polymarket-market-maker

name	polymarket-market-maker
description	Post two-sided GTC limit orders on Polymarket CLOB. Manages inventory skew, cancel/replace on price drift, and estimates rebate-eligible volume. Based on Akey et al. (2026): market-making reduces loss probability by 35.9 pp — the single strongest edge on Polymarket.

Polymarket Market Maker

Post two-sided GTC limit orders (bid + ask) on Polymarket CLOB to earn the bid-ask spread as a liquidity provider. This is a developer-only SDK primitive — not a retail "set and forget" skill. You will need to monitor inventory and tune spread/skew parameters to your specific markets.

Why this matters. Akey, Grégoire, Harvie & Martineau (2026) studied 1.4M Polymarket users and $20B in volume. Their headline finding: moving from pure taker to pure market-maker reduces the probability of losing money by 35.9 percentage points — more than 10× larger than any other behavioral predictor (overtrading: 5pp, category concentration: 13.6pp, extreme-price trading: 3pp). This is the single highest-edge behavior on Polymarket.

How it works

Each run (cron or manual):

  For each configured market:
    1. Fetch current mid price (external_price_yes)
    2. Compute bid/ask:
         bid = mid - spread/2  (round to 0.01 tick)
         ask = mid + spread/2  (round to 0.01 tick)
    3. Check inventory: if net_YES_skew > max_skew → pause bidding
                        if net_NO_skew  > max_skew → pause asking
    4. Check drift: if existing order price differs from new quote by > drift_threshold
                    → cancel and repost
    5. Post GTC orders:
         BID: client.trade("yes", amount=Q, price=bid, order_type="GTC")
         ASK: client.trade("no",  amount=Q, price=1-ask, order_type="GTC")
              ^ Synthetic sell: post NO buy at 1-ask_price instead of YES sell
                → earns maker rebates instead of paying taker fees on exits
    6. Log fee-equivalent volume for rebate estimation

Quote structure (synthetic sell on asks)

Polymarket YES and NO tokens are complementary (YES + NO = $1). Instead of posting a sell order on YES (which pays taker fees), this skill posts a buy order on NO at the complementary price:

Intent	Order posted	Price
Buy YES	BUY YES GTC	mid - spread/2
Sell YES (ask)	BUY NO GTC	1 - (mid + spread/2)

This earns maker rebates on both sides (Polymarket Maker Rebates Program). At a taker fee rate of 2%, the rebate formula is:

fee_equivalent = notional_USDC × taker_fee_rate × price × (1-price)

The --status and --cancel-all commands show cumulative fee-equivalent estimates.

Setup

1. Install SDK

   pip install -U 'simmer-sdk>=0.13.0'
   

2. Set API key

   export SIMMER_API_KEY=sk_live_...
   

3. Import markets to Simmer (markets must be imported before the skill can quote them)

   from simmer_sdk import SimmerClient
   client = SimmerClient(api_key=os.environ["SIMMER_API_KEY"], venue="polymarket")
   result = client.import_market("https://polymarket.com/event/will-btc-hit-100k-by-eoy-2026")
   print(result["market_id"])  # use this ID in MM_MARKETS
   

4. Configure markets

   export MM_MARKETS=abc123def456,xyz789...   # comma-separated Simmer market IDs
   export MM_SPREAD_PCT=0.04                  # 4% spread (2¢ each side at 50¢ mid)
   export MM_QUOTE_USDC=10                    # $10 USDC per side per market
   

5. Dry run (default — no orders placed)

   python market_maker.py
   

6. Live trading

   python market_maker.py --live
   

Parameters

Env Var	Default	Description
MM_MARKETS	—	Required. Comma-separated Simmer market IDs
MM_SPREAD_PCT	0.04	Spread as fraction of price (0.04 = 4 cents at 50¢); minimum 0.03 to avoid post-rounding quote collapse
MM_QUOTE_USDC	10.0	USDC per side per market
MM_MAX_SKEW_PCT	0.30	Pause one side when inventory skew exceeds this fraction
MM_DRIFT_THRESHOLD	0.02	Cancel/replace when quote drifts >2¢ from current mid
MM_TAKER_FEE_RATE	0.02	Taker fee rate for rebate estimation (sports: 0.03)
MM_MAX_MARKETS	5	Cap number of markets per run
MM_STATE_FILE	~/.simmer/market_maker_state.json	State persistence (order IDs + rebate log)

CLI commands

python market_maker.py              # Dry run
python market_maker.py --live       # Live trading
python market_maker.py --status     # Show active quotes + inventory
python market_maker.py --cancel-all # Cancel all open market-maker GTC orders
python market_maker.py --config     # Print active config

Cron / automaton mode

Run on a cron cadence (e.g., every 5 minutes) to keep quotes fresh. The skill emits {"automaton": {...}} when AUTOMATON_MANAGED=1 is set.

# Run every 5 minutes (OpenClaw / cron)
*/5 * * * * AUTOMATON_MANAGED=1 MM_MARKETS=abc123 MM_SPREAD_PCT=0.04 \
  python market_maker.py --live

Automaton output fields:

• markets — number of markets processed
• bids_placed / asks_placed — GTC orders submitted this run
• orders_cancelled — stale quotes cancelled
• fee_equiv_usd — estimated rebate-eligible volume this run
• cumulative_fee_equiv_usd — total across all runs

Market selection guidance

Good candidates for market-making:

• Active binary Polymarket markets with reasonable volume (>$5k/day)
• Prices between 5¢ and 95¢ (near-binary markets have poor risk/reward)
• Sports, crypto, and major political markets eligible for Maker Rebates Program

Poor candidates:

• Near-resolved markets (price < 2¢ or > 98¢) — high adverse selection, max loss is max
• Very low liquidity markets — spreads wide but fills rare; capital sits idle
• neg_risk (multi-outcome) markets — price arithmetic differs; this skill uses binary-only math

Inventory management

The skill tracks net inventory per market:

net_skew = shares_YES - shares_NO

When |net_skew| > max_skew_pct × (quote_usdc / price), the over-exposed side stops posting new quotes. This limits directional exposure but does not eliminate it — monitor inventory via --status.

Limitations (v1)

• Binary markets only. neg_risk markets have non-complementary token prices; the 1-ask_yes formula is incorrect for them. The skill will still quote them but the ask leg may misprice.
• No position close automation. The skill posts quotes but doesn't close inventory on resolution. Use client.auto_redeem() in a separate process or add --cancel-all before market resolution.
• State file is local. If running across multiple machines, the state file can desync. Each instance manages its own orders independently.

References

• Research: simmer-labs/shared-knowledge/research/2026-04-16-polymarket-winners-losers-study.md
• Synthetic sell: simmer-labs/shared-knowledge/research/2026-04-14-polymarket-quant-bot-copy-trading.md
• Risk disclaimer: DISCLAIMER.md