Title: Forecasting a latent variable: Application to VaR in stochastic volatilty models
Authors: Josu Arteche - University of the Basque Country UPV/EHU (Spain) [presenting]
Javier Garcia - University of the Basque Country (Spain)
Abstract: The existence of latent variables contaminated with an added noise is quite common in many areas, a typical example being the volatility component in Stochastic Volatility (SV) models. Prediction of these latent variables can be either in-sample, or, in a more etymological sense, out-of-sample. The former is related with signal extraction whereas the latter implies prediction of future values. We focus on out-of-sample predictions, analysing several forecasting techniques for prediction of latent variables. There is a controversy over whether these techniques should be applied on the contaminated series of observables or on in-sample predictions of the latent variable obtained by signal extraction. Some light is shed on this issue by implementing a Monte Carlo analysis with different forecasting techniques applied in models with low frequency behaviour, which are common in SV modelling. Their applicability to forecast the volatility in Stochastic Volatility models and its use for Value at Risk evaluation is also examined by an application to a daily series of SP500 returns.