Frisch-Waugh Theorem for different regressions

90 Views Asked by At

I'm stuck on the question when the residual vectors of different regressions are numerically equal according to the above mentioned theorem.

Let \begin{equation} y=Xb+e=X_{1}b_1+X_{2}b_2+e \label{eq:1} \end{equation} where $b_1, b_2$ are the OLS estimates of the "true" $\beta_1, \beta_2$, respectively. By pre-multiplying \ref{eq:1} by $M_1=X_1(X'_1X_1)^{-1}X'_1$ we obtain

\begin{equation}M_{1}y=M_{1}X_{2}b_2+M_{1}e \leftrightarrow \tilde{y}=\tilde{X_2}b_2+e\end{equation} Thus, the residuals of the regression of $\tilde{y}$ on $\tilde{X_2}$ are the same.

Considering the model $\tilde{y}=X_{1}b_1+X_{2}b_2+residual$, is it possible to say that $residual$ must equal $e$, because premultiplying this model by $M_1$ one obtains $\tilde{y}=\tilde{X_2}b_2+M_1*residual$, where $M_1*residual$ must be $e$ because we are regressing $\tilde{y}$ on $\tilde{X_2}$?

Pretty sure I'm making a mistake here. If this was true, then by the same reasoning, one would obtain that the residuals from $\tilde{y}$ on $X_2$ were also $e$, which is wrong I guess.

Thanks in advance!