Risk Estimation

Establish the magnitude of risk in a portfolio

The EMA system calculates both parametric and non-parametric estimates of the level of risk in a portfolio, either in absolute terms or relative to a benchmark.

The parametric measures – volatility and Tracking Error – can be estimated with user defined time horizons. The standard is medium term – based on a four-year history and designed to indicate risk over a three to twelve-month time frame. In addition, it is possible to use a much more extended lookback, of up to 40 years, to consider longer term patterns or, for example, to focus attention on crisis periods. For a near term forecast the GARCH based FASTVaR model projects forward to the next trading day and up to one month ahead.

It is increasingly recognised that both short and longer term measures of risk are required to properly represent the risk characteristics of a portfolio. Investors time-horizons are long-term and the real risk they face is of a long-term decline in capital value. In this context, risk also represents opportunity, as higher risk portfolios may be better placed to generate capital gains over the long-term. The value of risk measures relates to how they help the manager to construct a portfolio in the most efficient manner, and in ensuring that the manager is aware of the risks that they are taking at any time to keep them consistent with the mandate. For these purposes, the shorter term measures are useful as a guide to possible changes in the longer term risk patterns, while the long-term measure best matches typical investment horizons.

The non-parametric measures – VaR and CVaR – are estimated from a Monte Carlo model which performs full instrument repricing to build an accurate picture of the non-linear risks in the portfolio. It uses the EMA factor model to generate random shifts for each factor and each residual-risk component which are applied to all of the key instruments – equities, FX rates, interest rates and credit rates – in a way that is consistent with the market data covariance structure. Reporting shows the full returns distribution, VaR and expected shortfall (conditional VaR) for different confidence levels over the time horizon selected by the user.