CEX vs DEX

Two structurally different venues exist for trading digital assets, and knowing which one you are looking at changes what a price even means. A centralized exchange (CEX) operates much like a traditional venue, an order book matching buyers and sellers, custody of assets held by the exchange itself (the custodial risk covered in self-custody vs custodial). A decentralized exchange (DEX) instead prices trades algorithmically against a pool of assets locked in a smart contract, with no order book and no counterparty beyond the pool itself. The dominant DEX design, the constant-product automated market maker (AMM), was formalized in Uniswap's v2 whitepaper (Adams, Zinsmeister, and Robinson, 2020): a pool holds two assets whose reserves multiply to a constant, x × y = k, and every trade shifts the reserves along that curve.

How much slippage does a constant-product AMM cause?

Work a pool numerically to see what that means for execution quality. A pool holding 1,000 ETH and 3,000,000 USDC (implying a spot price of $3,000/ETH) has k = 3,000,000,000. Swapping in $3,000 of USDC (about 0.1% of the pool, net of Uniswap's standard 0.3% fee) buys 0.9960 ETH, an effective price of $3,012.03: 0.40% slippage above spot. Scale the trade to $30,000 (1.0% of the pool) and the effective price rises to $3,039.03 (1.30% slippage); scale it to $300,000 (10% of the pool) and the effective price jumps to $3,309.03: 10.30% slippage. Slippage on a constant-product AMM does not grow linearly with trade size; it accelerates, because every unit you buy makes the next unit more expensive along the same curve. The practical consequence: pool depth, not just "is there a market," determines whether a given trade size is even sensible, a fact traditional order-book intuition does not automatically transfer, and a real, computable reason why "the app shows a price" and "you can transact at that price in size" are two different claims.

How do perpetual futures and funding rates work?

Perpetual futures (a crypto-native derivative with no listed equivalent, since traditional futures expire) solve the no-expiry problem with a recurring funding-rate payment exchanged between longs and shorts, keeping the perpetual's price anchored near the spot/index price. A perpetual trading at $61,000 against a $60,700 index carries a 0.4942% premium; if the exchange's funding rate for that interval tracks that premium (subject to a typical exchange cap, here illustratively ±0.75%), a $100,000 notional long position pays $494.23 to the short side for that single interval. Funding intervals recur (commonly every eight hours, three times a day on major venues), were that specific rate to persist for a full year, the naive annualized carrying cost would run over 500%, which will never actually happen at that level for long, but illustrates why persistently bullish perpetual positioning is not "free leverage": it has a real, recurring cost, and that cost is precisely what funds a legitimate cash-and-carry trade, short the perpetual, hold the equivalent spot, and collect the funding as a market-neutral yield, with the two legs' price risk offsetting by construction.

Contrast this with the regulated, dated futures market DV1.01 already taught you the cost-of-carry framework for: CME's Bitcoin futures expire, settle, and converge to spot through ordinary cost-of-carry arbitrage, no perpetual funding mechanism exists there at all, because the contract simply expires. This is also where India's regulatory gap becomes concretely visible to a trader rather than an abstraction: India has no domestically regulated venue offering crypto derivatives, so Indian traders wanting leveraged crypto exposure mostly reach offshore, unregulated perpetual-swap venues, while a US-based investor wanting the same directional exposure through a regulated, exchange-cleared instrument can simply use CME's dated futures. Same underlying asset, structurally different market-access risk by jurisdiction, a fact worth stating plainly rather than assuming both investors face the same instrument.

What do regulators say about crypto market quality?

One further market-quality point the standard-setters have said plainly and you should be able to cite rather than merely suspect: IOSCO's own November 2023 Policy Recommendations for Crypto and Digital Asset Markets names, as core concerns, conflicts of interest from vertically integrated venues (a single entity acting as exchange, broker, custodian, and market-maker at once, precisely FTX's structure) and market manipulation, including wash trading, as first-order problems this asset class's market structure has not yet solved. That is a regulator's own stated concern, not this module's editorializing.

Modern Data Analysis callout. Every number in the correlation and stress-test work is built on a clearly labeled synthetic 60-month return series, exactly as R3 requires when a live, licensed dataset is not pulled in-session. The Python-literate version of this exact exercise, done properly, replaces that synthetic series with real pulled data: a public exchange or data-provider API (many offer free tiers for historical daily/monthly OHLCV data) fetched with requests or a wrapper library, loaded into a pandas DataFrame alongside your usual equity/bond/gold benchmark series, and run through the identical .pct_change(), .corr(), .rolling(12).apply(...), and cumulative-product max-drawdown code used to build this module's own tables. The mechanics (correlation matrix, portfolio-weighted return series, rolling-window worst-case, single-shock stress test) do not change one line between a synthetic and a real dataset; only the input does. The quant-code region's DA/QD branches own the full implementation depth (data pipelines, backtesting discipline, evaluation); this module's job is narrower and non-negotiable: never let a headline correlation or return figure into your own analysis without having pulled the series and run the computation yourself, once.