Autodeleveraging as Online Learning
New algorithm reduces autodeleveraging overshoot from $51.7M to just $3M during market crashes.
Deep Dive
Researchers Tarun Chitra and team published 'Autodeleveraging as Online Learning,' formalizing how crypto exchanges handle insolvency during crashes. Their optimized algorithm, tested on Hyperliquid's October 2025 stress event, reduced over-liquidation of trader profits from $51.7M to $3M—a 94% improvement. This provides exchanges with a simple, implementable framework to manage risk more fairly and efficiently during extreme volatility, treating ADL as a sequential control problem.
Why It Matters
Could prevent millions in unnecessary trader losses during future crypto market meltdowns, improving exchange stability.