Analyzes 7,292 resolved Polymarket markets and 28,793 on-chain trades to test whether yes bias exists in trade data. Finds that traders buy whichever token is cheaper, not whichever is labeled YES, and that the apparent bias is a compound effect of longshot preference channeled through Polymarket's "Will X happen?" question framing, which systematically assigns the longshot to the YES token.

Categorizes every active Polymarket wallet across trade frequency and volume tiers, producing a seven-persona map of platform participants. Finds that 2% of users (high-frequency, high-capital wallets) generate roughly 90% of all platform volume, with crypto markets dominated by algorithmic execution and politics markets driven by casual event-driven participants. Draws implications for fee design, category expansion, and why optimizing for user growth versus volume growth requires fundamentally different product decisions.

Wall Street equity research analysis of anonymized trading data comparing prediction market and sports betting returns. The median prediction market user has an ROI of -8%, worse than sports bettors at -5%, with only traders above $500K in volume achieving positive returns (+2.6%). Finds that prediction markets attract sharper competition than regulated sportsbooks, creating worse outcomes for casual retail participants.

Analyzes tick-level order flow across Polymarket and Kalshi to decompose market bias by trader type. Finds that the classic favorite-longshot bias may be a statistical artifact masking a pervasive "yes bias" driven by temporal volatility and incomplete controls for contract lifecycle. Also shows that whales are not the sharpest participants: heavily capitalized traders systematically bleed expected value to small-order traders, likely driven by ideological conviction rather than informational edge.

Argues that prediction markets face two structural problems preventing them from becoming transformative economic instruments: corporate hedging is impractical due to market fragmentation and basis risk, and insider trading undermines retail participation. Draws parallels to online poker and memecoins to suggest that without structural reforms, prediction markets will remain primarily a sports betting product.

Decomposes 222 million Polymarket trades into directional and execution components and finds that forecasting accuracy does not predict profitability. Traders who pick the right side still lose money because they arrive late and pay unfavorable prices, while automated traders with near-random directional skill profit by paying 2.52 cents less per contract.

Summarizes research analyzing 72 million Kalshi trades. Identifies three persistent biases: longshot bias (5c contracts win only 4.18% of the time), maker-taker asymmetry (makers outperform at 80 of 99 price levels), and YES/NO asymmetry (YES buyers average -1.02% returns vs +0.83% for NO buyers). Finance markets are most efficient (0.17% spread) while crypto is least (2.69%).

Analyzes 72.1 million trades ($18.26B volume) on Kalshi and documents a systematic wealth transfer from takers to makers averaging 1.12% excess returns on each side. Takers disproportionately buy YES longshots, accepting returns 64 percentage points lower than equivalent NO positions. Shows this transfer only emerged after Kalshi's October 2024 legal victory attracted professional algorithmic market makers, and that market efficiency varies sharply by category: finance markets are near-efficient while entertainment and media markets show gaps exceeding 7 percentage points.

Annual letter arguing that retail financial speculation has permanently shifted from investment to entertainment, driven by smartphone access and zero-commission trading. Documents the collapse of holding periods, zero-day options comprising 59% of options volume in 2025, and the proliferation of insider trading cases across sports and corporate prediction markets as event markets expand to cover everything. Frames this as a structural shift rather than a cyclical bubble.