Assessing Household Financial Vulnerability: Empirical evidence from the U.S. using machine learning

Household financial vulnerability has gained considerable attention since the global financial crisis and it concerns policymakers due to its impact on macroeconomic indicators of financial instability. However, financial vulnerability is a complex multi-dimensional concept and large gaps remain in the underpinning of a comprehensive financial vulnerability assessment at the micro level. In this paper, a new approach is proposed to assess financial vulnerability by employing an unsupervised machine learning technique. A two-step empirical strategy is used to conduct this analysis. First, Hierarchical Ascending Clustering (HAC) and K-means clustering analysis is undertaken to identify homogenous clusters of households, which are financially vulnerable. Afterwards, we estimate the probability of being financially vulnerable depending upon different household and geographical characteristics using a logistic regression. Data from the Survey of Consumer Finance (SCF) for the years 1998, 2007 and 2016 are used for this analysis. The empirical results show that about 28% of the households in United States are financially vulnerable as of 2016, which is 4% less as compared to 2007. The results of the econometric estimations highlight that Black and African Americans are 8% and Hispanic Americans are 6% more likely to be financially vulnerable than non-Hispanic white persons, after taking into account other household and regional level characteristics. Econometric estimations also highlight the existence of large gaps in household financial vulnerability across other household characteristics, such as education level, employment status, marital status, and age of the household head. Lastly, regional characteristics do not seem to have a significant impact on household’s financial vulnerability as long as the net-worth of the household is taken into account.