Munich Data Trading

Inspiration

Financial markets move fast — but news moves even faster. With modern AI capable of digesting language, patterns, and sentiment in milliseconds, we wanted to explore how far real-time decision-making could go. The Optiver challenge was the perfect playground to fuse market intuition with AI-driven signal extraction.

What It Does

Munich Data Trading is an AI-powered market-making and arbitrage engine that reacts to news and market data in real time.

It:

• Parses news for sentiment, shock factor, and directional bias
  
• Quotes dynamic bid/ask prices based on volatility signals
  
• Detects cross-product mispricings and executes arbitrage
  
• Continuously adapts its strategy as new data flows in

How We Built It

We designed a modular pipeline with three key layers:

1. Real-Time NLP Processing
   Lightweight transformer models tuned for sentiment and event detection.

2. Strategy Engine
   A custom market-making algorithm adjusting spreads, inventory, and risk thresholds depending on predicted impact.

3. Arbitrage Module
   A pricing consistency checker that acts when spreads open across related assets.

Challenges We Ran Into

• Finding the sweet spot between aggressive quoting and stable risk exposure
  
• Shrinking NLP latency enough to keep signals meaningful
  
• Dealing with noisy sentiment where “bad news” wasn’t actually bad
  
• Avoiding false arbitrage triggers during high-volatility bursts
  

Accomplishments That We’re Proud Of

• Achieving consistent profitability in simulation, even during chaotic news shocks
  
• Building a news-to-quote loop that reacts within milliseconds
  
• Creating a surprisingly effective cross-product arbitrage detector
  
• Outperforming several hand-tuned baseline strategies
  

What We Learned

• Real-time trading rewards speed and punishes complexity
  
• Market-making is mostly a risk game hiding behind predictions
  
• News sentiment is powerful but must be context-aware
  
• Clean architecture = easier iteration and better outcomes
  

What's Next for Munich Data Trading

• Expanding to multi-asset and cross-market arbitrage
  
• Integrating reinforcement learning for adaptive quoting
  
• Stress-testing under extreme and adversarial volatility
  
• Adding richer event classification beyond raw sentiment
  

Built With

• Python
  
• Transformer-based NLP models
  
• NumPy / Pandas
  
• Custom market-making logic
  
• Optiver simulation environment
  

Built With

• python