These Are the Sharps Actually Making Money on Prediction Markets

Episode Summary

Executive Summary: Odd Lots examines prediction markets through a conversation with reporter Adam, and sharp traders Brian Golden and Daniel Reichman. They argue the markets are beatable because many events are mispriced, especially politics, and that success comes from hard work, on-the-ground reporting, and model-building—not AI prompts or hype. They also debate efficiency, liquidity, insider trading, and whether these markets should focus on socially meaningful questions.

Main Topics: Prediction markets are beatable, but not easy (Priority: 5/5): The guests argue that consistent profits come from identifying mispricings in markets that are often still inefficient, especially when retail money is overconfident or misled by narratives and comments. Politics as the most mispriced category (Priority: 5/5): Brian and Daniel say elections are especially vulnerable to emotion, siloed media, and fast-moving vote-count noise, creating large opportunities for informed traders. How the trading edge is built (Priority: 5/5): Their edge comes from rigorous historical work, precinct-level modeling, polling, sourcing, and constant information-sharing in Discord rather than from intuition or AI-generated answers. Market structure, liquidity, and resolution mechanics (Priority: 4/5): The episode explores whether prediction market prices are informative, how liquidity affects price accuracy, and how traders decide whether to hold to resolution or exit earlier. Insider trading and information asymmetry (Priority: 4/5): The traders acknowledge some markets clearly move on inside information, raising ethical and regulatory concerns, though they argue a few insider events can be financially irrational to pursue. Comments, crowd psychology, and dumb money (Priority: 4/5): Participants discuss the value of betting against comment sections, the disappearance of open sharing as markets mature, and the way retail losses concentrate among a tiny set of accounts. Prediction markets vs. traditional forecasting (Priority: 3/5): They compare market prices with polls and expert forecasts, arguing markets can be useful because they impose consequences for being wrong, but are not automatically superior to other methods.

Key Arguments: Prediction markets are not fully efficient; persistent profits suggest exploitable mispricings remain, especially in politics and other low-liquidity or sentiment-driven markets. Politics is often the weakest-priced category because beliefs are highly siloed, emotionally charged, and shaped by partisan media rather than fundamentals. The real edge is labor-intensive: historical analysis, precinct-level modeling, polling, field reporting, and frequent recalibration when data changes. AI is useful for starting research or handling language barriers, but LLMs are backward-looking and mostly reflect whatever framing the user gives them. Crowded comment sections and obvious retail enthusiasm can be contrarian signals, because informed traders usually keep quiet. Prediction markets can be better than polls in some situations, but not simply because they are faster; their accuracy depends heavily on liquidity and who is trading. Insider trading is a serious problem in some contracts because it can dominate price discovery, though policing every borderline case may be impractical. The healthiest version of prediction markets is one focused on meaningful questions like elections and macroeconomic data, not trivial novelty bets.

Data Points: Book/article date: May 26 - The New York Times article discussed on the episode. Position limit on PredictIt: $850 - Daniel notes this old limit encouraged openness because traders could say when they were maxed out. Romanian election swing: about 20 points - Daniel describes the first-round margin for Simeon before the runoff. Comment-section profit concentration on Polymarket: 67% of profits go to 0.1% of accounts - Mentioned as evidence that most users lose money while a tiny elite captures gains. Time horizon for holding near-resolution markets: 48 hours - Brian says if a market resolves within about 48 hours, he usually holds from 99 to 100. Inflation model build time: about 3 months - Brian says it took him three months to reconstruct the BLS formula in Excel. BLS inflation categories: 200-plus subcategories - Brian explains the CPI basket uses many subcategories with changing weights. Los Angeles mayoral example: Spencer Pratt price moved to 27% - Daniel/Brian cite this as an example of a wildly mispriced election market. California vote timing issue: late vote returns - Used to explain why election-night prices can be noisy and misleading. Romanian election outcome: dark day / upset loss - The group lost money when the Romania runoff defied their expectations.

Pivotal Quotes: "Prediction markets are the future. They're this incredible price signal, and you know, they're more accurate than all these other places. I don't really believe that." — Daniel Reichman: He argues markets are useful but not automatically superior to all expert forecasting. "The real thing for me is, I just think that this says a lot more about the sort of softness of the people, the investment banks, and people who predict this in the institutional end than it does about me." — Brian Golden: He explains why he thinks his inflation forecasting beats institutional consensus. "Always bet against the comment section." — Brian Golden: A contrarian trading rule discussed as a way to identify overconfident retail sentiment.

Implications: Prediction markets may become more efficient as professionals enter, but for now they still reward deep research, sourcing, and discipline. The episode suggests their credibility depends on trading real-world questions well and policing insiders, not on hype or gambling-style marketing.