New social media index for inflation expectations
The Frankfurt School of Finance & Management (FS) is using posts on the social media platform X, formerly known as Twitter, to assess short-term inflation expectations of consumers in Germany in real-time. Researchers employ natural language processing and machine learning to analyze posts related to inflation, in order to gauge consumer sentiment. "This index can be an interesting addition to our toolkit for estimating inflation developments," said Johannes Müller, Global Head of Research at DWS, regarding the FS index during its introduction on Monday.
The project is currently in the development phase, and the university reports positive interim results. The available data from 2011 to May 2023 have shown a strong correlation with realized inflation rates. Additionally, the index responds to monetary policy measures. Unexpected easing by the European Central Bank (ECB) causes it to rise, while interest rate hikes lead to a decrease in the index, albeit with a delay of about a week. Researchers attribute this delay to users frequently discussing rising prices immediately after such a decision, which is the cause of the interest rate increase, and only later expressing expectations of falling prices. The index is calculated by subtracting posts on X about decreasing prices from those about rising prices.
The FS highlights the advantages of the index, including a larger sample size than the Bundesbank's surveys on consumer inflation expectations and the availability of real-time data. "We plan to expand our index analysis and, for example, investigate the relationship between inflation and private consumption. Furthermore, we can also set up the index for other countries," said Sascha Steffen, Professor of Finance and Director of the Centre for European Transformation at FS.
However, the index faces challenges as it currently cannot differentiate between short-term and medium-term inflation expectations of consumers due to technological limitations. To address this, researchers intend to manually feed the system with data to enable it to make such distinctions in the future. Additionally, users on X do not represent a cross-section of the population. The FS acknowledges this but points out that the index strongly correlates with the Bundesbank's representative survey, which is released with a time delay.