Household financial health: a machine learning approach for data-driven diagnosis and prescription

AbstractHousehold finances are being threatened by unprecedented social and economic upheavals, including an aging society and slow economic growth. Numerous researchers and practitioners have provided guidelines for improving the financial status of households; however, the challenge of handling heterogeneous households remains nontrivial. In this study, we propose a new data-driven framework for the financial health of households to address the needs for diagnosing and improving financial health. This research extends the concept of healthcare to household finance. We develop a novel deep learning-based diagnostic model for estimating household financial health risk scores from real-world household balance sheet data. The proposed model can successfully manage the heterogeneity of households by extracting useful latent representations of household balance sheet data while incorporating the risk information of each variable. That is, we guide the model to generate higher latent values for households with risky balance sheets. We also show that the gradient of the model can be utilized for prescribing recommendations for improving household financial health. The robustness and validity of the new framework are demonstrated using empirical analyses.Keywords: Household financeFinancial healthHeterogeneityRisk scoringDeep learning Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 Note that Indicator 4 follows the opposite direction of the other indicators. For Indicators 1 to 3, having a large value would increase financial risk, while it is the opposite for Indicator 4. Hence, stochastic dominance in Indicator 4 should also be interpreted in the opposite way from the other indicators.2 In Appendix C, we used the Bonferroni post-hoc test to assess the significance of the difference in risk information for each of the input variables to RI-HAE.3 To be more precise, the reciprocal of shadow price represents the amount of money required to increase the financial risk score by one unit estimated under first-order approximation because shadow price is a slope of the linear function tangent to RI-HAE.Additional informationFundingThis work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. NRF-2022R1I1A4069163 and No. NRF-2020R1C1C1011063).