In the Bank of Japan’s September 2026 Tankan survey, firms’ assessment of general business conditions improved from the previous survey. The all-enterprise, all-industry diffusion index rose from 18 in June to 21 in September. This does not mean production or sales grew by 3%. The Tankan “DI” is calculated by subtracting the share of respondents calling conditions unfavorable from the share calling them favorable; it is a balance measured in percentage points. Responses of “not so favorable” are not added to either side of that difference. September’s 21 therefore describes the net balance between favorable and unfavorable judgments by survey respondents about current conditions.

The September DI was above the June actual reading across all three enterprise-size groups. The all-industry balance rose from 29 to 30 among large firms, from 22 to 24 among medium-sized firms, and from 13 to 15 among small firms. Across enterprises of all sizes, manufacturing improved from 14 to 19, while nonmanufacturing edged down from 21 to 20. This variation means the overall 21-point result should not be read as the same movement occurring in every part of the economy. For example, the DI for large manufacturers rose from 22 to 24, while that for large nonmanufacturers declined from 37 to 35. The index tracks a sentiment balance derived from three response options; it is not a realised growth rate for output, turnover, profits or employment.

The survey also asked respondents to judge conditions three months ahead. The September forecast DI for all enterprises and all industries was 15, below the current-conditions reading of 21. Large firms reported 25, medium-sized firms 18 and small firms 11. These are not forecasts made by the BOJ or by Haber Paper; they are the balance of participating firms’ expectations. The September actual-result column records respondents’ assessment at the time of the survey, while the forecast column records their answers about conditions three months later. The table does not show whether those expectations were subsequently realised. The 15-point balance should therefore remain clearly labelled as a respondent outlook, with its forecast horizon stated, rather than presented as a settled outcome.

Responses were collected from 26 August to 30 September. The main all-enterprise sample contained 9,104 firms, and the BOJ reported a 99.4% response rate for this sample. Large enterprises accounted for 1,631 sampled firms and a 99.0% response rate; medium-sized firms, 2,589 and 99.1%; and small firms, 4,884 and 99.6%. The Tankan is not a census of every business in Japan but a stratified sample survey designed to meet statistical criteria. Its main target population is private enterprises in Japan with capital of at least 20 million yen. Financial institutions, business consultants and pure holding companies are covered by separate supplementary surveys, and some industries are excluded. Size groups are based on capital: large firms have at least 1 billion yen; medium-sized firms have 100 million yen to less than 1 billion; and small firms have 20 million yen to less than 100 million.

The BOJ says sample enterprises are generally revised every two or three years and statistical accuracy is regularly reviewed, in principle once a year. A high response rate does not turn the survey into a census or remove the role of nonresponse; the results remain firms’ own judgments. The bank also notes that DI shares are rounded to whole numbers, so changes in industry totals and subgroups may not reconcile exactly. Some long-run series have breaks following earlier methodological changes; users comparing older observations need to account for distinctions such as data before and after 2004. When the BOJ finds an error, it may release corrected data, but later revisions by responding firms do not generally cause the previously released figure to be corrected. The September summary PDF was released on 1 October, with the comprehensive data set following on 2 October.