A Predictable E-Process for Anytime-Valid Overidentification Monitoring in GMM

15 January 2026, Version 1
This content is an early or alternative research output and has not been peer-reviewed by Cambridge University Press at the time of posting.

Abstract

Classical Hansen–Sargan overidentification testing is not robust to optional stopping and repeated “peeking” in real-time monitoring. This manuscript intro- duces the S M Nazmuz Sakib eJ-Test and the associated Sakib Index SIt, an anytime-valid evidence stream for overidentified moment restrictions. The key idea is a predictable orthogonalization of GMM moments that removes parameter directions online, combined with a Gaussian mixture supermartingale to produce an e-process Et that is valid under optional continuation. All empirical illustrations are dataset-based and use open World Bank data on U.S. inflation and unemployment (1975–2024). In our inflation illustration, the resulting e-process crosses the 1/α threshold for α = 0.05 in year 2023, while an expanding-window J-path crosses much earlier, illustrating how repeated monitoring can distort classical fixed-sample testing.

Keywords

e-value
e-process
safe testing
optional stopping
generalized method of moments
Hansen J
online inference

Comments

Comments are not moderated before they are posted, but they can be removed by the site moderators if they are found to be in contravention of our Commenting and Discussion Policy [opens in a new tab] - please read this policy before you post. Comments should be used for scholarly discussion of the content in question. You can find more information about how to use the commenting feature here [opens in a new tab] .
This site is protected by reCAPTCHA and the Google Privacy Policy [opens in a new tab] and Terms of Service [opens in a new tab] apply.