Template-Type: ReDIF-Article 1.0

Author-Name: Csanad Temesvari
Author-Workplace-Name: Magyar Nemzeti Bank
Author-Email: temesvarics@mnb.hu
Author-Person:

Author-Name: Beata Horvath
Author-Workplace-Name: Magyar Nemzeti Bank
Author-Email: horvathbea@mnb.hu
Author-Person:

Author-Name: Livia Reka Onozo
Author-Workplace-Name: Magyar Nemzeti Bank; Budapest University of Technology and Economics
Author-Email: onozol@mnb.hu
Author-Person:

Title: Natural Language Processing-Driven Use-Cases for Economic Analysis Using Unstructured Data

Abstract: Economic text data, such as news articles or retail trade item names, are an
alternative, feature-rich, high frequency information source that can provide insight
into economic trends and generate timelier and more accurate estimates. We
trained multiple deep learning models for two distinct research tasks: 1) the creation
of a sentiment index derived from the categorisation of financial and economic
articles into three sentiment categories; and 2) the classification of retail trade item
names into appropriate tariff categories. Our models consistently outperformed
their baseline counterparts for retail trade item classification, while our sentiment
index was able to accurately predict economic downturns where high-frequency
data were not available.

Classification-JEL: C43, C45, C60
Keywords: Natural Language Processing, Deep Learning, macroeconomic nowcasting, classification
Pages: 27-52
Volume: 25
Issue: 1
Year: 2026
File-URL: https://hitelintezetiszemle.mnb.hu/sw/static/file/fer-25-1-st2-temesvari-horvath-onozo.pdf
File-Format: Application/pdf
Handle: RePEc:mnb:finrev:v:25:y:2026:i:1:p:27-52