In this paper an attempt has been made to analyze the Bank of Japan's business survey data with a view to learn something about how expectations on macro variables are formed and to estimate realization functions for a subset of the macro variables. We have found that there are pieces of evidence that expectation formations are affected by factors other than the past values of the own series, and that realization functions which include several variables other than the own expectation perform fairly well. An implication of these results will be that it is not appropriate to estimate simultaneously a behavior function and a simple expectation function by substituting the unobservable expectation variable into the behavior function, as is sometimes done in practice.estimate simultaneously a behavior function and a simple expectation function by substituting the unobservable expectation variable into the behavior function, as is sometimes done in practice.