Estimating Behavioral Agent-Based Models for Financial Markets through Machine Learning Surrogates

Traditional economic assumptions such as rational, representative agents and efficient market hypothesis failed to explain the macro-behavior of financial markets. On the other hand, agent-based approach proves high potentials in modeling bounded rational and heterogeneous micro-behaviors. This approach captures important stylized facts of financial markets. However, the high complexity of estimating agent-based models parameters precludes using these models in the forecasting process. This problem limits the applicability of agent-based models in decision making and policy formulation processes. Thereafter, this research aims at introducing a prospect for estimating agent-based models for financial markets through surrogate modeling approach. Surrogate models are considered as novel parameter estimation method in economics though it is a well-defined method in engineering. Few efforts have been spent to estimate parameters using surrogate models.