An Evaluation of Equity Premium Prediction Using Multiple Kernel Learning with Financial Features

This paper introduces and extensively explores a forecasting procedure based on multivariate dynamic kernels to re-examine—under a non-linear, kernel methods framework—the experimental tests reported by Welch and Goyal (Rev Financ Stud 21(4):1455–1508, 2008 ) showing that several variables proposed...

Full description

Saved in:
Bibliographic Details
Published in:Neural processing letters Vol. 52; no. 1; pp. 117 - 134
Main Authors: Arratia, Argimiro, Belanche, Lluís A., Fábregues, Luis
Format: Journal Article Publication
Language:English
Published: New York Springer US 01.08.2020
Springer Nature B.V
Subjects:
62 Statistics
62P Applications
Artificial Intelligence
Classificació AMS
Complex Systems
Computational Intelligence
Computer Science
Econometric models
Estadística matemàtica
Finance
Finances
Financial time series
Forecasting
Kernel functions
Kernel functions for time series
Learning
Matemàtiques i estadística
Mathematical models
Models economètrics
Multiple kernel learning
Ratios
Resampling
Securities markets
Support vector classification
Support vector regression
Time series
Valuation
Variables
Àrees temàtiques de la UPC
Multiple kernel learning
Support vector classification
Support vector regression
Financial time series
Kernel functions for time series
ISSN:1370-4621, 1573-773X, 1573-773X
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper introduces and extensively explores a forecasting procedure based on multivariate dynamic kernels to re-examine—under a non-linear, kernel methods framework—the experimental tests reported by Welch and Goyal (Rev Financ Stud 21(4):1455–1508, 2008 ) showing that several variables proposed in the finance literature are of no use as exogenous information to predict the equity premium under linear regressions. For this new approach to equity premium forecasting, kernel functions for time series are used with multiple kernel learning (MKL) in order to represent the relative importance of each of the variables. We find that, in general, the predictive capabilities of the MKL models do not improve consistently with the use of some or all of the variables, nor does the predictability by single kernels, as determined by different resampling procedures that we implement and compare. This fact tends to corroborate the instability already observed by Welch and Goyal for the predictive power of exogenous variables, now in a non-linear modelling framework.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1370-4621
1573-773X
1573-773X
DOI:10.1007/s11063-018-09971-7