Algebra of Integrated Time Series: Evidence from Unit Root Analysis

It is argued if x_t ~ I(1) and y_t ~ I(1), then running a regression x_t on y_t would produce spurious results because e _t would generally be I(1). However, there may exist a ‘b’ such that e _t  = x _t - by _t is I(0), then running a regression x _t on y _t would not produce spurious results. This special case of two integrated time series is known in the literature as cointegration. In this particular case, x _t and y _t are said to be cointegrated. In our review of the development of the concept of cointegration, we identified that the underlying reason for this special case to arise is the proposition that if x _t  ~ I(d _x ), y _t  ~ I(d _y ), then z _t  = bx _t  + cy _t  ~ I(max(d _x ,d _y )). In this research, we offer evidence against this proposition.