We propose exact tests and confidence sets for various structural models typically estimated by IV methods, such as models with unobserved regressors, which remain valid despite the presence of identification problems or weak instruments. Two approaches are considered: (1) an instrument substitution method, which generalizes the Anderson-Rubin procedure, and (2) a samplesplit method, that allows the use of "generated regressors." Projection techniques are also proposed for inference on general parameter transformations. The asymptotic theory of the tests under weaker assumptions is discussed. Simulation results are presented. The suggested techniques are applied to a model of Tobin's q and to a model of academic performance.
International audienceThis paper presents a new general class of compound autoregressive (Car) models for non-Gaussian time series. The distinctive feature of the class is that Car models are specified by means of the conditional Laplace transforms. This approach allows for simple derivation of the ergodicity conditions and ensures the existence of forecasting distributions in closed form, at any horizon. The last property is of particular interest for applications to finance and economics that investigate the term structure of variables and/or of their nonlinear transforms. The Car class includes a number of time-series models that already exist in the literature, as well as new models introduced in this paper. Their applications are illustrated by examples of portfolio management, term structure and extreme risk analysis
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.