Optimizing Regret
How to make better decisions by learning from past mistakes mathematically
When you make decisions based on costs—whether picking investments or allocating resources—your mistakes follow a pattern: you tend to regret choices most when costs were high. This paper builds a complete mathematical toolkit for minimizing regret by exploiting this pattern, showing that the best strategy is often contrarian (doing the opposite of what costs suggest) and proving these methods converge to optimal solutions quickly, even with limited real-world data.
Portfolio managers and AI systems that allocate resources can now use these mathematical rules to systematically reduce regret and improve returns without needing perfect information. The framework applies directly to investment tilting strategies and large language models choosing how to distribute computational resources, making it practical for anyone optimizing decisions under uncertainty.