Overview of These Are the Sharps Actually Making Money on Prediction Markets
This episode of Odd Lots explores how a small group of highly active prediction-market traders — the “sharps” — consistently beat the broader crowd by doing old-fashioned research: building models, calling experts, reading local media, and paying close attention to the details that most traders miss. The conversation centers on a Bloomberg article about a Discord group called the MAGA Kiwi Club, where specialists in politics, inflation, and other markets share information and help one another find edges on platforms like Polymarket and Kalshi.
Who the successful traders are
The Discord-based “Avengers” model
- The traders operate like a loose, distributed research desk.
- Each person tends to specialize in a different area:
- Brian Golden: inflation contracts
- Daniel Reitman aka Carnitas Taco: elections and politics
- They exchange tips, data, and context in a private Discord, effectively pooling expertise while trading their own books.
What makes them different
- They are not relying on vibes or surface-level headlines.
- They spend time:
- building models
- understanding official formulas
- calling experts
- doing field reporting
- updating beliefs when evidence changes
How they find an edge
Inflation markets
Brian explained that he reverse-engineered the BLS inflation formula in Excel and learned how the CPI basket is weighted and updated. His edge comes from:
- understanding the mechanics of the data release
- tracking underlying price trends
- knowing how each component contributes to the final number
He argued that if public forecasting institutions were truly doing great work, a theater kid with Excel shouldn’t be able to beat them so consistently.
Election markets
Daniel described election trading as a blend of:
- historical pattern recognition
- polling analysis
- on-the-ground reporting
- local/media context
- calibration about probabilities, not just outcomes
The group also used precinct-level and vote-timing models to anticipate how late-counted ballots might shift results, especially in places like California.
General principle
Their edge is not magic — it is work:
- gathering more and better information
- staying disciplined about probabilities
- avoiding emotional or socially reinforced narratives
Why prediction markets can be mispriced
The episode repeatedly comes back to the idea that prediction markets are not always efficient because:
- liquidity is uneven
- many participants are retail or emotionally driven
- social media and partisan echo chambers distort beliefs
- some people trade on “vibes” instead of evidence
Examples discussed:
- Spencer Pratt in the Los Angeles mayoral race
- the Romanian election, where the sharps got caught on the wrong side of a surprise upset
- markets that moved sharply on thin or misleading information
Comments, crowd sentiment, and dumb money
A recurring theme: the comment section is usually a bad indicator.
The traders said:
- more comments often mean more low-quality conviction
- sharps tend to stay quiet and trade privately
- as markets mature, easy money should get harder to find
This was compared to:
- the online poker boom
- daily fantasy sports
- any market where “soft” participants eventually get weeded out
Insider trading and market integrity
The group had a nuanced view of insider trading in prediction markets:
- It is bad to trade against someone who literally knows the answer.
- But prediction markets are also partly valuable because they surface information that might otherwise stay hidden.
- Enforcement should focus on serious, economically meaningful abuses, not trivial cases.
They noted that some markets are clearly vulnerable to insiders, especially:
- cabinet confirmations
- awards
- private political decisions
- event-specific outcomes where a small circle may know more than the public
What they think about AI
AI was described as useful, but not a source of real edge by itself.
Good uses
- starting research
- foreign-language search
- speeding up routine analysis
Weaknesses
- LLMs are backward-looking
- they can be steered by the user’s framing
- they do not replace proprietary information or real-world reporting
Their view: AI may help with productivity, but it does not substitute for the kind of ground truth gathering that actually wins markets.
Main takeaways
- Prediction markets can be beatable, but usually only with serious research and specialization.
- The best traders act more like investigative analysts than gamblers.
- Markets are most vulnerable when:
- liquidity is thin
- emotions run high
- insiders may be present
- public narratives diverge from reality
- As more professionals and informed traders enter, the easy edges should shrink.
- The long-term legitimacy of prediction markets depends on whether they are used for important questions, not just novelty betting.
Notable insight
A core idea from the episode was that prediction markets work best when the question itself matters. In that sense, the traders argued, a market’s usefulness depends not just on its price signal, but on whether it is pricing something real, measurable, and consequential.