Japan's August output fell 1.7% but manufacturers plan to raise it 3.2%. Do you trust survey forecasts or hard data?

Japan's industrial production index fell 1.7% in August, METI reported on 30 September, with motor vehicles and machinery the biggest drags. In the same release, its survey of manufacturers pointed to production rising 3.2% in September and 3.1% in October. In the August survey, firms had expected September output to fall 4.2%, so the plan changed sharply in a month (full report).

That is a clear example of two kinds of data pointing in different directions. Hard data, such as the production index itself, count what happened. Survey data count what firms or households say, think or intend. Surveys come earlier and can turn before hard data do, but they are less reliable, and plans are often revised.

We have the same split elsewhere this month. The ISM manufacturing index for September was 54.5 and its prices index 77.9 (report), while US factory data are mixed (durable goods). China's PMIs turned back above 50 (report).

How do you weigh them?

• When a survey and hard data disagree, which do you act on first, and which do you wait for?
• Do you track how well a particular survey has predicted the hard data over time?
• Do you treat forecast surveys like METI's as a plan, a hope or a data point?
• Which release gives you the earliest reliable read on manufacturing?

PMIs explained covers how survey indexes are built.

Please share your own approach and reasoning, not calls for others to copy. Posts that promise a direction or sell signals will be removed.