Why 87% of Polymarket Trades Are Sells

TL;DR: ~63% of the shift is crypto micro-markets taking over. The rest is newer, shorter markets replacing older ones. Three things are real: makers buying both sides for risk-free profit (96–100% consistency), a 20pp SELL gap between short and long-lived markets, and a two-week lifecycle every market goes through.

The top makers are playing a different game

We looked at 600M+ Polymarket fills from October 2024 to March 2026. The headline: 87% of taker fills are sells. But the more interesting question is — who's on the other side?

In crypto micro-markets (5-minute Up/Down bets that now make up 76% of all fills), 90%+ of trades are taker_side=SELL. Someone is buying all those tokens. We checked the top 10 makers: are they betting on a direction, or doing something else?

They're not betting. In 96.5–100% of their markets, these makers are net buyers on both Up and Down. They don't care which side wins.

How it works: mint-and-sell

Polymarket uses a Conditional Token Framework (CTF) — traders can mint complete token sets from USDC and sell the side they don't want. This creates a two-sided flow where takers sell and makers collect:

The maker posts resting bids on both outcomes. Different takers — each betting on a different side — sell their unwanted tokens into those bids. The maker ends up holding matched pairs (1 Up + 1 Down) that are worth $1.00 at resolution regardless of the outcome. The spread between what they paid and $1.00 is pure profit.

We can't directly observe the minting step — the data only records trades, not token creation. But the pattern fits: 90%+ taker_side=SELL, makers buying both sides at near-100% consistency, and a platform architecture that makes minting easy. This evidence covers June 2025 onward, when crypto micro-markets reached meaningful volume.

Where does the 87% come from?

Now that we understand the mechanism — back to the headline number. SELL% went from 59% to 87% in 18 months:

The red line (SELL%) and the blue line (crypto micro-market share) track almost perfectly. December 2024 is the one exception: BUY surged when the US presidential election settled.

Most of the trend is just market mix. We ran a shift-share decomposition — separating "the mix of markets changed" from "people started trading differently":

About 63% is mix-related. Here's what that looks like over time — the grey line is what SELL% would have been if the market mix stayed frozen at January 2025 levels:

The other 37%? It's 92% driven by traditional Yes/No markets, where SELL went from 59% to 71%. But that shift is itself explained by market turnover: each new batch of markets starts at a higher SELL rate, then decays toward 57–62%. No batch trends upward on its own. The reason newer batches start higher? They have shorter lifespans (median went from 37 days to 3.7 days). And shorter markets have more selling. It's mix effects all the way down.

Shorter markets = more selling

Each dot is a market. Short-lived markets are heavily SELL-dominated, long-lived ones much less so:

For traditional markets, SELL% drops from 81% to 61% — a 20pp gap. This isn't about token prices: in long-lived markets, SELL% is flat at ~61% whether tokens trade at $0.20 or $0.80. It only shifts at the extremes (<$0.10 and >$0.90).

The likely reason: when a market resolves soon, makers can post aggressive bids without much risk of getting stuck holding tokens. The order book fills up with buy orders, and takers sell into them.

Every market goes through the same lifecycle

Here's the one finding that isn't explained by market mix. When you track SELL% by how old a market is, long-lived traditional markets all follow the same curve:

Week one: 66% SELL. By week two: ~61%. Then it stays there for months. This holds regardless of how long the market ultimately lasts.

What we're seeing is a shift from position-building (people minting and selling to establish their bets) to steady-state trading (people trading tokens that already exist). That transition takes about two weeks. This is the only finding in the whole analysis that reflects actual behavior change rather than market mix.

What to watch out for

1. We can't see the minting directly. We see the sells that result from it, and we see makers collecting both sides. But the minting step itself happens on-chain and isn't in the fills data.
2. The maker data starts June 2025. Crypto micro-markets barely existed before that. The strategy is clear for the 10 months we can observe it.
3. The 87% number is fragile. If crypto micro-markets disappeared tomorrow, SELL% would drop back toward 70%. The underlying patterns would all still be there.
4. Kalshi's 100% BUY is a data format thing, not a real behavioral difference. Their system just doesn't record sells.
5. Fees are asymmetric. Same wallet pays 11x more when selling than buying (avg 0.56 vs 0.05 in traditional markets). BUY and SELL have different fee schedules regardless of who's trading.

So what?

If you're comparing Polymarket to Kalshi, taker_side doesn't mean the same thing on both platforms. Kalshi records 100% BUY — not because nobody sells, but because that's how their system works. The 87% vs 100% comparison is an architecture difference, not a behavior difference.

If you're building models on Polymarket data, the 87% is not a stable number — it would drop to ~70% overnight if the crypto micro-markets went away. The lifespan gradient and the two-week lifecycle are the durable patterns.

And if you're a maker: someone is already running the matched-pair strategy across 50,000+ markets with near-perfect consistency. The question isn't whether the strategy works — it's how thin the spread has gotten.

Data: 600M+ Polymarket fills, Oct 2024 – Mar 2026. Dataset by Probalytics