A new self-normalized forecast comparison test

This study develops a novel self-normalized Diebold–Mariano (DM) test for evaluating equal forecast accuracy. The proposed test offers several distinct advantages: it avoids bandwidth selection and bypasses direct estimation of long-run variances, both of which are typically required in conventional...

Full description

Saved in:
Bibliographic Details
Published in:Economics letters Vol. 256; p. 112646
Main Authors: Li, Haiqi, Zhang, Ni, Zhou, Jin
Format: Journal Article
Language:English
Published: Elsevier B.V 01.10.2025
Subjects:
Diebold–Mariano test
Functional central limit theorem
Predictive accuracy
Self-normalization
C22
Predictive accuracy
C12
Diebold–Mariano test
Functional central limit theorem
Self-normalization
ISSN:0165-1765
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study develops a novel self-normalized Diebold–Mariano (DM) test for evaluating equal forecast accuracy. The proposed test offers several distinct advantages: it avoids bandwidth selection and bypasses direct estimation of long-run variances, both of which are typically required in conventional forecast accuracy testing approaches. Under relatively mild regularity conditions, we show that the asymptotic null distribution of the self-normalized DM test statistics is pivotal, with corresponding critical values tabulated through simulations. Comprehensive Monte Carlo simulations confirm that our self-normalized DM test has superior finite-sample performances compared to the original and the existing modified DM tests. •Proposes a novel self-normalized DM test for evaluating equal forecast accuracy.•Derives pivotal asymptotic null distribution with tabulated critical values.•Outperforms original and the existing modified DM tests in simulations.
ISSN:0165-1765
DOI:10.1016/j.econlet.2025.112646