Daily Assessment of News Public Opinion Score
The Daily News Sentiment Index, a groundbreaking tool for gauging economic sentiment, has been unveiled by researchers at the Federal Reserve Bank of San Francisco. The index, based on lexical analysis of economics-related news articles from 24 major US newspapers, offers a high-frequency, objective gauge of investor and consumer mood.
The methodology behind the index is rooted in natural language processing (NLP) techniques and financial market data integration. The process begins with the collection and preprocessing of vast datasets of financial news headlines, which are then aligned with numerical market indicators like daily stock prices and trading volumes.
The core of the sentiment extraction process involves the use of a pre-trained financial domain-specific language model, FinBERT, fine-tuned for capturing sentiment in finance-related text. Each news headline is tokenized and passed through FinBERT, resulting in probabilities of the text being positive, neutral, or negative. These probabilities are then aggregated daily for each stock ticker to compute a daily sentiment score.
Beyond textual sentiment, the methodology also incorporates market price movements to create a Numerical Sentiment Index (NSI) that quantifies sentiment inferred directly from market behavior. The integration of textual and numerical sentiment data supports a robust measurement that reflects both reported sentiment and market reaction.
The Daily News Sentiment Index serves as a real-time proxy for economic sentiment, reflecting how news narratives and market participants’ perceptions evolve daily. It aligns with traditional consumer confidence and economic indicators by capturing the tone and outlook embedded in financial news, often preceding or corroborating shifts in economic sentiment measured by surveys or macroeconomic data.
By quantitatively summarizing positive versus negative news sentiment and combining it with market data, the index provides a high-frequency, objective gauge of investor and consumer mood. This information can be invaluable to policymakers, investors, and analysts seeking to assess economic conditions and forecast market trends.
The index also facilitates causal analysis of how news sentiment influences market volatility and economic confidence indirectly, helping to understand sentiment-driven market dynamics. The Daily News Sentiment Index is constructed as a trailing weighted-average of time series, with weights that decline geometrically with the length of time since article publication.
The research papers providing data and code for the Daily News Sentiment Index, including "Measuring News Sentiment" and "News Sentiment in the Time of COVID-19," were published by Adam Hale Shapiro, Moritz Sudhof, and Daniel J. Wilson. The data for the index will be regularly updated at a weekly frequency, and a Daily News Sentiment data Excel document and a replication code Zip file are available for download.
In summary, the Daily News Sentiment Index is a cutting-edge tool that offers a high-frequency, objective gauge of economic sentiment based on lexical analysis of economics-related news articles from major US newspapers. The index provides timely and nuanced insights into the mood of the economy as reflected through the lens of financial news and market responses.
The Index, rooted in natural language processing (NLP) techniques and financial market data integration, combines textual sentiment from finance-related news with market data to provide a high-frequency, objective gauge of investor and consumer mood in business. This information, derived from the Daily News Sentiment Index, can be utilized by policymakers, investors, and analysts to analyze economic conditions and forecast market trends.
