The analysis of tweets can predict changes in the volatility of equity markets. This is revealed by research conducted by the Centre for Experimental Research in Management and Economics at Ca’ Foscari University of Venice in collaboration with GAM.
To test how big linguistic data can help predict economic trends, the researchers analysed 1.2 million tweets that contained the word ‘uncertainty’. This covered over 3,500 tweets a day among 200 million English tweets between April and December 2016, a period marked by the Brexit referendum and the US presidential election.
The results of the study were presented in Milan at the GAM Italy SGR headquarters by Carlo Santagiustina, the lead author of the study and PhD student at the Department of Economics, Massimo Warglien, professor at the Department of Management of Ca’ Foscari and coordinator of the Laboratory of Experimental Economics GAM - Ca’ Foscari, Anthony Lawler, co-manager of GAM Systematic, and Riccardo Cevellin, Managing Director of GAM (Italy) SGR.
The researchers built an “Uncertainty Twitter Index” to predict the volatility indices of the British and American stock exchanges (VFTSE and VIX). A first aim of the research, in fact, was to identify predictive signals, in a VAR model with constant parameters, of the sign of the volatility of the stock markets.
The study shows that uncertainty in civil society, measured through the index, allows us to predict the sign of changes in volatility implied by stock markets with a high degree of accuracy: 79% for the US market and 84% the UK market.
There was also a reconstructed map of the contagion routes between markets, political uncertainty and options of civil society in the UK and the US. Finally, some dynamics of international propagation of uncertainty could be explored in greater detail.
Why analysing Twitter to try to predict economic facts? “Twitter is a solid and high frequency data source that allows us to track the patterns in the opinions of a broad ‘civil society’ online”, explained Carlo Santagiustina.