CMStatistics 2018: Start Registration
View Submission - CFE
A0884
Title: The role of market indices in forecasting stocks volatility: A HAR framework using a mixed sampling approach Authors:  Marwan Izzeldin - Lancaster University Management School (United Kingdom) [presenting]
Ingmar Nolte - Lancaster University (United Kingdom)
Vasileios Pappas - Kent Business School (United Kingdom)
Rodrigo Hizmeri - Lancaster University (United Kingdom)
Abstract: The aim is to examine the value added in forecasting high-frequency stock data using a Heterogeneous Autoregressive (HAR) model augmented with market indices (SPY and the S\&P 500). Our empirics are based on high-frequency data of 10 representative stocks, the S\&P 500 and SPY market indices for the period 2000 to 2016. We allow for different sampling frequencies on both sides of the HAR regression specification, different market regimes and signed realised variances and covariances. We find that, irrespectively of the specification adopted, the Market-Augmented HAR ($M\text{-}HAR$) specification always brings significant forecasting gains over the conventional HAR. Despite the high correlation between the S\&P 500 and SPY realised variances and covariances, both indices have different statistical features and patterns (i.e. periodicity, persistence, continuity and leverage) that results in differentiated forecasting gains. The S\&P 500 adds more than SPY at all sampling frequencies. Choosing an optimal sampling frequency is essential to maximise gains and these tend to vary across with the index in use. The gains from $M\text{-}HAR$ specification are regime sensitive where the highest gains are observed in the pre and post-crisis regimes. Adding the index during the crisis episode adversely affects the forecasts.