Risk management for asset managers: A test of relative VaR

The paper tests the accuracy of relative VaR and identifies some crucial methodological issues which are important when backtesting relative VaR models. In particular, while the eprformance for unconditional accuracy of relative VaR methods appears to be satisfactory for simple portfolios even with simlper measurement methods, the assessment of unconditional accuracy is hampered by negative tracking error autocorrelation. The extent of this effect, well known when daily relative returns are used, is shown to be still relevant with weekly data and to decline only on longer time horizons