Why do investment Euler equations fail?

This article isolates sources of misspecification in neoclassical investment Euler equations without ad hoc alterations of the basic model. First, allowing for nonlinear marginal investment adjustment costs improves model performance slightly. Some further improvement results from isolating firms whose optimality conditions hold even in the presence of fixed costs of adjustment or costly reversibility. Finally, I identify which instruments contribute to model failure via standard GMM-based tests and also via the empirical likelihood estimator of Imbens, which allows testing overidentifying restrictions individually. Both methods show that financial instruments contribute to rejection of the overidentifying restrictions for all firms; however, only the empirical likelihood estimator shows that they are a source of failure for firms that attain an interior optimum.