Forecast Encompassing in High Dimensional Predictive Models
Skip to main content
eScholarship
Open Access Publications from the University of California

UC Riverside

UC Riverside Electronic Theses and Dissertations bannerUC Riverside

Forecast Encompassing in High Dimensional Predictive Models

No data is associated with this publication.
Abstract

This thesis advances the field of econometric forecasting and predictor selection by developing a series of novel methods based on the forecast-encompassing principle and the higher-order elicitability concept. These methods enhance predictive regression models by improving predictor selection in the context of various financial and macroeconomic scenarios. Chapter 2 introduces two methods for predictor selection based on out-of-sample forecast encompassing principle in high-dimensional predictive regression models. This principle is applied to forecast U.S. output growth, inflation, and the probability of U.S. recession, demonstrating a lower forecast error loss. Chapter 3 addresses volatility forecast model comparison by proposing strictly consistent scoring functions based on the Bregman divergence. This new scoring function elicits the mean and variance without the zero mean assumption, facilitating more accurate predictive ability assessment. Chapter 4 delves into the International Monetary Fund’s Growth-at-Risk (GaR) and Growth Shortfall (GS) measures, and proposes new tests for predictor selection in GS and GaR using the Fissler-Ziegel (FZ) scoring function. Expectiles and their applications in financial risk measurement are explored in Chapter 5. The thesis proposes a framework for model selection and model averaging in predictive expectile regression models, providing a significant improvement over model selection. Chapter 6 constructs a new testing framework for Granger Causality in Expected Shortfalls, which previously didn’t exist due to the lack of an objective function to evaluate Expected Shortfalls. Chapter 7 proposes a Granger-Causality test in the predictive regression for the conditional mode and investigates the predictability of equity premium in mode regressions. Chapter 8 studies skewness and kurtosis in stock return distributions and introduces a novel Granger Causality test for these higher moments. The thesis thus significantly enhances the econometric toolbox for financial and macroeconomic forecasting and risk management.

Main Content

This item is under embargo until October 18, 2025.