Measurement of Intraday Volatility in Stock Markets

当投资于股票时，方差风险溢价通常被认为是对下行风险的补救措施。新冠肺炎大流行导致的不确定性增加，强调了投资者考虑方差风险的必要性，并为分析个人股票方差风险溢价的存在、横截面和时间变化提供了独特的环境。特别是，最近的事态发展表明，不同部门都不同程度地受到了新冠肺炎大流行对经济、监管和社会的影响。因此，我们将研究不同行业的个人股票方差风险溢价，为投资者对方差风险（不断变化的）估值提供有价值的见解，并特别强调Covid-19危机的影响。此外，我们通过将人工神经网络与基于注意的机制相结合，将深度学习方法应用于（下行）方差预测。注意机制是人类感知的一部分，意味着在每个时间点对特定信息的选择性注意。转移到方差预测的背景下，注意机制的应用可以为决策的关键变量的选择提供见解。人工神经网络在观察模式方面的灵活性为分析盘中股市数据提供了一种有趣的替代标准方法。考虑到金融市场在日内水平上的相互依赖性，关于期权市场在标的股票的价格发现过程中的作用存在矛盾的结论。因此，我们的目标是为时变价格发现措施开发一个理论框架和方法方法。我们计划将重点放在广义自回归评分模型上，该模型允许观测驱动的参数更新过程。这种方法有望进一步揭示，围绕盈利公告等公司特定信息事件，期权市场的重要性可能会发生变化。

The variance risk premium is commonly considered as a remedy against downside risk when investing in equity. The increased uncertainty due to the Covid-19 pandemic emphasizes the need for investors to account for variance risk and offers a unique setting to analyze the existence, cross-sectional and time variation of individual equity variance risk premia. In particular, as recent developments have shown that different sectors have been affected to a variable degree by the economic, regulatory and social impacts of the Covid-19 pandemic. We will therefore examine the individual equity variance risk premia across different sectors to provide valuable insights into the (changing) valuation of variance risk by investor with special emphasize on the impact of the Covid-19 crisis.In addition, we apply deep learning approaches for (downside) variance prediction by combining artificial neural networks with an attention-based mechanism. Attention mechanisms are part of human perception and imply a selective attention to specific pieces of information at each point in time. Transferred to the context of variance prediction, the application of an attention mechanism allows to provide insights into the selection of key variables for decision making. The flexibility of artificial neural networks with respect to the observed patterns offers an interesting alternative to standard approaches when analyzing intraday stock market data.Considering interdependencies among financial markets on an intraday level, conflicting conclusions exist with respect to the role of the option market within the price discovery process of the underlying stock. We therefore aim at developing a theoretical framework as well as a methodological approach for a time-varying price discovery measure. We plan to focus on generalized autoregressive score models, which allow an observation driven updating process of parameters. This approach is supposed to shed further light on the potentially changing importance of the option markets around firm-specific informational events such as earning announcements.