This paper evaluates fine-tuned large language models (LLMs) as probabilistic classifiers for Economic Policy Uncertainty (EPU), showing LLMs substantially outperform traditional keyword (BOW) and SVM approaches at the article level, that these gains materially change index construction through thresholding and aggregation choices, and that the same models enable historical (19th-century U.S.) and multilingual EPU measures—arguing LLMs should be treated as explicit, estimable parts of the data‑generating process .
元数据与概览(引言与报告概览)
逐节深度解读
图表与图像深度解读(关键环节) (展示并解析作者所有主要图表)
图 1(文章级别分类性能,图片文件页码 9)
图 2(从文章级到指数级:聚合与阈值影响,图片文件页码 12)
图 3(历史十九世纪美国 EPU 指数,图片文件页码 14)
图 4(多语言性能与跨国指数,图片文件页码 17)
估值/计量方法部分(本报告不适用)
风险因素评估(报告中识别的风险)
批判性视角与细微差别
结论性综合(总结性要点)
附注(方法学与图像复用)
总结性判断(客观)
若您希望,我可以:
2511.17866.pdf