In the original legacy RiskMetrics documentation from 1996, volatility is calculated using a simple exponentially weighted moving average with some decay factor to determine the weights. This would be used in the calculation of volatility for variance-covariance VaR for example. In RM2006, is volatility still calculated this way? If not, how does it differ and can the 1996 method be substituted?
Therefore, the new RM2006 methodology can be summarized as based on a ARCH-like process with long memory + Student distribution + residuals scale correction + lagged correlations between returns. All the ingredients contribute to the performances, albeit possibly at different risk horizons or according to different performance measures. At the end, the new RM2006 is clearly a more complicated methodology than the simple exponential moving average of RM1994. This is the price to pay for the increased accuracies and the possibility to reach long risk horizons