According to the Gauss-Markov Theorem, OLS estimators are BLUE if:

The Gauss-Markov theorem states that when the assumptions of the classical linear regression model are satisfied, the OLS estimators are the best linear unbiased estimators. The main assumptions are: (1) the model is linear in parameters, (2) the regressors are exogenous, E(u | X) = 0, (3) there is no perfect multicollinearity, (4) the errors are homoscedastic, Var(u) = σ², and (5) there is no autocorrelation, Cov(u_i, u_j) = 0 for i ≠ j. Normality of the errors is not required for the BLUE property. It is needed only for exact t and F tests in small samples. Option (A) refers to varying variances, which is heteroscedasticity, a violation of assumption (4). In option (C), perfect collinearity makes the OLS estimates impossible to compute. Option (D), conditional heteroscedasticity (ARCH), also violates assumption (4). In option (E), a binary dependent variable in a linear model produces heteroscedastic errors. Hence the correct option is (B).