Parameter Estimation of Linear Stochastic Differential Equations with Sparse Observations

Yuecai Han, Zhe Yin, Dingwen ZhangSymmetry 2022 · OpenReview

TL;DR

This paper tackles the challenge of estimating parameters in linear stochastic differential equations (LSDEs) when data is sparse and irregularly sampled, a common scenario in fields like pharmacokinetics and biometrics. They introduce a novel kernel-weighted maximum likelihood estimation method that provides a consistent and asymptotically normal estimator, even with limited observations.

What Problem Does This Paper Address?

Many real-world systems, from financial markets to biological processes, are modeled using linear stochastic differential equations (LSDEs). Accurately estimating the parameters that govern these systems is crucial for prediction and understanding. However, traditional estimation methods often assume continuous or regularly spaced observations. In practice, data is frequently sparse and collected at irregular intervals, making these standard methods ineffective. This paper addresses this gap by developing a robust parameter estimation technique specifically designed for such infrequent and irregularly sampled data, a common issue in fields like pharmacokinetics where patient data is collected sporadically.

Methods

The core of the paper's contribution lies in a kernel-weighted maximum likelihood estimation approach. Instead of directly using the likelihood function, which is impossible with sparse data, they construct a "smoothed" log-likelihood function. This smoothing is achieved by using a kernel function to weigh observations based on their proximity to a target time point. This allows them to leverage both "forward" and "lagged" observations to impute missing information. They then derive a score function from this smoothed likelihood and propose an estimator by solving the score equation. The paper rigorously proves the consistency and asymptotic normality of this estimator, providing a theoretical foundation for its reliability.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and simulation studies.

Key Insights and Ideas

  • Bridging the Gap for Sparse Data: The most significant insight is the development of a method that effectively handles sparse and irregularly sampled data for LSDE parameter estimation, a problem that has limited practical solutions.
  • Kernel Smoothing for Imputation: The clever use of kernel weighting to "smooth" observations and impute missing information within the likelihood framework is a key innovation. This allows for the utilization of all available data, even if it's not perfectly aligned.
  • Theoretical Guarantees: The paper doesn't just propose a method; it provides strong theoretical backing by proving the consistency and asymptotic normality of the proposed estimator, which is crucial for its adoption in scientific research.
  • Practical Applicability: The method is designed to be computationally efficient, making it suitable for real-world applications where large datasets might be involved.

Key Results

The paper demonstrates that the proposed kernel-weighted estimator is asymptotically consistent. Furthermore, it establishes the asymptotic normality of the estimator with a convergence rate of $(nh_n)^{1/2}$, which, while slower than the typical $n^{1/2}$ for dense data, is a significant achievement for sparse data scenarios. Simulation studies with moderate sample sizes show that the estimator performs well, with bias diminishing as sample size increases and bandwidths are appropriately chosen.

Limitations and Future Work

The authors acknowledge that the method's performance is dependent on the proper choice of bandwidth, which is a common challenge in kernel-based methods. Future work could explore more adaptive bandwidth selection strategies. Additionally, the paper focuses on linear SDEs; extending this methodology to nonlinear SDEs would be a valuable next step. The authors also suggest investigating parameter estimation for more generalized drift terms and exploring nonparametric estimation for drift functions in future research.

Summary updated

Abstract

We consider parameter estimation for linear stochastic differential equations with independent experiments observed at infrequent and irregularly spaced follow-up times. The maximum likelihood method is used to obtain an asymptotically consistent estimator. A kernel-weighted score function is proposed for the parameter in drift terms. The strong consistency and the rate of convergence of the estimator are obtained. The numerical results show that the proposed estimator performs well with moderate sample sizes.
BibTeX
@inproceedings{han2022parameter,
  author = {Yuecai Han and Zhe Yin and Dingwen Zhang},
  title = {{Parameter Estimation of Linear Stochastic Differential Equations with Sparse Observations}},
  booktitle = {Symmetry},
  year = {2022},
  url = {https://openreview.net/forum?id=k2LHNfNww7},
}
Figure