...information contained in job-related tweets provides new economic indicators.
Social media provide an enormous amount of high-frequency, real-time information which can potentially be used to complement or even substitute for traditional sources of economic data. In Using Social Media to Measure Labor Market Flows (NBER Working Paper No. 20010), Dolan Antenucci, Michael Cafarella, Margaret Levenstein, Christopher Ré, and Matthew Shapiro create indexes of job loss, job search, and job posting with data from Twitter posts and compare these indexes to official data. They conclude that information contained in job-related tweets provides new economic indicators.
To construct the indexes, the authors estimate the principal components of data series on the number of job-related phrases such as "lost my job" in tweets over a period of 28 months. The resulting job loss index, which is updated weekly by University of Michigan Economic Indicators from Social Media (http://econprediction.eecs.umich.edu/), tracks official data on claims for unemployment insurance well. While the two series exhibit the same general trend and similar spikes, the fit of the social media series is not meant to be perfect because of differences in population, timing, and the fact that some job losses do not lead to applications for unemployment insurance.
The authors compare their social media job loss index to the consensus forecast from macroeconomic forecasters on the eve of announcements of initial unemployment insurance claims. They find that variation in the social media index can account for about one-fifth of the prediction error variance in the consensus forecasts, so information from the job-related tweets could improve forecast accuracy.
The authors also use social media data to study domains such as job search and job postings, where official statistics are less timely and less frequent. By studying the relationship between their job loss and job posting indexes, they conclude that the Beveridge curve, which relates the unemployment rate to the job vacancy rate, has been shifting inward since 2011, implying that the number of job vacancies that is consistent with a given unemployment rate has been dropping.
-- Claire Brunel