) In this paper we propose a nonlinear long-memory time series model for realized
volatility that incorporates all well-known stylized facts from the (GARCH) volatility
literature, in particular level shifts, day-of-the-week e_ects, leverage e_ects and
volatility level e_ects. The model, as well as several restricted versions, are estimated
for the S&P 500 index and three exchange rates.
The in-sample results show that all nonlinearities are highly signi_cant and improve
the description of the data. The out-of-sample results show that for shorter
horizons, up to 10 days, accounting for these nonlinearities signi_cantly improves the
forecast performance compared to a linear ARFI model. Such short-term volatility
forecasts are especially useful for short-term risk management, including Value-at-
Risk. For longer horizons no bene_t is obtained from incorporating nonlinearities.
The most important nonlinearities are the leverage e_ect for the S&P 500 index,
and the leverage e_ect as well as the day-of-the-week e_ects for the exchange rates.
The best way to incorporate the e_ects of lagged daily returns is to include them
as exogenous regressors, i.e. outside the long memory _lter. Not important for the
forecast performance is allowing the persistence of shocks to depend on the level of
volatility, and modeling the level shifts for the S&P 500 index.
) In this paper we propose a nonlinear long-memory time series model for realizedvolatility that incorporates all well-known stylized facts from the (GARCH) volatilityliterature, in particular level shifts, day-of-the-week e_ects, leverage e_ects andvolatility level e_ects. The model, as well as several restricted versions, are estimatedfor the S&P 500 index and three exchange rates.The in-sample results show that all nonlinearities are highly signi_cant and improvethe description of the data. The out-of-sample results show that for shorterhorizons, up to 10 days, accounting for these nonlinearities signi_cantly improves theforecast performance compared to a linear ARFI model. Such short-term volatilityforecasts are especially useful for short-term risk management, including Value-at-Risk. For longer horizons no bene_t is obtained from incorporating nonlinearities.The most important nonlinearities are the leverage e_ect for the S&P 500 index,and the leverage e_ect as well as the day-of-the-week e_ects for the exchange rates.The best way to incorporate the e_ects of lagged daily returns is to include themas exogenous regressors, i.e. outside the long memory _lter. Not important for theforecast performance is allowing the persistence of shocks to depend on the level ofvolatility, and modeling the level shifts for the S&P 500 index.
การแปล กรุณารอสักครู่..
