Papers for

economic forecasting teams

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Language models remember future facts when reading past economic text

Reading or Remembering? Measuring Temporal Contamination in Language Models

Abstract: Language models are now used to read central-bank text for economic forecasting, but most were trained on text written after the documents they read. A model reading a 2021 FOMC statement may already know that inflation rose in 2022, so part of its apparent skill can be memory of the outcome, a temporal form of test-set contamination. We measure this memory on FOMC statements and minutes from 2000 to 2026 by comparing modern open models with chronologically consistent models, which are trained only on text from before a fixed date. Asked about past GDP outcomes given only a date, models that cannot know the answer still reach an AUC of up to 0.64 from priors on dates and eras, so chance is the wrong baseline. Against these memory-free models on the same questions, most modern models recall which years had strong growth. Membership-inference scores sort documents by era and document type rather than by whether a model saw them, as exact counts in an open pretraining corpus show. In real-time nowcasts, every lookahead premium we estimate on US GDP is smaller than the forecast comparison can detect at macroeconomic sample sizes, and none on GDP or inflation survives correction for multiple testing. Lookahead audits should therefore measure memory directly against memory-free models on the same items.

Mon 16 NovMachine LearningComputation and Language
The gist
Language models that analyze economic documents sometimes already know events that happened after those documents were written, like inflation rising after a 2021 report. The authors found this causes the models to appear better than they are because they 'remember' future outcomes instead of truly 'reading' the original text. They tested models trained only on past data and showed that some skill comes from general knowledge about dates, not from seeing the documents. This work helps distinguish whether models are understanding or just recalling information.
Open → 2611.21554v1