Type I Spurious Regression in Econometrics

Abstract

In applied econometrics researchers often infer the relation among nonstationary time series by regression of their differences. The aim of this paper is to show that in some circumstances regression of differenced time series tends to reject the relation among their levels. This phenomenon is known as type I spurious regression. Time series are dynamic processes, and the ignored system dynamics will become the systematic errors in regression equations. Differencing does not preserve the underlying relation among time series in regression due to systematic errors. This paper will outline how regression of differenced time series tends to reject the relation between their levels, and so potentially to incur type I spurious regression.

Private Higher Education and Economic Development

“…82.) Chiarella and Gao (2002, p. 1) sum up as follows: ‘To avoid spurious regression, a rule of the thumb is that for nonstationary time series regression of their differences is safe.’…”

Economic Segregation and Urban Growth

“…It is commonly believed that using the first difference of variables instead of the levels of variables in regressions could effectively remove the potential omitted variable bias, providing stronger evidence for the causal relationship of interest. However, a recent study (Chiarella and Gao, 2002) Table 3 runs a model of levels on the pooled sample using the same controls as those in regression 5 plus year dummies. All the estimated coefficients in regression 6 have around half the sizes compared to those in regression 5.…”

“…Therefore, discarding the connection between cross-sectional differences of urban population growth and urban characteristics in regression 7, based on (Chiarella and Gao, 2002), may generate Type I error. In this case, models of levels are more proper to use.…”

Continuous Time Model Estimation

“…When equation errors are not white noises, autocorrelation correction is called for. Such a procedure is the source of type I spurious regression in econometrics (see Chiarella and Gao, 2002a).…”