解析脉搏：用LLMs分解宏观经济情绪（英）

BIS Working Papers No 1294 Parsing the pulse: decomposing macroeconomic sentiment with LLMs by Byeungchun Kwon, Taejin Park, Phurichai Rungcharoenkitkul and Frank Smets Monetary and Economic Department October 2025 JEL classification: E30, E44, E60, C55, C82 Keywords: macroeconomic sentiment, growth, inflation, monetary policy, fiscal policy, LLMs, machine learning BIS Working Papers are written by members of the Monetary and Economic Department of the Bank for International Settlements, and from time to time by other economists, and are published by the Bank. The papers are on subjects of topical interest and are technical in character. The views expressed in this publication are those of the authors and do not necessarily reflect the views of the BIS or its member central banks.This publication is available on the BIS website (www.bis.org). be reproduced or translated provided the source is stated. ISSN 1020-0959 (print) ISSN 1682-7678 (online) © Bank for International Settlements 2025. All rights reserved. Brief excerpts may Parsing the Pulse:Decomposing Macroeconomic Sentiment with LLMs∗Byeungchun Kwon†Taejin Park‡Phurichai Rungcharoenkitkul§Frank Smets¶1 October 2025AbstractMacroeconomic indicators provide quantitative signals that must be pieced to-gether and interpreted by economists. We propose a reversed approach of parsingpress narratives directly using Large Language Models (LLM) to recover growthand inflation sentiment indices. A key advantage of this LLM-based approach isthe ability to decompose aggregate sentiment into its drivers, readily enabling aninterpretation of macroeconomic dynamics. Our sentiment indices track hard-datacounterparts closely, providing an accurate, near real-time picture of the macroe-conomy. Their components–demand, supply, and deeper structural forces–are in-tuitive and consistent with prior model-based studies.Incorporating sentimentindices improves the forecasting performance of simple statistical models, pointingto information unspanned by traditional data.JEL Classification: E30, E44, E60, C55, C82Keywords: macroeconomic sentiment, growth, inflation, monetary policy, fiscal policy,LLMs, machine learning∗We thank seminar participants at the Bank for International Settlements for helpful comments. Allremaining errors are ours. The views expressed are those of the authors and do not necessarily representthose of the Bank for International Settlements. The resulting indices constructed in this paper will bemade available on the BIS website.†Bank for International Settlements, byeungchun.kwon@bis.org‡Bank for International Settlements, taejin.park@bis.org§Bank for International Settlements, phurichai.rungcharoenkitkul@bis.org¶Bank for International Settlements, frank.smets@bis.org11IntroductionThe emergence of Large Language Models (LLMs), built on transformer architectures,marks a significant advance in textual analysis. Trained on massive corpora with billionsof parameters, these models excel