Papers
Topics
Authors
Recent
2000 character limit reached

Community Network Auto-Regression for High-Dimensional Time Series

Published 10 Jul 2020 in stat.ME | (2007.05521v1)

Abstract: Modeling responses on the nodes of a large-scale network is an important task that arises commonly in practice. This paper proposes a community network vector autoregressive (CNAR) model, which utilizes the network structure to characterize the dependence and intra-community homogeneity of the high dimensional time series. The CNAR model greatly increases the flexibility and generality of the network vector autoregressive (Zhu et al, 2017, NAR) model by allowing heterogeneous network effects across different network communities. In addition, the non-community-related latent factors are included to account for unknown cross-sectional dependence. The number of network communities can diverge as the network expands, which leads to estimating a diverging number of model parameters. We obtain a set of stationary conditions and develop an efficient two-step weighted least-squares estimator. The consistency and asymptotic normality properties of the estimators are established. The theoretical results show that the two-step estimator improves the one-step estimator by an order of magnitude when the error admits a factor structure. The advantages of the CNAR model are further illustrated on a variety of synthetic and real datasets.

Citations (26)

Summary

  • The paper presents the CNAR model as an extension of the NAR model by integrating community structure using spectral clustering and a two-step weighted least squares method.
  • It establishes rigorous theoretical properties, including consistency and asymptotic normality, across various network configurations.
  • Empirical simulations show that CNAR outperforms traditional models in estimation and prediction accuracy for complex network-based time series data.

Community Network Auto-Regression for High-Dimensional Time Series

This paper presents the Community Network Auto-Regression (CNAR) model, a sophisticated extension of the Network Autoregression (NAR) model, for modeling high-dimensional time series with intrinsic network structures. The CNAR model is designed to capture the community structure within the network, enhancing the model's flexibility by accounting for heterogeneous network effects across different communities. This work is particularly relevant for dynamic systems represented as networks, where entities are interconnected, such as in social networks, financial markets, or biological systems.

Model Formulation and Estimation

The CNAR model extends the conventional NAR model by incorporating community structures within the network through the Stochastic Block Model (SBM). The nodes are grouped into communities, with each community exhibiting distinct autoregressive effects. The model also accommodates non-community-related latent factors to capture unknown cross-sectional dependencies, thus addressing the limitations of the NAR model that assumes a uniform autoregression coefficient across the network.

A two-step weighted least-squares estimation method is introduced to estimate the model parameters efficiently. Initially, a spectral clustering approach is employed to derive a rotated community membership matrix, avoiding a direct estimation of community membership. The model parameters are then estimated using the least squares method followed by a refinement step that utilizes the precision matrix estimated via the POET method, leading to improved estimation accuracy as confirmed by both theoretical and simulation results.

Theoretical and Empirical Analysis

The paper provides a rigorous theoretical foundation for the CNAR model. Specifically, it establishes the consistency and asymptotic normality of the estimators, demonstrating the advantages of the CNAR model over the traditional NAR model under a broader set of conditions. The model's robustness is statistically validated through various network generation processes, including stochastic block models, general low-rank spectral networks, and power-law distributed networks.

Empirical findings on synthetic data consistently reveal that CNAR achieves superior estimation and prediction accuracy compared to the NAR model. The CNAR model's flexibility allows it to generalize better across different network structures and temporal dynamics, as evidenced by simulations involving different network topologies, such as clusters of power-law graphs and random partitions.

Implications and Future Directions

The CNAR model offers a substantial improvement in modeling network-based time series data, particularly in systems where the network structure is complex and communities exert distinct network effects. Its application spans across numerous domains, from predicting financial returns to understanding the dynamics of epidemiological spreads.

Future research directions include extending this static network framework to time-varying networks, where the network's community structure evolves over time. Another promising area is the integration of CNAR models with high-dimensional covariates, facilitating variable selection methodologies to handle the large data volumes typical in modern datasets. Exploring these avenues could further enhance the CNAR model's versatility and utility in various applications.

The advancements laid out in this work mark a significant step forward in network-based time series analysis, offering a refined approach to extract, model, and predict the temporal dynamics of interconnected systems, with community detection playing a central role. In summary, the CNAR model provides an intricate yet instrumental framework for exploring high-dimensional systems where network interactions fundamentally shape the temporal evolution of the entities involved.

Paper to Video (Beta)

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.

Collections

Sign up for free to add this paper to one or more collections.