An Alternative Estimation to Spurious Regression Model

In sturdy econometrics specification search problems of unit roots and multicollinearity are well documented since the inception of regression analysis. In examining the likely consequences of nonsense relationship Granger and Newbold (1974) make it clear that first differencing is not the universal sure fire solution to problem of spurious regression models. This has prompted the discovery of cointegration regression estimation by Engle and Granger (1987). In recent years applied econometricians are debating with the problem of spurious regression model when the co movements between the variables are different. If the variables of the model are not cointegrated, there is a question whether the background economic or financial theory is plausible with the data that we are analyzing. This paper reviews the debate and proposes an alternative solution to the problem. Our approach uses a suitable data transformation of the variables of the model based on Hendry (1995) and Phillips (1998) approaches to reduce the spurious correlation, stochastic means and variances in standard level. In a non cointegrated USA information processing investment model, we apply our technique and found a meaningful solution.