Why does Schweser say arithmetic mean, not geometric, is “statistically best estimator of the next year’s returns”?

From what I know, arithmetic mean is valid when events are independent from one another. But in the case of eg. stock returns, year 2 returns are related to year 1’s, hence dependent.

Given such setting + considering the compounding effect, I get that geometric mean is the right way to go about calculating the average returns for stock.

And so with this in mind, because I view stock returns as dependent events, I don’t know what to make of Schweser stating that arithmetic mean is the best estimator of the next year’s return. (I get the part where he says geometric mean is apt for calculating multi-year returns)

Does anyone know an appropriate example I could use to understand why? Appreciate your help in advance.