Sentiment index as a predictor of CPI: A lexicon-based approach using economic news data in Vietnam

Abstract

This study aims to construct a sentiment index for predicting CPI in Vietnam. Adopting a lexicon-based approach, the study utilized two widely recognized sentiment dictionaries, specialized in financial and economic contexts, to build the sentiment index. Data was mined from nine official economic and financial news websites for the period 2017–2024 using GDELT, resulting in approximately 200,000 URLs relevant to inflation sentiment. Our findings confirm two key points. First, sentiment factors are a significant predictor of inflation fluctuations, contributing 22.3 % to the variance. Second, by incorporating sentiment variables into Vector Autoregressive (VAR) and Artificial Neural Network (ANN) models, we achieved an acceptable accuracy in forecasting Vietnam's inflation. These findings, on the one hand, provide an empirical case for utilizing sentiment analysis in macroeconomic forecasting within developing countries. On the other hand, these findings suggest that policymakers should leverage sentiment analysis to enhance the effectiveness and timeliness of economic management in an increasingly volatile economy.