Score-driven exponential random graphs: A new class of time-varying parameter models for temporal networks.

Area of Science:

• Network Science
• Statistical Modeling
• Time Series Analysis

Background:

• Real-world networks increasingly exhibit dynamic features.
• Existing exponential random graph models (ERGMs) often assume static parameters.
• There is a need for models that can capture temporal variations in network structures.

Purpose of the Study:

• To extend exponential random graph models (ERGMs) to accommodate time-varying parameters.
• To introduce score-driven exponential random graph models (SD-ERGMs).
• To demonstrate the utility of SD-ERGMs for analyzing temporal networks.

Main Methods:

• Developed a novel extension of ERGMs incorporating dynamic conditional score principles.
• Each model parameter evolves based on the score of the ERGM distribution.
• Utilized SD-ERGMs as both data-generating processes and filters.

Main Results:

• Demonstrated the flexibility of SD-ERGMs in modeling dynamic network data.
• Showcased the advantages of the dynamic SD-ERGM approach over static models.
• Successfully applied SD-ERGMs to temporal networks in finance and politics.

Conclusions:

• SD-ERGMs provide a powerful framework for analyzing temporal network dynamics.
• The proposed models can effectively capture time-varying parameters in real-world systems.
• SD-ERGMs offer advantages in network prediction and parameter estimation for dynamic networks.