News Analytics and Content Analysis

What is news analytics

News analytics is the measurement of the various qualitative and quantitative attributes of textual news stories. Typical attributes include sentiment (whether the tone is positive or negative), relevance (how much a story concerns a given firm), and novelty (whether it reports something new). The work is usually done with automated text analysis, drawing on natural language processing and machine learning techniques such as latent semantic analysis, support vector machines, and "bag of words" representations.

Applied to finance and management research, news analytics lets you text-analyze the full sequence of a firm's press releases over time along dimensions such as sentiment, action types, or action patterns. Because the measurement is automated, these methods scale to many thousands of news items, which suits academic samples that span many firms over extended periods.

Content analysis at the core

At its heart, news analytics rests on content analysis: the structured coding of text into categories you define in advance. Instead of reading each document, you specify a scheme of categories and keywords, and a tool tags every text accordingly. The output is a set of variables, for example a sentiment score or an action-type flag, that you can merge with prices, accounting data, or other firm characteristics.

Dictionary-based approaches are common: a category is defined by a list of words, and the share of matching terms drives the score. Proximity rules can sharpen this, so a category fires only when two terms appear close together in the text. The result is a transparent, reproducible mapping from raw news to research-ready measures.

A practical workflow

A news analytics project typically runs in three stages. First, build a database of press releases or other news items for the firms and period you study. Second, lay the dataset out in time by identifying the announcement date of each item, so each text is anchored to an event date. Third, characterize the items along the dimensions you care about through text analysis.

For the date step, a regular-expression date finder reads each text and extracts the relevant announcement date, linked back to a text ID. For the analysis step, a content-analysis tool applies your category scheme to score or classify each text. You can then filter the results into the sub-samples you need, for instance keeping only items that fall into a particular action category.

How it feeds an event study

An event study measures how a security's price reacts around a specific event. The hard part is often defining the event cleanly: which announcement, on which date, of which type. News analytics supplies exactly that. By coding the news, you obtain the event date and a qualitative label, then estimate abnormal returns for each subgroup.

This pairing also supports longitudinal designs. Because text analysis can repeat across a firm's entire stream of releases, you can compare how markets respond to different action types, or track how a firm's communication patterns evolve, rather than studying a single one-off event.

Tools and next steps

EventStudyTools provides free apps for each step. The CATA app is a content-analysis text analyzer: you supply a category scheme and a set of texts, and it scores or classifies each one. CATA is a text-analysis tool, not itself an event study, but its output plugs directly into one. The EDI app identifies event dates in text so each item is correctly anchored in time.

To put the measures to work, continue with the event study methodology hub, or run the Abnormal Return Calculator once your events and dates are coded.