Machine Learning Surrogates for Parametric and Adaptive Optimal Execution
Abstract
Explore optimal order execution in the face of dynamic uncertainty with our model based on discrete time, stochastic transient price impact with Ludkovski's presentation. He, alongside collaborators Moritz Voss (UCLA) and Tao Chen (QRM), pioneer learning optimal strategies across various model configurations using a numerical algorithm that blends dynamic programming and deep learning. In this adaptive robust stochastic control framework, their controller dynamically learns model parameters, incorporating Bayesian uncertainty through a posterior uncertainty set. Witness a 3-way marriage between dynamic learning, dynamic robustness, and dynamic control within their proposed modeling framework. They extend their approach to address the 7-dimensional adaptive robust optimal order execution problem, presenting comparisons with alternative strategies.
Watch recording ↗About the speaker
Mike Ludkovski applies probability and statistical methods to large-scale societal issues, such as energy markets, computational finance and longevity analysis. His interests include Monte Carlo methods, stochastic games and stochastic control. Some of his recent work has focused on machine learning for financial derivatives, optimizing daily battery energy storage operations and designing groundwater trading markets.