Given two independent but integrated time series, classical linear regression models overestimate the true value of R^2 and tend to produce statistically significant regression coefficients. This could lead the researcher to assume he found evidence of a genuine relationship. This phenomenon, called spurious regression, still persists in modern academic work, although first words of caution in relating nonstationary processes were expressed more than a century ago. Early research, starting from the beginning of the twentieth century, will be revisited. The first analytic proof that spurious results can occur in classical OLS regression will be provided, revealing how the underlying distributional problem can be approached. Various extensions of this early work on spurious regressions will be discussed, such as the increasing scope of processes being investigated, as well as unit root tests and error correction models. Established methods to avoid spurious results, like differencing and detrending, will be reviewed. The theory of cointegration will be introduced and statistical procedures to test for cointegration will be studied.     «

Given two independent but integrated time series, classical linear regression models overestimate the true value of R^2 and tend to produce statistically significant regression coefficients. This could lead the researcher to assume he found evidence of a genuine relationship. This phenomenon, called spurious regression, still persists in modern academic work, although first words of caution in relating nonstationary processes were expressed more than a century ago. Early research, starting from th...     »