McLovin

Inspiration

We wanted to play market maker for all 8 IMC markets at once, without clicking ourselves to death in the GUI. So we built one "slightly overconfident" bot that does the market‐making for us.

What it does

• Connects to the IMC exchange via SSE + REST.
  
• Computes a fair value per market:
  • M1–M2: Eisbach water level & flow (HND Bayern)
  • M3–M4: temperature & humidity (timeanddate)
  • M5–M6: arrivals/departures (Aerodatabox)
  • M7–M8: ETF & 80–100 strangle built from all of the above
    
• Trades around the mid price with position limits, profit-take and stop-loss.

How we built it

• A single generic 'MarketBot' class from 'imcity_template.py' (SSE listener + REST orderbook polling).
  
• Separate metric modules per market (Eisbach, weather, flights, ETF) that implement the official payoff formulas.
  
• Per-market parameters ('max_position', 'threshold', 'profit_take', 'order_volume') to tune risk/aggro level.

Challenges we ran into

• HTML scraping of public data.
  
• Keeping external APIs stable under timeouts and rate limits.
  
• Multiple bots polling and trading in parallel without overloading the exchange or ourselves.

Accomplishments that we're proud of

• One unified bot controlling all 8 markets instead of 8 half-broken scripts.
  
• Metric calculations that actually match the IMC slides (including the funky weather window and ETF/strangle).
  
• A risk setup that can go up to ±80 contracts per market without (well, or with) instantly exploding the PnL.

What we learned

• Market making is mostly fair-value estimation + inventory risk management + knowing when to stop averaging down.
• Real-world data pipelines are messier than toy examples, and they directly hit your PnL.
• Good modularization (one bot core, many metric modules) saves a ton of debugging time under hackathon pressure.

What's next for McLovin

• A small dashboard for live PnL, positions and metrics.
• Maybe a v2 that uses bandits/RL to learn how much Elon-Musk-level risk is actually optimal.

Built With

• chatgpt
• python