} The book is neither complete nor binding. On electronic markets the overwhelming majority of placed orders are cancelled rather than executed, because market makers from §4 reprice continuously as conditions move, so the book turns over many times between two trades. Large interest hides in iceberg orders showing a fraction of true size, or sits off the book entirely at OTC desks. And a fat wall of bids may stand precisely in order to be seen, which §8 revisits under the name spoofing. ::: :::

:::further{label="Further Reading"}

• Binance Academy: What Is an Order Book and How Does It Work? | academy.binance.com :::

:::quiz{id=q-5-4-s3 label="Check yourself" correct=0 explanation="A limit sell at $100.00 doesn't cross the best bid of $99.95, so no trade occurs. It joins the book as the new best ask, narrowing the spread from $0.10 to $0.05. Trades only happen when an order crosses the other side's price. Selling "at market" here would mean accepting $99.95, and the mid price is a reference point, not a place where orders magically meet."} The best bid is $99.95 and the best ask is $100.05. A new limit sell order arrives at $100.00. What happens?

• It joins the book as the new best ask, the spread narrows to $0.05, no trade yet
• It executes immediately against the best bid, and the seller receives $99.95
• It executes at the mid price of $100.00 because that's the fair value
• The exchange rejects it because limit prices inside the spread are not allowed :::

:::section{mins=5} Market Makers: Paid to Stand on Both Sides :::

:::lead A quote of $100.00 to buy and $100.02 to sell means somebody agreed to stand on both sides at once, around the clock. Wintermute alone runs over $10 billion of daily volume across more than 60 venues as of 2026. Why would anyone volunteer for a job where every counterparty might know more than you? :::

:::p The job is called market making. The firm keeps standing quotes, its posted buy and sell prices, on both sides of the book and manages its own inventory of the asset. On a calm market it buys slightly below and sells slightly above, and the spread pays for the service. But the spread is not guaranteed profit. If the price drops right after a purchase, the firm holds a depreciating inventory. And if a better-informed trader consistently deals with you just before the market moves, you lose on every such fill. That losing pattern has a name, adverse selection, and pricing it is most of the craft. :::

:::p{html} So the work is much more than posting two numbers. The firm tracks its position in real time, widens or narrows quotes, resizes them, hedges on other venues, and moves inventory between markets. Some of the informed flow it fears comes from MEV-style strategies, the world of 5.6 MEV. Among active crypto liquidity providers in 2026 are Wintermute, GSR, and Cumberland, named here as examples rather than a ranking. :::

:::p{html} There's also a cautionary tale about what happens when the market maker and the exchange share an owner. Alameda Research was a trading firm founded by Sam Bankman-Fried, FTX was his exchange, and the 2022 collapse revealed that customer funds had flowed to Alameda. A market maker profits from flow it trades against, an exchange holds custody of that flow, and putting both under one roof removed the wall that was supposed to separate them. Exchanges since then have been much more eager to show that separation exists, which loops back to the transparency toolkit from §2. :::

:::callout A scrap metal yard on the edge of town runs exactly like a market maker. Two prices are chalked on the board by the scales, copper bought from the public at $7.20 a kilo and sold on to the smelter at $7.60. Those forty cents are the spread, the fee for being open to anyone who walks in, a plumber with offcuts of copper pipe or a demolition crew with a full truck. The margin only looks easy from a distance. A warehouse full of copper is a bet on the copper price, and if it slides while the stock waits for the smelter run, a month of earnings evaporates in a day. Now the nastiest part. A dealer who watches metal quotes learns about tomorrow's drop before the yard owner does, and today of all days he delivers three tons at the old price. The yard has bought dear what turns cheap tomorrow, and that is adverse selection. Which is why, at the sight of suspiciously large lots, the owner lowers his buying price and cuts the size he accepts, not out of spite, but because the person across the scales may be trading on tomorrow's news. :::

:::keypoints{label="KEY POINTS"}

• A tight spread on a major pair doesn't occur naturally; market makers maintain it because it pays them to. The moment the pay disappears or the risk jumps, the firms step back, quotes drift apart, and the same market becomes expensive to trade.
• Quote width is the market maker's running estimate of how likely the next counterparty knows more. The spread works as an insurance premium against informed flow, which is why it widens around news before anything visibly happens.
• Liquidity is not a property of the market but a per-second decision by firms. In stress a rational market maker widens quotes and cuts size, so liquidity evaporates exactly when it's needed most.
• Nearly every exchange trades for its own account; the question is against whom. Hedging house risk is normal, trading against client flow on the venue's own data is the conflict, and from outside the two are almost indistinguishable — which is why the §2 questions about entity and disclosure carry the weight here. :::

:::details{badge=Advanced title="How market-making deals are structured"} :::p{html} On the exchange :::

The mildest form is fee economics, reduced fees or rebates for orders that add liquidity. Designated programs go further, the venue or a token issuer agrees on measurable targets, maximum spread, minimum depth, minimum share of time in the book, and the firm reports against them. These deals mostly align incentives, the firm earns from providing exactly the liquidity everyone else consumes.

:::p{html} Loan plus option, with numbers :::

Token projects often can't pay cash, so they hand the market maker inventory instead, and the common shape is a loan with a call option attached. A stylized example shows the incentives. A project lends 1 million tokens for 12 months, and the firm receives the right to buy them at 125% of the listing price. If the token rises above the strike, the firm exercises and keeps the difference; if it doesn't, the tokens go back. On paper this pays for liquidity without cash. In practice the firm now holds a cheap call, and its economics reward a price above the strike far more than a tight spread, which is exactly why "market making" in a term sheet can mean anything from neutral liquidity provision to a broad trading mandate over the project's own token. Token teams have learned to read that clause the expensive way.

:::p{html} Where the industry is heading :::

:::p{html} After FTX, disclosure became part of the product, exchanges publish more about their own trading desks and their separation from custody, and token deals increasingly specify what the market maker may and may not do. A separate line for the record books, in August 2026 Wintermute registered a U.S. broker-dealer with the SEC and FINRA, a crypto-native market maker formally entering regulated equities. The convergence thread continues in §9. ::: :::

:::further{label="Further Reading"}

• Wintermute Launches U.S. Broker-Dealer | wintermute.com :::

:::quiz{id=q-5-4-s4 label="Check yourself" correct=2 explanation="The two structural risks are inventory risk and adverse selection. A falling price devalues whatever the firm just bought, and informed counterparties systematically trade with the quotes that are about to become stale. The spread compensates for these risks on average; on bad days it doesn't. There are no doubled fees, no regulator-fixed spreads, and no minimum holding periods."} Why can a market maker lose money even while "earning the spread" on every round trip?

• Because exchanges charge market makers double fees, one for each side of the quote
• Because the width of the spread is fixed by the regulator and lags behind the rising costs
• Because inventory loses value as the market moves, and informed players hit the quotes just before it
• Because exchange rules require market makers to hold purchased positions for at least 24 hours :::

:::section{mins=6} Orders, Spot, Margin, and Derivatives :::

:::lead The Buy button hides a choice with real money attached. Buy 1 ETH at market and you're asking for execution now at whatever prices the book offers. Set a limit at $2,000 and you control the price but lose any promise the trade happens at all. Every order type trades one certainty for another, and so does every market type stacked on top. :::

:::p Three basic order types cover most buttons you'll meet. A market order prioritizes execution over price. A limit order sets the worst price you accept, buy no higher, sell no lower, and waits in the book if nobody meets it. A stop order sleeps until a trigger price is reached, then turns into something else, and that "something else" is where people get hurt. A stop-market becomes a market order and can fill far from the trigger during a sharp move. A stop-limit becomes a limit order that controls price but is vulnerable to a gap. A gap is a jump where price leaps over a level with no trades at the values in between, and a limit order left above such a jump may simply never fill. So a "stop-loss" is a mechanism, not an insurance policy, and which mechanism you picked decides what happens on the worst day. :::

:::p{html} The market type changes risk again. Spot means immediate exchange, you pay and receive the asset itself right away, so the worst case on spot is bounded by the price of what you bought. Margin trading adds the exchange's borrowed funds to your own money, and your own share serves as the collateral, which is what the margin actually is. Gains and losses scale together, and a deep enough move ends in liquidation, the forced closing of the position. A futures contract works differently, it's a contract tied to the asset's price without holding the asset itself, and a dated future expires on a set day. A perpetual futures contract never expires and instead uses funding, periodic payments between longs and shorts, that is between bets on the price rising and falling, which pull the contract's price toward spot. Perpetuals dominate crypto derivatives volume, and in Q2 2026 the top ten perpetual CEXes handled $12.7 trillion, more than six times the spot figure. Leverage and its liquidation mechanics get a full treatment in 6.6 Derivatives. :::

:::p{html}

:::callout A smart home on a variable electricity tariff. The boost button on the water heater is a market order, hot water now, at whatever the grid charges this minute. The washing machine is set differently, run only when electricity is under 20 cents, a limit order that might wait until 3 a.m. or, on an expensive week, never run at all. And the frost guard in the garage is a stop, asleep for months, armed to act the moment the thermometer dips below 5 degrees. Set it to "heat at any cost" and a brutal cold snap meets a brutal bill, the stop-market way. Set it to "heat only if power is cheap" and one icy night the pipes may lose the bet, the stop-limit way. Same house, same buttons, very different worst days. :::

:::keypoints{label="KEY POINTS"}

• The simple buy mode on exchanges hides the order type and usually fills at market with a wider effective cost; the advanced interface hands the choice back to you. The difference between the two modes is paid in spread and conversion fees.
• A stop fires on a touch, not on a "real" move. On a thin pair one wick through an obvious level, a round number or a recent low, collects the stops parked there and lets price come back — so where you place a stop matters as much as whether you place one.
• Leverage is asymmetric in time. A spot position can wait out a drawdown; a leveraged one gets liquidated before the thesis has time to come true, so on margin "right, but later" settles as "wrong".
• The choice between stop-market and stop-limit comes down to which failure is more acceptable, a bad fill or no fill. There is no universal answer, and that single choice makes up most of practical risk control on a CEX. :::

:::details{badge=Advanced title="Maker, taker, and the execution toolbox"} :::p{html} The fee grid :::

Exchanges price orders by their effect on liquidity. An order that rests in the book adds liquidity and pays the maker fee, an order that consumes resting orders pays the taker fee, which is usually higher. Base spot rates on large venues sit around 0.1% and fall along VIP tiers driven by 30-day volume, down to near-zero maker rates at the top. Note the terminology trap, a limit order is not automatically a maker. If your limit crosses an existing quote, it executes immediately as a taker, and the post-only flag exists precisely for this, cancelling the order instead of letting it take.

:::p{html} Time in force and safety flags :::

Every order carries a time-in-force setting. GTC (good till cancelled) waits indefinitely, IOC (immediate or cancel) fills whatever it can right now and drops the rest, FOK (fill or kill) executes entirely or not at all. On derivatives, reduce-only guarantees an order can shrink a position but never accidentally open or flip one. OCO (one-cancels-other) links two orders so that filling one cancels the other, the standard way to hold a take-profit and a stop simultaneously, and a trailing stop follows the price at a set distance, converting an unrealized move into a trigger level that ratchets along.

:::p{html} Execution algorithms :::

:::p{html} Large orders get sliced. TWAP cuts volume evenly over time, an order of $1.2 million over four hours becomes roughly $5,000 per minute, small enough for each slice to hide inside normal flow. VWAP follows the market's own volume pattern instead, trading more when the market trades more. Iceberg shows only a fraction of the true size in the book and refills as slices execute. All of them manage the order's market impact from §6; none of them removes it, and an algorithm stretched over four hours accepts four hours of market risk in exchange. ::: :::

:::further{label="Further Reading"}

• Binance Academy: Understanding the Different Order Types | binance.com :::

:::quiz{id=q-5-4-s5 label="Check yourself" correct=3 explanation="A gap through the trigger activates the stop, which places a limit sell at $2,395. But the market is already at $2,300, and nobody buys at $2,395 when the going price is lower. The order rests unfilled, and the position is still open with a deeper loss. A stop-market would have sold, at some unpredictable price on the way down. That's the trade-off, price control versus execution certainty, and gaps are where it bites."} ETH trades at $2,500. You set a stop-limit sell with trigger $2,400 and limit $2,395. Overnight the price gaps from $2,450 straight to $2,300. What most likely happened to your position?

• It sold at exactly $2,400, a stop executes at its own trigger price
• It sold at $2,395, a limit price guarantees execution at itself
• It sold at $2,300 because stop orders always execute at the market price
• The stop triggered, but the $2,395 limit sits unfilled while the market trades at $2,300 :::

:::section{mins=4} Spread, Slippage, and Depth :::

:::lead Two venues show the same last price, you send the same $100,000 order to each, and the results differ by hundreds of dollars. Nothing malfunctioned. One book was deep and absorbed the order near the top quote, the other was thin and your own buying pushed the average fill price up. The difference has names, and they're measurable before you click. :::

:::p{html} Three terms, three different things. The spread is the distance between the best bid and the best ask. It is exactly what someone loses by buying at market right now and selling the minimum size straight back. Price impact is what your own volume does as it walks through available liquidity, the effect from §3 scaled to your order. Slippage compares the price you expected with the price you got, so it includes your impact plus whatever the market did between sending and execution. :::

:::p{html} How do you see liquidity before trading? Daily volume alone won't tell you. More useful is the pair of spread and depth within a band around the current price, many data services quote depth within 2% of mid for exactly this reason. The size of the spread itself works as a quick thermometer of the market. On the most liquid pairs of the largest venues the BTC spread is measured in hundredths of a percent, on small altcoins and weak venues it stretches to whole percents, and the same pair showing different spreads on two exchanges instantly reveals where the market is more alive and where the market makers of §4 have gathered. This is the execution quality that §2 proposed as one of the four yardsticks for an exchange, next to reserves. The thermometer's reading still gets checked against depth, because a tight spread over a nearly empty book flatters small orders and punishes large ones, so execution quality is a property of a specific size, not of the pair in the abstract. :::

:::p{html} On a DEX the same question routes through different plumbing. The result depends on the AMM curve of the specific pool, the pool's state at the moment your transaction lands in a block, and MEV, where a sandwich bot can manufacture your slippage on purpose. The mechanics live in 5.5 DEX & AMM and 5.6 MEV. :::

:::callout A power grid on a hot afternoon. Base demand is served by the big hydro dam, the cheapest electricity there is, humming away at 3 cents. As air conditioners switch on across the city, dispatch starts waking up pricier plants, gas turbines at 8 cents, then the old diesel peakers at 20, their fans roaring in the heat. Boil one kettle and you pay the hydro rate, nobody notices you. Now connect an aluminum smelter. Your own demand marches up the supply curve, and the average price of your power rises with every megawatt, because you personally forced the expensive generators online. Depth is how much cheap capacity exists at each price step. Impact is what happens when your size outgrows a step. :::

:::keypoints{label="KEY POINTS"}

• Spread and depth both measure liquidity, but spread is far cheaper to fake. Keeping two tiny quotes close together costs almost nothing, while filling a book with real size puts capital at risk — so of the two signals, depth is the honest one.
• Price impact depends on your order size relative to available liquidity, which makes it a property of the specific order in the specific book. The same pair is liquid for $1,000 and illiquid for $1,000,000.
• The slippage-tolerance field in DEX and swap-widget interfaces is not a forecast but a damage limit you authorize in advance. Anything inside it executes silently, so a generous tolerance on a thin pair works as a pre-signed check for its full amount.
• Ten seconds in the order book routinely saves more than fee optimization ever will. Fees are measured in tenths and hundredths of a percent; bad execution on a thin pair is measured in whole percents. :::

:::details{badge=Advanced title="Kyle's lambda: impact as a coefficient"} :::p{html} The model :::

Market microstructure formalizes impact with Kyle's lambda, from Albert Kyle's 1985 model. In simplified form, the price change is proportional to signed volume, and the coefficient λ measures how strongly price reacts to additional flow. High λ means a fragile price, low λ means a market that absorbs flow calmly, and empirical work estimates λ from trade and quote data rather than taking it from theory.

:::p{html} The empirical square root :::

:::p{html} Real markets bend the straight line. A result confirmed repeatedly on equities, futures, and crypto, known as the square-root law, says impact grows roughly with volatility times the square root of order size over daily volume. Two consequences are worth carrying around. An order near 1% of a pair's daily volume, at a daily volatility of 3%, moves price on the order of tenths of a percent, and because of the square root, quadrupling the order size only doubles the impact, while splitting an order across time or venues cuts it. That arithmetic is the entire reason the execution algorithms of §5 exist. :::

:::p{html} Orders of magnitude :::

Aggregated depth numbers put scale on the picture, with dates attached. By Kaiko's data, Bitcoin's 2% market depth across 33 centralized exchanges stood near $539 million in February 2024, still below the $800+ million seen before the FTX collapse in November 2022. The structure proved more stable than the level, in Kaiko's 2023 concentration study eight venues held over 90% of global market depth, and 72% of ETH depth sat on just five exchanges. Depth numbers float with the cycle; the concentration of liquidity in a handful of books is the durable fact, and it's why one venue's outage instantly makes the whole market feel thin. :::

:::further{label="Further Reading"}

• Kaiko Research: Bitcoin Liquidity Rebounds as Price Soars | research.kaiko.com :::

:::quiz{id=q-5-4-s6 label="Check yourself" correct=1 explanation="The spread prices the smallest trade; depth decides everything larger. A $50,000 market buy against $8,000 of nearby liquidity walks deep into the book, filling at progressively worse prices, and likely moves the pair's printed price in the process. Exchanges don't reject oversized orders, they execute them expensively. And slippage exists anywhere execution can differ from expectation, CEX and DEX alike."} A small altcoin pair shows a spread of $0.01, but the whole book within 2% of mid holds only $8,000. What does this mean for a $50,000 buy order?

• The tight spread guarantees the whole order fills near the current price
• The order will eat through the visible book and fill well above the quote
• The exchange will reject the order because it exceeds the depth visible in the book
• Slippage will be zero, since it never happens on a CEX and is a DEX property :::

:::section{mins=5} Why Crypto Has No Single Price :::

:::lead Bitcoin has no central board that every venue must copy. The same BTC trades on hundreds of CEXes, on DEXes, at OTC desks, and in derivatives, all at once, and at any given second the quotes disagree slightly. In July 2026 spot volume on top DEXes reached 24% of top CEX volume, an all-time high, so the disagreement now spans two different kinds of market plumbing. :::

:::p Why doesn't the disagreement grow without limit? Arbitrage. A price gap between venues is an invitation to buy where it's cheaper and sell where it's dearer, and the traders who accept the invitation pull the prices back together. But arbitrage is not free. Fees, transfer times, capital parked on both venues, withdrawal limits, and the risk that the gap closes mid-operation all eat into it. So small gaps persist, because closing them costs more than they're worth. The gap you see is roughly the market's estimate of the cost of moving money between those two places. :::

:::p{html} The DEX side of the split deserves its own number. By CoinGecko's data, monthly DEX spot participation roughly doubled in two years, from 6.9% of the market in January 2024 to 13.6% in January 2026, and the DEX-to-CEX ratio touched 24.1% in July 2026. Any figure for the CEX-to-DEX split is therefore a photograph with a date stamp, not a constant. Which venues quote which stablecoin pairs matters too, since most crypto trading prices against USDT or USDC rather than dollars in a bank, a nuance 5.2 Stablecoins explains. :::

:::p{html} One more piece of plumbing hides from the public books entirely. OTC (over-the-counter) desks arrange large trades directly between two parties, at a negotiated price anchored to the market but executed without pushing an eight-figure order through a public book. The buyer avoids the price impact from §6, the seller avoids advertising intent, and public volume statistics quietly miss the whole thing. :::

:::callout Honey fairs in three neighboring towns on the same Saturday. Same linden honey, same harvest, but each square has its own crowd, its own stalls, its own weather. In one town a jar goes for 9 euros, in the next for 10.50. A dealer with a van notices, loads up in the cheap square, sells in the dear one, and by afternoon the gap narrows to the price of his diesel, tolls, and the risk that prices move while he's on the road. The gap never quite reaches zero, it settles at his cost of doing the round. Meanwhile a restaurant chain that needs 300 jars doesn't visit any square at all. It calls the beekeeper directly, agrees on a price near the fair's level, and the fair statistics never learn the town's biggest honey deal happened. :::

:::keypoints{label="KEY POINTS"}

• Each venue is its own auction with its own crowd, so at any moment "the BTC price" is a family of close numbers. A single number only appears inside someone's averaging methodology.
• The width of cross-venue gaps measures the friction of the whole system. Cheaper capital, faster transfers, and more trusted venues pull prices tighter, so gaps shrinking over the years amount to an infrastructure report card nobody has to publish.
• DEX-share figures differ across reports for structural reasons, not sloppiness. You can count pairs, venues, or whole chains, and every provider filters wash volume differently, so only numbers from one methodology are comparable with each other.
• An OTC desk exists because a public eight-figure order is a signal the market trades against before the fill completes. Privacy of execution is part of the price, and size pays it willingly. :::

:::details{badge=Advanced title="Indexes, premiums, and where price discovery lives"} :::p{html} How indexes are built :::

Because the market is fragmented, serious systems rarely trust one venue's last trade. Derivatives platforms mark positions and compute liquidations against indexes assembled from several constituent exchanges, with eligibility criteria, weights, and outlier handling, so a single bad print on one venue doesn't cascade into forced closures everywhere. Price discovery, the question of where new information turns into price first, migrates between venues with time of day, liquidity, and where fresh flow arrives, so "venue X always leads" is too strong without a defined period and a study behind it.

:::p{html} Kimchi premium in numbers :::

:::p{html} The Korean premium is the best-documented case of arbitrage costs taken to the extreme. A University of Calgary study measured an average premium of 4.73% between January 2016 and February 2018, with a peak of 54.48% in January 2018; Arcane Research's series records about 63% in May 2017 and a return to roughly 23% in April 2021. The premium is two-sided, in February 2021 it sat at a discount of −6.5%. And on May 19, 2021, the crash day from §9, it spiked to 20.8%. The mechanism is capital controls, moving won abroad at scale is slow and restricted, banks resist crypto-linked transfers, so the arbitrage loop that would flatten the gap can't close, and local demand prices itself. :::

:::p{html} Coinbase premium :::

A milder sibling lives between Coinbase's USD price and Binance's USDT price. CryptoQuant popularized it as the Coinbase premium, an indicator of relative demand pressure on the U.S. venue. In the ETF era the premium is sometimes used as an indirect proxy for fund inflows, since authorized participants source coins on regulated U.S. venues, but causality doesn't follow from a single number. Like any single-number gauge, it works better as context than as a trigger. :::

:::further{label="Further Reading"}

• CoinGecko: 2026 Q2 Crypto Industry Report | coingecko.com :::

:::quiz{id=q-5-4-s7 label="Check yourself" correct=2 explanation="A persistent gap is usually a priced-in cost, not free money. To capture $120 you'd pay trading fees twice, move funds between venues, hold capital on both, and carry the risk that the gap closes mid-flight. Professional arbitrageurs watch these gaps continuously; the ones that survive an hour are the ones not worth closing. Extreme cases like the kimchi premium show what happens when the closing cost becomes near-infinite."} BTC trades $120 higher on exchange A than on exchange B, and the gap has held for an hour. What's the most likely explanation?

• Exchange A shows the true bitcoin price, and exchange B's data is lagging or wrong
• Arbitrageurs haven't noticed the gap yet, and free money has sat on the table for an hour
• Closing the gap costs about the same $120, fees, transfers, capital, and risk eat the profit
• The exchanges run on different blockchains, so their assets aren't formally comparable :::

:::section{mins=5} Manipulation and Bad Market Data :::

:::lead In 2019 Bitwise examined reported Bitcoin volume across 81 exchanges and concluded that roughly 95% of it was fake or non-economic. That number describes 2019, not today's market, but the lesson survived the data. Reported volume is a claim somebody typed into an API, and claims can be manufactured. :::

:::p{html} Three manufacturing techniques cover most of what you'll encounter. Wash trading creates the appearance of activity through trades between the same or economically connected parties, volume without economic substance. Spoofing works on perception instead. A trader places orders without intent to execute, lets other participants read them as supply or demand, then cancels. Pump-and-dump schemes need the least infrastructure of all, organizers quietly build a position in a thin asset, manufacture attention around it, and sell into the demand they created. Understanding that structure matters more than memorizing a stage template, because real schemes improvise, and the thin markets they prefer are the same ones where §6 showed small money moves price a lot. The retail-facing cousins of these schemes live in 4.1 Scams. :::

:::p Where does that leave "crypto is unregulated"? As a blanket statement, it has expired. Under MiCA in the EU, firms professionally arranging or executing crypto trades must run market-abuse detection and report suspicious orders. In the U.S., manipulation cases have been brought under existing commodities and securities law. Enforcement is uneven across jurisdictions, and thin offshore corners remain thin and offshore, but the era when the whole market was one lawless zone is over. :::

:::p{html} For the reader the practical defense is data hygiene. Aggregators that filter or adjust reported volume give a more honest picture than a venue's self-reported numbers, which remain marketing until independently corroborated. And as §2 showed, volume was never a solvency metric to begin with. :::

:::callout A publishing house that wants a bestseller by Friday. The blunt method is buying its own print run, vans quietly ferry boxes from the warehouse through the tills of friendly bookstores and back, sales charts light up, readers see "national bestseller" on a book nobody has read. That's wash trading. The subtler method is announcing a monstrous pre-order from a fictitious retail chain, competitors reprice their launches around it, then the order quietly evaporates. That's spoofing. And for the full pump, the house buys up a forgotten novel for pennies, seeds glowing reviews and morning-show mentions, waits for genuine readers to storm the shops, and sells its warehouse into their enthusiasm. Different tools, one target, everyone else's beliefs about real demand. :::

:::keypoints{label="KEY POINTS"}

• Fake volume has customers. It lifts an exchange's rank in volume-sorted listings and gives a token the look of liquidity for a listing pitch, so as long as rankings sort by volume, demand for manufacturing it stays built into the system.
• Wash trading manufactures volume through economically connected trades; spoofing manufactures perceived supply and demand through orders meant to be cancelled. One fakes the past, the other fakes the future.
• A pump doesn't end at the peak, it starts there. Organizers need buyers to exit into, not a pretty chart, which is why the loudest marketing phase of a scheme coincides with their selling.
• A regulated venue gives a manipulation complaint a defined path, a procedure and a supervisory body. With an offshore one, courts and arbitration at its place of registration remain, but the path there is longer and hazier — a difference that gets chosen at deposit time and felt at incident time. :::

:::details{badge=Advanced title="How researchers read suspicious activity"} :::p{html} Statistical fingerprints :::

Authentic trading leaves regularities that fabricated trading struggles to imitate. In the trading data of organic markets, first-digit distributions often come close to Benford's law. Under that regularity, numbers starting with 1 occur roughly six times more often than numbers starting with 9. Substantial deviations are used as one diagnostic signal rather than proof, and bot-generated volume produces such deviations routinely. Humans round, so genuine markets show clustering at sizes like 0.5 and 1, while generated flow is either too uniform or too perfectly rounded. There is a trace in the very shape of the distribution too. On a live market most trades are small while the occasional one is very large, statisticians call that shape a heavy tail, and generated flow usually comes out too even, missing the natural rare giants.

The last two fingerprints look outward. Real activity attracts market makers, which tightens the spread and fills the book, so enormous reported volume next to a wide spread and an empty book contradicts itself. And on on-chain venues the blockchain shows where each wallet got its funds, so when dozens of "different" counterparties turn out to be funded from one source, a single actor stands behind them.

:::p{html} The Cong–Li–Tang–Yang numbers :::

The reference study here is "Crypto Wash Trading" by Cong, Li, Tang, and Yang, published in Management Science in 2023. Applying the same checks, Benford’s law, size clustering, and tail shapes, to data from 29 centralized exchanges, the authors found that regulated venues look statistically like ordinary financial markets, while on unregulated ones wash trading averaged over 70% of reported volume, approaching 80% on the second tier. Their estimate for the first quarter of 2020 alone exceeded $4.5 trillion of fabricated spot volume, and the paper documents the motive directly, fabricated volume measurably lifts an exchange's position in volume-sorted rankings. The Bitwise figure from the section body was not an anomaly of 2019 but one snapshot of a persistent industry.

:::p{html} The mirror caution :::

:::p{html} The same signals cut the other way. A high cancellation rate alone doesn't prove spoofing, because legitimate market makers from §4 also reprice constantly as the market moves, cancelling thousands of quotes a minute in the ordinary course of business. Intent separates manipulation from ordinary quoting, which is precisely what makes these cases hard to prosecute and easy to allege, and why serious research leans on converging evidence from several independent fingerprints rather than any single metric. ::: :::

:::further{label="Further Reading"}

• Cong, Li, Tang, Yang — Crypto Wash Trading (NBER w30783) | nber.org :::

:::quiz{id=q-5-4-s8 label="Check yourself" correct=0 explanation="Real volume and market quality travel together. Heavy genuine trading attracts market makers, tightens spreads, and builds depth. Volume claims that coexist with a wide spread and an empty book point to wash trading, exactly the mismatch Bitwise used in 2019 to separate the 5% of real Bitcoin volume from the rest. This is why researchers cross-check volume against spread and depth rather than reading it alone."} A small exchange reports daily volume comparable to Coinbase, but the spread on its BTC pair is five times wider and the order book is nearly empty. What's the most reasonable reading?

• The volume is likely manufactured, real activity of that size would tighten the spread and fill the book
• The exchange is simply more popular with traders who find wide spreads easier to trade
• High volume always comes with wide spreads, that is how all large markets work
• The book is empty because all traders use market orders there :::

:::section{mins=6} Crypto Markets vs TradFi :::

:::lead "Crypto settles instantly, TradFi takes days" stopped being accurate on May 28, 2024, when U.S. securities moved from T+2 to T+1 settlement. Meanwhile an "instant" fill on a CEX is a database row, not blockchain settlement. The honest comparison runs function by function, and each side wins some rows. :::

:::p First, the notation from the headline. The letter T stands for the trade date, and the number next to it shows how many business days later settlement happens, the moment securities and money actually change hands. Under T+2 a share bought on Monday became yours and paid for on Wednesday; under T+1 it does on Tuesday. :::

:::p Trading hours still differ visibly. Crypto venues run around the clock, while traditional exchanges hold regular sessions plus extended windows. Less visible but more consequential is what happens in a crash. U.S. market-wide circuit breakers halt trading at S&P 500 declines of 7%, 13%, and 20%. Crypto has no global equivalent, and May 19, 2021 showed the difference in action. Bitcoin fell about 30% intraday, liquidations across derivatives venues exceeded $7 billion in a day as leveraged positions force-closed into a falling market, and Binance's futures platform suffered outages mid-crash. Nobody paused anything, because there was nobody positioned to pause it. An individual venue can halt a product under its own rules, but a market-wide stop doesn't exist. :::

:::p{html} Custody separates the systems even more sharply. Crypto can move into self-custody, where the key holder controls the asset directly, no broker, no depository — a skill set covered in 3.1 Wallets. Keeping the same asset on a CEX reintroduces the counterparty risk this card opened with. Traditional markets spread client rights across brokers, custodians, clearing, and settlement infrastructure, with protections that depend on instrument and jurisdiction. Regulation follows the same pattern of partial convergence, MiCA in the EU requires authorization for crypto services and sets market-conduct rules, while other jurisdictions run different frameworks, so the same token can sit under different regimes on the same day. :::

:::p{html}

:::p And the border keeps blurring. Traditional firms sell crypto-linked products and tokenized assets, while crypto companies collect licenses and build custody and reporting that institutions recognize. Wintermute's U.S. broker-dealer registration in August 2026, a crypto market maker formally entering equities, is one tile in that mosaic. The systems aren't merging into one, but "regulated TradFi versus lawless crypto" now describes a historical period rather than the present. :::

:::callout Controlled airspace versus open sky. Around a major airport, radar screens glow in a dim room, every plane files a flight plan, departures queue for slots, and when a storm cell rolls in, a controller's calm voice grounds everything. Flights stop, passengers grumble in the terminal, nobody flies into the wall of cloud. Now picture the open sky over remote bush country. Anyone with a plane takes off whenever they like, midnight included, no slots, no clearances, total freedom. But when the storm comes, there is no voice on the radio and no one to close the sky. Each pilot decides alone, in the dark, with the instruments they brought. Both systems move people through the air. They differ in who, if anyone, can say stop. :::

:::keypoints{label="KEY POINTS"}

• TradFi's acceleration isn't finished. Behind T+1 the industry is already debating T+0 and settlement on tokenized rails, so "we're faster" is a melting advantage for crypto rather than a permanent one.
• A 24/7 market means risk events don't wait for your morning. A gap at 3 a.m. is a routine occurrence, and the stop orders from §5 stand watch in your place — or nothing does.
• No market-wide circuit breaker means the breaker's job moves to you. In TradFi the system enforces the pause during panic; in crypto the pause is your discipline, your leverage limits, and orders placed in advance.
• In practice the two systems have long been compatible rather than opposed. Brokerage and ETFs in TradFi, execution on a CEX, and storage in self-custody coexist in one portfolio, and the choice of infrastructure follows the task and the risks a user is willing to accept, not loyalty to one camp. :::

:::details{badge=Advanced title="Funding rates: the mechanism and the temptation to read it"} :::p{html} The mechanism :::

A dated future has an expiry, and the expiry itself pulls its price toward spot. A perpetual has no date, so the anchoring is done with money. First, two words we have only touched in passing. A long is the position of someone betting the price rises, a short of someone betting it falls. Funding is a periodic payment between these two camps, on most venues every 8 hours, and the market itself sets the direction. If the perpetual trades above spot, demand is tilted toward the longs, and for that interval longs pay shorts. Paying every 8 hours just to keep a position open is unpleasant, some longs close, buyers thin out, and the perpetual's price presses back toward spot. When the perpetual trades below spot, everything mirrors. In a calm market the rate sits near the baseline of about 0.01% per interval.

:::p{html} The arithmetic of extremes :::

:::p{html} Small numbers deceive here, because the payment repeats three times a day all year round. The 0.01% baseline per interval adds up to about 11% annualized, and a 0.1% reading per interval already means roughly 110% a year. That is the cost of merely staying in the long position, before earning anything on the price itself. In the spring of 2021 rates on major venues held several times above baseline for weeks, and that meant exactly one thing. Too many leveraged longs had piled up, and they were paying dearly for the right to stay. That positioning became the fuel of May 19, the falling price triggered liquidations, the forced sales pushed price lower, and the wave collected the next in line. The overheating showed on several independent gauges at once, the kimchi premium from §7 hit 20.8% that same day. :::

:::p{html} Who pulls the rate back :::

:::p{html} Extreme funding carries its own counterforce. While the longs are paying heavily, an arbitrageur can buy the coin on spot and simultaneously open a short in the perpetual for the same amount. The price direction doesn't matter to him, profit on one half of the trade cancels the loss on the other, while funding drips his way every 8 hours. In the hot stretches of 2021 and again 2024 that combination paid double-digit annualized returns, capital flowed in, the short side grew heavier, and rates compressed back toward baseline. This is also where the signal ends. Funding honestly shows which side the crowd is on and who pays whom, but not when it ends, rates can stay extreme for weeks while price grinds on. The continuation, basis and the liquidation math, waits in 6.6 Derivatives. ::: :::

:::further{label="Further Reading"}

• SEC Investor Bulletin: The New T+1 Settlement Cycle | investor.gov
• NYSE: Trading Information, Circuit Breakers | nyse.com :::

:::quiz{id=q-5-4-s9 label="Check yourself" correct=1 explanation="The instant position is a row in the exchange's database, execution without on-chain settlement. The blockchain gets involved when you withdraw, and only then does the asset move to an address you control. T+1 is the U.S. securities cycle, not a crypto rule, and block production doesn't settle trades that never touched the chain. This split between internal execution and on-chain settlement is the core difference this whole card is built around."} You buy BTC on a CEX and the interface shows the position instantly. At what moment does blockchain settlement of your purchase actually happen?

• At the instant of the trade itself, which is exactly why the position shows at once
• Only when the BTC is withdrawn to your own address, until then it lives in the exchange ledger
• One business day later, since the crypto market also adopted the T+1 cycle in 2024
• When the next Bitcoin block is mined, roughly ten minutes later :::