The errors $\epsilon_i$ must be normally distributed.

$E[\epsilon|X] = 0$ (Exogeneity).

The number of predictors $k$ must be less than 5.

The variance of the error term $\sigma^2$ must be 1.