Use this URL to cite or link to this record in EThOS:
Title: Forecasting with large datasets
Author: Furman, Yoel Avraham
ISNI:       0000 0004 5778 2904
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
This thesis analyzes estimation methods and testing procedures for handling large data series. The first chapter introduces the use of the adaptive elastic net, and the penalized regression methods nested within it, for estimating sparse vector autoregressions. That chapter shows that under suitable conditions on the data generating process this estimation method satisfies an oracle property. Furthermore, it is shown that the bootstrap can be used to accurately conduct inference on the estimated parameters. These properties are used to show that structural VAR analysis can also be validly conducted, allowing for accurate measures of policy response. The strength of these estimation methods is demonstrated in a numerical study and on U.S. macroeconomic data. The second chapter continues in a similar vein, using the elastic net to estimate sparse vector autoregressions of realized variances to construct volatility forecasts. It is shown that the use of volatility spillovers estimated by the elastic net delivers substantial improvements in forecast ability, and can be used to indicate systemic risk among a group of assets. The model is estimated on realized variances of equities of U.S. financial institutions, where it is shown that the estimated parameters translate into two novel indicators of systemic risk. The third chapter discusses the use of the bootstrap as an alternative to asymptotic Wald-type tests. It is shown that the bootstrap is particularly useful in situations with many restrictions, such as tests of equal conditional predictive ability that make use of many orthogonal variables, or `test functions'. The testing procedure is analyzed in a Monte Carlo study and is used to test the relevance of real variables in forecasting U.S. inflation.
Supervisor: Sheppard, Kevin ; Mavroeidis, Sophocles Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Economic forecasting ; Finance--Forecasting ; Econometrics