Modeling of high-frequency financial data

This thesis addresses the exploitation of high-frequency financial data through the modeling of realized covariances, with the purpose of forecasting the covariance matrix of the selected assets. Generally, the analysis of realized covariances is carried out with models using a Wishart distribution whose scale matrix is described by an autoregressive moving average structure. The first chapter introduces a novel distribution used in place of the Wishart, which can be employed to identify a common factor in the assets behaviour, representing the systematic risk in the market. The model is empirically tested in an asset allocation framework with the objective of tracking the market index of reference. Therefore, the purpose of this chapter is not to improve the conventional models based on the Wishart distribution, but rather to expand their usual scope of application. Another common issue stems from the presence of data retrieved at multiple frequencies and the ensuing need to properly exploit the available information content. The second chapter briefly reviews the literature concerning the modeling mixed-frequency data, and formulates a way to integrate the high-frequency information in a low-frequency-based model, using the former as "views" about the assets. Thus, the approach proposed in this chapter requires the derivation of a posterior equation, which depends upon the distributional assumptions for the high and low-frequency data. The resulting expression for the posterior distribution effectively allows to operate with data retrieved at different frequencies, and therefore to provide an alternative to the common approaches used in this framework. The final chapter is intended to outline a possible way of exploiting high-frequency data in the field of risk spillover analysis, starting from recent results concerning the structured specifications of conventional models employing low-frequency data.