This paper is concerned with the question in which way the allocation of regulatory capital to positions in a given market can reflect the extent to which this market can be reliably modeled. One may argue in general that a relatively greater amount of regulatory capital should be associated to positions in ``unstable'' markets. We provide a framework through which this idea can be given a quantitative form. Our approach is based on computing worst-case risk measures over sets of models that are in some appropriate sense close to a nominal model. The method is general in the sense that it can be applied with any of the usual risk measures such as Value-at-Risk and Tail Conditional Expectation. We present applications both to stock portfolios and to derivative products; we find that (nonparametric) misspecification risk is typically much more important than (parametric) estimation risk.