Query Indices and a 2008 Downturn: Israeli Data

"Predicting the Present with Google Trends" (Choi, Varian, 2008) offers a nowcasting model, suggesting that the popularity of web searches tracked by Google over time may be applied as an indicator of contemporaneous economic activity, before the official data become available and/or are revised. Such evidence was found in different countries with respect to automobile sales, home sales, retail trade and travel. Using Google's Insights for Search application, which aggregates search terms by large socioeconomic categories, I test the predictive ability of Israeli query indices. At a monthly frequency, six predictive categories were found: Human Resources (Recruitment and Staffing), Home Appliances, Travel, Real Estate, Food and Drink and Beauty and Personal Care. According to the pair-wise Granger causality tests, the strongest predictor is in the Human Resources category (Recruitment and Staffing). The latter, taken at a quarterly frequency, is a leading indicator with regard to the job openings ratio, currently surveyed by the Ministry of Industry, Trade and Labor, and consequently may help in drawing monthly inferences about the unemployment rate. The large fraction of innovation variance explained by other selected categories is related to the employment rate. It suggests an attitudinal interpretation of the queries effect, which may be conveyed through the employment channel. In-sample Bayesian probabilities of a downturn computed between 2004:2 and 2009:2, peak in the first half of 2007 and since March 2008, conforming official assessments of a slowdown in economic growth and an economic decline, respectively. Real-time simulations made since December 2007 signal a downturn likely to occur from April 2008. For monitoring purposes, an index of Home Appliances queries has been incorporated as an instrumental variable in the State-of-the-Economy composite index, while current assessments of private consumption (trade and services revenue) have been improved, in terms of RMSE. The operative use of query data, however, may encounter problems that deserve attention. First, query indices may appear not to be stationary, due to alternative social search which is not tracked by the Google engine. Second, the predictive ability of query indices may vary over time.