These Are the Sharps Actually Making Money on Prediction Markets

Summary

This episode examines the mechanics and incentives behind prediction markets through the lens of a private Discord of specialist traders, Maga Kiwi Club. Brian Golden and Daniel Reichman argue that consistent profits come from deep domain work: reconstructing CPI methodology in spreadsheets, gathering ground truth from experts and reporters, and revising priors faster than the crowd. The conversation also dwells on market structure, including zero-sum dynamics, liquidity, oracle/resolution rules, position limits, and the practical difficulty of defining and policing insider trading. A recurring theme is that elections and other “soft” markets remain beatable because they are driven by emotion, while local information can overwhelm broad internet consensus, as shown by losses in the Romanian election. The episode closes by suggesting that durable edge in prediction markets comes from finding new information and turning points, not from generic AI prompts or trivial novelty bets.

Notes

• Prediction markets are treated as close to zero-sum, with the hosts contrasting them against equity markets that can compound through cash flows and dividends.
• The hosts explicitly reject get-rich-quick framing and gambling-style ads that imply easy wins.
• Maga Kiwi Club is framed as an ironic name, not a political organization, and functions like a specialist network with multiple trading domains.
• The group traces its roots to PredictIt-era political markets in 2016-2017.
• The traders emphasize that specialist sharing happens alongside separate portfolios, not as one pooled book.
• Brian Golden says his inflation edge came from reverse-engineering the BLS methodology in Excel.
• The CPI basket has 200-plus subcategories, and weights shift monthly with spending patterns.
• He claims the work took about three months before he could reproduce the calculation reliably.
• Brian says his average absolute inflation error over two years beat Bloomberg consensus.
• Adam Iscoe says sharps do not rely on screens alone; they call experts, reporters, and specialists.
• The information process is explicitly multimodal: on-the-ground checking, expert calls, and reporter conversations.
• Daniel Reichman says Polymarket’s oracle/resolution layer is problematic compared with Kalshi’s.
• A major trading distinction is whether a 98-to-99 price gap reflects true probability or just time-to-resolution.
• Daniel cites the New Jersey governor’s race as a case where channel priors diverged from polls.
• They believed the race was 12-14 points despite polls showing only a 2-5 point contest.
• Comments sections can serve as a contrarian indicator because sharps usually stay quiet publicly.
• PredictIt’s $850 cap made it easier to share views; Kalshi’s looser limits changed incentives.
• Daniel Reichman says prediction markets may get harder as the field professionalizes, like poker and DFS did.
• Elections still look beatable to him because they are driven by vibes and emotions.
• Romania is presented as a sharp loss caused by missing local sentiment.
• The failure mode was underestimating how quickly the candidate became a joke in Romania.
• Brian says prediction markets can surface private information into public prices, but retail traders may be on the wrong side of truly informed flow.
• He argues that the value of prediction markets is in converting hidden information into visible price signals.
• Brian says he can often spot insider-like trading by abnormal volume and a sudden jump at a never-traded price.
• He cites the Jacob Elordi Critics Choice market as an example, with price moving from $0.01 to $0.40 on heavy volume.
• Daniel Reichman argues serious insider trading must be made economically irrational.
• He cites cases where traders lost jobs or faced charges after trading on forbidden information.
• Ambiguous insider definitions remain hard to police in practice.
• Prediction-market legitimacy, in Brian’s view, depends more on serious reporting than novelty prop bets.
• He says those trivial markets are not what prediction markets should be judged by.
• Daniel says AI helps most with international research and foreign-language search.
• He says very few traders in the group use AI much for modeling.
• Brian mocks ChatGPT as highly prompt-sensitive and willing to reverse itself on demand.
• The broader edge, Tracy Alloway argues, is still human data collection and spotting turning points, not LLM output.
• Joe Weisenthal compares the market’s evolution to poker, where a visible boom attracts participants before sharps dominate.
• The episode’s bottom line is that durable alpha comes from proprietary information, fieldwork, and fast priors revision.