In this article
After this article, you can
- Match a Tankan forecast to the period subsequently assessed.
- Separate the prior forecast error from the expected change for next quarter.
Terms at hand: Tankan Survey (BoJ Business Sentiment) · Economic Calendar · Balance · Central banks · Monetary Policy · Inflation
Tankan is the Bank of Japan’s enterprise survey. To read it, keep three questions separate: what firms report now, what they expect for a later period, and how the published result compares with an external forecast recorded beforehand. Our fictional examples show why improvement, a negative DI and disappointment against expectations can occur together.
What the Tankan actually is
The Tankan is the Bank of Japan's Short-Term Economic Survey of Enterprises in Japan. Surveys are conducted in March, June, September and December, with results normally released in early April, July and October and in mid-December. Confirm actual release dates in the BoJ schedule. Distinguish the survey period from its publication month.
Business Conditions DI asks firms to choose favourable, not so favourable or unfavourable. The middle answer does not mean that conditions are unchanged. The report also contains annual figures and plans, and firms’ inflation expectations; they are different measures. Record the complete row label, enterprise group, reference period, unit and data vintage rather than writing only “Tankan was 10”. Nonmanufacturing covers more than services.
How to read the diffusion index
For Business Conditions DI, subtract the percentage of unfavourable responses from the percentage of favourable responses. The result is in percentage points, not production growth. BoJ simply aggregates these responses rather than weighting the DI by enterprise size; its published percentage shares are rounded. The examples below use shares that need no rounding.
| Responses / measure | A | B |
|---|---|---|
| Favourable | 40 = 20% | 50 = 25% |
| Middle answer | 80 = 40% | 90 = 45% |
| Unfavourable | 80 = 40% | 60 = 30% |
| Total | 200 = 100% | 200 = 100% |
| DI | 20 − 40 = −20 pp | 25 − 30 = −5 pp |
B improves by 15 points, yet more firms still answer unfavourable than favourable. This is not 15% output growth or a favourable majority. Separately, both 20% / 60% / 20% and 40% / 20% / 40% give DI zero. The same balance does not recover the distribution of extreme answers. These invented counts are not historical Tankan results.
Business Conditions DI is reported separately by enterprise size and industry. Large-manufacturer DI is widely followed, but there is no single undifferentiated small-firm index. Specify the exact series, population and data vintage when comparing manufacturers, non-manufacturers and investment plans.
A new forecast does not assess last quarter's forecast
Tankan reports current conditions and expected conditions three months ahead separately. Hypothetical Q1 report forecast DI = 12 for Q2. The Q2 outcome was 8; its new Q3 forecast was 14. These are two comparisons, not one continuous forecast-versus-forecast line.
Scroll sideways to see the full diagram
For every forecast, record both the publication and target quarters. The difference 14 - 12 = 2 compares forecasts for different periods. It does not replace Q2's error of -4 or represent a realized improvement in Q3.
Assumptions: 12, 8 and 14 are fictional readings of the same Business Conditions DI series in percentage points for the same enterprise group. Sample composition and data vintage are unchanged. Error means outcome minus forecast for the same period. Method source: BoJ: explanation of the Tankan survey.
Separate three reference values
The Forecast column for business conditions describes surveyed firms’ expectations three months ahead. It is separate from the prior current result, an analyst consensus and a BoJ policy forecast. If an external consensus cannot be recovered from before the release, record that comparison as unavailable; do not substitute the firms’ Forecast column.
| Reference | Value | Difference / meaning |
|---|---|---|
| Prior current result | −20 | −5 − (−20) = +15 pp: improvement |
| External consensus recorded before release | −2 | −5 − (−2) = −3 pp: below expectation |
| New firms’ forecast for next period | +1 | 1 − (−5) = +6 pp: expected improvement |
All three descriptions are consistent in this model. None establishes a buy/sell instruction for JPY. Preserve the forecast source and timestamp; a table reconstructed later can include information unavailable at the event.
Investment plans are intended spending, not completed expenditure. Inflation expectations are firms’ views, not guaranteed future inflation. For annual plans, distinguish year-on-year growth from revision of the same plan. In our arithmetic example, 110 against a prior-year base of 100 gives +10%; a later plan of 108 against a revised base of 90 gives +20%. The plan falls by 2 units while its growth rate rises. Record both bases and their publication vintages; do not add different Tankan measures into an undefined overall score.
When and how the Tankan reaches the market
The BoJ FAQ specifies release at 08:50 Japan Standard Time (UTC+9). Convert the actual date to your zone, accounting for any local daylight-saving change. Summary and Outline are released together; the current index shows the September 2026 survey on October 1 and the comprehensive set on October 2. Use the official schedule for your chosen release and record the survey period separately from the publication date.
Plan your record before publication: identify the series, reference period, earlier same-period firms’ forecast and independently sourced external consensus. Keep the prior data vintage too. The same observation can improve over the prior period yet be below expectations, as the table above shows. The guide on using an economic calendar step by step explains how to organize these releases.
Separate the survey from a currency-price observation
A hypothesis about policy expectations is different from an observed currency response. Tankan aggregates enterprise answers; the central bank’s policy decision and the exchange rate are separate outcomes. A high DI does not fix the next interest rate or guarantee JPY strengthening. To investigate a reaction, define the pair, price source and observation window in advance; preserve quiet events and moves against the hypothesis.
Do not attribute a JPY move to Tankan merely because the timestamps are close. Other currencies and concurrent events also matter. Record the start/end quote and, if modelling a transaction, its execution prices and costs separately from the survey comparison. The guide to watching the Fed, the ECB and the Bank of Japan as one system and our fundamental analysis section provide related context.
Could sample composition explain the change?
Check who is represented in the compared samples. BoJ describes regular sample revisions, normally every two or three years; its March 2026 notice links pre/post-revision figures for the same December 2025 survey. Our extreme model below is not that procedure: every A firm always answers favourable and every B firm always answers unfavourable, with no middle answers or changed opinions.
| Fictional sample | First | Second |
|---|---|---|
| A: always favourable | 10 | 90 |
| B: always unfavourable | 90 | 10 |
| Total | 100 | 100 |
| DI | −80 pp | +80 pp |
DI changes by 160 points solely through the mixture. With the first proportions held constant, it remains −80. Saying that these firms improved their assessments would be false. This demonstrates a possibility, not the size of a real Tankan sample effect. Read BoJ’s comparability notes instead of subtracting an invented composition correction; a real DI change is not automatically an artefact.
What now
- Record exact release times. Use the official schedule for the chosen date, convert 08:50 JST into your zone and keep the reference quarter. A month-only reminder does not identify the event.
- Label each selected row. Write “Business Conditions DI, large manufacturers, current”, for example. Record size and industry separately for other rows, and preserve units/year for investment plans. An unspecified “small-firm index” cannot be matched reliably.
- Compare matched series and periods. Record the enterprise group, current DI, prior DI and any external forecast source separately. Use the earlier same-period forecast when assessing forecast error. These differences can inform expectations but do not alone decide the strength or direction of a yen reaction.
Check your understanding
One short exercise
About 2 min · you can try again
Q1 forecast DI = 10 for Q2. The Q2 outcome was 7 and the new Q3 forecast was 9. What is Q2 forecast error, calculated as outcome minus forecast?
- -1 pp, because 9 - 10 = -1.
- -3 pp, because 7 - 10 = -3.
- +2 pp, because 9 - 7 = 2.
Check the answer and explanation
-3 pp, because 7 - 10 = -3.
Correct. Both figures refer to Q2. The new 9 is a Q3 expectation and does not change Q2 forecast error.
Frequently asked
Can a negative Tankan DI improve?
For Business Conditions DI, subtract unfavourable from favourable response shares. The middle option is “Not so favorable”; it does not mean unchanged conditions. In our 200-answer model, 40 favourable and 80 unfavourable give −20 pp; 50 and 60 give −5 pp. The improvement is 15 points despite a still-negative balance, not 15% production growth. A zero balance can also conceal different extreme-response shares.
How do I read the Tankan diffusion index?
Keep the complete series label, enterprise group, period and data vintage. In our example, current −5 versus prior −20 gives +15 pp; versus external consensus −2 it gives −3 pp. A new firms’ forecast of +1 is +6 pp above current conditions. Those differences answer different questions. The Forecast column is firms’ three-month expectation, not the analyst consensus; without a pre-release external forecast, mark surprise unavailable.
Can higher plan growth mean a smaller plan?
Because it reports enterprise activities and expectations, not because one figure dictates policy. Plans are not completed spending. In our invented annual-plan test, 110 against 100 gives +10%, while a later 108 against a revised 90 gives +20%. A higher growth rate can accompany a smaller plan. Keep numerator, denominator and publication vintage before comparing revisions or relating them to other economic data.
How does the Tankan affect the yen and the dollar against the yen?
The survey does not establish a fixed currency direction. If testing a hypothesis, specify the pair, price source, observation window and concurrent events before comparing actual responses. Retain quiet and opposite-direction cases. Separately check sample comparability: our extreme 10/90→90/10 respondent model changes DI from −80 to +80 with no changed opinion. It is a mathematical counterexample, not an estimate of BoJ’s revision effect.
Sources & bibliography
-
Bank of Japan Tankan (Short-Term Economic Survey of Enterprises in Japan) · Official Tankan survey data, methodology, diffusion indicators and publication schedules for Japanese enterprises. www.boj.or.jp ↗
-
Bank of Japan Outline of the Tankan · Explains the Tankan survey's structure, sector coverage and questions about investment plans and firms' inflation expectations. www.boj.or.jp ↗
-
Bank for International Settlements Triennial Central Bank Survey of Foreign Exchange Markets · The 2022 survey reports currency-market turnover, including USD/JPY and the yen's global role. www.bis.org ↗
-
Bank of Japan Tankan FAQ — response shares, forecasts and annual-plan revisions · Definitions separate current conditions, firms’ expectations, response shares and the reference bases for annual plans. The article’s worked counts are fictional. www.boj.or.jp ↗
-
Bank of Japan Regular revision of Tankan sample enterprises — March 6, 2026 outline · Describes periodic respondent updates and provides a same-period pre/post comparison. The extreme respondent-mix model does not estimate the actual revision effect. www.boj.or.jp ↗
