Twitter a goldmine for tracking consumer mood on prices, Bank of Italy finds

The analysis started by collecting 11.1 million tweets posted in Italian between June 2013 and December 2019 containing at least one of a set of previously selected words related to inflation, prices and price dynamics.

"The results suggest that Twitter can be a new timely source for devising a method to elicit beliefs," the authors of the 107-page study said. (REUTERS)

The Bank of Italy said on Monday a set of experimental indicators it created from the content of millions of tweets accurately tracked consumer mood on price, offering scope for a powerful new monetary policy tool.

The effort comes as economists and policy-makers around the world increasingly turn to social media and other unconventional sources to measure consumer behaviour and as inflation continues to defy targets set by many leading central banks.

Researchers found their indicators, based on millions of tweets, not only tallied with final inflation read-outs and existing measures of price expectations by Italy's national statistics office, financial markets and other forecasters but were also in real-time and provided more granular detail.

"The results suggest that Twitter can be a new timely source for devising a method to elicit beliefs," the authors of the 107-page study said, adding they believed the Italy-focused research could be replicated elsewhere.

Twitter has roughly 200 million monthly active users worldwide and had around 10 million active users in Italy in 2019, the authors said.

The analysis started by collecting 11.1 million tweets posted in Italian between June 2013 and December 2019 containing at least one of a set of previously selected words related to inflation, prices and price dynamics.

"The rationale for focusing on pure raw tweets count is the intuitive notion that the more people talk about something, the larger is the probability it reflects their opinion and that their view can influence other people's expectations," it said.

Then the dataset was "cleaned" to remove advertisements or tweets that use the word inflation in an unrelated context.

In this way, tweets such as "#Draghi: 'We saved Europe from deflation.' Do not count your chickens before they are hatched!" were kept, while others, such as "Only at Baby Glamour if you buy three items the least expensive is free. Promotional sales until October 10" were filtered out.

The remaining dataset was used to build two indexes on expectations of increasing or decreasing inflation by measuring the daily volume of tweets containing previously selected word combinations such as "bargain price" or "very high price".

"The fact that economic agents talk about expensive bills should reflect expectations of higher inflation," the report said. "On the other hand, people discussing declining oil prices should correspond to expectations of lower inflation."

The final set of indicators was then created based on divergence between the two indexes.

The authors said their work underscored the significance and policy implications of information contained on social networks but acknowledged further study was needed to interpret the data.

They also noted that there were a few cases of a Twitter-based indicator been thrown off course by a viral social media event, for example when the sale of an apartment for a record-breaking $236 million in 2014 led to a flurry of tweets containing variants of the phrase "more expensive".