Prediction market liquidity benchmark

Research question

If a trader attempted to buy the Up or Yes outcome in a BTC 15-minute prediction market, how often could the displayed book absorb $100, $1,000, or $10,000, and how far would the average execution price move beyond the best ask? The benchmark deliberately answers this narrow question. It does not use reported volume as a proxy for liquidity, because past turnover cannot show whether a new order could execute at the observation time.

Comparable universe

The balanced panel contains every scheduled BTC 15-minute Up/Down contract in the shared 30-day window from 26 June 2026 at 08:00 UTC through 26 July at 08:00 UTC: 2,880 contracts per venue and three lifecycle checkpoints per contract. Predict.fun is the underlying third-party protocol distributed through Binance Wallet; Binance Wallet is the access layer rather than a separate event creator or counterparty.

Sampling and side normalization

DepthFeed selected the venue-native full-depth observation nearest 04:30, 08:30, and 12:30 after each contract opened. This gives all venues equal statistical weight regardless of collector speed. Polymarket’s separate outcome-token books were restricted to the Up token. Kalshi publishes Yes and No bid ladders; executable Yes asks were reconstructed as one minus the corresponding No bid, preserving size. Polymarket, Limitless, and Predict.fun provide direct ladders. Missing checkpoints remain explicit empty rows in the downloadable panel.

Execution test

For each lifecycle checkpoint, the simulator starts at the lowest displayed ask and spends the requested dollars level by level. A level’s displayed notional is price multiplied by shares. The observation counts as a full fill only when the entire target could execute. One-sided or askless books remain in the denominator with a zero fill; excluding them would reward venues for disappearing precisely when execution was impossible. Median price impact is the volume-weighted average fill price minus the best ask, expressed in cents, and is calculated only for complete fills.

Uncertainty and ranking

The displayed 95% intervals use 4,000 deterministic UTC-day block bootstrap samples. Complete trading days—not individual checkpoints— are resampled, preserving within-market and intraday dependence. The primary leaderboard orders venues by $1,000 full-fill rate and uses median price impact only as a numerical tie-breaker. Overlapping intervals are described as a statistical tie even when the table must place one row first.

Limitations

This is a historical displayed-liquidity study, not a promise of future execution. Resting orders can be cancelled before a live order arrives, and the benchmark does not model queue position, network latency, minimum order constraints, market-order protection, or venue-specific taker fees. Fees are excluded because the goal is to isolate the book; a production strategy must add the fee schedule in effect for its account and jurisdiction. Results cover BTC 15-minute contracts during one common 30-day period and should not be generalized to politics, sports, longer expiries, other assets, or future market regimes without a new run.

Primary venue references

The normalization follows the venues’ documented book conventions: Polymarket order books, Kalshi Yes/No bids, Limitless book updates, and Binance Wallet Prediction Markets. Predict.fun archive retrieval uses Predexon orderbook history, while market identity is matched to the public Predict.fun catalogue. DepthFeed independently normalized the observations and is not affiliated with or endorsed by any venue or data provider. The public reproduction repository contains the immutable panel, verifier, methods, citation metadata, and release assets without requiring a DepthFeed account or API key.