Constructing networks by filtering correlation matrices: a null model approach.

Area of Science:

• Network science
• Data analysis
• Economic modeling

Background:

• Network analysis commonly uses correlation matrix thresholding.
• This method faces criticism for failing to filter spurious correlations.
• Alternative methods are needed to improve network construction from correlation data.

Purpose of the Study:

• To propose a novel method for creating networks from correlation matrices.
• To address limitations of traditional thresholding techniques.
• To develop a robust approach for identifying significant correlations.

Main Methods:

• A new algorithm based on optimization with regularization is proposed.
• Edges are formed when pairwise correlations are unexpected from a null model.
• The method incorporates model selection to identify the most plausible null model.

Main Results:

• The proposed method effectively creates networks by identifying unexpected correlations.
• The configuration model for correlation matrices is frequently preferred over standard null models.
• For country-level export data, the method demonstrates superior prediction of main exported products.

Conclusions:

• The optimization-based network construction method offers an improvement over simple thresholding.
• This approach provides a flexible framework adaptable to various null models.
• The method shows practical utility in economic data analysis and prediction.