Revision of the Sample Design of the Tankan Using the Economic Census of Japan

In this paper we discuss how to achieve a more efficient sample design of the Tankan (Short-Term Economic Survey of Enterprises in Japan), making use of the Economic Census of Japan. The Bank of Japan currently forms strata for population estimation using three bases -- industry, capital size, and the number of employees -- and extracts sample enterprises from each stratum. Employing the Economic Census newly available as population information allows the Bank to have direct access to population data regarding quantitative survey items of the Tankan such as sales. This helps to measure the statistical accuracy of the Tankan more precisely as well as to improve its accuracy through revisions of the sample design, such as stratification using different bases. When measuring the statistical accuracy of the Tankan based on the Economic Census data, the standard error ratios of total sales are generally low, although some industries show somewhat high ratios. By contrast, those of fixed investment are higher than those of sales, particularly for small enterprises. Moreover, while the Tankan estimates of sales and fixed investment are somewhat larger in terms of levels than the aggregated population values, their rates of change exhibit only a small bias, which means no difficulties for users. Based on this quantitative evaluation, we propose revisions to our sampling methods, considering a balance between an improvement in the statistical accuracy and a reduction in the total burden on respondents. On this basis, changing a basis for stratification from the number of employees to sales and eliminating a fraction of sample enterprises lead to further improvements in the statistical accuracy of the Tankan. The Bank will finalize details of the new sampling methodology based on the results of this paper, and implement them in the upcoming revision of sample enterprises around 2018.