Dados do Trabalho

Título

MODELING AND FORECASTING LONG MEMORY STOCHASTIC VOLATILITY

Resumo

This is an ongoing work where our main objective is to propose a new procedure for estimating stochastic volatility models with long memory latent volatility. Our proposal is based on a state space approximation of the log variance process, and the disturbance error is approximated by a mixture of normals. All parameters are estimated by Maximum Likelihood, with the likelihood function expressed in terms of the innovations of a Kalman Filter algorithm. The performance of our proposal is evaluated by a Monte Carlo experiment when the log variance follows an ARFIMA(0,d,0) process

Palavras-chave

mixtures, non-gaussian errors, long memory