When two time series are integrated but not causally related, conventional tests reject up to 80% under the null, at a 5% nominal level. This ‘nonsense regressions’ phenomenon is analysed, and detrending is shown not to solve the problem. Integrated variables that are connected are cointegrated. Since dynamics and cross‐variable interdependence interact, both sequential and conditional factorizations of data‐density functions are needed.
Keywords: cointegration; conditional factorization; detrending; integrated time series; nonsense regression; sequential factorization
Chapter. 16795 words. Illustrated.
Subjects: Econometrics and Mathematical Economics
Full text: subscription required