Can We Measure Inflation Expectations Using Twitter?

Using Italian data from Twitter, we employ textual and machine learning techniques to build new real-time measures of consumers' inflation expectations. First, select some relevant keywords identify tweets related prices expectations thereof. Second, a set daily on the selected combining Latent Dirichlet Allocation (LDA) with dictionary-based approach, using manually labelled bi-grams tri-grams. Finally, show that Twitter-based indicators are highly correlated both monthly survey-based market-based expectations. Our provide additional information beyond expectations, professional forecasts, realized inflation, anticipate proving be good proxy. Results suggest Twitter can timely source elicit beliefs.