Rational expectations modeling with seasonally adjusted data

In a world where time series show clear seasonal fluctuations, rational agents will take account of those fluctuations in planning their own behavior. Using seasonally adjusted data to model behavior of such agents throws away information and introduces possibly severe bias. Nonetheless it may be true fairly often that rational expectations modeling with seasonally adjusted data, treating the adjusted data as if it were actual data, gives approximately correct results; and naive extensions of standard modeling techniques to seasonally unadjusted data may give worse results than naive use of adjusted data. This paper justifies these claims with examples and detailed arguments.