Strategy · Low risk · Intermediate

Spread farming

Passive market making on prediction markets. Post limit orders on both sides of the book and earn the bid-ask spread.

Published Apr 10, 2026

The idea

Every trade has a maker and a taker. Takers pay the spread; makers earn it. Over many trades, if you can keep inventory balanced, you harvest the spread as profit — even with zero directional view.

Polymarket’s CLOB doesn’t pay maker rebates, but the spread itself is 2–5 cents on most markets — far wider than equity spreads. That’s the edge.

How polybot implements it

spread_farm posts two-sided limit orders on a universe of markets. The core loop:

1. Subscribe to the top-of-book for each target market.
2. Compute a fair-value estimate (last trade, volume-weighted, or from an AI plugin).
3. Post bid at fair - half_spread, ask at fair + half_spread.
4. On fill, re-quote immediately with updated inventory.
5. Every refresh_seconds, reprice to stay current even without fills.

Inventory is the key risk. polybot tracks net delta per market and skews quotes to encourage unwinding: if you’re long 100 shares of YES, bid lower and ask more aggressively.

# from src/polybot/strategies/spread_farm.py — simplified
fair = fair_value_estimator(market)
inventory_skew = -self.alpha * self.inventory[market.id] / self.max_inventory
half = self.config.half_spread + inventory_skew
self.quote(market.id, bid=fair - half, ask=fair + half, size=self.config.quote_size)

Where this works

• High-volume, mid-lived markets. Markets that trade for days or weeks, with at least a few dozen daily fills.
• Tight books. Where the existing spread is narrow, you’ll compete on price; you need to be willing to quote the inside or skip the market.
• Capital-rich operators. Spread farming wants a lot of markets, small size each — the edge is volume × spread, and volume needs breadth.

Where it fails

• Toxic flow. A whale with private information runs through your quotes; you pick up the wrong side. Mitigation: quote-pull on sudden volume spikes; polybot’s toxicity_monitor disables markets with adverse-selection metrics above threshold.
• One-sided markets. Near resolution, all trades go one way. Your YES asks get lifted; NO asks don’t. Inventory skew should un-balance but may not be enough — polybot’s late-market shutdown disables quoting in the last T hours.
• Gas costs on cancels. Polymarket is on-chain. Every re-quote costs gas. polybot batches re-quotes and prefers modify-in-place where the venue supports it.

Configuration

polybot strategy enable spread_farm
polybot strategy config spread_farm \
    --half-spread 0.02 \
    --quote-size 50 \
    --max-inventory 300 \
    --refresh-seconds 30 \
    --cutoff-hours-before-resolution 6 \
    --markets-include "crypto,politics"

Risk profile

FAQ

Do I need an AI model for this? No, but it helps. A better fair-value estimate (e.g. Perplexity checking for just-breaking news) means your spread gets lifted less often by informed flow.

How much capital is this worth on? Scales sublinearly. $5k → $50k is roughly 5× revenue; $50k → $500k is maybe 2×, because you start running out of markets with room to quote inside the book.

Does polybot compete with itself if I run spread_farm and ai_model? Yes, if they target the same market. Use --markets-exclude to partition, or run them in different categories.

Source: src/polybot/strategies/spread_farm.py.