Related Market Spread Agent Skill

Strategy: Polymarket · Related-Market Spread

When to use

Use this when the user asks for relative-value trades, discrepancy checks, linked outcomes, or questions like "this market is mispriced versus that one."

What the agent should look for

• At least two exact market slugs from POST /v3/markets/search.
• Resolution logic that is genuinely comparable across the two markets.
• Spread between filled-trade prices that is wide versus historical behavior.
• Enough historical filled data on both markets for the requested backtest window.
• Ability to enter and exit both legs (or acknowledge unavailable liquidities before entering).

Backtest fit with filled data

Partial fit. Backtests built from filled TradeTick data are useful for checking whether the spread relationship has widened and then compressed in history. They cannot guarantee simultaneous fills or full cross-book liquidity.

Use this as a generated archetype: the strategy should be adapted in code before use, not treated as a finished edge.

Strategy logic

Track the latest filled price for both legs. Enter when spread is above a threshold and exits when spread narrows or max holding duration expires.

Nautilus strategy shape

• Subscribe to TradeTick for both instruments.
• Keep the latest filled price per instrument.
• Compute spread each tick using the latest values.
• Enter and flatten only when spread thresholds are hit.
• Manage stale-state safety (e.g., skip entries until both legs have recent fills).

Example strategyConfig

{
  "primary_instrument_id": "AUTO_FROM_MARKET_A",
  "secondary_instrument_id": "AUTO_FROM_MARKET_B",
  "spread_entry": 0.08,
  "spread_exit": 0.03,
  "order_size": 10,
  "max_holding_ticks": 120
}

Iteration knobs

| Knob | Effect |

|---|---|

| spread_entry | Higher = only trade stronger relative-value discrepancies. |

| spread_exit | Lower = wait for fuller normalization before exiting. |

| max_holding_ticks | Lower = tighter time risk control if convergence does not happen. |

Failure modes

• Markets with different outcome mappings can produce a real structural spread, not an inefficiency.
• Resolution rules can shift, changing what "discrepancy" means.
• One leg may look tradable in backtest and be unavailable or illiquid live.
• TradeTick replay cannot prove simultaneous execution across legs.

User-facing framing

"This is a relative-value archetype, not a guaranteed edge. It is built from filled TradeTick backtests and exact market slug discovery first, then adapted into custom NautilusTrader code before any deployment."