I am wondering why in linear regression, if you have a categorical variable like months of the year, and you represent it by dummy variables, such as: $X_1 = 1$ if it is January, $X_2 = 2$ if it is February, $\ldots$, $X_{12}=1$ if it is Decemeber, why dropping one variable for the sake of collinearity doesn't make you lose any information?
How does the linear regression model "know" that you are dropping one variable?