Abstract:
This paper estimates the daily market risk of Italian bank securities portfolios under different model assumptions, using granular data on all banks and exposures from 2008 to 2023. Market risk is measured via value-at-risk and expected shortfall, estimated with three approaches: (1) non-parametric historical simulation, (2) multivariate normal GARCH, and (3) a multivariate parametric model capturing heavy tails, negative skewness, asymmetric dependence, and volatility clustering. We empirically examine the characteristics of each approach and compare them through extensive backtesting. Results show that parametric models generally outperform the non-parametric method, though the latter remains viable. Using actual bank portfolio data introduces unique challenges, requiring careful treatment in risk analysis. Finally, we discuss which approach is most suitable for financial stability purposes, informing system-wide market risk indicators and stress-testing frameworks.
