This paper formalizes autodeleveraging as online learning, providing robustness results and algorithms for better performance.
problem Autodeleveraging as a mechanism to restore solvency in perpetual futures markets when liquidation and insurance buffers are insufficient.
method Formalizes autodeleveraging as online learning on a PNL-haircut domain, using an algorithm to recover solvency.
result The optimized algorithm achieves about 2.6% of an upper bound on regret, reducing overshoot to $3M.