[LudwigNagasena]

> Because you've assumed a linear relationship with normally distributed errors.

Normality will give you stronger results, but generally linear regression doesn’t require normality. You only need errors to be uncorrelated for OLS to be the best linear unbiased estimator.