The paper suggests new daily indicators of consumers' inflation expectations built using machine learning techniques and textual analysis drawn from the microblogging and social networking service, Twitter. Trends in these indicators are compared both with monthly inflation expectations based on the Istat surveys and with those obtained daily from the market prices of derivative contracts on inflation (inflation swaps), which measure the expectations only indirectly, because they include unobservable risk premia.
Adopting the most recent specific algorithms for dealing with big and voluminous data (Big Data Analytics), the paper shows that the Twitter-based indicators are significantly correlated with the other existing measures of inflation expectations. Furthermore, these indicators have a superior forecasting power for survey-based monthly inflation expectations than all the other available sources. Therefore, they represent an additional source of information available in real time.
Published in: Journal of Econometrics, v. 228, 2, pp. 259-277
