An experimental sentiment indicator for the euro area – the relevance of broad - based sectoral shifts in economic sentiment

The Economic Sentiment Indicator (ESI), published by the European Commission on a monthly basis, is a powerful tool for tracking year on year GDP growth. However, its performance is weaker when GDP growth is expressed in quarter on quarter changes. This paper investigates whether cross-sector survey data gathered in the framework of the harmonised EU Business and Consumer Survey (EU BCS) can be combined in a way different from the ESI construction method, with the explicit aim of boosting correlation with q-o-q GDP growth. The construction method under investigation deviates from the ESI in that i) only the survey questions best correlated with q-o-q GDP growth are used and ii) the q-o-q change of the selected questions is amplified through multiplication with a constant in case a critical amount of the selected questions (at least 8 out of 11) changes in the same direction as the average. The logic of the latter is that changes in the average of the survey questions should be taken more "seriously" (i.e. amplified) in case they are broad-based, i.e. reflected in many questions (and sectors) underlying the indicator. The resulting experimental indicator indeed achieves promising results: Its coincident and leading correlation with q-o-q GDP growth improves significantly compared to the ESI. In a subsequent step, the paper examines the merits of the experimental sentiment indicator for the purpose of nowcasting q-o-q GDP growth in the euro area at the end of the third month of a given quarter. Departing from an autoregressive benchmark model, the performance of a bi-variate model containing only the experimental indicator is tested. Subsequently, slightly more complex models, which add some "hard" data as predictor variables, are presented. The hard data include macro-economic variables (unemployment, etc.) as well as sector-specific variables (index of industrial production, etc.) which are sufficiently timely available at euro area level. Using a top-down testing approach, the non-significant variables are removed to produce parsimonious models. All models, including the simple bi-variate one, are shown to outperform the benchmark model in terms of MAE, RMSE and different hit-ratios. The differences are most pronounced from 2008q2 onwards (i.e. from the onset of the financial crisis). The best-performing of the new models combines the sentiment indicator with industrial production growth and a measure of volatility on the European stock