This article considers the model Y = M ( X , U ) where U is an unobservable continuously distributed scalar and M is monotonic with respect to U. It is assumed there is an observable scalar W satisfying the restriction T F W | X , U = T F W | U almost surely where T is a known functional and F A|B denotes the distribution of A|B. This article shows that M can be identified under a mild monotonicity condition. This result requires neither statistical independence between X and U nor X to be continuously distributed. The estimation problem is treated when T F A | B ≡ E [ A | B ] . The proposed pointwise estimator of M is asymptotically normally distributed under weak technical conditions. Furthermore, the rate of convergence in probability is equal to n − r / ( 2 r + d + 1 ) where d denotes the dimension of the continuously distributed components of X and r is a positive integer which relates to the smoothness of certain functions. A Monte Carlo experiment is conducted and reveals the benefits of the estimator in the presence of endogeneity. We apply our estimator to estimate the returns to schooling.
Read full abstract