An exogenous impact function is defined as the derivative of a structural function with respect to an endogenous variable, other variables, including unobservable variables held fixed. Unobservable variables are fixed at specific quantiles of their marginal distributions. Exogenous impact functions reveal the impact of an exogenous shift in a variable perhaps determined endogenously in the data generating process. They provide information about the variation in exogenous impacts across quantiles of the distributions of the unobservable variables that appear in the structural model. This paper considers nonparametric identification of exogenous impact functions under quantile independence conditions. It is shown that, when valid instrumental variables are present, exogenous impact functions can be identified as functionals of conditional quantile functions that involve only observable random variables. This suggests parametric, semiparametric and nonparametric strategies for estimating exogenous impact functions.
Centre for microdata methods and practice