The article discusses a problem with using cointegration techniques in regression models with nonstationary variables. If important factors are left out of the error term, the variables in the model won't be cointegrated. The researchers suggest using time-varying coefficient modeling as a solution to this issue.