Abstract
    In the analysis of economic times series it was recognized a long time ago that they are not stationary, but that their differences (increments) could often be described by stationary processes. Such processes are called difference stationary.
    Cointegration is the phenomenon that two such non-stationary time series can have the property that a linear combination is stationary. Such linear combinations (cointegrating relations) are found in economics to describe long-run relations and the usual regression analysis is replaced by cointegration analysis in order to estimate and test hypotheses on the coefficients. We discuss cointegration of multivariate autoregressive time series and indicate how the usual statistical theory for inference has to be modified to take into account the non-stationarity.
    We illustrate with the two time series of global averages of surface temperature and sea level, both of which have risen through the past century.
    A few theoretical exercises have been collected in a special file. They can be discussed if time permits.