6.11.5 Failure Modes & Practice

Basic Prerequisites: 04 · MEV & Sandwiching.

---

1. Intuition & Practical Objective

Every structural advantage of the AMM (passivity, no inventory management, no quote management) is also its structural weakness. This page names the failures precisely so a practitioner can decide, in money terms, whether to provide liquidity at all. The objective is not cynicism - it is knowing exactly where the “free money from fees” story breaks.

The five failures, in one line each:

1. Fees rarely cover divergence loss on volatile assets. The LP earns fee income but loses the path-dependent loss-versus-rebalancing (LVR) to arbitrageurs; on volatile, low-turnover pools all standard fee tiers lose net.
2. 24/7, no circuit breakers, no halts. A crash that would halt a CEX is just another (faster) arbitrage trade on-chain - and the LP cannot exit faster than the arbitrageur can trade.
3. Protocol risk compounds market risk. Smart-contract bugs, oracle manipulation, and bridge hacks are tail risks with no exchange backstop or settlement guarantee.
4. Gas and priority fees are a fixed-cost tax. Rebalancing concentrated liquidity, hedging, and withdrawing all burn gas; below a scale threshold, market making is uneconomic.
5. The AMM has no inventory or adverse-selection controls. Everything a human maker does with a reservation price and toxicity gating (Avellaneda–Stoikov, VPIN) is structurally absent.

---

2. Mathematical Ground Truth & Derivations

Where the break-even lives. The LP’s net P&L (in numeraire, ignoring fees already described by the divergence-loss of 02) is

$$\text{net}=\text{fee income}-\text{LVR},$$

where LVR is the divergence loss as a running, path-dependent cost (Milionis et al. 2022), and fee income $=f\times$ (traded volume). Providing liquidity is profitable iff

$$\boxed{f\times \text{volume}>\mathrm{LVR}}.$$

The instantaneous LVR rate. Milionis, Moallemi, Roughgarden & Zhang (2022) prove that for a constant-product AMM the instantaneous rate of LVR is

$$\frac{d\mathrm{LVR}}{dt}=\frac{1}{8}{\sigma }^{2}V,$$

with $V$ the pool value in numeraire and $\sigma$ the volatility of the price process - the same $\frac{1}{8}{\sigma }^2$ that the small-move expansion of divergence loss ($\mathrm{DL}\approx s^2/8$) in 02 seeded. Over a finite horizon the expected total LVR is $V_0(1-e^{-{\sigma }^{2}T/8})\approx \frac{1}{8}{\sigma }^2{V}_{0}T$ for small ${\sigma }^{2}T$. Since fee income scales with volume (roughly $\propto \sigma$) but LVR scales with ${\sigma }^2$, the required fee tier grows with volatility - high-vol pools need high fee tiers and high turnover to break even, which is precisely why volatile tokens shed liquidity.

Gas as a fixed cost. On Ethereum-style chains, each on-chain action costs a base fee plus a priority fee (EIP-1559); MEV competition bids the priority fee up exactly when rebalancing matters most. Gas therefore behaves as a per-action fixed cost that shifts the break-even further toward large, infrequent, high-edge market making - the opposite of the “always-on passive” promise.

---

3. Computational Implementation - do the fees cover the LVR?

We simulate a constant-product pool over one year (drift-free geometric Brownian motion on the price, volatility $\sigma =0.6$ annualized). At every step an arbitrageur rebalances the pool to the fair price; we accumulate fee income at each fee tier against the realized LVR (the pool’s mark-to-market divergence loss over the path).

Reading the numbers. With $\sigma =0.6$/yr and only $1.66\times$ annual turnover, the realized LVR is $7.55$ (per $200$ of pool value) while fee income at every Uniswap tier - $0.01\%$, $0.05\%$, $0.30\%$ - is a fraction of that. Net P&L is negative for all three tiers; the breakeven fee is $\approx 2.27\%$, far above any standard tier. This is the LVR literature’s core empirical claim (Milionis et al. 2022; a16z LVR): low-fee, volatile, moderate-turnover pools transfer value from LPs to arbitrageurs. It also explains the failure mode list that follows - LPing is only a positive-expectation business at high turnover (where $f\times \text{volume}$ overtakes $\frac{1}{8}{\sigma }^{2}V$) or on low-volatility pairs.

---

4. Failure Modes & First-Principles Breakdowns (numbered)

1. LVR swamps fees on volatile pools (the #1 failure). At $\sigma =0.6$/yr the required fee is $\approx 2.3\%$ (§3); real pools on volatile tokens sit at $0.01$–$0.30\%$ tiers and bleed. Before adding liquidity, check whether $f\times \text{volume}>\frac{1}{8}{\sigma }^{2}V$ - this is the single most important go/no-go test in DeFi market making.
2. 24/7 markets, no circuit breakers, no price bands. A CEX halts, widens bands, or invokes a kill switch; an AMM can do none of these. In a flash crash the arbitrageur still trades and the divergence loss is realized instantly. Liquidity “always available” is also liquidity “always exposed.”
3. Protocol risk is additive and unhedgeable. Smart-contract vulnerabilities, oracle manipulation (a wrong price feeds a liquidation or a swap), and bridge hacks can wipe the entire deposit, not just the divergence loss. There is no exchange guarantee and no settlement protection; the “risk-free spread” does not exist on-chain.
4. Gas + priority fees are a per-action tax. Rebalancing concentrated liquidity (03), hedging, and withdrawing all consume gas, which MEV competition inflates exactly when action is most urgent. Below a size threshold the fixed costs exceed the edge - DeFi market making is inherently large-scale.
5. No inventory or toxicity control. The AMM can neither skew away from inventory nor widen against toxic flow. Every mitigation must be built outside the pool (hedging, range migration, exit rules) - the on-chain mirror of Avellaneda–Stoikov’s reservation-price discipline, minus the ability to actually control quotes.

---

5. References

• Milionis, Moallemi, Roughgarden & Zhang (2022), Automated Market Making and Loss-Versus-Rebalancing, arXiv:2208.06046
• a16z Crypto, LVR: Quantifying the Cost of Providing Liquidity to Automated Market Makers
• Clark, Joseph (2020), The Replicating Portfolio of a Constant Product Market, SSRN 3550601
• Lehar & Parlour (2023), Decentralized Exchange: The Uniswap Automated Market Maker
• Daian et al. (2019), Flash Boys 2.0, arXiv:1904.05234

---

6. Connected Graph Bridges

• Back: 04 · MEV & Sandwiching · Index Hub
• Forward: 06 · Advanced Extensions
• Sibling: Adverse Selection & Glosten–Milgrom · Toxic Order Flow & VPIN · Liquidity Risk & Funding