Singular Spectrum Analysis for signal extraction in Stochastic Volatility models

Estimating the in-sample volatility is one of the main difficulties that face Stochastic Volatility models when applied to financial time series. A non-parametric strategy based on Singular Spectrum Analysis is proposed to solve this problem. Its main advantage is its generality as it does not impose any parametric restriction on the volatility component and only some spectral structure is needed to identify it separately from noisy components. Its convincing performance is shown in an extensive Monte Carlo analysis that includes stationary and nonstationary long memory, short memory and level shifts in the volatility component, which are models often used for financial time series. Its applicability is finally illustrated in a daily Dow Jones Industrial index series and an intraday series from the Spanish Ibex35 stock index.