Estimation for single-index models via martingale difference divergence

In this paper, we focus on the estimation of the index coefficients in single-index models and develop a new procedure based on martingale difference divergence. Since the proposed procedure can capture automatically the conditional mean dependence of the response variable on the covariates, it does not involve smoothing techniques or require the commonly used assumptions in the literature of single-index models, such as smooth link functions and at least one continuous covariate. Under some mild conditions, we establish the asymptotic normality of the estimators. We assess the finite sample performance of the proposed procedure by Monte Carlo simulation studies. We further illustrate the proposed method through empirical analyses of a real dataset.