Asymmetric Adjustment of Realized Volatility

Using the multiple threshold autoregressive and moving average (TARMA) model we analyze the nonlinearities in the dynamics of realized volatilities of daily stock returns of 30 companies in the Dow Jones index. We find that the realized volatility processes can be characterized by the high, moderate, and low regimes and that the persistence, variance and ARMA error term change with each regime. Also, we find that spiked jumps in realized volatilities are better captured by the TARMA model than by ARMA or fractional ARMA models. Forecasts by the TARMA model tend to outperform those by the ARMA and fractional ARMA models.