TY - BOOK

T1 - The impact of intraday periodicity and news announcements on high-frequency stock volatility

AU - Guan, Yanying

PY - 2021

Y1 - 2021

N2 - High-frequency intraday financial data are commonly used in stock market volatility estimation and forecasting because they produce accurate results. However, little work to date has focused on the stylised facts of high-frequency returns, such as their tail properties, autocorrelations and leverage effects. One of the most discussed features of high-frequency returns is intraday periodicity, yet it is not well known how this feature operates in returns from data with different sampling schemes and frequencies. In addition, macroeconomic news announcements have been shown to have a large impact on first-moment and second-moment responses in financial markets. However, few existing models consider the effect of news on volatility estimation and forecasting, and those that do tend to treat it as a dummy variable, limiting its analytical power. This thesis addresses these issues by reporting a study of the stylised facts of returns from S&P 500 stocks and the SPY index, and standardised returns from the latter, using various volatility measures in different financial regimes (i.e. before, during and after the 2008 financial crisis). It presents a comparison of the intraday patterns, jump frequencies, jump components and volatility forecasting of stock returns from calendar-time and business-time sampling schemes, as well as how these features are affected by intraday periodicity. It assesses the direct impact of macroeconomic news announcements on volatility estimation and forecasting for stock returns by incorporating significant news announcements as an index to identify the jumps caused by news in heterogeneous autoregressive (HAR) class models. The results suggest that absolute intraday returns for high-frequency data exhibit autocorrelations and that aggregated returns display heavy tails. Standardising the returns of the SPY index using eleven different volatility measures produces distributions that are closer to a normal distribution. We find that various volatility measures are significantly correlated with trading volume, and hence that HAR-class models that include trading volume yield better volatility forecasting results than existing models. However, this effect may be limited to data from the relatively non-volatile pre-crisis and post-crisis periods. High-frequency returns based on business-time sampling have smaller jump frequencies, jump components and intraday periodicity patterns, than calendar-time data, which may be useful for volatility analysis. Intraday periodicity has a notable impact on jumps for both sampling schemes, however, and adjusting for intraday periodicity produces fewer jumps for all returns and smaller jump components for the majority. We also find that the forecasting results for less volatile data, such as healthcare stocks and data from the post-crisis period, improved after filtering for intraday periodicity. Finally, macroeconomic news announcements can affect jump components, and considering news outlets in HAR models can improve the forecasting results. The thesis thus contributes to our understanding of the factors affecting stock market volatility by providing evidence in support of including trading volume, efficient intraday periodicity estimators and news surprise in volatility estimation and forecasting models.

AB - High-frequency intraday financial data are commonly used in stock market volatility estimation and forecasting because they produce accurate results. However, little work to date has focused on the stylised facts of high-frequency returns, such as their tail properties, autocorrelations and leverage effects. One of the most discussed features of high-frequency returns is intraday periodicity, yet it is not well known how this feature operates in returns from data with different sampling schemes and frequencies. In addition, macroeconomic news announcements have been shown to have a large impact on first-moment and second-moment responses in financial markets. However, few existing models consider the effect of news on volatility estimation and forecasting, and those that do tend to treat it as a dummy variable, limiting its analytical power. This thesis addresses these issues by reporting a study of the stylised facts of returns from S&P 500 stocks and the SPY index, and standardised returns from the latter, using various volatility measures in different financial regimes (i.e. before, during and after the 2008 financial crisis). It presents a comparison of the intraday patterns, jump frequencies, jump components and volatility forecasting of stock returns from calendar-time and business-time sampling schemes, as well as how these features are affected by intraday periodicity. It assesses the direct impact of macroeconomic news announcements on volatility estimation and forecasting for stock returns by incorporating significant news announcements as an index to identify the jumps caused by news in heterogeneous autoregressive (HAR) class models. The results suggest that absolute intraday returns for high-frequency data exhibit autocorrelations and that aggregated returns display heavy tails. Standardising the returns of the SPY index using eleven different volatility measures produces distributions that are closer to a normal distribution. We find that various volatility measures are significantly correlated with trading volume, and hence that HAR-class models that include trading volume yield better volatility forecasting results than existing models. However, this effect may be limited to data from the relatively non-volatile pre-crisis and post-crisis periods. High-frequency returns based on business-time sampling have smaller jump frequencies, jump components and intraday periodicity patterns, than calendar-time data, which may be useful for volatility analysis. Intraday periodicity has a notable impact on jumps for both sampling schemes, however, and adjusting for intraday periodicity produces fewer jumps for all returns and smaller jump components for the majority. We also find that the forecasting results for less volatile data, such as healthcare stocks and data from the post-crisis period, improved after filtering for intraday periodicity. Finally, macroeconomic news announcements can affect jump components, and considering news outlets in HAR models can improve the forecasting results. The thesis thus contributes to our understanding of the factors affecting stock market volatility by providing evidence in support of including trading volume, efficient intraday periodicity estimators and news surprise in volatility estimation and forecasting models.

KW - volatility forecasting

KW - news announcements

KW - intraday periodicity

KW - high-frequency data

U2 - 10.17635/lancaster/thesis/1258

DO - 10.17635/lancaster/thesis/1258

M3 - Doctoral Thesis

PB - Lancaster University

ER -