Model-based stress tests: linking stress tests to VaR for market risk

Carol Alexander, Elizabeth Sheedy

Research output: Chapter in Book/Report/Conference proceedingConference proceeding contributionpeer-review


Under the new capital accord stress tests are to be included in market risk regulatory capital calculations. This development necessitates a coherent and objective framework for stress testing portfolios exposed to market risk. Following recent criticism of stress testing methods our tests are conducted in the context of risk models, building on the VaR literature. First, to identify the most suitable risk models for stress testing, we apply an extensive back testing procedure that focuses on extreme market movements. We consider eight possible risk models including both conditional and unconditional models and four possible return distributions (normal, Student’s t, empirical and normal mixture) applied to three heavily traded currency pairs using a sample of daily data spanning more than 20 years. Finding that risk models accommodating both volatility clustering and heavy tails are the most accurate predictors of extreme returns, we develop a corresponding model-based stress testing methodology. Our results are compared with traditional stress tests and we assess the implications for capital adequacy. On the basis of our results we conclude that the new recommendations for market risk regulatory capital calculation will have little impact on current levels of foreign exchange regulatory capital.
Original languageEnglish
Title of host publicationThe 20th Annual Australasian Finance and Banking Conference
Subtitle of host publicationconference papers
Publication statusPublished - 2007
EventAustralasian Finance and Banking Conference (20th : 2007) - Sydney
Duration: 12 Dec 200714 Dec 2007


ConferenceAustralasian Finance and Banking Conference (20th : 2007)


  • value-at-risk models
  • stress testing
  • market risk
  • exchange rates


Dive into the research topics of 'Model-based stress tests: linking stress tests to VaR for market risk'. Together they form a unique fingerprint.

Cite this