Sparse identification of nonlinear economic dynamical model

Reliable modeling of economic systems is essential for understanding their dynamic behavior. Traditional economic models rely on theoretical assumptions, often lacking accuracy in capturing real-world complexities. Existing studies focus primarily on data-driven analysis of microeconomic problems, with limited attention to modeling complex macroeconomic systems. We investigate the use of the Sparse Identification of Nonlinear Dynamics (SINDy) algorithm to identify governing equations of economic systems. Under ideal conditions, SINDy successfully reconstructs system dynamics, but its performance degrades when faced with noisy or limited data, as commonly found in economic systems. We show that two extended SINDy methods, SINDy-SR3 and E-SINDy, improve identification accuracy under these constraints. Additionally, noise filtering techniques, including the Kalman filter and Savitzky-Golay filter, enhance model robustness. Finally, we identify an investment–interest rate dynamical system from real-world economic data that aligns with established economic principles, demonstrating the feasibility of applying SINDy to practical economic modeling. Our results highlight the potential of SINDy for economic system modeling and provide guidelines for handling data quality challenges in real-world applications.