Producing a forward-looking covariance matrix
FASTVaR is a unique combination of a robust longer term linear statistical factor model with a daily GARCH derived volatility forecast. FASTVaR with Excerpt delivers the responsiveness of a daily data set combined with the attribution and stability strengths of a long-term factor model.
FASTVaR is an adjustment to EMA’s standard risk model that produces VaR and volatility estimates over short-term horizons typically from 1 day to 1 month. It starts by generating a six-year daily returns set for the robust systematic factors produced from the long-term EM analysis. Next a univariate GARCH model is estimated for each of the systematic factors. The GARCH estimation incorporates a skewed T-distribution error model to more closely approximate the observed “fat-tailed” characteristic of security returns and more accurately model larger down-market returns. The variance in residual returns is estimated at the aggregate level with the estimation of a scaling factor.
The key advantages over an exponentially weighted moving average (EWMA) approach are that GARCH allows for a “reactivity” term which takes account of the tendency of volatility to return to its long-term trend, and the incorporation of a skew-T distribution to account for the fat-tailed distribution of asset returns.
The result is a dynamic variance forecast model that responds rapidly to changes in the level of market volatility. Its independence from market traded measures of implied volatility make it immune to trade flow distortions.
With the incorporation of FASTVaR into Excerpt, EM Applications’ clients are able to measure and forecast both longer term intrinsic risk and very short term risk derived from a consistent underlying methodology.