Considerações sobre a relação entre eficiência estatística versus desempenho gerencial na seleção e validação de modelos de VaR

Abstract

The most widespread and used risk measure is Value at Risk (VaR), which gained popularity after 1994 with the RiskMetrics methodology and adoption by the Basel accords. This paper results from internal studies at BNDES and debates with risk management actors who use this risk metric. Several classic models were revisited to select and validate the best models, using adherence tests and other several criteria that consider both the statistics and management issues. A statistically efficient VaR model can lead to increased regulatory capital, implementation difficulties and other barriers that risk managers may face. Moreover, this study provides an in-depth look into the impacts of changing their parameters, which should be periodically revisited by Institutions.