Measure cross-sectoral structural similarities from financial networks

Area of Science:

• Computational finance
• Network science
• Auditing analytics

Background:

• Auditing plays a critical role in financial stability within a dynamic global economy.
• Assessing the trustworthiness of financial data is paramount for market integrity.
• Existing methods may not fully capture complex inter-firm relationships.

Purpose of the Study:

• To develop a novel method for measuring cross-sectoral structural similarities between firms.
• To leverage microscopic real-world transaction data for auditing insights.
• To explore the application of network representations and embeddings in computational audit.

Main Methods:

• Deriving network representations of companies from transaction datasets.
• Computing network embedding vectors for each company's transaction network.
• Analyzing over 300 real transaction datasets to identify patterns and similarities.

Main Results:

• Significant changes in bookkeeping structures and client similarities were detected.
• High classification accuracy was achieved for various auditing-related tasks.
• The embedding space effectively clustered similar companies while separating different industries.

Conclusions:

• The developed network-based approach provides valuable insights for computational auditing.
• The method accurately captures relevant structural aspects of financial networks.
• This approach has potential applications beyond individual firms, including country-level risk assessment.