Stress testing household balance sheets in Luxembourg

DateJuly 2018
AuteurGiordana, G. and M. Ziegelmeyer

This paper uses individual household data from Luxembourg to evaluate how severe economic conditions could affect bank exposure to the household sector. Using data from a representative survey, information on household income, expenses and liquid assets are used to calculate a household-specific probability of default (PD). Aggregate bank exposure at default (EAD) is obtained by multiplying these household-level PDs by their corresponding volume of outstanding loans and summing across the population of households. Aggregate bank loss given default (LGD) is calculated by assuming that banks recover real estate assets from defaulting households and liquidate them with a haircut. To simulate adverse economic conditions, the exercise is repeated with scenarios combining severe but plausible shocks (i.e. tail risk) to real estate prices, bonds and stocks, household income and interest rates. Compared to the no-shock baseline, the LGD rises by a multiple of eight, reaching 4.2% of total bank exposure to the household sector. Thus, bank losses appear to be quite sensitive to severe stress. The high-stress scenario also generates a relatively high percentage of defaults among socioeconomically disadvantaged households (i.e. low net wealth, low income, low education, three or more dependent children). For instance, households in the lowest income quintile see their PD rise from 9.3% in the no-shock baseline to 14.8% in the most severe scenario. Our main conclusion is that bank losses appear to be quite sensitive to financial stress, despite three mitigating factors in Luxembourg: indebted households tend to hold liquid assets that can help smooth shocks, household leverage tends to decline rapidly once mortgages have been serviced several years, and loan-to-value ratios at origination appear not to be excessive.
JEL-codes: D10, D14, E44, G01, G21
Keywords: Stress testing; Financial stability; HFCS; Household finance

Téléchargement Cahier d'étude 121 (pdf, 3 MByte)