Resolution-Aware Perpetual Futures on Binary Prediction Markets: Failure Modes and Mechanical Stress Tests Using Polymarket Data

Battle Count: 50

The paper is primarily about derivative product design and exchange risk management for prediction markets rather than about developing trading strategies or alpha signals. However, it is relevant to quantitative trading in several ways: (1) it characterizes microstructure features of Polymarket (spreads, depth, terminal jumps, boundary-region behavior) that inform event-driven trading strategies; (2) it identifies failure modes of leveraged synthetic positions that affect risk management for traders using prediction market derivatives; (3) the funding trilemma and margin decomposition inform how a trader would evaluate the cost and risk of synthetic exposure; (4) the mechanical replay protocol is a reusable stress-testing methodology. The paper does not propose or evaluate any trading strategy, signal, or portfolio construction method.

The formal proofs (Propositions 1 and 3) are mathematically straightforward balance-sheet identities and a simple divergence argument. The mechanical replay pipeline is more complex: it requires processing 13.69 billion raw events, applying stratified enrichment, passing adequacy gates, implementing multiple engine configurations (C0-C2, M0-M3) with dynamic margin, leverage compression, funding corrections, and staged halts, and running deterministic position grids. The pipeline separates raw ingest, cleaning, feature construction, and deterministic engine evaluation. However, no ML model training or complex optimization is involved; the complexity is in data engineering and rule-based simulation rather than in algorithmic sophistication.

The paper describes a companion release plan including code commits, input hashes, parameters, seeds, row counts, output hashes, schemas, methodology, processing ledger, and cryptographic manifests (pmxt-stylized-facts-v1, pmxt-counterfactual-replay-v1). However, the analysis plan was only internally version-locked (pre-specified), not externally pre-registered with an independent timestamped registry. The revision itself performs no new empirical calculation; all values are copied from archived publication outputs. The PMXT v2 Polymarket archive is publicly accessible, but reconstruction instructions are needed for source data that cannot be redistributed. Zenodo archival versions are planned but not yet confirmed as released.