Generalized Correlations and Instantaneous Causality for Data Pairs Benchmark

Abstract

Usual correlations assume linearity. If new generalized correlations satisfy r*(Y |X) > r*(X|Y ), X better predicts Y than vice versa. Then we say that X "causes" Y . Thus, Vinod (2013) revives Granger's instantaneous causality concept. Mooij et al. (2014) and their references seem unaware of causality in econometrics. We propose and illustrate new generalized correlations using benchmark data set Cause Effect Pairs (CEP) that consists of 88 different binary "cause effect pairs" from 31 real world data sources, where the cause is presumed known. Our ability to successfully identify the cause in some 75% of cases means researchers can benefit from using our fairly simple data- analytic R software tools as a first step to save time and expense.

Keywords: Generalized measure of correlation, non-parametric regression, R software

JEL Classification: C10, C20, A20