“…Errors-in-variables can be interpreted as missing variables : if the noise were observable, it could be included as an additional variable, and different conclusions can emerge. As previously observed by several authors (see Hsiao, 1982;Lütktepohl, 1982;Dufour and Renault, 1998;Triacca, 1998Triacca, , 2000, causality properties in the sense of Wiener-Granger depend crucially on the information set considered, which can affect both the sheer presence of causality (or non-causality) and causality measures (Geweke, 1982;Dufour and Taamouti, 2010;Dufour et al, 2012). Of course, the central difficulty remains that noise is typically unobserved.…”