The "spurious regression problem" in the classical regression model framework

I analyse the "spurious regression problem" from the Classical Regression Model (CRM) point of view. Simulations show that the autocorrelation corrections suggested by the CRM, e.g., feasible generalised least squares, solve the problem. Estimators are unbiased, consistent, efficient and deliver correctly sized tests. Conversely, first differencing the data results in inefficiencies when the autoregressive parameter in the error process is less than one. I offer practical recommendations for handling cases suspected to be in the "spurious regression" class.