Title: Detecting Recessions Through Real-Time GDP Figures

Abstract

This study investigates the efficacy of real-time GDP reports in accurately pinpointing the onset and termination of business cycles. By leveraging U.S. real-time GDP data spanning from 1947 to 2021, we developed 4,356 distinct recession indicators. These were created by applying various scaling adjustments and smoothing techniques. Subsequently, by pairing these indicators with diverse threshold levels, we generated a pool of 137,457 perfect recession classifiers. The classifiers we selected successfully detected all 12 historical recessions, achieving a flawless record with zero false positives or false negatives. When focusing on the high-precision subset of these models, two specific classifiers demonstrated detection error standard deviations under three months. On average, the chosen ensemble signals a recession 3.04 months after it officially begins. The framework’s ability to consistently identify recession periods across different data vintages suggests that inconsistencies in earlier research might stem from the constraints of conventional dating methods and data revisions. Ultimately, the findings confirm that real-time GDP announcements serve as a viable and practical alternative for NBER-style recession dating.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC