Cointegration in a macro-economic system

This paper represents an exploratory study that investigates the vector autoregressive properties of a time series system containing six main indicators of the austrian economy (gross domestic product gdp, private consumption, investment, gdp deflator, interest rate, wages). interest focuses on cointegrating structures in the system, i.e. on linear combinations of some (trending) variates that generate stationary series. johansen's (1987) procedure identifies three relations of this kind. it is also tried to give an economic interpretation to these relations. more extensively, a variety of sensitivity experiments are performed which make somehow original use of the technique of canonical correlations. contrary to the sensitivity experiments, a critical assessment of the identified structure by medium-term forecasting (until 1999) sheds considerable doubt on its correctness. at most one or two cointegrating relations may help to increase forecasting precision relative to traditional vector autoregression which do not use the idea of cointegration at all. on the other hand, using all three identified relations seems to impose "excess stationarity" on the system which is not replicated by actual data behavior.