Analyzing Twitter allows … to anticipate inflation in a country

Does the social network Twitter have any other interest than bitching on certain politicians or selling products? According to the Bank of Italy, yes. The institution published on February 15, 2021 a major study by which researchers seem to have discovered that Twitter can analyze inflation trends.

11 million Italian tweets studied

Researchers recorded 11.1 million tweets posted on Twitter in Italy between June 2013 and December 2019. The tweets were selected based on different keywords related to inflation and price trends, then were “cleaned” to eliminate parasitic tweets, such as promotional tweets, or uses of terms outside the context sought.

For their study, the researchers therefore ultimately saved 1.5 million tweets, i.e. a little less than 15% of the initial set, which they then analyzed at the level of the comments. Their (condensed) hypothesis was as follows: if at any time t the tweets were more generally speaking of low prices, inflation must be falling, while if at any other time the tweets appeared to show high prices, l inflation should be on an uptrend.

A reliable … and reproducible measurement

Once the researchers were able to identify trends in price changes through analysis of the tweets, they had no than to compare them with the official statistics of the Bank of Italy and the various forecasts of Istat, the Italian statistical institute, over the same period from mid-2013 to the end of 2019.

The trends identified on Twitter were not only accurate, but sometimes even more precise than those of the classic indicators used in statistics. If, of course, Twitter is not enough, the researchers believe that by improving their analysis and search algorithm, Twitter could eventually become a tool to integrate into their official statistics.

And, they point out, all statistical institutes could do it, since it would suffice to sort the tweets by the country of publication (even if Italian being a language little spoken outside Italy, there is theoretically less risk of parasitic tweets than in French or English).

Design by NewsLax