Authors
Ross Metusalem
JMP
Muralidhara A
JMP
Objective
Apply Sentiment analysis to quantify the emotion in unstructured text.
Background
The Federal Reserve System is the central bank of the United States. It is comprised of 12 districts, each overseen by a board of directors. Eight times per year, each board of directors issues a narrative report of the state of the economy in their district. These reports, collectively known as the Beige Book for their traditional beige-colored covers, summarize the conditions in many economic sectors, such as housing, retail, agriculture, tourism, and finance. Each report presents anecdotal information gleaned through interviews with business leaders, industry experts, and other sources.
The reports frequently use language that conveys sentiment – feelings, attitudes, or emotions – about the economy; for example, a report might state that “retailers are generally upbeat” (positive sentiment) or “commercial construction continued to report dismal conditions” (negative sentiment). The Beige Book thus contains valuable information about how experts feel about the state of the economy, a valuable source of information not readily available in other economic indicators. Yet, this sentiment is contained within unstructured text data, which is not amenable to quantitative analysis in its raw form.
Lexical sentiment analysis
In text mining, sentiment analysis refers to a collection of techniques for quantifying sentiment in unstructured text data. The goal is to assign a sentiment label or score to each document, where a document can be a product review, social media post, email, newspaper article, or in this case, a economic narrative report. Documents typically are labeled or scored for positive versus negative sentiment. The sentiment classifications or scores then form the basis for quantitative analysis.
Numerous methods exist for assigning sentiment scores to raw text. This case study utilizes a common technique known as lexical sentiment analysis. In this method, the researcher or analyst specifies a dictionary (aka lexicon) of sentiment terms and corresponding numeric scores, and the values in this dictionary are used as the basis for calculating an overall sentiment score for each document. We discuss the dictionary and document scoring in detail in the following sections.
The Task
We use lexical sentiment analysis to explore changes in sentiment in the Beige Book reports before and after the Great Recession of December 2007-June 2009. Precipitated by a complex set of factors, most notably the collapse of the U.S. housing market, the Great Recession saw a loss of $2 trillion USD from the global economy. Bankruptcies, business closures, unemployment, and personal hardship spread across the world, and governments took drastic measures to avert a global economic collapse. By many metrics, the recession was the most severe economic downturn since the Great Depression of the 1930s.
We use sentiment analysis to determine the timing and extent of changes in Beige Book sentiment in the time period encompassing the Great Recession. Exploring this relationship represents a first step in understanding how the sentiment conveyed in economic reports might be useful as an indicator of economic recession and recovery.