In a multiple regression model $Y = \beta_0 + \beta_1 X_1 + \beta_2 X_2 + \epsilon$, if $X_1$ and $X_2$ are perfectly collinear, what happens to the OLS estimator $\hat{\beta}$?