Nearest neighbors estimation for long memory functional data


In this paper, we consider the asymptotic properties of the nearest neighbors estimation for long memory functional data. Under some regularity assumptions, we investigate the asymptotic normality and the uniform consistency of the nearest neighbors estimators for the nonparametric regression models when the explanatory variable and the errors are of long memory and the explanatory variable takes values in some abstract functional space. The finite sample performance of the proposed estimator is discussed through simulation studies.

This is a preview of subscription content, access via your institution.

Fig. 1


  1. Al-Awadhi FA, Kaid Z, Laksaci A, Ouassou I, Rachdi M (2018) Functional data analysis: local linear estimation of the \(L_1\)-conditional quantiles. Stat Methods Appl.

    Article  MATH  Google Scholar 

  2. Benhenni K, Hedli-Griche S, Rachdi M, Vieu P (2008) Consistency of the regression estimator with functional data under long memory conditions. Statist Probab Lett 78:1043–1049

    MathSciNet  Article  Google Scholar 

  3. Benhenni K, Hedli-Griche S, Rachdi M (2010) Estimation of the regression operator from functional fixed-design with correlated errors. J Multivar Anal 101:476–490

    MathSciNet  Article  Google Scholar 

  4. Benhenni K, Hedli-Griche S, Rachdi M (2017) Regression models with correlated errors based on functional random design. TEST 26:1–21

    MathSciNet  Article  Google Scholar 

  5. Beran J, Feng Y, Ghosh S, Kulik R (2013) Long-memory processes. Probabilistic properties and statistical methods. Springer, Hiedelberg

    Google Scholar 

  6. Beran J, Feng Y, Ghosh S (2015) Modelling long-range dependence and trends in duration series: an approach based on EFARIMA and ESEMIFAR models. Statist Pap 56:431–451

    MathSciNet  Article  Google Scholar 

  7. Bertram P, Kruse R, Sibbertsen P (2013) Fractional integration versus level shifts: the case of realized asset correlations. Statist Pap 54:977–991

    MathSciNet  Article  Google Scholar 

  8. Bongiorno EG, Salinelli E, Goia A, Vieu P (2014) Contributions in infinite-dimensional statistics and related topics. Società Editrice Esculapio, Bologna

    Google Scholar 

  9. Burba F, Ferraty F, Vieu P (2009) k-Nearest Neighbour method in functional nonparametric regression. J Nonparametr Stat 21:453–469

    MathSciNet  Article  Google Scholar 

  10. Chagny G, Roche A (2016) Adaptive estimation in the functional nonparametric regression model. J Multivar Anal 146:105–118

    MathSciNet  Article  Google Scholar 

  11. Chen J, Zhang L (2009) Asymptotic properties of nonparametric M-estimation for mixing functional data. J Statist Plan Inference 139:533–546

    MathSciNet  Article  Google Scholar 

  12. Csörgő S, Mielniczuk J (1995) Nonparametric regression under long-range dependent normal errors. Ann Stat 23:1000–1014

    MathSciNet  Article  Google Scholar 

  13. Csörgő M, Szyszkowicz B, Wang L (2005) Limit theorems for nearest-neighbor density estimation under long-range dependence. J Statist Res 39:121–138

    MathSciNet  Google Scholar 

  14. Cuevas A (2014) A partial overview of the theory of statistics with functional data. J Statist Plan Inference 147:1–23

    MathSciNet  Article  Google Scholar 

  15. Doukhan P, Lang G, Surgailis D (2002) Asymptotics of weighted empirical processes of linear fields with long-range dependence. Ann Inst H Poincaré Probab Statist 38:879–896

    MathSciNet  Article  Google Scholar 

  16. Doukhan P, Oppenheim G, Taqqu MS (2003) Theory and applications of long-range dependence. Birkhäuser, Boston

    Google Scholar 

  17. Ferraty F, Vieu P (2004) Nonparametric models for functional data, with application in regression, time-series prediction and curve discrimination. J Nonparametr Stat 16:111–125

    MathSciNet  Article  Google Scholar 

  18. Ferraty F, Vieu P (2006) Nonparametric functional data analysis. Springer, New York

    Google Scholar 

  19. Ferraty F, Vieu P (2008) Erratum of: ’non-parametric models for functional data, with application in regression, time-series prediction and curve discrimination’. J Nonparametr Stat 20:187–189

    MathSciNet  Article  Google Scholar 

  20. Gasser T, Hall P, Presnell B (1998) Nonparametric estimation of the mode of a distribution of random curves. J R Stat Soc Ser B Stat Methodol 60:681–691

    MathSciNet  Article  Google Scholar 

  21. Giraitis L, Koul HL, Surgailis D (2012) Large sample inference for long memory processes. Imperial College Press, London

    Google Scholar 

  22. Goia A, Vieu P (2016) An introduction to recent advances in high/infinite dimensional statistics. J Multivar Anal 146:1–6

    MathSciNet  Article  Google Scholar 

  23. Guégan D (2005) How can we define the concept of long memory? An econometric survey. Econom Rev 24:113–149

    MathSciNet  Article  Google Scholar 

  24. Guo H, Koul HL (2007) Nonparametric regression with heteroscedastic long memory errors. J Statist Plan Inference 137:379–404

    MathSciNet  Article  Google Scholar 

  25. Hassler U, Scheithauer J (2011) Detecting changes from short to long memory. Statist Pap 52:847–870

    MathSciNet  Article  Google Scholar 

  26. Horváth L, Kokoszka P (2012) Inference for functional data with applications, vol 200. Springer, Berlin

    Google Scholar 

  27. Hsing T, Eubank R (2015) Theoretical foundations of functional data analysis, with an introduction to linear operators. Wiley series in probability and statistics. Wiley, Chichester

    Google Scholar 

  28. Kara-Zaitri L, Laksaci A, Rachdi M, Vieu P (2017a) Uniform in bandwidth consistency for various kernel estimators involving functional data. J Nonparametr Stat 29:85–107

    MathSciNet  Article  Google Scholar 

  29. Kara-Zaitri L, Laksaci A, Rachdi M, Vieu P (2017b) Data-driven kNN estimation in nonparametric functional data analysis. J Multivar Anal 153:176–188

    Article  Google Scholar 

  30. Kudraszowa NL, Vieu P (2013) Uniform consistency of kNN regressors for functional variables. Statist Probab Lett 83:1863–1870

    MathSciNet  Article  Google Scholar 

  31. Lian H (2011) Convergence of functional k-nearest neighbor regression estimate with functional responses. Electron J Stat 5:31–40

    MathSciNet  Article  Google Scholar 

  32. Masry E (2005) Nonparametric regression estimation for dependent functional data: asymptotic normality. Stoch Process Appl 115:155–177

    MathSciNet  Article  Google Scholar 

  33. Masry E, Mielniczuk J (1999) Local linear regression estimation for time series with long-range dependence. Stoch Process Appl 82:173–193

    MathSciNet  Article  Google Scholar 

  34. Messaci F, Nemouchi N, Ouassou I, Rachdi M (2015) Local polynomial modelling of the conditional quantile for functional data. Stat Methods Appl 24:597–622

    MathSciNet  Article  Google Scholar 

  35. Robinson PM (2011) Asymptotic theory for nonparametric regression with spatial data. J Econom 165:5–19

    MathSciNet  Article  Google Scholar 

  36. Taqqu MS (1975) Weak convergence to fractional brownian motion and to the rosenblatt process. Z. Wahrscheinlichkeitstheorie verw. Gebiete 31:287–302

    MathSciNet  Article  Google Scholar 

Download references

Author information



Corresponding author

Correspondence to Lihong Wang.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

This work was supported by National Natural Science Foundation of China (NSFC) Grants 11671194 and 11501287.



Proof of Lemma 1

It suffices to show that, as \(n\rightarrow \infty \),

$$\begin{aligned} \text {E}[g_n(x,H_n)]\longrightarrow 0 \end{aligned}$$


$$\begin{aligned} \text {Var}[g_n(x,H_n)]\longrightarrow 0. \end{aligned}$$

Let \(\xi _{n1}=\min (H_n, h_n)\) and \(\xi _{n2}=\max (H_n, h_n)\). Then, by (8), \(\xi _{n1}=h_n(1+ o(n^{-\rho }))\), \(\xi _{n2}=h_n(1+ o(n^{-\rho }))\) and \(\xi _{n2}-\xi _{n1}=o(h_nn^{-\rho })\) a.s.

Since E\(\varepsilon _0=0\) and \(k_n=nh_nf_x(0)\), by Assumption (A4),

$$\begin{aligned} \text {E}[g_n(x, H_n)]= & {} \text {E}\big [n^{1/2-\beta }k_n^{-1}\sum _{i=1}^n(K(d(x,X_i)/H_n)-K(d(x,X_i)/h_n))(Y_i-r(x))\big ]\nonumber \\= & {} n^{3/2-\beta }k_n^{-1} \text {E}[(K(d(x,X_1)/H_n)-K(d(x,X_1)/h_n))(r(X_i)-r(x))]\nonumber \\\le & {} Cn^{3/2-\beta }k_n^{-1}\text {E}\Big [\int _{\xi _{n1}}^{\xi _{n2}}u dF_x(u)\Big ]\nonumber \\= & {} Cn^{3/2-\beta }k_n^{-1} \text {E}[(\xi _{n2}-\xi _{n1})\zeta _n f_x(\zeta _n)]\nonumber \\= & {} C n^{3/2-\beta }k_n^{-1}h_n o(n^{-\rho }) \text {E}[\zeta _n f_x(\zeta _n)]\nonumber \\= & {} o(n^{\alpha -\beta -1/2-\rho })\text {E}[f_x(\zeta _n)]/f^2_x(0), \end{aligned}$$

where \(\xi _{n1}<\zeta _n<\xi _{n2}\). Hence (12), together with Assumption (A2) and the fact that \(\zeta _n=h_n(1+o(1))\) a.s. implies (10).

Let \(Z_{i1}=(K(d(x,X_i)/H_n)-K(d(x,X_i)/h_n))(r(X_i)-r(x))\) and \(Z_{i2}=(K(d(x,X_i)/H_n)-K(d(x,X_i)/h_n))\varepsilon _i\). To prove (11), it is enough to show

$$\begin{aligned} \text {Var}\Big (n^{1/2-\beta -\alpha }\sum _{i=1}^n Z_{i1}\Big )\longrightarrow 0, \quad \text {and} \quad \text {Var}\Big (n^{1/2-\beta -\alpha }\sum _{i=1}^n Z_{i2}\Big )\longrightarrow 0. \end{aligned}$$

Note that

$$\begin{aligned} \text {Var}\Big (n^{1/2-\beta -\alpha }\sum _{i=1}^nZ_{i1}\Big )= & {} n^{2-2\beta -2\alpha }\text {Var}(Z_{01})+n^{1-2\beta -2\alpha }\sum _{i\ne j}\text {Cov}(Z_{i1},Z_{j1}). \end{aligned}$$

For the first term of the above variance, similarly to (12), we obtain

$$\begin{aligned} \text {Var}(Z_{01})\le & {} \text {E}\big [(K(d(x,X_0)/H_n)-K(d(x,X_0)/h_n))(r(X_0)-r(x))\big ]^2\\\le & {} C\text {E}\Big [\int _{\xi _{n1}}^{\xi _{n2}}u^2f_x(u)du\Big ]= C\text {E} [(\xi _{n2}-\xi _{n1})\zeta _n^2 f_x(\zeta _n)]\nonumber \\= & {} C h_n^3 o(n^{-\rho })\text {E}[f_x(\zeta _n)]. \end{aligned}$$

This means that

$$\begin{aligned} n^{2-2\beta -2\alpha }\text {Var}(Z_{01})=o(n^{\alpha -2\beta -1-\rho })=o(1). \end{aligned}$$

In addition, by mean value theorem, \(F_x(\xi _{n2})-F_x(\xi _{n1})=f_x(\zeta _n) (\xi _{n2}-\xi _{n1})\) for some \(\xi _{n1}< \zeta _n<\xi _{n2}\). Moreover, by Assumption (A5),

$$\begin{aligned}&P(u_1\le d(x, X_i)\le u_2, u_1\le d(x, X_j)\le u_2)\\&\quad =P(d(x, X_i)\le u_2, d(x, X_j)\le u_2)+P(d(x, X_i)\le u_1, d(x, X_j)\le u_1)\\&\qquad -\,P( d(x, X_i)\le u_2, d(x, X_j)\le u_1)-P( d(x, X_i)\le u_1, d(x, X_j)\le u_2)\\&\quad \le F_x^2(u_2)+F_x^2(u_1)-2F_x(u_2)F_x(u_1)+C\gamma _x(i-j)\\&\quad =\big (F_x(u_2)-F_x(u_1)\big )^2+C\gamma _x(i-j), \end{aligned}$$

for any \(u_1\), \(u_2\) close to 0. Thus, by (8) and Assumption (A4),

$$\begin{aligned} \text {Cov}(Z_{i1},Z_{j1})= & {} \text {Cov}\left[ (r(X_i)-r(x))(K(d(x,X_i)/H_n)-K(d(x,X_i)/h_n)),\right. \nonumber \\&\left. (r(X_j)-r(x))(K(d(x,X_j)/H_n)-K(d(x,X_j)/h_n))\right] \nonumber \\\le & {} C\text {E}\Big [d(x, X_i)d(x, X_j)\big |K(d(x,X_i)/H_n)-K(d(x,X_i)/h_n)\big |\nonumber \\&\cdot \,\big |K(d(x,X_j)/H_n)-K(d(x,X_j)/h_n)\big |\Big ]\nonumber \\&+\,C\Big (\text {E} \big [d(x, X_i)\big |K(d(x,X_i)/H_n)-K(d(x,X_i)/h_n)\big |\big ]\Big )^2\nonumber \\\le & {} C\text {E}\big [\xi _{n2}^2P(\xi _{n1}\le d(x, X_i)\le \xi _{n2}, \xi _{n1}\le d(x, X_j)\le \xi _{n2})\big ]\nonumber \\&+\,C\Big (\text {E}\big [\xi _{n2} P(\xi _{n1}\le d(x, X_0)\le \xi _{n2})\big ]\Big )^2\nonumber \\\le & {} C h_n^2(1+o(1))\Big \{\gamma _x(i-j)+\text {E}\big [F_x(\xi _{n2})-F_x(\xi _{n1})\big ]^2\Big \}\nonumber \\= & {} C h_n^2(1+o(1))\Big \{\gamma _x(i-j)+o(h_n^2 n^{-2\rho })\Big \}. \end{aligned}$$

Now using (7) and Assumption (A2), we arrive at, for some large enough N,

$$\begin{aligned} n^{1-2\beta -2\alpha }\sum _{i\ne j}\text {Cov}(Z_{i1},Z_{j1})= & {} Cn^{1-2\beta -2\alpha }h_n^2\Big \{\sum _{i\ne j}\gamma _x(i-j)+o(n^{2\alpha -2\rho })\Big \}\\= & {} O(n^{-2\beta -1})\Big \{n\sum _{k=1}^n\gamma _x(k)+o(n^{2\alpha -2\rho })\Big \}\\= & {} O(n^{-2\beta -1})\Big \{n\big (\sum _{k=1}^N+\sum _{k=N+1}^n\big )\gamma _x(k)+o(n^{2\alpha -2\rho })\Big \}\\\le & {} O(n^{-2\beta -1})\Big \{n\sum _{k=1}^n k^{-\tau _x D}+o(n^{2\alpha -2\rho })\Big \}\\= & {} O(n^{-2\beta -1})\Big \{n^{2-\tau _x D}+o(n^{2\alpha -2\rho })\Big \}=o(1). \end{aligned}$$

These bounds imply the weak convergence of \(n^{1/2-\beta -\alpha }\sum _{i=1}^n Z_{i1}\). To derive the same result for \(n^{1/2-\beta -\alpha }\sum _{i=1}^n Z_{i2}\), note that, by (2),

$$\begin{aligned}&\text {Var}\Big (n^{1/2-\beta -\alpha }\sum _{i=1}^nZ_{i2}\Big )\\&\quad = n^{2-2\beta -2\alpha }\text {Var}(Z_{02})+n^{1-2\beta -2\alpha }\sum _{i\ne j}\text {Cov}(Z_{i2},Z_{j2})\\&\quad = n^{2-2\beta -2\alpha }\text {E}\varepsilon _0^2\text {E}\big (K(d(x,X_0)/H_n)-K(d(x,X_0)/h_n)\big )^2\\&\qquad +\,n^{1-2\beta -2\alpha }\sum _{i\ne j}\gamma _\varepsilon (i-j)\text {E}\big [(K(d(x,X_i)/H_n)-K(d(x,X_i)/h_n))\\&\qquad \cdot \, (K(d(x,X_j)/H_n)-K(d(x,X_j)/h_n))\big ]\\&\quad \le Cn^{2-2\beta -2\alpha }\text {E}\big [P(\xi _{n1}\le d(x, X_0)\le \xi _{n2})\big ]\\&\qquad +\,n^{1-2\beta -2\alpha }\sum _{i\ne j}\gamma _\varepsilon (i-j)\text {E}\big [P(\xi _{n1}\le d(x, X_i)\le \xi _{n2}, \xi _{n1}\le d(x, X_j)\le \xi _{n2})\big ]\\&\quad \le C n^{2-2\beta -2\alpha }\text {E}(\xi _{n2}-\xi _{n1})\\&\qquad +\,Cn^{1-2\beta -2\alpha }\sum _{i\ne j}\gamma _\varepsilon (i-j)\big (\gamma _x(i-j)+\text {E} [F_x(\xi _{n2})-F_x(\xi _{n1})]^2\big )\\&\quad = o(n^{1-2\beta -\alpha -\rho })+O(n^{1-2\beta -2\alpha })\big (O(n^{2\beta +1-\tau _x D})+o(n^{2\beta +2\alpha -1-2\rho })\big )\\&\quad =o(n^{1-2\beta -\alpha -\rho })+O(n^{2-2\alpha -\tau _x D})+o(n^{-2\rho }). \end{aligned}$$

Since \(\rho >0\) and \(\tau _x\ge 1\), Assumption (A2) implies that \(1-2\beta -\alpha -\rho <0\) and \(2-2\alpha -\tau _x D<0\). That is,

$$\begin{aligned} \text {Var}\Big (n^{1/2-\beta -\alpha }\sum _{i=1}^nZ_{i2}\Big )=o(1). \end{aligned}$$

This completes the proof of Lemma 1. \(\square \)

Proof of Theorem 1

By (1) and (6),

$$\begin{aligned}&n^{1/2-\beta }(r_n(x,H_n)-r(x))\nonumber \\&\quad =n^{1/2-\beta }k_n^{-1}\sum _{i=1}^n K(d(x,X_i)/h_n)(Y_i-r(x))+g_n(x, H_n)\nonumber \\&\quad =n^{1/2-\beta }k_n^{-1}\sum _{i=1}^n K(d(x,X_i)/h_n)(r(X_i)-r(x)+\varepsilon _i)+g_n(x, H_n)\nonumber \\&\quad =I_{n1}+I_{n2}+g_n(x, H_n) \end{aligned}$$

where \(I_{n1}=n^{1/2-\beta }k_n^{-1}\sum _{i=1}^n K(d(x,X_i)/h_n)(r(X_i)-r(x))\) and \(I_{n2}=n^{1/2-\beta }\) \(k_n^{-1}\) \(\sum _{i=1}^n K(d(x,X_i)/h_n)\varepsilon _i\).

By Lemma 1, it suffices to show that

$$\begin{aligned} I_{n1}{\mathop {\longrightarrow }\limits ^\mathcal{P}} 0, \quad \text {and}\quad I_{n2}{\mathop {\longrightarrow }\limits ^\mathcal{D}} c_0 Z. \end{aligned}$$

Along the similar lines of the Proof of Lemma 1, we obtain

$$\begin{aligned} \text {E}I_{n1}= & {} n^{1/2-\beta }k_n^{-1}n \text {E}[K(d(x,X_0)/h_n)(r(X_0)-r(x))]\nonumber \\\le & {} Cn^{3/2-\beta -\alpha }\int _0^{h_n} u dF_x(u)=Cn^{3/2-\beta -\alpha } h_n\zeta _nf_x(\zeta _n)\nonumber \\= & {} O(n^{\alpha -\beta -1/2})=o(1), \end{aligned}$$

where \(0<\zeta _n<h_n\). Moreover, again by Assumption (A2),

$$\begin{aligned} \text {Var}(I_{n1})\le & {} n^{2-2\beta -2\alpha }\text {E}\big (K(d(x,X_0)/h_n)(r(X_0)-r(x))\big )^2+n^{1-2\beta -2\alpha }\sum _{i\ne j}\nonumber \\&\text {Cov}\left[ (r(X_i)-r(x))K(d(x,X_i)/h_n), (r(X_j)-r(x))K(d(x,X_j)/h_n)\right] \nonumber \\= & {} n^{2-2\beta -2\alpha }O(h_n^3)+n^{1-2\beta -2\alpha }\Big \{O(h_n^2)\sum _{i\ne j}\gamma _x(i-j)+O(n^2h_n^4)\Big \}\nonumber \\= & {} O(n^{\alpha -2\beta -1})+n^{1-2\beta -2\alpha }\big (O(n^{2\alpha -\tau _x D})+O(n^{4\alpha -2})\big )\nonumber \\= & {} O(n^{\alpha -2\beta -1})+O(n^{1-2\beta -\tau _x D})+O(n^{2\alpha -2\beta -1})=o(1). \end{aligned}$$

Then we arrive at \(I_{n1}{\mathop {\longrightarrow }\limits ^\mathcal{P}} 0\). For \(I_{n2}\), note that

$$\begin{aligned} I_{n2}= & {} n^{1/2-\beta }k_n^{-1}\sum _{i=1}^n \big (K(d(x,X_i)/h_n)-\text {E}(K(d(x,X_i)/h_n))\big )\varepsilon _i\nonumber \\&+\,n^{1/2-\beta }k_n^{-1}\sum _{i=1}^n \text {E}(K(d(x,X_i)/h_n))\varepsilon _i. \end{aligned}$$


$$\begin{aligned} \text {E}(K(d(x,X_i)/h_n))=\int _0^{h_n} dF_x(u)=F_x(h_n)=h_nf_x(\zeta _n)=h_n(f_x(0)+o(1)), \end{aligned}$$

where \(0<\zeta _n<h_n\), by (9), we have

$$\begin{aligned} n^{1/2-\beta }k_n^{-1}\sum _{i=1}^n \text {E}(K(d(x,X_i)/h_n))\varepsilon _i {\mathop {\longrightarrow }\limits ^\mathcal{D}} c_0 Z. \end{aligned}$$

Therefore, to complete the proof, it suffices to show the first term of (15) tends to 0 in probability. In a similar way as in (14), the variance for the first term of (15) is bounded by

$$\begin{aligned}&n^{2-2\beta -2\alpha }\text {E}\big (K(d(x,X_0)/h_n)\big )^2\text {E}\varepsilon _0^2\\&\qquad +\,n^{1-2\beta -2\alpha }\sum _{i\ne j}\gamma _{\varepsilon }(i-j) \text {Cov}\left[ K(d(x,X_i)/h_n), K(d(x,X_j)/h_n)\right] \\&\quad \le n^{2-2\beta -2\alpha }O(h_n)+Cn^{1-2\beta -2\alpha }\sum _{i\ne j}\gamma _\varepsilon (i-j)\gamma _x(i-j)\\&\quad =O(n^{1-2\beta -\alpha })+O(n^{2-2\alpha -\tau _x D})=o(1). \end{aligned}$$

This concludes the Proof of Theorem 1. \(\square \)

Proof of Theorem 2

From (6), we obtain

$$\begin{aligned}&r_n(x, H_n)-r(x)\\&\quad =k_n^{-1}\sum _{i=1}^n K(d(x, X_i)/H_n)(r(X_i)-r(x))+k_n^{-1}\sum _{i=1}^n K(d(x, X_i)/H_n)\varepsilon _i\\&\quad :=R_1(x)+R_2(x), \quad \text {say}. \end{aligned}$$

It suffices to show that

$$\begin{aligned} \sup _{x\in S} |R_1(x)|=o_P(1),\quad \text {and}\quad \sup _{x\in S} |R_2(x)|=o_P(1). \end{aligned}$$

By (8) and Assumption (A4), we have

$$\begin{aligned} \sup _{x\in S} |R_1(x)|\le \sup _{x\in S} \sup _{y\in B(x, H_n)}|r(y)-r(x)|\le CH_n=O(h_n)\quad \text {a.s.} \end{aligned}$$

This is enough to prove the first claim of (16). It just remains to check the second result of (16). Note that we can write

$$\begin{aligned} \sup _{x\in S} |R_2(x)|\le \sup _{x\in S} |R_2(x)-R_2(t_x)|+\sup _{x\in S} |R_2(t_x)|, \end{aligned}$$


$$\begin{aligned} \sup _{x\in S} |R_2(x)-R_2(t_x)|\le \sup _{x\in S} k_n^{-1}\sum _{i=1}^n \big |K(d(x, X_i)/H_n)-K(d(t_x, X_i)/H_n)\big ||\varepsilon _i|. \end{aligned}$$

By (8) and Assumption (B2), we have

$$\begin{aligned}&\text {E} \Big [\big |K(d(x, X_0)/H_n)-K(d(t_x, X_0)/H_n)\big |\Big ]\nonumber \\&\quad = \text {P}((X_0\in B(x, h_n)\bigcap \bar{B}(t_x, h_n))+\text {P}(X_0\in \bar{B}(x, h_n)\bigcap B(t_x, h_n))\nonumber \\&\quad =O(l_n) \end{aligned}$$

for any \(x\in S\). This result implies that, for any \(x\in S\) and any \(1\le i\le n\),

$$\begin{aligned} \big |K(d(x, X_i)/H_n)-K(d(t_x, X_i)/H_n)\big |=O_P(l_n). \end{aligned}$$

Moreover, for any \(\delta >0\), there exists an \(x^*\in S\) with

$$\begin{aligned}&\sup _{x\in S}\big |K(d(x, X_i)/H_n)-K(d(t_x, X_i)/H_n)\big |\\&\quad < |K(d(x^*, X_i)/H_n)-K(d(t_{x^*}, X_i)/H_n)\big |+\delta . \end{aligned}$$

Let \(\delta \rightarrow 0\), we obtain

$$\begin{aligned} \sup _{x\in S}\big |K(d(x, X_i)/H_n)-K(d(t_x, X_i)/H_n)\big |=O_P(l_n). \end{aligned}$$

This leads directly to

$$\begin{aligned} \sup _{x\in S} |R_2(x)-R_2(t_x)|=O_P(l_nk_n^{-1}n)=O_P(n^{1-\alpha -\xi })=o_P(1). \end{aligned}$$

Looking at the second term on the RHS of (17), we have, for any \(\varepsilon >0\),

$$\begin{aligned}&\text {P}\big (\sup _{x\in S} |R_2(t_x)|>\varepsilon \big )\\&\quad =\text {P}\big (\max _{k\in \{1,\ldots , {z_n}\} } |R_2(t_k)-\text {E}(R_2(t_k))|>\varepsilon \big )\\&\quad \le z_n\max _{k\in \{1,\ldots , {z_n}\} }\text {P}\big ( |R_2(t_k)-\text {E}(R_2(t_k))|>\varepsilon \big ). \end{aligned}$$

By using the similar lines of the proof of (13), and by Assumptions (A2), (B1) and (B3), one gets directly for any \(\varepsilon >0\),

$$\begin{aligned} \text {P}\big (\sup _{x\in S} |R_2(t_x)|>\varepsilon \big )\le z_n\text {Var}\big ( R_2(t_1)\big )/\varepsilon ^2 =O(l_n^{-1})O(n^{2\beta -1})=o(1). \end{aligned}$$

This leads to

$$\begin{aligned} \sup _{x\in S} |R_2(t_x)|=o_P(1). \end{aligned}$$

This, together with (18), is enough to get

$$\begin{aligned}\sup _{x\in S} |R_2(x)|=o_P(1).\end{aligned}$$

This allows us to finish the Proof of Theorem 2. \(\square \)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Wang, L. Nearest neighbors estimation for long memory functional data. Stat Methods Appl 29, 709–725 (2020).

Download citation


  • Asymptotic normality
  • Functional data
  • Long memory
  • Nearest neighbors estimation
  • Uniform consistency

Mathematics Subject Classification

  • 62M10