Title: Exploit market microstructure noise in volatility forecasting
Authors: Ye Zeng - CREATES, School of Business and Social Science, Aarhus University, Denmark (Denmark) [presenting]
Abstract: In volatility forecasting, realized volatility as an estimator for the latent quadratic variation of asset price unavoidably begets the measurement error in variable problem. In a similar vein to the HARQ model which tackles the problem by attenuating realized volatility according to its asymptotic variance, another simple extension of the heterogenous autoregressive (HAR) model, named HARN, is proposed which exploits size of market microstructure noise to gauge reliability of realized volatility. Empirical analysis on datasets covering 29 stocks listed in NYSE show that realized volatility is always attenuated in response to the noisiness of data. Improved forecasting accuracy is also documented, both in-sample and out-of-sample, in comparison with results of the standard HAR model. In addition, empirical results show that the HARN model utilizing a simple estimator for the size of noise computed with data sampled every 5 minutes outperforms, or at least is on par with, models incorporating more sophisticated estimators. Thus, by augmenting HAR model with an extra simple noisiness measure, we obtain a parsimonious extension which improves volatility forecasting.