摘要
A new methodology to evaluate market risks is introduced. It is designed to be more accurate than the existing
methodologies, and to be able to reach long risk horizons, up to one year. Consistency across risk horizons is
obtained by building the methodology using a long memory ARCH process to compute the required forecasts.
A large data set covering the main asset classes and geographical areas is used to validate the various subcomponents of the methodology. Extensive backtesting using probtiles is done to assess the final performance,
as well as the contributions of the various parts. One key quantitative result is that the new methodology applied
to a risk horizon of three months is more accurate than the exponential moving average scheme at a risk horizon
of one day. This quantitative improvement allows us to analyse risks in a portfolio both at tactical and strategic
time horizons