Format | |
---|---|
BibTeX | |
MARCXML | |
TextMARC | |
MARC | |
DataCite | |
DublinCore | |
EndNote | |
NLM | |
RefWorks | |
RIS |
Files
Abstract
Using natural language processing methods with Twitter data, I estimate a daily series of short-run US inflation expectations for the 2020-2022 period. Treating survey-based sources as a benchmark, I show that Twitter series reveals a similar pattern with a correlation of 89% with the survey data. Using tweet geolocation, I document state-level heterogeneity in inflation expectations that may be explained via differences in regional cost of living. Finally, as an application for macroeconomic forecasting, I show that the daily Twitter series is predictive of movements in future inflation, making it a relevant input in (now)forecasting exercises. More generally, this paper illustrates a textual analysis pipeline that can be used to extract inflation expectations almost concurrently and cost-effectively and highlights the potential of using unstructured text to obtain critical information about the perceived current state of the economy.