Abstract:We develop a modeling framework for dynamic function-on-scalars regression, in which a time series of functional data is regressed on a time series of scalar predictors. The regression coefficient function for each predictor is allowed to be dynamic, which is essential for applications where the association between predictors and a (functional) response is time-varying. For greater modeling flexibility, we design a nonparametric reduced-rank functional data model with an unknown functional basis expansion, which is data-adaptive and, unlike most existing methods, modeled as unknown for appropriate uncertainty quantification. Within a Bayesian framework, we introduce shrinkage priors that simultaneously (i) regularize time-varying regression coefficient functions to be locally static, (ii) effectively remove unimportant predictor variables from the model, and (iii) reduce sensitivity to the dimension of the functional basis. A simulation analysis confirms the importance of these shrinkage priors, with notable improvements over existing alternatives. We develop a novel projection-based Gibbs sampling algorithm, which offers unrivaled computational scalability for fully Bayesian functional regression. We apply the proposed methodology (i) to analyze the time-varying impact of macroeconomic variables on the U.S. yield curve and (ii) to characterize the effects of socioeconomic and demographic predictors on age-specific fertility rates in South and Southeast Asia.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to model the relationship between the functional data of time series and the scalar predictor variables of time series in dynamic function - on - scalars regression (DFOSR). Specifically, the paper focuses on the fact that in applications, the association between predictor variables and functional responses may change over time. Therefore, the main research objectives include: 1. **Modeling time - dependence and time - variation**: The paper proposes a Bayesian dynamic function - on - scalars regression model, aiming to simultaneously model the within - curve (functional) dependence, between - curve (time) dependence, and the dynamic association between scalar predictor variables and functional responses. This solves the problem that existing methods fail to fully consider time - dependence and time - variation when dealing with time - series functional data. 2. **Improving modeling flexibility and uncertainty quantification**: By designing a non - parametric low - rank functional data model and allowing for unknown functional basis expansion, the model can adapt to a wider range of application scenarios while providing appropriate uncertainty quantification. This is crucial for accurately assessing the uncertainty of model parameters. 3. **Introducing shrinkage priors to prevent overfitting**: In order to prevent the model from overfitting while maintaining its flexibility, the paper introduces shrinkage priors, which can simultaneously (i) make the time - varying regression coefficient functions locally static; (ii) effectively remove unimportant predictor variables from the model; (iii) reduce the sensitivity to the dimension of the functional basis. 4. **Developing an efficient posterior inference algorithm**: To achieve the computational scalability of full - Bayesian functional regression, the paper develops a novel projection - based Gibbs sampling algorithm. This enables efficient posterior inference even in high - dimensional data. 5. **Verification in practical applications**: The paper verifies the effectiveness of the proposed model through two practical cases: one is to analyze the time - varying impact of macro - economic variables on the US yield curve; the other is to characterize the impact of socio - economic and demographic predictor variables in South and Southeast Asia on age - specific fertility rates. Overall, the paper aims to address the challenges encountered in dealing with time - series functional data by proposing a new dynamic function - on - scalars regression model and its corresponding computational methods, especially in cases where the association between predictor variables and functional responses may change over time.

Dynamic Function-on-Scalars Regression

Bayesian Function-on-Scalars Regression for High-Dimensional Data

Longitudinal Dynamic Functional Regression

Scalar-on-function regression: Estimation and inference under complex survey designs

Large-Scale Dynamic Predictive Regressions

Functional Linear Regression with Mixed Predictors

Functional Autoregression for Sparsely Sampled Data

Partially Functional Linear Regression in High Dimensions

Dynamic Shrinkage Processes

Dynamic sparsity on dynamic regression models

Bayesian Function-on-Function Regression for Spatial Functional Data

Regression with Functional Errors-in-Predictors: A Generalized Method-of-Moments Approach

Unified statistical inference for a nonlinear dynamic functional/longitudinal data model

Dynamic Single-Index Model for Functional Data

Functional linear regression analysis for longitudinal data

Robust Functional Regression with Discretely Sampled Predictors

Ultra-efficient MCMC for Bayesian longitudinal functional data analysis

Nonlinear Multivariate Function-on-function Regression with Variable Selection

High-dimensional functional time series forecasting: An application to age-specific mortality rates

A Bayesian approach to functional regression: theory and computation

Dynamic Multivariate Functional Data Modeling Via Sparse Subspace Learning