This paper investigates a dataset that provides information about assets held

by U.S. equity mutual funds, but are not U.S. equities (`unmapped holdings'). I

show the widespread presence of these assets and investigate how they are used

within mutual fund portfolios. I find that their effects are statistically significant

upon both portfolio risk and return. They can either hedge or complement mapped

asset returns. I show that predictability of mutual fund returns are reduced when

unmapped holdings returns are controlled. Since unmapped holdings returns are not

observable, I define an econometric technique that in chapter two that can control

for their effect. This technique uses an average return (an `endogenous benchmark')

to control for common but immeasurable or unobservable characteristics in a group

of funds. I find that an `endogenous benchmark' alone produces estimates nearly as

good as those using common risk factor regression models. By combining an endoge-

nous bechmark with other risk factors in regression models, I find that estimates

are improved.