w.p.a.1., where the equality follows from a second-order expansion with _ on the l<strong>in</strong>e jo<strong>in</strong><strong>in</strong>g <strong>and</strong> 0,<strong>and</strong> the <strong>in</strong>equality follows from 1 P nnh i=1 g i (^) g i (^) 0 p! V , boundedness <strong>of</strong> V , <strong>and</strong> Lemma 2. Also notethatsup ^P (^; ) sup ^P ( 0 ; ) = O p(nh) 1 ; (24)2 n(^)2 n( 0 )w.p.a.1., where the <strong>in</strong>equality follows from the de…nition <strong>of</strong> ^, <strong>and</strong> the equality follows from Lemma 3with = 0 <strong>and</strong> 1 P nnh i=1 g i ( 0 ) = O p (nh) 1=2 (by Lemma 1). S<strong>in</strong>ce ~ 2 n , Lemma 2 guarantees~ 2 n (^), w.p.a.1., which implies ^P^; ~ sup ^P 2n(^) (^; ). Thus, comb<strong>in</strong><strong>in</strong>g (23) <strong>and</strong> (24),n j^gj Cn 2 O p(nh) 1 ; (25)w.p.a.1. S<strong>in</strong>ce we chose to satisfy (nh) 1=2 n ! 0, we have j^gj = O p n . Now, pick any n ! 0<strong>and</strong> de…ne = n^g. From j^gj = O p n , we have = o p n <strong>and</strong> 2 n n (^). Thus, we applythe same argument to (23)-(25) after replac<strong>in</strong>g ~ with . Then we obta<strong>in</strong> n j^gj 2 C 2 n j^gj 2 ^P ^; sup ^P (^; ) = O p(nh) 1 ;2 n(^)which implies n j^gj 2 = O p(nh) 1 : S<strong>in</strong>ce this results holds for any n ! 0, we obta<strong>in</strong> the conclusion.20
References[1] Angrist, J. D. <strong>and</strong> V. Lavy (1999) Us<strong>in</strong>g Maimonides’rule to estimate the e¤ect <strong>of</strong> class size onscholastic achievement, Quarterly Journal <strong>of</strong> Economics, 114, 533-575.[2] Cheng, M. Y. (1994) On boundary e¤ects <strong>of</strong> smooth curve estimators (dissertation), Unpublishedmanuscript series # 2319, University <strong>of</strong> North Carol<strong>in</strong>a.[3] Cheng, M. Y. (1997) A b<strong>and</strong>width selector for local l<strong>in</strong>ear density estimators, Annals <strong>of</strong> Statistics,25, 1001-1013.[4] Cheng, M. Y., Fan, J. <strong>and</strong> J. S. Marron (1997) On automatic boundary corrections, Annals <strong>of</strong>Statistics, 25, 1691-1708.[5] Cl<strong>in</strong>e, D. B. <strong>and</strong> J. D. Hart (1991) Kernel estimation <strong>of</strong> densities with discont<strong>in</strong>uities or discont<strong>in</strong>uousderivatives, Statistics, 22, 69-84.[6] Copas, J. B. (1995) Local likelihood based on kernel censor<strong>in</strong>g. Journal <strong>of</strong> the Royal StatisticalSociety, B, 57, 221-235.[7] Davidson, R. <strong>and</strong> MacK<strong>in</strong>non, J. G. (1998) Graphical methods for <strong>in</strong>vestigat<strong>in</strong>g the size <strong>and</strong> power<strong>of</strong> test statistics. The Manchester School, 66, 1-26.[8] Fan, J. <strong>and</strong> I. Gijbels (1996) Local Polynomial Modell<strong>in</strong>g <strong>and</strong> Its Applications, Chapman & Hall.[9] Fan, J., Zhang, C. <strong>and</strong> J. Zhang (2001) Generalized likelihood ratio statistics <strong>and</strong> Wilks phenomenon,Annals <strong>of</strong> Statistics, 29, 153-193.[10] Gregory, A. W. <strong>and</strong> M. R. Veal (1985) Formulat<strong>in</strong>g Wald tests <strong>of</strong> nonl<strong>in</strong>ear restrictions, Econometrica,53, 1465-68.[11] Hahn, J., Todd, P. <strong>and</strong> W. van der Klaauw (2001) Identi…cation <strong>and</strong> estimation <strong>of</strong> treatment e¤ectswith a regression discont<strong>in</strong>uity design, Econometrica, 69, 201-209.[12] Hjort, N. L. <strong>and</strong> M. C. Jones (1996) Locally parametric nonparametric density estimation. Annals<strong>of</strong> Statistics, 24, 1619-1647.[13] Imbens, G. W. <strong>and</strong> T. Lemieux (2008) Regression discont<strong>in</strong>uity designs: a guide to practice,Journal <strong>of</strong> Econometrics, 142, 615-635.[14] Kitamura, Y. (1997) Empirical likelihood methods with weakly dependent data, Annals <strong>of</strong> Statistics,25, 2084-2102.[15] Loader, C. R. (1996) Local likelihood density estimation, Annals <strong>of</strong> Statistics, 24, 1602-1618.21