NLP-enhanced inflation measurement using BERT and web scraping

Area of Science:

• Computational Economics
• Natural Language Processing
• Econometrics

Background:

• Traditional inflation metrics may not capture granular market dynamics.
• The Harmonized Index of Consumer Prices (HICP) serves as a key economic indicator.
• Consumer electronics prices exhibit complex fluctuations.

Purpose of the Study:

• To develop a custom inflation index using NLP and web scraping.
• To benchmark the custom index against the official HICP.
• To analyze detailed price variations in consumer electronics.

Main Methods:

• Web scraping was employed to gather a database of consumer electronics products.
• The BERT model was utilized for high-performance classification of items into COICOP categories.
• Weighted and median methodologies were applied to construct the custom price index.

Main Results:

• Achieved 94.56% accuracy in classifying approximately 10,000 items into COICOP categories using BERT.
• The custom index, especially with weighted/median methods, aligned closely with the official HICP.
• Observed greater variability in monthly inflation trends within the COICOP 091 category compared to the HICP.

Conclusions:

• NLP and web scraping offer a powerful approach for alternative inflation measurement.
• Computational methods can enhance economic analysis by providing more detailed insights.
• The custom index demonstrates potential for a more nuanced understanding of inflation.