Stress-Testing in a Lucky Country

Original by T. Huughes et D. Melser, Moody’s, 2014 

This summary note was Posted on

  • Use regression models of the 30+-days, approach commonly used in banking
  • Standard models would have underestimated the extent of the credit losses, often significantly
  • Need to leverage some sort of external crisis in order to give an accurate reflection of likely stressed performance
  • Benchmark how Australia’s mortgagors would perform given severe macroeconomic difficulty
  • Result: Stressed delinquency rates for the Australia from models calibrated to US data are significantly higher than those estimated from Australian data alone
  • A very high proportion of mortgages are kept on the bank’s books in Australia relative to other countries (62% vs 40% in US or 32% in CH, 23% in NL)
  • The standards approach to stress testing involves finding some data on credit performance, modeling it as a  function of economic variables, then forecasting under various stressed scenarios
  • Delinquency is significantly underestimated using pre-crisis data in the US
  • Modeling three key elements: the maturation effect, the time effect, the vintage effect
  • Use current employment rate lags, annual change in house prices, unemployment rate at origination
  • They suppose that new pools are created over the forecast horizon at the same rate as they have been over the past 24 months
  • Correlation and feedback mechanisms that drive activity are likely to shift during severe recessions
  • Recessions associated with banking crises and large falls in house prices such envisaged are quite distinct and much worse than garden variety recessions
  • The lesson of the US crisis are that using economic and credit performance feedbacks benchmarked on sunny economic times will tend to give overly optimistic projections which are found wanting when the storms clouds hit