Detecting structural breaks in realized volatility

This paper considers the detection of structural changes in realized volatility based on HAR–GARCH models. For this, we propose a quasi-likelihood based score test for parameter changes in HAR–GARCH models. We derive the limiting null distribution of the score test by first introducing the quasi-maximum likelihood estimator to the HAR–GARCH model and establishing its asymptotic properties. The proposed test statistic is shown to converge weakly to a function of the Brownian bridge under the null of no structural change. Our simulations study shows reasonable sizes and powers of the test, even for non-Gaussian innovations. A real data application to S&P 500 realized volatility over the last 12 years coincides with three waves of financial crisis, namely the US housing, European sovereign debt, and emerging market crisis.