Since deploying inventory-mm-live (the real two-sided maker for 5m/15m Up/Down markets), it has been running with solid fill rates and controlled inventory. The core quoting brain (skewed inventory logic + seven anti-pick-off conditions) continues to perform as it did in shadow testing.
However, live execution has revealed several areas worth upgrading. Here are the most promising potential update paths I’m exploring for v2.
Current Strengths Recap
- Real GTC posting with safe cancel-then-repost repricing
- Strict capital + inventory guards
- Clean tail flattening via FOK at round end
- Full integration with the modular bot lifecycle (real PnL tracking)
Potential v2 Upgrades
1. Dynamic Skew Intensity & Volatility-Adaptive Spreads
Currently the skew is relatively static. In v2 I want to make it adaptive:
def get_target_inventory(self, vol: float, imbalance: float):
base_skew = self.base_inventory_target
vol_factor = max(0.3, 1.0 - (vol / self.high_vol_threshold))
imbalance_factor = imbalance * self.imbalance_sensitivity
return base_skew * vol_factor + imbalance_factor
This would tighten spreads in low-vol chop and widen them (with stronger skew) during high volatility to reduce adverse selection.
2. Multi-Level Anti-Pick-Off Engine
Expand the current seven pull conditions with:
- Order book imbalance delta (sudden bid/ask wall removal)
- CEX correlation break (if Binance moves sharply against our quotes)
- Time-to-expiry aggression (tighter quotes near round end)
- Cross-market regime correlation (if related 5m/15m markets diverge)
3. Hybrid Maker + Sweep Logic
Combine inventory-MM with elements of the Fair Value Sweep:
- When a strong stale ask appears (CEX fair value edge), take it as a taker while still maintaining the maker ladder.
- This turns the bot into a true hybrid: earn spread on normal flow + capture occasional high-conviction stale quotes.
4. Advanced Tail Management
Instead of always FOK-flattening the unpaired tail:
- Intelligent partial tail hedging using correlated markets
- Dynamic hold vs flatten decision based on remaining time and volatility
- Optional micro-hedge on Binance/Bybit for large unpaired positions
5. Machine-Learning Regime Classifier
Replace the current rule-based regime detection with a lightweight model trained on:
- Realized volatility clusters
- Order flow features
- Historical fill quality
- CEX lead/lag patterns
This would allow smarter quoting frequency and skew in different market regimes.
6. Capital Efficiency Improvements
- Introduce temporary leverage via flash loans or spot borrowing for short-duration rounds (with strict repayment before round end)
- Better pair-completion logic to reduce idle capital
Expected Impact
These upgrades aim to:
- Increase fill rate and spread capture in favorable regimes
- Reduce tail risk and adverse selection
- Improve overall Sharpe while keeping maximum drawdown tightly controlled
- Make the strategy more robust across different volatility environments
Current Status
The base inventory-mm-live remains live on conservative size. I’m implementing v2 components one by one in a staging environment before full rollout.
The modular architecture makes these upgrades relatively clean — most changes live in the quoting brain and risk module without touching the core lifecycle or execution engine.
Market making on Polymarket short markets continues to be one of the more sustainable edges when executed with proper inventory control and risk discipline. These upgrades should push performance even further.
I’ll share detailed results once v2 has sufficient live data.
If you have more questions, please feel free to contact me at any time: https://t.me/FatherSon97
Top comments (0)