Abstract
In this paper, a robust, efficient and easily implemented estimation procedure for single-index varying-coefficient models is proposed by combining minimum average variance estimation (MAVE) with exponential squared loss. The merit of the proposed method is robust against outliers or heavy-tailed error distributions while asymptotically efficient as the original MAVE under the normal error case. A practical minorization–maximization algorithm is proposed for implementation. Under some regularity conditions, asymptotic distributions of the resulting estimators are derived. Simulation studies and a real data example are conducted to examine the finite sample performance of the proposed method. Both theoretical and empirical findings confirm that our proposed method works very well.
Similar content being viewed by others
Change history
15 September 2022
Equations in the appendix have been corrected.
16 September 2022
A Correction to this paper has been published: https://doi.org/10.1007/s42952-022-00190-4
References
Fan, J. Q., Yao, Q. W., & Cai, Z. W. (2003). Adaptive varying-coefficient linear models. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 65, 57–80.
Fan, J. Q., & Zhang, W. Y. (2008). Statistical methods with varying coefficient models. Statistics and Its Interface, 1, 179–195.
Feng, S. Y., Tian, P., Hu, Y. P., & Li, G. R. (2021). Estimation in functional single-index varying coefficient model. Journal of Statistical Planning and Inference, 214, 62–75.
Feng, S. Y., & Xue, L. G. (2013). Variable selection for single-index varying-coefficient model. Frontiers of Mathematics in China, 8, 541–565.
Feng, S. Y., & Xue, L. G. (2015). Model detection and estimation for single-index varying coefficient model. Journal of Multivariate Analysis, 139, 227–244.
Friedman, J., Hastie, T., & Tibshirani, R. (2000). Additive logistic regression: A statistical view of boosting. The Annals of Statistics, 28, 337–407.
Härdle, W., Hall, P., & Ichimura, H. (1993). Optimal smoothing in single-index models. The Annals of Statistics, 21, 157–178.
Harrison, D., & Rubinfeld, D. L. (1978). Hedonic housing pries and the demand for clean air. Journal of Environmental Economics and Management, 5, 81–102.
Hu, T., & Xia, Y. C. (2012). Adaptive semi-varying coefficient model selection. Statistica Sinica, 22, 575–599.
Huang, Z. S., Pang, Z., Lin, B. Q., & Shao, Q. X. (2014). Model structure selection in single-index-coefficient regression models. Journal of Multivariate Analysis, 125, 159–175.
Jiang, Y. L. (2015). Robust estimation in partially linear regression models. Journal of Applied Statistics, 42, 2497–2508.
Jiang, Y. L., Ji, Q. H., & Xie, B. J. (2017). Robust estimation for the varying coefficient partially nonlinear models. Journal of Computational and Applied Mathematics, 326, 31–43.
Lai, P., Zhang, Q. Z., Lian, H., & Wang, Q. H. (2016). Efficient estimation for the heteroscedastic single-index varying coefficient models. Statistics and Probability Letters, 110, 84–93.
Li, G. R., Peng, H., Dong, K., & Tong, T. J. (2014). Simultaneous confidence bands and hypothesis testing in single-index models. Statistica Sinica, 24, 937–955.
Lian, H., Liang, H., & Carroll, R. J. (2015). Variance function partially linear single-index models. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 77, 171–194.
Liu, J. C., Xu, P. R., & Lian, H. (2019). Estimation for single-index models via martingale difference divergence. Computational Statistics and Data Analysis, 137, 271–284.
Peng, H., & Huang, T. (2011). Penalized least squares for single index models. Journal of Statistical Planning and Inference, 141, 1362–1379.
Powell, J. L., Stock, J. H., & Stoker, T. M. (1989). Semiparametric estimation of index coefficients. Econometrica, 57, 1403–1430.
Shi, J. H., Yang, Q., Li, X. Y., & Song, W. X. (2017). Effects of measurement error on a class of single-index varying coefficient regression models. Computational Statistics, 32, 977–1001.
Song, Y. Q., Jian, L., & Lin, L. (2016). Robust exponential squared loss-based variable selection for high-dimensional single-index varying-coefficient model. Journal of Computational and Applied Mathematics, 308, 330–345.
Wang, J. L., Xue, L. G., Zhu, L. X., & Chong, Y. S. (2010). Estimation for a partial-linear single-index model. The Annals of Statistics, 38, 246–274.
Wang, Q. H., & Xue, L. G. (2011). Statistical inference in partially-varying-coefficient single-index model. Journal of Multivariate Analysis, 102, 1–19.
Wang, X. Q., Jiang, Y. L., Huang, M., & Zhang, H. P. (2013). Robust variable selection with exponential squared loss. Journal of the American Statistical Association, 108, 632–643.
Xia, Y. C. (2006). Asymptotic distributions for two estimators of the single-index model. Econometric Theory, 22, 1112–1137.
Xia, Y. C., & Li, W. K. (1999). On single-index coefficient regression models. Journal of the American Statistical Association, 94, 1275–1285.
Xia, Y. C., Tong, H., Li, W. K., & Zhu, L. X. (2002). An adaptive estimation of dimension reduction space. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 64, 363–410.
Xue, L. G., & Pang, Z. (2013). Statistical inference for a single-index varying-coefficient model. Statistics and Computing, 23, 589–599.
Yu, P., Zhu, Z. Y., & Zhang, Z. Z. (2019). Robust exponential squared loss-based estimation in semi-functional linear regression models. Computational Statistics, 34, 503–525.
Yue, L. L., Li, G. R., & Lian, H. (2019). Identification and estimation in quantile varying-coefficient models with unknown link function. Test, 28(4), 1251–1275.
Zhang, W. Y., Li, D. G., & Xia, Y. C. (2015). Estimation in generalised varying-coefficient models with unspecified link functions. Journal of Econometrics, 187, 238–255.
Zhang, W. Y., & Zhang, H. (2010). Simultaneous confidence band and hypothesis test in generalised varying-coefficient models. Journal of Multivariate Analysis, 101, 1656–1680.
Zhao, Y., Xue, L. G., & Feng, S. Y. (2017). Semiparametric estimation of the single-index varying-coefficient model. Communications in Statistics-Theory and Methods, 46, 4311–4326.
Acknowledgements
The authors would like to thank the Editor, Associate Editor, and two anonymous referees for their constructive comments that have led to a substantial improvement of this paper. Yang Zhao’s research was supported by the National Natural Science Foundation of China (12061044, 62163027). Lili Yue’s research was supported by the National Natural Science Foundation of China (12001277). Gaorong Li’s research was supported by the National Natural Science Foundation of China (11871001, 11971001 and 12131006) and the Natural Science Foundations of Shaanxi Province of China (2020JM-571).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The original online version of this article was revised: Due to missing a part of equations in the Appendix.
Appendices
Appendix: Proofs of main results
For convenience of presentation, we denote \(\varphi _{\gamma }(t)=\exp (-t^2/\gamma )\), \(\varvec{\mu }_{\varvec{\theta }}({\varvec{x}})=E({\varvec{X}} \vert \varvec{\theta }^{\mathrm{T}}{\varvec{X}}=\varvec{\theta }^{\mathrm{T}}{\varvec{x}})\), \(\delta _{\varvec{\theta }}=\Vert \varvec{\theta }-\varvec{\theta }_{0} \Vert\) and \(\delta _{n}=\{ \log n /(nh)\}^{1/2}\).
Proof of Theorem 1
Proof
From (7), we directly consider the maximization of the following objective function
Let \((\tilde{\varvec{a}},\tilde{\varvec{d}})\) be the maximizer of (13), and it satisfies the following equation
Denote \(r_{i}={\varvec{g}}_{0}^{\mathrm{T}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}_{i}){\varvec{Z}}_{i} -{\varvec{g}}_{0}^{\mathrm{T}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{x}}){\varvec{Z}}_{i}-{\varvec{g}}_{0}'^{\mathrm{T}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{x}}){\varvec{Z}}_{i}\varvec{\theta }^{\mathrm{T}}{\varvec{X}}_{i0}\), we have
By Taylor expansion and (14), it follows that
where
By some direct calculations, we obtain
where \(R(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}_{i},\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{x}},{\varvec{Z}}_{i})\) is defined as remainder.
Using the fact \((\mathbf{A}+\mathbf{B}h)^{-1}=\mathbf{A}^{-1}-h\mathbf{A}^{-1}{} \mathbf{B}{} \mathbf{A}^{-1}+O(h^2)\), for the third term on the right-hand side of (15), we have
For the second term on the right-hand side of (15), we can write
where \({\varvec{r}}_{n,1}({\varvec{x}})=\big [n f_{\varvec{\theta }}(\varvec{\theta }^\mathrm{T}{\varvec{x}}) \mathbf{D}_{\varvec{\theta }}(\varvec{\theta }^{\mathrm{T}}{\varvec{x}}) E\{ \varphi _{\gamma }''(\epsilon ) \} \big ]^{-1} \sum _{i=1}^{n} K_{h}(\varvec{\theta }^{\mathrm{T}}{\varvec{X}}_{i0}) \varphi '_{\gamma }(\epsilon _{i}) {\varvec{Z}}_{i},\) \({\varvec{r}}_{n,2}({\varvec{x}})=\big [n f_{\varvec{\theta }}(\varvec{\theta }^{\mathrm{T}}{\varvec{x}}) \mathbf{D}_{\varvec{\theta }}(\varvec{\theta }^{\mathrm{T}}{\varvec{x}}) E\{ \varphi _{\gamma }''(\epsilon ) \} \big ]^{-1} \sum _{i=1}^{n} K_{h}(\varvec{\theta }^{\mathrm{T}}{\varvec{X}}_{i0}) \varphi '_{\gamma }(\epsilon _{i}) \big ( \varvec{\theta }^{\mathrm{T}}{\varvec{X}}_{ix}/h \big ) {\varvec{Z}}_{i}.\)
Combining the above equations, we have
Let \(\tilde{\varvec{a}}_{j}\) and \(\tilde{\varvec{d}}_{j}\) be, respectively, the values of \(\tilde{\varvec{a}}\) and \(\tilde{\varvec{d}}\) with \({\varvec{x}}\) replaced by \({\varvec{X}}_{j}\). Replacing \(({\varvec{a}}_{j},{\varvec{d}}_{j})\) in (6) with \((\tilde{\varvec{a}}_{j}\), \(\tilde{\varvec{d}}_{j})\), and denoting \(\Delta _{ij}={\varvec{g}}_{0}^{\mathrm{T}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}_{i}){\varvec{Z}}_{i}-\tilde{\varvec{a}}_{j}^{\mathrm{T}}{\varvec{Z}}_{i}-\tilde{\varvec{d}}_{j}^{\mathrm{T}}{\varvec{Z}}_{i}\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}_{ij}\), we obtain
where \(\varsigma _{\varvec{\theta }}({\varvec{x}})=n^{-1}\sum _{i=1}^{n}K_{h}(\varvec{\theta }^{\mathrm{T}}{\varvec{X}}_{i0})\).
Let
Using the condition \(\varvec{\theta }\in {\varvec{\Theta }}_{n}\) and exchanging the order of the summation, we have
where
By the expansions of \(\tilde{\varvec{a}}\) and \(\tilde{\varvec{d}}\), we have
where \(R(\varvec{\theta }_{0}^{\mathrm{T}}\mathrm{X}_{i},\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}_{j},{\varvec{Z}}_{i})\) is defined in (16). For (18), we obtain
where \(\mathbf{W}_{1}=E\big [ \mathbf{T}_{\varvec{\theta }_{0}}(\varvec{\theta }_{0}^\mathrm{T}{\varvec{X}}) \mathbf{D}_{\varvec{\theta }_{0}}^{-1}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}) \mathbf{T}_{\varvec{\theta }_{0}}^{\mathrm{T}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}) \big ]\) and \(\mathbf{T}_{\varvec{\theta }_{0}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}) =\mathbf{C}_{\varvec{\theta }_{0}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}) -{\varvec{X}}{\varvec{g}}'^{\mathrm{T}}_{0}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}) \mathbf{D}_{\varvec{\theta }_{0}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}})\). Then, we have
From the above equations and Taylor expansion for (17), we have
Following the proof of Theorem 4.2 in Xia (2006), we obtain
where \({\varvec{\Sigma }}_{0}=\mathbf{W}_{0}-\mathbf{W}_{1}\). Under the assumptions of Theorem 1, it then follows from the above equation and the central limit theorem. \(\square\)
Proof of Theorem 2
Proof
From (15) in Theorem 1, we have
where \(r_{i}(u)={\varvec{g}}_{0}^{\mathrm{T}}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}_{i}){\varvec{Z}}_{i} -{\varvec{g}}_{0}^{\mathrm{T}}(u){\varvec{Z}}_{i}-{\varvec{g}}_{0}'^{\mathrm{T}}(u){\varvec{Z}}_{i}(\varvec{\theta }_{0}^{\mathrm{T}}{\varvec{X}}_{i}-u)\), and
Note that Theorem 1 implies \(\Vert \hat{\varvec{\theta }}-\varvec{\theta }_{0} \Vert =O_{p}(n^{-1/2})\). Therefore, we have
The proof of Theorem 2 is completed by applying Slutsky’s theorem and the central limit theorem. \(\square\)
Rights and permissions
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Zhao, Y., Yue, L. & Li, G. Robust MAVE for single-index varying-coefficient models. J. Korean Stat. Soc. 51, 1302–1325 (2022). https://doi.org/10.1007/s42952-022-00187-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s42952-022-00187-z