These Are the Sharps Actually Making Money on Prediction Markets

What It Covers

Odd Lots interviews prediction market "sharps" Brian Golden and Daniel Reitman, members of the MAGA Kiwi Club Discord, alongside journalist Adam, exploring how a small group of disciplined traders consistently outperforms institutional forecasters on platforms like Kalshi and Polymarket using original research, historical modeling, and deliberate information networks.

Key Questions Answered

• •Inflation Forecasting via BLS Reconstruction: Brian Golden, a theater graduate with no economics degree, rebuilt the Bureau of Labor Statistics inflation formula in Excel over three months, then called BLS employees directly for clarification. His resulting bottom-up model achieves lower average absolute error than the Bloomberg consensus, suggesting institutional forecasters with unlimited resources are systematically underinvesting in basic methodology.
• •Comment Section Contrarian Signal: On Kalshi and Polymarket, heavy comment-section activity advocating one side reliably indicates the wrong side. Sharps deliberately stay silent on public boards to protect their edge, meaning visible retail enthusiasm functions as a consistent fade signal. Monitoring comment volume and sentiment before sizing a position provides a low-effort directional filter.
• •Political Market Mispricing via Siloed Media: Elections remain among the most mispriced prediction markets because participants bet emotional convictions rather than data. The LA mayoral race saw Spencer Pratt priced at 27% despite Los Angeles being a Democrat-plus-42 city. Anchoring to base rates and historical partisan margins, rather than social media narratives, consistently identifies overpriced longshots.
• •Election Night Intraday Edge via Precinct Modeling: The sharps built precinct-level historical vote models for the New Jersey governor's race, projecting a 12-to-14-point margin months before polls showed a close race. On election nights, real-time incoming batch data can be compared against these prebuilt models to trade mispriced contracts before the broader market updates, compressing the resolution window into minutes.
• •LLM Limitations as a Competitive Moat: LLMs produce sycophantic, backward-looking forecasts that mirror the same expert consensus already priced into markets. When Brian reversed his own stated inflation assumption mid-conversation with ChatGPT, the model immediately agreed both times. Proprietary data collection — door-to-door polling, direct source calls, commissioned phone surveys — remains the scarce input that LLMs cannot replicate.