Nonparametric model reconstruction for stochastic differential equations from discretely observed time-series data.

Area of Science:

• Stochastic processes
• Time-series analysis
• Statistical modeling

Background:

• Stochastic differential equations (SDEs) model systems with inherent randomness.
• Estimating state-dependent drift and diffusion coefficients is crucial for understanding SDE dynamics.
• Existing methods often require pre-defined parametric forms for these coefficients.

Purpose of the Study:

• To develop a novel nonparametric scheme for estimating state-dependent drift and diffusion coefficients in SDEs.
• To provide a method that does not necessitate prior specification of parametric forms for coefficients.
• To enable fast and accurate estimation from potentially sparse or discrete time-series data.

Main Methods:

• Combines maximum likelihood estimation with kernel density estimation for nonparametric inference.
• Employs a local linearization method to handle discrete or sparse time-series observations.
• The scheme allows for efficient computation without assuming specific coefficient functions.

Main Results:

• Successfully developed a flexible scheme for estimating drift and diffusion coefficients.
• The method demonstrated effectiveness even with limited or irregularly sampled data.
• Achieved fast estimation through the local linearization technique.

Conclusions:

• The proposed nonparametric scheme offers a powerful tool for analyzing SDEs from time-series data.
• It overcomes limitations of parametric approaches by not requiring pre-specified coefficient forms.
• The method is efficient and robust, particularly for discrete or sparse data scenarios.