Abstract
Summary: It has been argued that whenever regression models involve nonstationary and trending variables, estimation methods appropriate to stationary series cannot be applied to such models and instead require cointegration techniques. Unfortunately, extant methodologies applied to cointegration are a trap: if the error term of a cointegration regression consists of omitted relevant regressors, then, even though they are integrated to the same order, the dependent and the independent variables of the regression are not cointegrated! This paper proposes a remedy based on time-varying coefficient (TVC) modeling that overcomes all the shortcomings described in the paper.