Nonparametric regression with parametric help (Q2209833)

For random variables \(X,Y\), consider the estimation of the conditional expectation function \(m(x) = E(Y|X=x)\), \(x \in \mathbb{R}\). With a twice continuously differentiable function \(g=g(x)\) having a positive expectation \(Eg{''}(X)\), the function \(m(\cdot)\) is represented by \(m(x)=\theta_0 g(x) + m_0(x)\) with a parameter \(\theta_0\) and a function \(m_0=m_0(x)\). This model, having a projection interpretation, is estimated by using profiling techniques, where in the first step the parameter \(\theta_0\) and in the second step the function \(m_0(\cdot)\) are estimated. Under certain assumptions, the properties of this method are evaluated, where an improved estimation performance (bias reduction) can be shown. Simulation results and comparisons with other, related estimators are presented.