Detecting misinformation and quantifying its business impact: a machine learning analysis of financial volatility and retail demand shocks

Abstract

Financial markets and retail supply chains are in-creakingly exposed to rapid information flows, making them vulnerable to misinformation and disinformation circulating through news and social media. This paper examines how misleading information translates into measurable business risk by linking automated misinformation detection with observed anomalies in stock prices and retail demand. We develop and evaluate machine learning models to classify misleading news content and align detected misinformation events with real-world indicators of financial volatility and demand forecast error (DFE). Using a human-curated dataset of 45,000 real news articles labeled for veracity, we compare Logistic Regression, Gradient Boosting, and Bidirectional LSTM models for misinformation detection. Gradient Boosting achieves the highest classification accuracy (99.44%), followed closely by Logistic Regression (98.85%), while the BiLSTM performs substantially worse on this dataset. We then examine documented misinformation events and show their association with extreme stock price movements, abnormal trading volumes, and sharp retail demand surges that lead to large forecasting errors. Overall, the results suggest that integrating misinformation detection with time-series monitoring can help firms recognize information-driven disruptions earlier and manage financial and operational risk more effectively.