Linear Model Regression on Time-series Data: Non-asymptotic Error Bounds and Applications

TL;DR

This paper derives deterministic, non-asymptotic error bounds for linear regression models applied to time-series data, emphasizing the influence of system symmetry and eigenvalue multiplicity on model accuracy.

Contribution

It introduces the first deterministic error bounds for linear time-series regression, highlighting the effects of eigenvalue properties on model fitting.

Findings

Error bounds depend on eigenvalue multiplicity and symmetry.

Provides insights into modal properties of the underlying system.

Enhances understanding of model accuracy in deterministic settings.

Abstract

Data-driven methods for modeling dynamic systems have received considerable attention as they provide a mechanism for control synthesis directly from the observed time-series data. In the absence of prior assumptions on how the time-series had been generated, regression on the system model has been particularly popular. In the linear case, the resulting least squares setup for model regression, not only provides a computationally viable method to fit a model to the data, but also provides useful insights into the modal properties of the underlying dynamics. Although probabilistic estimates for this model regression have been reported, deterministic error bounds have not been examined in the literature, particularly as they pertain to the properties of the underlying system. In this paper, we provide deterministic non-asymptotic error bounds for fitting a linear model to the observed time-series data, with a particular attention to the role of symmetry and eigenvalue multiplicity in the underlying system matrix.