On-Demand Combinatorial Event Markets on Kalshi: Instantiation, Concentration, and Effective Market Breadth

Battle Count: 51

The paper is primarily about market architecture and microstructure of prediction markets, not about trading strategies, signal generation, or portfolio construction. However, understanding the concentration of market objects (94.82% in one collection), the effective primitive count (~720), and the low activity conversion rate (35.38%) is relevant for market-makers designing automated parlay pricing systems and for traders assessing liquidity and breadth in Kalshi's MVE markets. The findings caution against using ticker count as a proxy for tradable economic breadth, which is relevant for any quantitative strategy operating in prediction markets.

The empirical pipeline involves: (1) authenticated REST API calls to enumerate MVE market objects over a seven-day window with exact timestamp filtering; (2) WebSocket stream capture for creation notifications and lifecycle events; (3) deterministic signature construction from event keys, collection keys, and sorted selected-leg multisets; (4) hierarchical concentration analysis (HHI, Gini, Lorenz) across collections, events, and primitives; (5) REST-WebSocket crosswalk with exact ticker matching; (6) 14-day downstream lifecycle follow-up with endpoint and two-clock path verification. The data volume (7.6M+ market objects, 66M+ primitive occurrences) requires careful deduplication and boundary handling. No ML model training is involved; the complexity is in data engineering, API interaction, and precise structural analysis.

The registered population and hierarchy data are publicly available as the Kalshi Multivariate Event Market Materialization Dataset (KMVE) on Mendeley Data (doi: 10.17632/fn65786cg6.1). The methodology is fully specified with exact signature construction formulas, non-intervention criteria, and deterministic deduplication rules. However, the analysis depends on Kalshi's live REST and WebSocket APIs, which may change over time. The 190-leaf membership proof and boundary audit are documented. No raw authenticated payloads are included. The paper is version r0.7.10, suggesting iterative refinement.