Author: Eutteum Park(Bank of Korea), Kyu Ho Kang(Korea Univ.)

This study develops a Large Bayesian VAR (LBVAR) model with 22 macro-financial variables to forecast household and corporate loan delinquency rates in the Korean banking sector and assess related upside risks. Using quarterly data from 2010Q1 to 2025Q4 and an expanding-window out-of-sample evaluation, the LBVAR outperforms the random-walk benchmark across forecast horizons. The estimates indicate that delinquency rates are positively related to lagged borrowing-cost variables, including interest rates and inflation, while real activity indicators and housing prices show negative contemporaneous correlations. Density forecasts suggest a moderate rise in household and corporate delinquency rates to about 0.41% and 0.68%, respectively, with the upper bound of the 90% credible interval indicating persistent upside risk. Stress-scenario results reveal sector-specific vulnerabilities: corporate loans are more sensitive to a housing-market downturn, whereas household loans respond more strongly to an external shock. The high-inflation and high-interest-rate scenario produces only limited additional risk in both sectors. These findings provide empirical support for macroprudential risk monitoring and preemptive policy responses.