Abstract:This thesis investigates the economic value of incorporating intraday volatility estimators into the volatility forecasting process. The increased reliance on volatility forecasting in the financial industry has intensified the need for more rigorous analysis from an economic perspective instead of merely statistical point of view. A better understanding of the available methods has implications for portfolio optimization, volatility trading and risk management. More recently, volatility of asset returns was once again under spotlight during the 2008-2009 financial crisis. The study contributes to the extant volatility forecasting literature in three areas. First, it addresses the question of how to practically and effectively exploit intraday price information for variance and covariance modelling and forecasting. Second, it addresses the development of an 'optimal' intraday volatility model that accommodates market practitioners preferences. Third, it evaluates the economic value of combining realized (intraday) volatility estimators for utilizing unique information embedded in each estimator. The thesis is organised as follows. One of the most visible indicators of the crisis that captured the attention of the financial industry was the extremely high level of asset return volatility. This uncertainty prompted much interest for a more accurate, yet practically applicable approach for volatility forecasting. Chapter 2 introduces the various realized volatility estimators, volatility forecasting procedures and their corresponding realized extensions used in our subsequent empirical investigations. Chapter 3 evaluates the economic value of various intraday covariance estimation approaches for mean-variance portfolio optimization. Economic loss functions overwhelmingly favour intraday covariance matrix models instead of their daily counterparts. The constant conditional correlation (CCC) augmented with realized volatility produces the highest economic value when applied with a time-varying volatility timing strategy. Chapter 4 compares the practical value of intraday based single index (univariate) and portfolio (multivariate) models through the lens of Value-at-Risk (VaR) forecasting. VaR predictions are generated from standard daily univariate or multivariate GARCH models, as well as GARCH models extended with ARFIMA forecasted realized measures. Conditional coverage test results indicate that intraday models, both univariate and multivariate ones, outperform their daily counterparts by providing more accurate VaR forecasts. Chapter 5 investigates the economic value of combining intraday volatility estimators for volatility trading. The simulated option trading results indicate that a naive combination of an intraday estimator and implied volatility cannot be outperformed by the best individual estimator. In addition, trading performance can be further boosted by applying more complex combination models such as a regression based combination of 42 single volatility estimators.

Value of Information in the Mean-Square Case and its Application to the Analysis of Financial Time-Series Forecast

Statistical Hypothesis Testing for Information Value (IV)

Uncertainty and Value of Information in Risk Prediction Modeling

Rethinking Mean Square Error: Why Information is a Superior Assessment of Estimators

Volatility Information in High-Frequency Financial Interval-Valued Time Series: A Direct Modeling Pattern

Measuring Financial Time Series Similarity With a View to Identifying Profitable Stock Market Opportunities

Enhancing Multivariate Time Series Forecasting with Mutual Information-driven Cross-Variable and Temporal Modeling

Financial Time-Series Forecasting: Towards Synergizing Performance And Interpretability Within a Hybrid Machine Learning Approach

The Inclusion of the Volume-Price-Product Factor for the Trend Forecasting of Futures Time Series Data

Minimization of Information Loss Based Hybrid Short-term Load Forecasting, Part II:Algorithm and Case Study

Bayesian forecasting with information theory

Information-Theoretic Foundations for Machine Learning

Efficient Estimation of the Value of Information in Monte Carlo Models

Implied volatility directional forecasting: a machine learning approach

Essays on the Economic Value of Intraday Covariation Estimators for Risk Prediction

A novel approach to quantify volatility prediction

A Value of Information Framework for Latent Variable Models

Evaluating the Performance of Machine Learning Algorithms in Financial Market Forecasting: A Comprehensive Survey

On Information-Theoretic Measures of Predictive Uncertainty

QUANTIFYING THE IMPACT OF PARTIAL INFORMATION ON SHARPE RATIO OPTIMIZATION

Development of a Time Series Forecasting Method Based on Entropy Measures of Diagnostic Information