Interview
Michael Kearns: Algorithmic Trading and the Role of AI in Investment at Different Time Scales
- Algorithmic adoption in finance has progressed through stages mirroring broader societal automation, beginning with the electronic conversion of exchanges that enabled computers to submit orders via API faster and more efficiently than humans.
- Algorithms initially dominated "optimized execution" problems, where large brokerage firms manage complex trades for institutional clients (e.g., buying massive stakes in Apple over a day) to minimize market impact and price manipulation across multiple electronic venues.
- High-frequency trading (HFT) represents a related but distinct algorithmic capability focused on identifying temporary mispricings or predicting directional moves based on granular, low-level exchange data.
- Statistical arbitrage ("stat arb") currently serves as the effective "sweet spot" for quantitative trading, relying on directional price predictions with validity periods ranging from a few seconds to a few days.
- Long-term investment strategies (e.g., Warren Buffett-style horizons of 10–20 years) remain resistant to full algorithmic automation because they require navigating economic cycles, recessions, wars, and geopolitical events.
- Long-term investing necessitates a deep understanding of human nature and the integration of diverse data sources across wildly different time scales, which machines currently struggle to synthesize compared to humans.
- A workshop co-sponsored by the speaker and the Federal Reserve Bank of Philadelphia on machine learning for macroeconomic prediction indicated that the field is still in early stages regarding long-term forecasting capabilities.
- The speaker concludes that roles requiring long-term risk appetites and nuanced views on political and economic landscapes are secure from algorithmic displacement, noting there is no imminent "robo Warren Buffett."