DeFi, AMMs, Flash Loans & MEV

DeFi = financial primitives (exchange, lending, derivatives, stablecoins) implemented as composable smart contracts ("money legos"). Composability is the superpower and the main systemic risk.

1. Automated Market Makers (AMM)

An order book needs market makers quoting prices. An AMM replaces them with a pool and a pricing invariant.

Constant Product (Uniswap v2): x * y = k

Pool holds x of token A and y of token B. Selling dx of A for B, with fee f (0.3%):

dx_eff = dx * (1 - f)
dy     = y - k / (x + dx_eff)         // = y * dx_eff / (x + dx_eff)
price impact grows with trade size relative to reserves

import java.math.BigInteger;

public final class ConstantProduct {
    /** Uniswap v2 getAmountOut: integer math, fee 0.3% -> 997/1000. */
    public static BigInteger amountOut(BigInteger amountIn, BigInteger reserveIn, BigInteger reserveOut) {
        BigInteger inWithFee = amountIn.multiply(BigInteger.valueOf(997));
        BigInteger numerator = inWithFee.multiply(reserveOut);
        BigInteger denominator = reserveIn.multiply(BigInteger.valueOf(1000)).add(inWithFee);
        return numerator.divide(denominator);
    }
}
// Pool 1000 ETH / 2,000,000 USDC; swap 10 ETH -> about 19,743 USDC (spot value 20,000): ~1.3% (0.3% fee + ~1% price impact)

Always integer math on-chain (no floats); rounding must favour the pool.

Impermanent Loss (IL)

If price ratio changes by factor r, an LP's position value vs holding is 2*sqrt(r)/(1+r) - 1. Example: price x2 -> about -5.7%; x4 -> -20%; x10 -> -42%. Fees must outweigh IL for profitability.

AMM Evolution

Version	Idea	Trade-off
v2	Uniform liquidity over (0, infinity)	Capital inefficient, simple, fungible LP tokens
v3	Concentrated liquidity in tick ranges; LP = NFT position	Up to ~4000x capital efficiency; active management, more IL risk
Curve (stableswap)	Hybrid invariant flat near peg	Great for stable pairs; depeg risk
Balancer	Weighted multi-asset pools	Flexible; complex math
v4	Singleton contract + hooks (custom logic), flash accounting	Power vs hook-security risk
RFQ / intent-based (CoW, UniswapX)	Solvers compete off-chain to fill orders	MEV protection, batch auctions

2. Lending Protocols (Aave, Compound, Morpho)

Supply collateral -> borrow up to LTV (e.g. 80% for ETH)
Health Factor HF = sum(collateral_i * liquidationThreshold_i) / totalDebt
HF < 1  ->  liquidatable: anyone repays up to 50% of debt and seizes collateral + bonus (5-10%)
Interest rate = f(utilisation U)   (kinked curve: low slope under optimal U ~80%, steep above)

Risks: oracle lag/manipulation, bad debt in fast crashes (Black Thursday 2020, congestion blocks liquidations), liquidation cascades, governance attacks on risk parameters, and wrapped-asset depeg (stETH).

3. Stablecoins

Type	Mechanism	Failure mode
Fiat-backed (USDC, USDT)	Reserves at custodians	Custodian/regulatory risk, blacklist function
Crypto-collateralised (DAI/USDS, LUSD)	Over-collateralised vaults, liquidations	Collateral crash, oracle failure
Algorithmic (UST)	Reflexive mint/burn with a volatile sibling	Death spiral ($40B collapse, 2022)
Synthetic delta-neutral (USDe)	Spot + short perps funding	Funding rate flip, exchange risk

4. Flash Loans

Uncollateralised loan that must be repaid within the same transaction, otherwise the whole transaction reverts. Legit uses: arbitrage, collateral swaps, self-liquidation. Attack use: temporarily acquire huge capital to manipulate a spot oracle or governance vote.

function executeOperation(address asset, uint256 amount, uint256 premium, address, bytes calldata)
    external returns (bool)
{
    require(msg.sender == address(POOL), "only pool");
    // ... use funds: arbitrage, swap collateral ...
    IERC20(asset).approve(address(POOL), amount + premium);   // pool pulls repayment
    return true;
}

Atomicity = no credit risk for the lender; the victim protocol's invariants are what break.

5. MEV - Maximal Extractable Value

Profit from ordering, including, or censoring transactions in a block.

Strategy	Description	Who loses
DEX arbitrage	Equalise prices across pools	Nobody (price discovery)
Liquidation	Be first to liquidate positions	Borrower pays bonus
Front-running	Copy and outbid a profitable pending tx	Original trader
Sandwich	Buy before victim's swap, sell after	Victim gets worse price up to slippage
JIT liquidity	Add/remove concentrated liquidity around a big swap	Passive LPs
Time-bandit	Reorg to capture past MEV	Network stability

Mempool: victim swap 100 ETH -> USDC (slippage tolerance 3%)
Bot:  tx1 buy USDC (pushes price up)  | victim swap executes at worse price | tx2 sell USDC (profit)
Block order controlled by (priority fee / bundle bid to builder)

Supply Chain Post-Merge: Proposer-Builder Separation (PBS)

Searchers --bundles--> Builders --blocks+bids--> Relays (MEV-Boost) --> Proposer (validator) picks highest bid

~90% of Ethereum blocks come via MEV-Boost; a few builders dominate (centralisation + censorship concerns - OFAC-compliant relays). Roadmap: enshrined PBS, inclusion lists (FOCIL), encrypted mempools.

User & Protocol Mitigations

• Tight slippage, limit orders, private RPC (Flashbots Protect, MEV Blocker - refunds backrun profit).
• Batch auctions (CoW) give uniform clearing price - no ordering advantage.
• Commit-reveal / threshold encryption for sealed bids.
• Design protocols so that order of transactions within a block does not affect fairness.

6. Systemic Risk Checklist

• Composability depth: how many protocols fail if one oracle/stablecoin breaks?
• Admin keys and upgradeability: who can change parameters or mint?
• Liquidity depth vs TVL: can collateral be sold at stress prices?
• Oracle design: TWAP window vs attack cost; fallback sources.
• Governance: token concentration, flash-loan voting defence, timelocks.

Next: Production Engineering.