High-Frequency Convex Optimization Arbitrage

How It Works This strategy frames multi-market arbitrage as a convex optimization problem. Given a set of related Polymarket markets with interdependent probabilities (e.g., correlated political or sports outcomes), the Frank-Wolfe algorithm iteratively identifies the portfolio of positions that maximizes expected profit subject to probability consistency constraints and capital limits.

Entry Conditions Positions are entered when the optimizer identifies a feasible portfolio where the current market prices are inconsistent with each other in a way that guarantees positive expected value regardless of outcome. The Frank-Wolfe algorithm converges toward the optimal allocation iteratively, and trades are executed when a minimum profit threshold is exceeded.

Exit / Resolution As a high-frequency system, positions may be exited before resolution if prices converge. The optimization framework continuously re-runs, and positions are adjusted or closed as the opportunity disappears.

Verified Results No empirical results are disclosed in the repo description. The use of a well-known convex optimization algorithm adds theoretical credibility, but the absence of backtesting data or live results limits confidence.

Implementation Notes & Risks The Frank-Wolfe algorithm is well-suited to large-scale constrained optimization but can be slow to converge in highly non-linear settings. Key risks include latency (optimization takes compute time, during which prices may move), model risk (incorrect market correlation assumptions), and the inherent competitive nature of HFT arbitrage on a public blockchain where all trades are visible.