Stable Multiple Time Step Simulation/Prediction from Lagged Dynamic Network Regression Models

TL;DR

This paper introduces a novel algorithm for simulating and predicting dynamic networks over multiple time steps, addressing the common issue of model degeneration in existing methods, and demonstrates improved performance over standard approaches.

Contribution

The authors develop a smoothing-based algorithm for stable multi-step simulation of lagged dynamic network regression models, enhancing prediction accuracy and scalability.

Findings

Significantly improves multi-step prediction over standard DNR(V) models.

Performs comparably to complex frameworks like SAOM and STERGMs for small networks.

Outperforms existing methods over long time intervals and large networks.

Abstract

Recent developments in computers and automated data collection strategies have greatly increased the interest in statistical modeling of dynamic networks. Many of the statistical models employed for inference on large-scale dynamic networks suffer from limited forward simulation/prediction ability. A major problem with many of the forward simulation procedures is the tendency for the model to become degenerate in only a few time steps, i.e., the simulation/prediction procedure results in either null graphs or complete graphs. Here, we describe an algorithm for simulating a sequence of networks generated from lagged dynamic network regression models DNR(V), a sub-family of TERGMs. We introduce a smoothed estimator for forward prediction based on smoothing of the change statistics obtained for a dynamic network regression model. We focus on the implementation of the algorithm, providing a series of motivating examples with comparisons to dynamic network models from the literature. We find that our algorithm significantly improves multi-step prediction/simulation over standard DNR(V) forecasting. Furthermore, we show that our method performs comparably to existing more complex dynamic network analysis frameworks (SAOM and STERGMs) for small networks over short time periods, and significantly outperforms these approaches over long time time intervals and/or large networks.