The rational expectations-permanent income hypothesis (REPIH) may imply the cointegration relationship between nondurable consumption including services and income with the 'time-varying' cointegrating variable rather than the 'constant' one. Previous empirical rejection of REPIH seems to be the result of this unreal constancy assumption.
Estimating the time-varying parameter(TVP) cointegrating variable by Kalman Filtering is not meaningful because it gives 'the perfect equilibrium, solution' or 'the unstable equilibrium solution'. Hence, as an alternative, we suggest "ARCH-type heteroskedastic cointegration" to estimate the expected value of cointegrating variable instead of the recursive value.
Although empirical test of TVP model using the Family Income \& Expenditure Survey data in Korea does not give significantly different results from the constancy-assumption model, the degree of rejection of REPIH is weakened in the TVP mode. Our result may be interpreted as the weak evidence of TVP model in REPIH.