Why is $\operatorname{Cov}(X,U) = E[XU]=0$?

11k Views Asked by At

In my econometric course, when talking about zero conditional mean in the context of simple linear regression $E[U \mid X]=0$, it says $\operatorname{Cov}(X,U) = E[XU]=0$ on lecture slides.

Why is $\operatorname{Cov}(X,U) = E[XU]=0$?

X is the independent variable; U is the error term.

1

There are 1 best solutions below

3
On

As @SangchulLee commented, if you have zero conditional mean then you consequently have zero covariance.

If we may assume $\mathsf E(U\mid X)=0$, therefore:

$\begin{align}\mathsf {Cov}(X,U)&=\mathsf E((X-\mathsf E(X))(U-\mathsf E(U))) &&\textsf{by Definition} \\ &= \mathsf E(\,XU-X\,\mathsf E(U)-\mathsf E(X)\,U+\mathsf E(X)\,\mathsf E(U)\,) &&\textsf{by Distribution} \\ &= \mathsf E(XU)-\mathsf E(X)\,\mathsf E(U)-\mathsf E(X)\,\mathsf E(U)+\mathsf E(X)\,\mathsf E(U)&&\textsf{by Linearity of Expectation} \\ &=\mathsf E(XU)-\mathsf E(X)\,\mathsf E(U) &&\textsf{by Algebraic simplification} \\ &= \mathsf E(XU)-\mathsf E(X)\,\mathsf E(\mathsf E(U\mid X)) &&\textsf{by Law of Total Expectation} \\ &= \mathsf E(XU) - \mathsf E(X)\,\mathsf E(0) && \textsf{by the Assumption} \\ &= \mathsf E(XU) && \textsf{by Definition of Zero} \\[2ex] &= \mathsf E(\mathsf E(XU\mid X)) &&\textsf{by Law of Total Expectation} \\ &= \mathsf E(X\,\mathsf E(U\mid X)) &&\textsf{by Linearity of Expectation} \\ &= \mathsf E(X\cdot 0) &&\textsf{by the Assumption} \\ &= 0 &&\textsf{by Definition of Zero} \end{align}$