Research
Counterfactual Regret Evolutionary Agents (CREA): A Mathematical Framework for Adaptive Decision-Making in Financial Systems under Deep Uncertainty
ABSTRACT
A rigorous mathematical framework for adaptive decision-making under deep uncertainty. CREA integrates probability spaces, stochastic price dynamics, structural causal discovery (NOTEARS, DAG-GNN, PCMCI), fuzzy ambiguity modeling, CVaR-based tail-risk control, Wasserstein distributionally robust optimization, transformer memory architectures, and meta-evolutionary optimization to discover strategies that remain stable across multiple counterfactual market futures.
Causal Discovery
Structural equation learning with NOTEARS continuous acyclicity and PCMCI time-lagged graph induction.
Counterfactual Regret
Branching scenario evaluation with regret minimization guarantees across diverging market regimes.
Wasserstein DRO
Worst-case distribution optimization over Wasserstein metric balls with certified radius \(\epsilon\).
Tail-Risk Control
Rockafellar-Uryasev Conditional Value-at-Risk (CVaR) bounds controlling severe tail drawdowns.