Fill-Side Non-Retail Trading on Polymarket: An Empirical Study of Behavioral Tiers and Microstructure Signatures Under Quote-Attribution Constraints
This empirical study of Polymarket's non-retail trading reveals that the platform's off-chain order book architecture permanently prevents address-level quote attribution, resulting in a uni-modal fill-side behavior distribution that contradicts pre-registered archetype hypotheses while confirming that a small tier of high-volume addresses dominates 81.4% of total notional volume.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Picture: Who is Playing the Game?
Imagine Polymarket as a giant, digital casino where people bet on future events (like "Will it rain tomorrow?" or "Who will win the election?").
For this casino to work, you need two types of people:
- The Gamblers (Retail): Regular people placing small bets.
- The Dealers & Whales (Non-Retail): The people providing the cash, setting the odds, and making sure there is enough money for everyone to bet against.
This paper is a scientific study of the Dealers and Whales. The researchers wanted to answer: Who are these big players? Are they organized into distinct groups (like "The Speed Traders" vs. "The Long-Term Investors")? And how do they behave?
The Main Problem: The "Black Box" Casino
The researchers hit a major roadblock immediately. In a normal casino, you can see the dealer's hands. You can see when they place a card on the table, when they take it back, and how long they hold it.
On Polymarket, the "Dealers" (Market Makers) do their work in a private, off-chain room.
- What the researchers could see: Only the final result of a bet (the "Fill"). They could see who bought, who sold, how much, and when.
- What they could NOT see: The "Quote Lifecycle." They couldn't see when a dealer offered a price, when they changed it, or when they cancelled an offer without it being filled.
The Analogy: Imagine trying to study a chef's cooking style, but you are only allowed to look at the empty plates after the food is eaten. You can see what was ordered and how much was eaten, but you can't see the chef chopping vegetables, tasting the sauce, or deciding to throw a dish away before serving it.
Because of this "Black Box" limitation, the researchers had to throw out their original plan to study the chefs' movements and instead focus entirely on the eating habits (the trades that actually happened).
The Big Discovery: One Big Crowd, Not Distinct Groups
The researchers had a hypothesis: "We bet that the big players fall into 4 or 5 distinct groups, like different species of animals in a zoo." They used a computer algorithm (DBSCAN) to try to sort the 77,000+ active addresses into these groups.
The Result: The algorithm found zero groups.
Instead of finding distinct "species," the data looked like one giant, continuous cloud of fog. The "Dealers," "Whales," and "Arbitrageurs" weren't separate islands; they were just different points on a single, smooth slope. The "Whales" were just the people at the very top of the slope, and the "Retail" traders were at the bottom, but they were all part of the same continuous distribution.
The Analogy: Imagine trying to sort a pile of sand into "Small Grains," "Medium Grains," and "Large Grains." You try to use a sieve, but you realize the sand is just one continuous mix of sizes. You can't draw a hard line where "Small" ends and "Medium" begins.
The New Solution: Sorting by "Weight" Instead of "Type"
Since they couldn't sort the players by behavior type (because they all looked the same), the researchers switched to sorting them by how much money they moved. They created a "Tier System" based on the size of their bets:
- The Whales (0.1% of people): These are the giants. Just 68 addresses. They hold 28% of all the money in the market.
- The High-Frequency Operators (3.8%): These are the speed demons, making thousands of trades.
- The Power Traders (8.7%): The heavy hitters who trade large amounts.
- The Episodic Retail (82%): The vast majority of people. They make up 82% of the players but only hold 6.8% of the total money.
The Takeaway: The market is extremely unequal. A tiny handful of people (the top 12.6%) control 81.4% of all the money being traded.
What About Manipulation?
The researchers also looked for "bad behavior," like Wash Trading (trading with yourself to fake volume) or Spoofing (placing fake orders to scare people).
- Spoofing: They couldn't find this. Remember the "Black Box"? To catch a spoofer, you need to see them place a fake order and then cancel it. Since that data is hidden, they had to admit: "We can't see this."
- Wash Trading: They found 3,980 addresses that bought and sold roughly the same amount, ending up with zero net change. This looks like wash trading, but they couldn't prove it was the same person doing it. They labeled these as "suspects" but couldn't convict them.
The "Aha!" Moment for Future Designs
The paper ends by talking to the authors of a previous paper (Paper 1) that designed the rules for this casino.
The Surprise: The previous paper assumed that a "regular person" (retail trader) usually bets about $1,000 at a time.
The Reality Check: The new study found that the average retail bet is actually only $4.77.
The Analogy: It's like a city planner designing a bridge assuming everyone drives a 20-ton truck. They build the bridge super strong. But then, a new study reveals that 99% of the traffic is actually just people walking with backpacks. The bridge is way over-engineered for the actual traffic.
The researchers are telling the designers: "You need to redesign your rules because the 'average' player is much smaller than you thought."
Summary of Findings
- We can't see the "Offer" phase: Polymarket hides the "placing of orders" data, so we can only study the "filling of orders."
- No distinct groups: The big players don't fall into neat categories; they are one big, continuous group.
- Extreme concentration: A tiny elite (Whales + Power Traders) controls the vast majority of the money.
- Retail is tiny: The average bet is $4.77, not $1,000.
- Manipulation is hard to prove: We can see suspicious patterns, but without the "off-chain" data, we can't prove who is doing what.
This paper is a "reality check" for anyone trying to understand or regulate these prediction markets: The data is limited, the players are more uniform than we thought, and the money is concentrated in very few hands.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.