Identifying household finance heterogeneity via deep clustering

Area of Science:

• Household finance
• Econometrics
• Data science

Background:

• Household heterogeneity is complex and traditionally studied with limited variables.
• Previous methods struggle to capture the full picture of household diversity.

Purpose of the Study:

• To investigate high-dimensional household heterogeneity using advanced clustering techniques.
• To demonstrate the utility of deep learning for analyzing complex financial data.

Main Methods:

• Utilized a deep learning-based clustering method on high-dimensional balance sheet data (approx. 50,000 households).
• Employed dimension-reduction techniques to integrate the full joint distribution of financial variables.

Main Results:

• Identified distinct household clusters based on financial variables.
• Asset variables are critical for understanding wealthy households, while debt variables are key for non-wealthy households.
• Clusters derived from financial data show strong correlations with sociodemographic variables (age, education, family size).

Conclusions:

• Simultaneous analysis of multiple variables is essential for a detailed understanding of household heterogeneity.
• Financial variables, particularly assets and debts, are powerful indicators of household segmentation.
• The identified financial clusters have significant implications for sociodemographic characteristics.