Testing for Curvature in Economic Applications

University Of Chicago, Chicago IL

Investigators

Abstract

This project will focus on statistical testing of a function's curvature in common econometric models. The research will provide practitioners with a simple method for testing linearity, concavity, and/or convexity in situations where the shape of the relationship between variables may have important consequences for the empirical results. The test statistics, which are based on a simple idea related to pair-wise slopes in the data, do not require parametric specification (or estimation) of the function of interest. Unlike many nonparametric methods, the test statistics do not require the choice of bandwidths or other smoothing parameters. The tests can be applied in a variety of econometric models of interest, including the nonparametric regression model, the semilinear regression model, the single-index model, and fixed-effects models. Throughout the course of the project, computer routines will be clearly written and documented so that they may be made publicly available to researchers.