Quality Enhancements in Experimental Statistics: The Italian Social Mood on Economy Index

Istat, since October 2018, publishes an experimental daily index, i.e., the Social Mood on Economy Index (SMEI). This index is derived from real-time samples of public Italian tweets, containing at least one word belonging to a set of keywords. The sentiment scores are obtained by means of an Italian sentiment lexicon, namely, a vocabulary whose lemmas are associated to pre-computed sentiment values. Two core methodological aspects have been recently analyzed and redesigned: the filter and the vocabulary. The scope was to answer two main questions: was the set of keywords properly grasp economic conversations and was it available, another lexicon, which could improve the quality of tweets’ scoring and therefore the dynamics of the daily series? To answer the first question, a word embedding analysis has been carried out. This allowed elaborating a new sub-filter. Peaks observed in the sub-filtered series seem to be more consistent and more related to economy. Concerning the vocabulary, comparisons between the dynamics of the time series, obtained by using two different lexicons, namely, Sentix and DPL, show better results in favor of the latter. More precisely using DPL, we observe (i) more coherent downward trend during Covid waves and (ii) an increase in the correlation with economic series. Overall, the current methodology for the SME index has been revised to embody these findings and improvements.