Data from the quarterly Employers Survey of the Ministry of Industry, Trade and Labor--which include retrospective data on the manning of employee posts and terminations of employment as well as employers' expectations regarding payroll levels in the following quarter--are published months before the release of the official data by the Central Bureau of Statistics (CBS). In this study we examine whether the information in the Employers Survey is leading information on which a forecast for employment can be based. For this purpose, we test a number of hypotheses. First, we show that the relative employment balance, defined as the difference between posts filled and employment terminations divided by the total number of employees in the previous quarter, provides a consistent and unbiased estimate of the actual rate of change in employee posts in the business sector published later by the CBS. Second, we test the hypothesis of rationality of employers' expectations.
On the basis of Employers Survey panel data for the period 1998 to 2008, we examine different methods of calculating a one-quarter-ahead leading index of employment. We compare the traditional aggregate method--based on the difference between the proportion of companies with optimistic and the proportion of those with pessimistic expectations--with the disaggregate model, which exploits companies' heterogeneity by using firm-level data. We estimate parameters of the disaggregate model by applying the discrete ordered choice model, which relates firm-level expectations to the official data, in order to quantify individual employers' expectations in terms of rates of change in employee posts in the next quarter. These estimates were obtained for only 20 percent of firms, but the forecasting ability of this model was better than that of the traditional method.
A leading index with the best performance combines two components: the relative employment balance derived from the Employers Survey, which is an estimate of the current employment index, and the expected change in employment in the next quarter derived from employers' expectations, using the disaggregate model. No evidence of selective bias (particularly, in the direction of large companies) was found in the sub-sample of companies filtered by the disaggregate model, providing additional support for the use of the leading index for short-term forecasting.
