Abstract
The exponentially weighted moving average (EWMA) model is a particular modelling scheme supported by RiskMetrics that is capable of predicting the current level of financial time series volatility. It is designed to track changes in conditional variance of financial returns by assigning exponentially decreasing weights to the observed past squared measurements.
Recently, several on-line (i.e. recursive) estimation techniques suitable for this class of stochastic models have been introduced. These methods undoubtedly represent attractive alternatives to the common identification and calibration procedures (i.e. off-line or batch); they can estimate and control the process behaviour in real time.
The aim of the paper is to examine different EWMA model estimators by using financial data. For instance, one might consider the Value at Risk (VaR) backtesting approach since Value at Risk predictions are relevant outputs of the RiskMetrics EWMA modelling framework (especially from the practical point of view).
Therefore, various VaR backtests can be used to study the adequacy of different EWMA model estimators.