2008年中国数量经济学会年会论文
表 3 各分位数下固定效应分位数回归估计结果 分位数 IP 常数项 判定系数（R2） 系数 T统计量 系数 T统计量 0.2 0.6799 36.28 490.5411 5.56 0.9096 0.5 0.6715 19.91 597.5509 3.27 0.9212 0.75 0.7328 25.23 351.1442 2.22 0.9329 0.8 0.7408 24.76 309.1915 1.90 0.9359 比较分析上述两种回归方法的统计结果，发现在固定效应情形下两种方法回归效果均比合并数据情形下更好；在同一情形下做回归分析，显然分位数回归分析结果更加稳定，各系数估计显著程度更高。因而，分位数回归估计在PANEL DATA模型中可以发挥重要作用。
四、结论
本文在对分位数回归方法的含义和基本原理进行全面分析说明的基础上，对分位数回归方法在PANEL DATA模型中的应用作了深入分析，并对不同回归估计方法在PANEL DATA模型中的估计效果进行了比较分析。一般而言，分位数估计方法在估计具有非正态分布的误差项或不可观察的随机效应时具有一定优势。本文在理论分析之后，提供了一个应用案例分析，通过对我国人均收入和人均消费的各种回归分析，充分证明了分位数回归的较好效果。当然，由于样本数据的不足，缺乏对在随机效应情形下两种回归方法估计的效果比较，以及在不同调节系数下对惩罚分位数回归估计效果的考察。这将在分位数回归方法的应用分析中进一步开展深入探讨。
参考文献
1．Emanuel Parzen, Quantile Probability and Statistical Data Modeling, [J].Statistical Science, Vol. 19, No. 4, (Nov., 2004), pp. 652-662.
2．Roger Koenker and Zhijie Xiao, Inference on the Quantile Regression Process, [J].Econometrica, Vol. 70, No. 4, (Jul., 2002), pp. 1583-1612.
3．Keming Yu, Zudi Lu, Julian Stander, Quantile Regression: Applications and
Current Research Areas,[J]. The Statistician, Vol. 52, No. 3, (2003), pp. 331-350.
4. Roger Koenker and Jose A. F. Machado, Goodness of Fit and Related Inference Processes for Quantile Regression,[J].Journal of the American Statistical Association, Vol. 94, No. 448, (Dec., 1999), pp. 1296 –1310.
5. Keming Yu and M. C. Jones, Local Linear Quantile Regression, [J].Journal of the American Statistical Association, Vol. 93, No. 441, (Mar., 1998), pp. 228 –237. 6. James W. Taylor and Derek W. Bunn, A Quantile Regression Approach to Generating Prediction Intervals,[J] .Management Science, Vol. 45, No. 2, (Feb., 1999), pp.
6
2008年中国数量经济学会年会论文
225-237.
7. Rene Gelinas, Alain Martel, Pierre Lefrancois, SOS: A Quantile Estimation Procedure for Dynamic Lot-Sizing Problems,[J]. The Journal of the Operational Research Society, Vol. 46, No. 11, (Nov., 1995), pp. 1337-1351.
8. P. K. Bhattacharya and Ashis K. Gangopadhyay, Kernel and Nearest-Neighbor Estimation of a Conditional Quantile,[J]. The Annals of Statistics, Vol. 18, No. 3, (Sep., 1990), pp. 1400-1415.
9. PORTNOY S. KOENKER R., The Gaussian Hare and the Laplacian Tortoise :Computability of Squared - error Versus Absolute- error Estimators[J ] . Statistical Science ,1997(12) :279 - 300.
10. Koenker , R. and Bassett , G. 1978.“Regression Quantiles. ”[J ] . Econometrica. Vol. 46 , pp. 33 - 50.
11. Koenker, R. (2004): “Quantile Regression for Longitudinal Data,” [J],Journal of Multivariate Analysis, 91,74–89.
12.吴建南、马伟，估计极端行为模型:分位数回归方法及其实现与应用，[J]数理统计与管理，2006、9月，第25卷第5期，536-543。
Research on Quantile Regression Modeling and Its Application
Wang guisheng
(capital university of economics and business,Beijing,100026)
Abstract: This article principally introduces some basic principles of all kinds of
quantile regressions and their application. After stating the concepts and estimation equations of general least square method and quantile regression, we especially analyze the principle of penalized quantile regression and its application in Panel data modeling. We contrast the estimation efficiency of two kinds of methods by theoretical and empirical analyses: LS,PLS and QR,PQR. We find that PQR or QR has certain relative advantages over LS or PLS on some estimation of models in which there are non-Gauss distributional stochastic error terms or stochastic effects.
Keywords: quantile regression, Panel data model, penalized quantile regression
estimation
7