Machine Learning - Special issue on learning with probabilistic representations
Time Series Analysis, Forecasting and Control
Time Series Analysis, Forecasting and Control
Naive (Bayes) at Forty: The Independence Assumption in Information Retrieval
ECML '98 Proceedings of the 10th European Conference on Machine Learning
Credit Rating Change Modeling Using News and Financial Ratios
ACM Transactions on Management Information Systems (TMIS)
| Hi-index | 0.00 |
We introduce a novel dataset with a news sentiment index that was constructed from a selection of over 300,000 newspaper articles from five of the top ten U.S. newspapers by circulation. By constructing ARMA models, we show that news and consumer sentiment, when combined with other macroeconomic variables, achieve statistically significant results to explain changes in private consumption. We make three distinct findings with respect to sentiment in consumption behavior models: first, both consumer and news sentiment add explanatory power and statistical significance to conventional consumer behavior models. Second, consumer sentiment, measured by the University of Michigan Index of Consumer Sentiment, adds more explanatory power and statistical significance than news sentiment when tested individually. Third, news sentiment is able to determine the signs of all coefficients in the model correctly, whereas consumer sentiment does not. In general, we conclude that news sentiment is a useful variable to add in consumer behavior models, especially when coupled with consumer sentiment and other macroeconomic variables. Tested individually, news sentiment is as good a proxy as personal income for explaining private consumption growth when tested individually.