With the recent availability of high-frequency nancial data the longrange dependence of volatility regained researchers' interest and has leadto the consideration of long memory models for realized volatility. Thelong range diagnosis of volatility, however, is usually stated for long sampleperiods, while for small sample sizes, such as e.g. one year, the volatilitydynamics appears to be better described by short-memory processes. Theensemble of these seemingly contradictory phenomena point towards shortmemory models of volatility with nonstationarities, such as structuralbreaks or regime switches, that spuriously generate a long memory pattern(see e.g. Diebold and Inoue, 2001; Mikosch and Starica, 2004b). In thispaper we adopt this view on the dependence structure of volatility andpropose a localized procedure for modeling realized volatility. That isat each point in time we determine a past interval over which volatilityis approximated by a local linear process. Using S&P500 data we ndthat our local approach outperforms long memory type models in termsof predictability.
C14 - Semiparametric and Nonparametric Methods ; C51 - Model Construction and Estimation ; G17 - Financial Forecasting ; Accounting and auditing. Other aspects ; Individual Working Papers, Preprints ; No country specification
