Factor-Augmented Forecasting for High-Dimensional Data
2016 (English)Licentiate thesis, comprehensive summary (Other academic)
In this thesis, we take a critical look at the factor-augmented forecast models, when a large number of time series variables available can provide the vital information for prediction. We discuss how to describe the commonality and idiosyncrasy of high-dimensional data by a handful of factors in various levels, and how to improve the predictive performance using these factors as augmented predictors. Moreover, this thesis consists of two papers. In the first paper, we propose an extended factor-augmented vector autoregression for macroeconomic forecasting, which models the joint dynamics of the variables to be forecast, factor components and a large number of observed predictors, and we construct the forecasts based on estimated model using least absolute shrinkage and selection operator (Lasso) together with cross validation as model validation technique. In the second paper, we analyze the regional population dynamics for several ungulate species, and forecast the population abundance using multi-level factors as augmented predictors. In summary, the improvement of predictive performance can be achieved in both two papers.
Place, publisher, year, edition, pages
Stockholm: Institute of Mathematical Statistics, 2016.
High-Dimensional Data, Factor-Augmented Forecasting, Principal Component, Lasso, Cross Validation, Multi-Level Factor, Predictive Performance
Research subject Mathematical Statistics
IdentifiersURN: urn:nbn:se:su:diva-129971OAI: oai:DiVA.org:su-129971DiVA: diva2:926499
2016-06-07, Seminarium, Matematisk statistik (Room 306), House 6, Kräftriket, Department of Mathematics, Stockholm University, Stockholm, 18:51 (English)
Singull, Martin, Universitetslektor
Sköld, Martin, UniversitetslektorDocent