Given a very simple linear projection model $Y=X\beta+e$ with $E(e|X)=0$ and $X$ a scalar. Notice that this is a simple linear model with no intercept.

Then $\beta = E(XY)/E(X^2)$ from the least squares formula.

On the other hand, if I take expectation on both sides of the original model, I get $E(Y)=\beta E(X)+E(e)=\beta E(X)$. Therefore, $\beta=E(Y)/E(X)$. (Of course, suppose $E(X)\neq 0$.)

I just don't see how the two quantities could be identical. Did I do something wrong?

  • $\begingroup$ If the model were to assume a non-zero intercept, $\alpha$, then the OLS regression results in a estimate for $\alpha$ given by $$\hat \alpha =E(Y)-\hat \beta E(X)$$From this, note that a zero value of $\hat \alpha$ would correspond with an estimate for $\beta$ given by $$\hat \beta =E(Y)/E(X)$$ So, this is consistent with your result. $\endgroup$
    – Mark Viola
    Feb 10 at 15:46

$E[f(X)Y]=E[f(X)X]\beta+E[f(X)e]=E[f(X)X]\beta$, so $\beta=\frac{E[f(X)Y]}{E[f(X)X]}=\frac{E[Y]}{E[X]}=E[\frac{Y}{X}]=E[\frac{f(X)}{E[f(X)]}\frac{Y}{X}]$ ...

Strange ...

  • $\begingroup$ Sorry I don’t follow. Is this an answer? $\endgroup$
    – Bunbury
    Feb 10 at 15:27
  • $\begingroup$ I think so, but I'm not sure how to interpret it. $\endgroup$
    – Ian Zhang
    Feb 10 at 15:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.