HRT's Dunning on Market Worry & AI Upskilling

🔑 Enhanced Key Takeaways

• •Iain Dunning previously worked at DeepMind before joining Hudson River Trading as head of AI research[1][2].
• •Hudson River Trading applies AI for short-term price predictions and more efficient trading execution, distinguishing it from traditional machine learning approaches[1][2].
• •HRT has shifted toward medium-frequency trading with average holding times around five minutes, holding about 25% of trading capital overnight[3].
• •HRT decomposes AI trading challenges into subproblems like prediction (e.g., forecasting security prices over next five days) and optimization for portfolio allocation under risk constraints[4].

🛠️ Technical Deep Dive

• •AI systems at HRT are framed as optimization problems to maximize trading revenue while respecting constraints like risk limits, regulatory compliance, and good market participant behavior[4].
• •Decomposition approach: Separate into prediction AI (forecasts future prices, e.g., security prices at end of next five days without considering market reactions) and optimization AI (allocates capital based on predictions, diversifying to manage risk using financial mathematics)[4].
• •Uses hundreds or thousands of GPUs to crunch data quickly for execution efficiency and spotting sophisticated patterns or discrepancies in trading[2].

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline