The CoinAlg Bind: Profitability-Fairness Tradeoffs in Collective Investment Algorithms
Samuel Breckenridge
Session summary
The presenter analyzes “coin algs,” collective investment algorithms where many users pool assets and a shared strategy trades on their behalf. They contrast private and transparent designs, arguing that private algorithms can enable unfairness if insiders or covert channels leak strategic information to a single advantaged adversary, who can then consistently outperform other contributors. To avoid this, a fully transparent algorithm might reveal its logic, but the speaker contends that this exposes it to value-extraction tactics such as strategy cloning and particularly sandwich attacks, where an informed trader buys before and sells after the algorithm’s trades to capture profit with limited risk.
The authors study this interaction through both a game-theoretic model and simulations. In a one-shot setting with limit prices, they find the algorithm can deter sandwiching, but in repeated play an adversary using a “grim trigger” strategy can force profit sharing, creating a transparency cost. Using historical Uniswap pool data, they simulate an idealized, oracle-aided coin alg and a front runner. Across different transparency assumptions—including secret trading times and a one-bit leak of trade direction—the front runner still siphons profit and reduces the coin alg’s returns. The talk closes with proposed mitigations, such as adding randomized wrappers to private strategies and trustless, TEE-based audits of black-box code.
Key takeaways
- The authors argue that private collective investment algorithms can create structural unfairness when leaked strategy information lets one adversary systematically extract more value than other participants.
- They model transparent coin algs facing sandwich attacks and show that, in repeated interactions, a committed adversary can enforce a profit split, implying a persistent cost of transparency.
- Simulations on historical Uniswap pools suggest that even partial transparency, such as hidden trading times or a one-bit leak of trade direction, still allows front runners to siphon profits from an otherwise idealized algorithm.