Abstract:This paper is concerned with estimation and inference for ultrahigh dimensional partially linear single-index models. The presence of high dimensional nuisance parameter and nuisance unknown function makes the estimation and inference problem very challenging. In this paper, we first propose a profile partial penalized least squares estimator and establish the sparsity, consistency and asymptotic representation of the proposed estimator in ultrahigh dimensional setting. We then propose an $F$-type test statistic for parameters of primary interest and show that the limiting null distribution of the test statistic is $\chi^2$ distribution, and the test statistic can detect local alternatives, which converge to the null hypothesis at the root-$n$ rate. We further propose a new test for the specification testing problem of the nonparametric function. The test statistic is shown to be asymptotically normal. Simulation studies are conducted to examine the finite sample performance of the proposed estimators and tests. A real data example is used to illustrate the proposed procedures.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the challenge of estimation and inference in the ultra - high - dimensional partially linear single - index model. Specifically, the paper focuses on how to effectively estimate the model parameters and conduct statistical inference on these parameters in the presence of high - dimensional nuisance parameters and unknown functions. The main research contents include:
1. **Estimation method**: A profile partial penalized least squares estimator is proposed, and the sparsity, consistency and asymptotic representation of this estimator in the ultra - high - dimensional setting are established.
2. **Hypothesis testing**:
- For the hypothesis testing of the parameter part \(\beta\), an F - type test statistic is proposed, and it is proved that its limiting distribution under the null hypothesis is the \(\chi^2\) distribution, and it can detect local alternative hypotheses that converge to the null hypothesis at the root - \(n\) rate.
- For the hypothesis testing of the non - parametric function \(\eta(\cdot)\), a new test method is proposed, and it is proved that this test statistic is asymptotically normal under the null hypothesis.
3. **Numerical simulation**: The performance of the proposed estimator and test statistic in finite samples is verified through simulation studies.
### Specific problem description
#### Model setting
Consider the partially linear single - index model (PLSIM):
\[ Y=\eta(\alpha^{\top}X)+\beta^{\top}Z +\epsilon, \]
where:
- \(Y\) is the response variable,
- \(X\) and \(Z\) are covariates of dimension \(p\) and \(q\) respectively,
- \(\alpha\) and \(\beta\) are unknown parameters,
- \(\eta(\cdot)\) is an unknown smooth function,
- \(\epsilon\) is the error term, satisfying \(E(\epsilon|X,Z) = 0\) and \(E(\epsilon^{2}|X,Z)=\sigma^{2}\).
For model identification, assume \(\|\alpha\|_{2}=1\) and the first element of \(\alpha\) is positive.
#### Main challenges
- The existence of the high - dimensional nuisance parameter \(\alpha\) makes the estimation and inference problems very challenging.
- In the ultra - high - dimensional setting (i.e., \(p\) can be exponential in the sample size \(n\)), the sparsity problem of high - dimensional data needs to be dealt with.
#### Solutions
- **Estimation method**: The profile partial penalized least squares method is adopted, only penalizing the nuisance parameter \(\alpha\), not the parameter of interest \(\beta\).
- **Hypothesis testing**:
- For the hypothesis testing of the parameter \(\beta\), \(H_{0}:\beta = 0\) vs \(H_{1}:\beta\neq 0\), an F - type test statistic is proposed.
- For the hypothesis testing of the non - parametric function \(\eta(\cdot)\), \(H_{0}:\eta(t)=g(t,\zeta)\) vs \(H_{1}:\eta(t)\neq g(t,\zeta)\), a test statistic based on the kernel function is proposed.
### Theoretical contributions
- The asymptotic properties of the partially penalized least squares estimator in the ultra - high - dimensional setting are established.
- Effective hypothesis testing methods are proposed and their limiting distributions under the null hypothesis are proved.
### Numerical simulation
The performance of the proposed method in finite samples, including estimation accuracy and test power, is verified through simulation studies.
In conclusion, this paper makes important theoretical and methodological contributions to the estimation and inference in the ultra - high - dimensional partially linear single - index model.