Evaluating Firm-Level Expected-Return Proxies: Implications for Estitating Treatment Effects