摘要：In this paper a potential problem with tests for Granger-causality is investigated. If one of the two variables under study, but not the other, is measured with error the consequence is that tests of forecastablity of the variable without measurement error by the variable with measurement error will be rejected less often than it should. Since this is not the case for the test of forecastability of the variable with measurement error by the one without there is a danger of concluding that one variable leads the other while it is in fact a feed-back relationship. The problem is illustrated by an example.
