Sentiment Indicators Based on a Short Business Tendency Survey

18/09/2019 |  Suhoy Tanya, Roash Daniel
All Press Releases In Subject:
The Economy and Economic Activity


The monthly frequency of the Business Tendency Survey, launched in 2011, sectoral representativeness, and early availability have created new opportunities for nowcasting. However, Israeli data confirm growing concerns that the aggregate balance of opinions has become less correlated with macroeconomic indicators in the post-crisis period.

We test this relationship using firm-level and macro (time-series) data.

At the firm-level, logit checks of qualitative evaluations of past domestic sales in the manufacturing, retail trade and services sectors in 2013–17 revealed significant cross-sectional correlations with corresponding revenue data, matched from administrative records; however, comovement between the qualitative evaluations and the aggregate sectoral index was documented only since the questionnaire wording was changed to focus on the specific month, instead of a three-month evaluation. Although this change has amplified seasonal variation in the categorical answers, correlations between qualitative and quantitative data remain (weakly) significant even after seasonal effects are controlled for. We find also that firms' heterogeneity has an effect on the reliability of qualitative evaluations, particularly in the services industry.

At the macro level, we are looking for a composite sentiment indicator that aggregates sectoral balances of opinions and tracks real growth at a monthly frequency. We suggest an indicator with time-varying weights, evaluated through Partial Least Squares regression with respect to GDP growth. As GDP is measured quarterly, we simulate intra-quarter GDP-changes from monthly interpolated and bootstrapped seasonally-adjusted GDP-levels. This sentiment indicator performs better than an overall balance of opinions, calculated as a composition of sectoral balances with predefined weights based on industrial GDP-shares. In most (about 85%) simulations the short-term forecasts outperform the benchmark of mean growth. The out-of-sample error is larger when the sentiment indicator forecast is compared to later GDP estimates published by the CBS than with the first estimate.

To the Full Article as PDF file ​