Doctoral thesis

Latent factor models for large and mixed-frequency data in finance and macroeconomics


233 p

Thèse de doctorat: Università della Svizzera italiana, 2016

English My thesis considers new latent factor models, and their estimation methodologies, suitable for settings relatively unexplored in the econometric literature as (i) a nonlinear model for the joint dynamics of a large cross-sectional distribution of asset returns, and the persistence of the ranks of the individuals inside it; (ii) approximate linear latent factor models for large panels of mixed-frequency data; and (iii) small scale state space models featuring multiple time series with stochastic volatility and observed at different frequencies. The thesis is articulated in four chapters: Chapter 1 summarizes the motivation and the objectives of the thesis, while the remaining three chapters correspond to three different articles. Chapter 2 presents a new type of asset allocation strategies based on a novel dynamic model of the cross-sectional distribution of returns and the ranks of assets inside the same cross-sectional distribution. These strategies are implemented on a large panel of US stocks, and are shown to perform well compared to traditional asset allocation strategies. Chapter 3 proposes a new class of approximate latent factor models suitable for large panels of data observed at different frequencies. An empirical application uncovers the common components of monthly data on output growth rates of the US industrial production sectors, and the yearly output growth rates of all the remaining sectors of the US economy, mainly services. Chapter 4 introduces indirect inference estimators for state space models featuring mixed frequency observables and stochastic volatility, and considers an application to forecasting quarterly European GDP using monthly macroeconomic indicators.
  • English
License undefined
Persistent URL

Document views: 87 File downloads:
  • 2016ECO002.pdf: 74