DEV Community

FatherSon
FatherSon

Posted on

Shadow Two-Sided Inventory Market Maker v2: Realistic Simulation for Polymarket Trading Bots

Market making on Polymarket (and similar prediction markets) is brutally unforgiving. Version 1 of most inventory-based MM engines fail silently because their simulation engines introduce systematic adverse selection. This new Inventory-MM v2 Shadow approach solves that by going fully shadow — no orders on the book, only hypothetical fills derived from real trade prints.

The Core Problem with v1 Simulation

Traditional backtesters/simulators use an isSimFilled model that only registers a fill when price crosses the resting quote in the adverse direction:

  • Your bid gets filled only when the market trades down into it.
  • Your ask gets filled only when the market trades up into it.

This creates pure adverse selection in simulation: you never capture the benign two-way flow that real market makers earn from natural buyer/seller crossing. The result? Paper PnL looks terrible even when the strategy logic is sound. You cannot honestly evaluate a maker strategy under such a model.

Inventory-MM v2 Shadow Architecture

Key Principle: Place zero orders on the engine. Maintain a virtual quote ladder and infer fills from real-time trade streams.

Quote Generation Logic (desired quotes):

  • Base: mid - halfSpread
  • Dynamic inventory skew: skew bid/ask based on current net position (push quotes away from inventory to encourage mean-reversion)
  • Pair-margin guard: prevent simultaneous long/short exposure on highly correlated outcomes
  • Stale book / feed divergence pull: widen or cancel quotes if order book depth or price feeds diverge beyond thresholds
  • Toxic trend filter: pause or skew aggressively during detected momentum / information events

Fill Inference Engine (the shadow magic):

# Pseudocode - real trade print handler
def on_real_trade_print(token, price, size, side, timestamp):
    our_bid = current_desired_bid_price
    our_ask = current_desired_ask_price

    if side == "SELL" and price <= our_bid:
        # A real seller hit our hypothetical bid
        filled_size = min(size, remaining_bid_liquidity)
        process_shadow_fill(token, our_bid, filled_size, "BUY", timestamp, is_shadow=True)

    elif side == "BUY" and price >= our_ask:
        # A real buyer hit our hypothetical ask
        filled_size = min(size, remaining_ask_liquidity)
        process_shadow_fill(token, our_ask, filled_size, "SELL", timestamp, is_shadow=True)
Enter fullscreen mode Exit fullscreen mode

This captures actual observed flow at your quoted levels — not artificial crossing.

Full Shadow PnL & Inventory Tracking

Every fill event carries a shadowPnl flag. The system maintains:

  • Real-time hypothetical inventory across outcomes
  • Pair-locking constraints (e.g., election winner + popular vote)
  • End-of-window inventory dump simulation
  • Resolution PnL calculation (critical on Polymarket binary/ multi-outcome markets)

All metrics are instrumented identically to a live runner, allowing direct apples-to-apples comparison between shadow and production readiness.

Why This Matters for Polymarket Trading Bots

Polymarket’s order book + trade print feed (via RTDS) provides the perfect data source for shadow market making. You can now:

  • Test two-sided liquidity provision strategies on high-volume election, crypto, or sports markets without risking a single USDC.
  • Measure true adverse selection vs. benign flow in live conditions.
  • Iterate quote skew, spread curves, and risk parameters with statistical confidence.
  • Build readiness scoring that only promotes strategies to live capital once shadow performance clears strict gates.

This approach bridges the massive gap between naive backtesting and production market making. Once shadow readiness metrics (Sharpe, max drawdown, win rate on inventory cycles, etc.) look robust, the transition to live quoting becomes a deliberate, lower-risk engineering step.

Technical Advantages Over Traditional Simulators:

  • Zero look-ahead bias in fill modeling
  • Captures real microstructure (trade sizes, clustering, toxic flow)
  • Naturally incorporates Polymarket’s discrete tick sizes and liquidity dynamics
  • Enables proper evaluation of inventory control logic under real volatility regimes

The shadow paradigm turns market making from a high-risk guessing game into a measurable, iteratively improvable quantitative strategy — exactly what serious Polymarket trading bots require in 2026.

If you have more questions, please feel free to contact me at any time: https://t.me/FatherSon97

#PolymarketTradingBot #TradingBot #CryptoTradingBot #PolymarketBot #DeFiTrading #MarketMaking #InventoryMM #ShadowMarketMaker #PredictionMarkets #DeFiBots #QuantTrading #AutomatedMarketMaking #PolymarketStrategy #CryptoDev #HighFrequencyTrading

Top comments (0)