Changes in seasonal patterns. Are they cyclical? (Q1342433)

In economics seasonal components in time series are mostly modelled by the use of dummy variables or by using seasonally adjusted data. This common practice is criticized by the authors on two grounds: First, modern dynamic business cycle theories explicitly link seasonal and business cycle by cross-frequency restrictions to obtain information about the interaction of both components. Second, several researchers noted that the seasonal patterns of many macroeconomic variables appear to evolve over time. Therefore, there is a need to model seasonals stochastically. But prior to modelling one has to check whether the seasonal pattern of macroeconomic time series vary with expansions and contractions. To proceed along this line the authors suggest several graphical techniques and generalized predictive tests for structural stability to identify and test patterns of changing seasonality. Two types of predictive tests are suggested. One examines individual values of a residual based test statistics for evidence of structral instability and the other examines several values of the test statistics. The whole sample is splitted into two parts and a least squares regression with seasonal dummies and a linear trend variable as well as lagged endogenous variables is estimated. The residuals of the first subset regression are used to construct the test statistic which then serves as a guideline for the second subset residuals. For US macroeconomic time series the authors find that most seasonal patterns are unstable and that they are linked to the stages of the business cycles. Furthermore, the cost of treating seasonality as constant is evaluated by a forecasting experiment.