B0256
Title: Uniform inference for conditional deconvolution estimation
Authors: Stefan Hubner - University of Bristol (United Kingdom) [presenting]
Abstract: A uniform inference method is developed for conditional deconvolution estimators. For this, we rewrite Kotlarskis identity as a system of linear conditional moment inequalities by approximating the space of conditional distributions of the mismeasured function of interest by a multi-dimensional separable Hermite sieve basis. Separability has the advantage that partial application of the Fourier transform again forms a separable, orthonormal basis of the resulting space of conditional characteristic functions, which are implicitly defined by the conditional moments. By appropriately choosing instrument functions, we can then rewrite the conditional moments as unconditional ones without losing identifying power. Based on the latter, using generalised moment selection, we obtain uniform confidence bands for the mismeasured function of interest.