I have the true model set as $y = b_0+b_1 x+u$. Now supposing that I'm running the model without the intercept term,

Under what circumstances would the coefficient term in the model (without the intercept term) be an unbiased estimator?

Personally, I like this model choice because the conditional variance of my estimator is the most efficient (between the two model choices). Would this be a correct evaluation on my part?