Abstract
Using micro- and macro-level reference statistics, we assess the information content of qualitative (soft) data, reported monthly by the Israeli Business Tendency Survey since 2011.
By matching firms' qualitative evaluations of past sales to corresponding administrative records of sales turnover, we show that the logistic relationship between soft and hard data has strengthened since 2015, when the questionnaire’s wording was simplified and focused on month-over-month evaluation. Estimated marginal effects show that the probabilities of the "Up" response have been influenced the most. Incorporating firms' random effects suggests significant intrasectoral firm heterogeneity, particularly in Services.
At the macro level, pairwise correlations between timely available sectoral balances of opinions and corresponding industrial indices, available with publication lags, allow us to gain end-of-series information for monthly projections of GDP growth by hard and survey-based (“sentiment”) covariates included in the PLS regression. We employ Chow-Lin time disaggregation to deal with monthly-frequency GDP data.
Due to the short sample and low volatility of real growth in recent years we have failed to beat the average growth assumption over the test period between 2016 and 2019. Since 2020, the contribution of the sentiment indicators has increased, and the model’s out-of-sample performance has improved.
Keywords: Business-tendency survey, Sentiment indicator, Partial Least Squares, Monthly GDP
JEL: E32, E37