Skip to main content

Kolmogorov–Smirnov simultaneous confidence bands for time series distribution function

Abstract

Claims about distributions of time series are often unproven assertions instead of substantiated conclusions for lack of hypotheses testing tools. In this work, Kolmogorov–Smirnov type simultaneous confidence bands (SCBs) are constructed based on simple random samples (SRSs) drawn from realizations of time series, together with smooth SCBs using kernel distribution estimator (KDE) instead of empirical cumulative distribution function of the SRS. All SCBs are shown to enjoy the same limiting distribution as the standard Kolmogorov–Smirnov for i.i.d. sample, which is validated in simulation experiments on various time series. Computing these SCBs for the standardized S&P 500 daily returns data leads to some rather unexpected findings, i.e., student’s t-distributions with degrees of freedom no less than 3 and the normal distribution are all acceptable versions of the standardized daily returns series’ distribution, with proper rescaling. These findings present challenges to the long held belief that daily financial returns distribution is fat-tailed and leptokurtic.

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

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6

References

  • Billingsley P (1999) Convergence of probability measures, 2nd edn. Wiley, New York

    Book  Google Scholar 

  • Brockwell PJ, Davis RA (1991) Time series: theory and methods, 2nd edn. Springer, New York

    Book  Google Scholar 

  • Cai L, Yang L (2015) A smooth simultaneous confidence band for conditional variance function. TEST 24:632–655

    MathSciNet  Article  Google Scholar 

  • Cao G, Wang L, Li Y, Yang L (2016) Oracle-efficient confidence envelopes for covariance functions in dense functional data. Stat Sin 26:359–383

    MathSciNet  MATH  Google Scholar 

  • Cao G, Yang L, Todem D (2012) Simultaneous inference for the mean function based on dense functional data. J Nonparametr Stat 24:359–377

    MathSciNet  Article  Google Scholar 

  • Cardot H, Degras D, Josserand E (2013) Confidence bands for Horvitz–Thompson estimators using sampled noisy functional data. Bernoulli 19:2067–2097

    MathSciNet  MATH  Google Scholar 

  • Cardot H, Josserand E (2011) Horvitz–Thompson estimators for functional data: asymptotic confidence bands and optimal allocation for stratified sampling. Biometrika 98:107–118

    MathSciNet  Article  Google Scholar 

  • Cheng M, Peng L (2002) Regression modeling for nonparametric estimation of distribution and quantile functions. Stat Sin 12:1043–1060

    MathSciNet  MATH  Google Scholar 

  • Deo CM (1973) A note on empirical process of strong-mixing sequences. Ann Probab 5:870–875

    MATH  Google Scholar 

  • Degras D (2011) Simultaneous confidence bands for nonparametric regression with functional data. Stat Sin 21:1735–1765

    MathSciNet  Article  Google Scholar 

  • Falk M (1985) Asymptotic normality of the kernel quantile estimator. Ann Stat 13:428–433

    MathSciNet  Article  Google Scholar 

  • Francq C, Zakoian J (2010) GARCH models: structure, statistical inference and financial applications. Wiley, Paris

    Book  Google Scholar 

  • Gu L, Wang L, Härdle W, Yang L (2014) A simultaneous confidence corridor for varying coefficient regression with sparse functional data. TEST 23:806–843

    MathSciNet  Article  Google Scholar 

  • Gu L, Yang L (2015) Oracally efficient estimation for single-index link function with simultaneous band. Electron J Stat 9:1540–1561

    MathSciNet  Article  Google Scholar 

  • Kong J, Gu L, Yang L (2018) Prediction interval for autoregressive time series via oracally efficient estimation of multi-step ahead innovation distribution function. J Time Ser Anal 39:690–708

    MathSciNet  Article  Google Scholar 

  • Liu R, Yang L (2008) Kernel estimation of multivariate cumulative distribution function. J Nonparametr Stat 20:661–677

    MathSciNet  Article  Google Scholar 

  • Ma S, Yang L, Carroll R (2012) A simultaneous confidence band for sparse longitudinal regression. Stat Sin 22:95–122

    MathSciNet  MATH  Google Scholar 

  • Reiss R (1981) Nonparametric estimation of smooth distribution functions. Scand J Stat 8:116–119

    MathSciNet  MATH  Google Scholar 

  • Rosén B (1964) Limit theorems for sampling from finite populations. Ark Mat 5:383–424

    MathSciNet  Article  Google Scholar 

  • Shao Q, Yang L (2017) Oracally efficient estimation and consistent model selection for auto-regressive moving average time series with trend. J R Stat Soc Ser B Stat Methodol 79:507–524

    MathSciNet  Article  Google Scholar 

  • Song Q, Liu R, Shao Q, Yang L (2014) A simultaneous confidence band for dense longitudinal regression. Commun Stat A-Theory Methods 43:5195–5210

    MathSciNet  Article  Google Scholar 

  • Song Q, Yang L (2009) Spline confidence bands for variance functions. J Nonparametr Stat 5:589–609

    MathSciNet  Article  Google Scholar 

  • Wang J, Cheng F, Yang L (2013) Smooth simultaneous confidence bands for cumulative distribution functions. J Nonparametr Stat 25:395–407

    MathSciNet  Article  Google Scholar 

  • Wang J, Liu R, Cheng F, Yang L (2014) Oracally efficient estimation of autoregressive error distribution with simultaneous confidence band. Ann Stat 42:654–668

    MathSciNet  MATH  Google Scholar 

  • Wang J, Wang S, Yang L (2016) Simultaneous confidence bands for the distribution function of a finite population and its superpopulation. TEST 25:692–709

    MathSciNet  Article  Google Scholar 

  • Wang J, Yang L (2009) Polynomial spline confidence bands for regression curves. Stat Sin 19:325–342

    MathSciNet  MATH  Google Scholar 

  • Xue L, Wang J (2010) Distribution function estimation by constrained polynomial spline regression. J Nonparametr Stat 22:443–457

    MathSciNet  Article  Google Scholar 

  • Yamato H (1973) Uniform convergence of an estimator of a distribution function. Bull Math Stat 15:69–78

    MathSciNet  Article  Google Scholar 

  • Zheng S, Liu R, Yang L, Härdle W (2016) Statistical inference for generalized additive models: simultaneous confidence corridors and variable selection. TEST 25:607–626

    MathSciNet  Article  Google Scholar 

  • Zheng S, Yang L, Härdle W (2014) A smooth simultaneous confidence corridor for the mean of sparse functional data. J Am Stat Assoc 109:661–673

    MathSciNet  Article  Google Scholar 

  • Zhang Y, Liu R, Shao Q, Yang L (2020) Two-step estimation for time varying arch models. J Time Ser Anal 41:551–570

    MathSciNet  Article  Google Scholar 

  • Zhang Y, Yang L (2018) A smooth simultaneous confidence band for correlation curve. TEST 27:247–269

    MathSciNet  Article  Google Scholar 

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Lijian Yang.

Additional information

Publisher's Note

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

Supported by National Natural Science Foundation of China awards 11771240, 11801272, 12026242, Natural Science Foundation of Jiangsu Province of China BK20180820, Qinglan Project of Jiangsu Province of China, Philosophy and Social Science Research in Colleges and Universities of Jiangsu Province 2019SJA0353.

Appendix

Appendix

Throughout this section, c denotes any positive constant and \(\mathcal {O}_{p}\) (or \( o_{p}\)) a sequence of random variables of certain order in probability. In addition, \(u_{p}\) denotes a sequence of random functions which are \(o_{p}\) uniformly defined in the domain. For any continuous function \(\phi \) defined on an interval \(\mathcal {I}\), the modulus of continuity is defined as \(\omega \left( \phi ,\varDelta \right) =\sup _{x,x^{\prime }\in \mathcal {I} ,\left| x-x^{\prime }\right| \le \varDelta }\left| \phi \left( x^{\prime }\right) -\phi \left( x\right) \right| \)

Lemma 1

(Theorem 7.1,2, Brockwell and Davis (1991)) If \(\left\{ X_t\right\} _{t=1}^n\) is the stationary process,

$$\begin{aligned} X_t = \mu +\sum _{j=-\infty }^{\infty }\psi _j \varepsilon _{t-j}, \quad \varepsilon _{t}\sim \text {IID}(0,\sigma ^2) \end{aligned}$$

with \(\sum _{j=-\infty }^{\infty }\left| \psi _j\right| <\infty \) and \(\sum _{j=-\infty }^{\infty }\left| \psi _j\right| \ne 0\), then \(\overline{X}_n\) is AN\(\left( \mu ,n^{-1}v\right) \), where \(\overline{X}_n=n^{-1}\sum _{t=1}^n X_t\), \(v=\sum _{h=-\infty }^{\infty }\gamma (h)=\sigma ^2\left( \sum _{j=-\infty }^{\infty }\psi _j \right) ^2\), and \(\gamma (\cdot )\) is the autocovariance function of \(\left\{ X_t\right\} _{t=1}^n\).

Lemma 2

(Theorem 8.1, Brockwell and Davis (1991)) If \(\left\{ X_t\right\} _{t=1}^n\) is the zero-mean causal AR(p) process,

$$\begin{aligned} X_t -\phi _1X_{t-1}-\cdots -\phi _pX_{t-p}=Z_t, \quad Z_{t}\sim \text {IID}(0,\sigma ^2) \end{aligned}$$

and \(\hat{\varvec{\phi }}\) is the Yule-Walker estimator of \({\varvec{\phi }}\), that is \(\hat{\varvec{\phi }}={\varvec{\varGamma }}_p^{-1}\hat{\varvec{\gamma }}_p\) with \(\hat{\varvec{\varGamma }}_p=\left\{ {\hat{\gamma }}(i-j)\right\} _{i,j=1}^p\) and \(\hat{\varvec{\gamma }}_p=\left( {\hat{\gamma }}(1),\ldots ,{\hat{\gamma }}(p)\right) ^{\top }\), then

$$\begin{aligned} \sqrt{n}\left( \hat{\varvec{\phi }}-{\varvec{\phi }}\right) \overset{d}{ \rightarrow } N\left( 0, \sigma ^2 {\varvec{\varGamma }}_p^{-1}\right) \end{aligned}$$

where \({\varvec{\varGamma }}_p\) is the covariance matrix with \({\varvec{\varGamma }}_p=\left\{ \gamma (i-j)\right\} _{i,j=1}^p\). Moreover,

$$\begin{aligned} {\hat{\sigma }} ^2 \overset{p}{ \rightarrow } \sigma ^2, \end{aligned}$$

where \({\hat{\sigma }}^2={\hat{\gamma }}_0-\hat{\varvec{\phi }}^{\top }\hat{{\varvec{\gamma }}}_p\).

A.1 Preliminary results on weak convergence

The next weak convergence result extends (1) of the Donsker’s Theorem to strongly mixing time series.

Lemma 3

(Deo 1973) Let \(\{\xi _n:-\infty<n <\infty \}\) be a strictly stationary sequence of random variables, \(\{F_n\left( t\right) : 0\le t\le 1\}\) be the empirical process for \(\xi _1, \xi _2, \ldots , \xi _n\), i.e., \(F_n(t) =n^{-1}\sum _{i=1}^nI_{\left[ 0,t\right] }\left( \xi _i\right) \) where \(I_{\left[ 0,t\right] }\left( \cdot \right) \) is the indicator function of the interval [0, t]. Suppose that \(0\le \xi _0 \le 1\) and \(\xi _0\) have continuous distribution function F with \(F(0)=0\) and \(F(1)=1\). Normalize \(F_n(t)\) as

$$\begin{aligned} Y_n(t)=n^{1/2}\left( F_n\left( t\right) -F\left( t\right) \right) , \quad \quad 0\le t\le 1. \end{aligned}$$

For \(0\le t\le 1\), define the function \(g_t\) by

$$\begin{aligned} g_t(x)=I_{\left[ 0,t\right] }\left( x\right) -F(t), \end{aligned}$$

and suppose further that \(\{\xi _n\}\) satisfies the mixing condition

$$\begin{aligned} \sum _{n=1}^{\infty } n^2\alpha (n)^{1/2-\tau }<\infty \quad \quad for ~ some\quad \tau \in \left( 0,1/2\right) . \end{aligned}$$

Then the sequence \(\{Y_n(t): 0\le t\le 1\}\) of normalized empirical processes converges weakly in \(\mathcal {D}[0,1]\) to a Gaussian random function \(\{Y(t): 0\le t\le 1\}\) specified by \(\mathbb {E}\left( Y\left( t\right) \right) =0\) and

$$\begin{aligned} \begin{aligned} \mathbb {E}\{Y(s)Y(t)\}&=\mathbb {E}\{g_s\left( \xi _0\right) g_t\left( \xi _0\right) \}+\sum _{k=1}^{\infty } \mathbb {E}\{g_s\left( \xi _0\right) g_t\left( \xi _k\right) \}\\&\quad +\sum _{k=1}^{\infty } \mathbb {E}\{g_s\left( \xi _k\right) g_t\left( \xi _0\right) \} \end{aligned} \end{aligned}$$
(14)

Furthermore, the series in (14) converges absolutely and the sample paths of Y are continuous with probability one.

The following lemma yields a uniformly continuous Gaussian limiting process \(\zeta \left( \cdot \right) \) on \(\mathbb {R}\) for the empirical process \(N_{k}^{1/2}\left\{ F_{N_{k}}\left( \cdot \right) -F\left( \cdot \right) \right\} \), which is used in the proof of Theorem 2.

Lemma 4

Under Assumptions (A1) and (A2), there exists a mean-zero Gaussian process \(Y\left( \cdot \right) \) whose sample path is continuous on \(\left[ 0,1\right] \) with probability one such that as \(k\rightarrow \infty \), \(N_{k}^{1/2}\left\{ F_{N_{k}}\left( \cdot \right) -F\left( \cdot \right) \right\} \overset{d}{ \rightarrow }\zeta \left( \cdot \right) =Y\left( F\left( \cdot \right) \right) \). Furthermore, the process \(\zeta \left( \cdot \right) \) is uniformly continuous on \(\mathbb {R}\) with modulus of continuity \(\omega \left( \zeta , \varDelta \right) \le \omega \left( Y, \omega \left( F,\varDelta \right) \right) \rightarrow 0 \ a.s.\ \text { as } \varDelta \rightarrow 0\).

Proof. Define a transformed time series \(u_{i}=F\left( x_{t}\right) \), \(i=0,\pm 1,\pm 2,\ldots \). For any \(x\in \mathbb {R}\), let \(t=F(x)\in [0,1]\), then \(F_{N_{k}}(x)=F_{U,N_{k}}(t)\), in which

$$\begin{aligned} F_{U,N_{k}}(t)=N_{k}^{-1}\sum _{i=1}^{N_{k}}I\left\{ u_{i}\le t\right\} . \end{aligned}$$

The \(\alpha \)-mixing coefficients for \(\left\{ u_{i}\right\} _{i=-\infty }^{\infty }\) is the same as those for \(\left\{ x_{i}\right\} _{i=-\infty }^{\infty }\), which satisfy Assumption (A1) that \(\alpha (n) \ll n^{ -6 -\epsilon }\), hence there exists \(\tau \in \left( 0,1/2\right) \) such that \(\alpha (n)^{1/2-\tau } \ll n^{-3}\), and thus \(\sum _{n=1}^{\infty } n^2\alpha (n)^{1/2-\tau }<\infty \). Then applying Lemma 3 with \(\xi _{i}\) replaced by \(u_{i}\), one has \(N_k^{1/2}\{F_{U,N_{k}}(t)-t\}\rightarrow Y(t)\).

Define \(\zeta (x)=Y\left( F\left( x\right) \right) \), then \(N_{k}^{1/2}\left\{ F_{N_{k}}\left( \cdot \right) -F\left( \cdot \right) \right\} \overset{d}{ \rightarrow }\zeta \left( \cdot \right) \) as \(k\rightarrow \infty \) and

$$\begin{aligned}&\sup _{x, x'\in \mathbb {R}, \left| x-x'\right| \le \varDelta } \Big \vert \zeta \left( x\right) -\zeta \left( x'\right) \Big \vert =\sup _{x, x'\in \mathbb {R}, \left| x-x'\right| \le \varDelta } \Big \vert Y\left( F\left( x\right) \right) -Y\left( F\left( x'\right) \right) \Big \vert \\&\quad \le \sup _{t, t'\in \left[ 0,1 \right] , \left| t-t'\right| \le \omega \left( F,\varDelta \right) } \Big \vert Y\left( t\right) -Y\left( t'\right) \Big \vert \le \omega \left( Y, \omega \left( F,\varDelta \right) \right) \end{aligned}$$

The uniform continuity of \(F(\cdot )\) is guaranteed by Assumption (A2), and almost sure uniform continuity of \(Y(\cdot )\) by the fact that sample paths of \(Y(\cdot )\) are almost surely continuous over the compact interval \(\left[ 0,1\right] \). These facts imply that \(\omega \left( Y, \omega \left( F,\varDelta \right) \right) \rightarrow 0 \ a.s.\ \text { as } \varDelta \rightarrow 0\), thus \(\zeta \) is continuous with probability one and \(\omega \left( \zeta , \varDelta \right) \le \omega \left( Y, \omega \left( F,\varDelta \right) \right) \).

A.2 Proof of Theorem 1

Define a transformed time series \(u_{i}=F\left( x_{i}\right) \), \(i=0,\pm 1,\pm 2,\ldots \) and for any \(k=1,2,\ldots \), a finite population \(\pi _{k,U}=\{u_{1},u_{2},\ldots \text {,}\) \(u_{N_{k}}\}\) together with a simple random sample \(U_{i}=F\left( X_{i}\right) \), \(1\le i\le n_{k}\) from population \(\pi _{k,U}\). For any \(x\in \mathbb {R}\), let \(t=F(x)\in \left[ 0,1 \right] \), then

$$\begin{aligned} F_{N_{k}}(x)=F_{U,N_{k}}(t),F_{n_{k}}(x)=F_{U,n_{k}}(t), \end{aligned}$$
(15)

in which

$$\begin{aligned} F_{U,N_{k}}(t)= & {} N_{k}^{-1}\sum _{i=1}^{N_{k}}I\left\{ u_{i}\le t\right\} , \end{aligned}$$
(16)
$$\begin{aligned} F_{U,n_{k}}(t)= & {} n_{k}^{-1}\sum _{i=1}^{n_{k}}I\left\{ U_{i}\le t\right\} . \end{aligned}$$
(17)

By Assumption (A1), the time series \(\left\{ u_{t},t=0,\pm 1,\pm 2,\ldots \right\} \) is ergodic and has stationary distribution \(\mathcal {U}\left( 0,1\right) \), hence almost surely \(\lim _{k\rightarrow \infty }F_{U,N_{k}}(t)=t\) for \(0\le t\le 1\). As \(\lim _{k\rightarrow \infty }\min \left( n_{k},N_{k}-n_{k}\right) =\infty \) is contained in Assumption (A3), applying Theorem 14.1 of Rosén (1964), one obtains that as random elements taking values in the space \(\mathcal {D}\left[ 0,1\right] \) of cadlag functions:

$$\begin{aligned} \lambda _{k}\left\{ F_{U,n_{k}}(t)-F_{U,N_{k}}(t)\right\} \overset{d}{ \rightarrow }B(t) \end{aligned}$$

almost surely. Lastly, Skorohod’s Representation Theorem (Theorem 6.7, Billingsley 1999) provides versions \(B_{k}^{*}\) of Brownian bridge such that

$$\begin{aligned} \sup _{t\in \left[ 0,1\right] }\left| \lambda _{k}\left\{ F_{U,n_{k}}(t)-F_{U,N_{k}}(t)\right\} -B_{k}^{*}(t)\right| \rightarrow 0,\,a.s. \end{aligned}$$

which implies that

$$\begin{aligned} \sup _{x\in \mathbb {R}}\left| l_{k}\left\{ F_{n_{k}}(x)-F_{N_{k}}(x)\right\} -B_{k}^{*}(F\left( x\right) )\right| \rightarrow 0,\,a.s. \end{aligned}$$

The Theorem 1 is proved.

A.3 Proof of Theorem 2

Lemma 5

Under Assumptions (A1) to (A3), (A5), as \(k \rightarrow \infty \),

$$\begin{aligned} \sup _{w\in \left[ -1,1\right] ,x\in \mathbb {R}}\Big \vert \left\{ F_{n_{k}}\left( x-hw\right) -F_{n_{k}}\left( x\right) \right\} -\left\{ F_{N_{k}}\left( x-hw\right) -F_{N_{k}}(x)\right\} \Big \vert =o_{p}\left( l_{k}^{-1}\right) . \end{aligned}$$

Proof

For the Brownian bridges \(B_{k}^{*}\left\{ \cdot \right\} \) in Theorem 1,

$$\begin{aligned}&\sup _{w\in \left[ -1,1\right] ,x\in \mathbb {R}}\left| l_{k}\left\{ F_{n_{k}}\left( x-hw\right) -F_{N_{k}}\left( x-hw\right) \right\} -l_{k}\left\{ F_{n_{k}}(x)-F_{N_{k}}(x)\right\} \right| \nonumber \\&\quad \le \sup _{x,x^{\prime }\in \mathbb {R},\left| x-x^{\prime }\right| \le h}\left| l_{k}\left\{ F_{n_{k}}\left( x^{\prime }\right) -F_{N_{k}}\left( x^{\prime }\right) \right\} -l_{k}\left\{ F_{n_{k}}(x)-F_{N_{k}}\left( x\right) \right\} \right| \nonumber \\&\quad \le 2\sup _{x}\left| l_{k}\left\{ F_{n_{k}}\left( x\right) -F_{N_{k}}(x)\right\} -B_{k}^{*}\left\{ F(x)\right\} \right| \nonumber \\&\qquad \text { }+\sup _{x,x^{\prime }\in \mathbb {R},\left| x-x^{\prime }\right| \le h}\left| B_{k}^{*}\left\{ F\left( x^{\prime }\right) \right\} -B_{k}^{*}\left\{ F(x)\right\} \right| . \end{aligned}$$
(18)

Since \(F\left( \cdot \right) \) is uniformly continuous by Assumption (A2), and Assumption (A5) implies that \(h \rightarrow 0\) as \(k\rightarrow \infty \), so \(\omega \left( F,h\right) \rightarrow 0\) as \(k\rightarrow \infty \). Assumptions (A1), (A3) ensure Theorem 1, so \(l_{k}\left\{ F_{n_{k}}\left( \cdot \right) -F_{N_{k}}\left( \cdot \right) \right\} -B_{k}^{*}\left\{ F\left( \cdot \right) \right\} \rightarrow 0\ a.s. \) as \(k\rightarrow \infty \). Thus, the expression in (6) is bounded by

$$\begin{aligned}&2\sup _{x\in \mathbb {R}}\left| l_{k}\left\{ F_{n_{k}}(x)-F_{N_{k}}(x)\right\} -B_{k}^{*}\left\{ F(x)\right\} \right| +\sup _{t,t^{\prime }\in \left[ 0,1\right] ,\left| t-t^{\prime }\right| \le \omega \left( F,h\right) }\left| B_{k}^{*}\left( t^{\prime }\right) -B_{k}^{*}\left( t\right) \right| \\&\quad =o_{a.s.}\left( 1\right) +o_{p}\left( 1\right) =o_{p}\left( 1\right) \text {.} \end{aligned}$$

In other words,

$$\begin{aligned} \sup _{w\in \left[ -1,1\right] ,x\in \mathbb {R}}\Big \vert \left\{ F_{n_{k}}\left( x-hw\right) -F_{n_{k}}\left( x\right) \right\} -\left\{ F_{N_{k}}\left( x-hw\right) -F_{N_{k}}(x)\right\} \Big \vert =o_{p}\left( l_{k}^{-1}\right) . \end{aligned}$$
(19)

\(\square \)

Lemma 6

Under Assumptions (A1), (A2), (A5), as \(k \rightarrow \infty \),

$$\begin{aligned} \sup _{w\in \left[ -1,1\right] ,x\in \mathbb {R}}\Big \vert \left\{ F_{N_{k}}\left( x-hw\right) -F_{N_{k}}\left( x\right) \right\} -\left\{ F\left( x-hw\right) -F(x)\right\} \Big \vert =o_{p}\left( N_{k}^{-1/2}\right) . \end{aligned}$$

Proof

Next, since Lemma 4 implies that \(N_{k}^{1/2}\left\{ F_{N_{k}}\left( \cdot \right) -F\left( \cdot \right) \right\} \overset{d}{ \rightarrow }\zeta \left( \cdot \right) \), Skorohod’s Representation Theorem (Theorem 6.7, Billingsley 1999) provides versions \(\zeta _{k}\left( \cdot \right) \) of \(\zeta \left( \cdot \right) \) such that

$$\begin{aligned} \sup _{x\in \mathbb {R} }\left| N_{k}^{1/2}\left\{ F_{N_{k}}\left( x \right) - F\left( x \right) \right\} -\zeta _{k}\left( x \right) \right| \rightarrow 0, \ a.s. \end{aligned}$$

Consequently

$$\begin{aligned}&\sup _{w\in \left[ -1,1\right] ,x\in \mathbb {R}}\left| N_{k}^{1/2}\left\{ F_{N_{k}}\left( x-hw\right) -F\left( x-hw\right) \right\} -N_{k}^{1/2}\left\{ F_{N_{k}}\left( x\right) -F\left( x\right) \right\} \right| \\&\quad \le \sup _{x,x^{\prime }\in \mathbb {R},\left| x-x^{\prime }\right| \le h}\left| N_{k}^{1/2}\left\{ F_{N_{k}}\left( x^{\prime }\right) -F\left( x^{\prime }\right) \right\} -N_{k}^{1/2}\left\{ F_{N_{k}}\left( x\right) -F\left( x\right) \right\} \right| \\&\quad \le 2\sup _{x\in \mathbb {R}}\left| N_{k}^{1/2}\left\{ F_{N_{k}}\left( x\right) -F\left( x\right) \right\} -\zeta _{k}\left( x\right) \right| +\omega \left( \zeta _{k},h\right) =o_{p}\left( 1\right) . \end{aligned}$$

Hence the following holds

$$\begin{aligned} \sup _{w\in \left[ -1,1\right] ,x\in \mathbb {R}}\Big \vert \left\{ F_{N_{k}}\left( x-hw\right) -F_{N_{k}}\left( x\right) \right\} -\left\{ F\left( x-hw\right) -F(x)\right\} \Big \vert =o_{p}\left( N_{k}^{-1/2}\right) . \end{aligned}$$
(20)

Lemma 7

Under the Assumptions (A2), (A5) and (A6), as \(k\rightarrow \infty \),

$$\begin{aligned} \sup _{x\in \mathbb {R}}\left| \int _{-1}^{1}\{F\left( x-hw\right) -F\left( x\right) \} K\left( w\right) dw\right| =o\left( l_k^{-1}\right) . \end{aligned}$$

Proof

According to the assumptions of the cumulative distribution function, we discuss the problem in two cases.

Case 1: \(\nu \ge 1\). Note that by Assumption (A6) \( \int _{-1}^{1}K\left( w\right) w^{r}dw\equiv 0,r=1,\ldots ,l-1\), and by Assumption (A2) \(F(\cdot )\in C^{\left( \nu ,\mu \right) }\left( \mathbb {R} \right) \). Hence

$$\begin{aligned}&\int _{-1}^{1}\left\{ F\left( x-hw\right) -F(x)\right\} K\left( w\right) dw\\&\quad =\int _{-1}^{1}\left\{ F\left( x-hw\right) -\sum _{r=0}^{\nu -1}\frac{ F^{\left( r\right) }(x)}{r!}\left( -hw\right) ^{r}\right\} K\left( w\right) dw\text {{}} \\&\quad =\int _{-1}^{1}\left\{ \int _{x}^{x-hw}\frac{F^{\left( \nu \right) }(t)}{ (\nu -1)!}\left( x-hw-t\right) ^{\nu -1}dt\right\} K\left( w\right) dw \\&\quad =\int _{-1}^{1}\left\{ \frac{F^{\left( \nu \right) }(x)}{(\nu -1)!}\left( -hw\right) ^{\nu }+\int _{x}^{x-hw}\frac{F^{\left( \nu \right) }(t)-F^{\left( \nu \right) }(x)}{(\nu -1)!}\left( x-hw-t\right) ^{\nu -1}dt\right\} K\left( w\right) dw \\&\quad =\int _{-1}^{1}\left\{ \int _{x}^{x-hw}\frac{F^{\left( \nu \right) }(t)-F^{\left( \nu \right) }(x)}{(\nu -1)!}\left( x-hw-t\right) ^{\nu -1}dt\right\} K\left( w\right) dw \end{aligned}$$

Furthermore, by Assumption (A2) \(F^{\left( \nu \right) }(\cdot )\in C^{\left( 0,\mu \right) }\left( \mathbb {R}\right) \) and

$$\begin{aligned}&\sup _{x\in \mathbb {R}}\left| \int _{-1}^{1}\left\{ F\left( x-hw\right) -F\left( x\right) \right\} K\left( w\right) dw\right| \nonumber \\&\quad \le \sup _{x\in \mathbb {R}}\int _{-1}^{1}\left| \int _{x}^{x-hw}\frac{ F^{\left( \nu \right) }(t)-F^{\left( \nu \right) }(x)}{(\nu -1)!}\left( x-hw-t\right) ^{\nu -1}dt\right| K\left( w\right) dw \nonumber \\&\quad \le \sup _{x\in \mathbb {R}}\int _{-1}^{1}\left| (hw)^{\nu }\sup _{x\le t\le x-hw}\frac{\left| F^{\left( \nu \right) }(t)-F^{\left( \nu \right) }(x)\right| }{(\nu -1)!}\right| K\left( w\right) dw \nonumber \\&\quad \le \sup _{x\in \mathbb {R}}\int _{-1}^{1}\left| (hw)^{\nu }\sup _{x\le t\le x-hw}\frac{C\left| t-x\right| ^{\mu }}{(\nu -1)!}\right| K\left( w\right) dw \nonumber \\&\quad \le \sup _{x\in \mathbb {R}}\int _{-1}^{1}\left| (hw)^{\nu }\frac{ C(hw)^{\mu }}{(\nu -1)!}\right| K\left( w\right) dw \nonumber \\&\quad \le \sup _{x\in \mathbb {R}}\int _{-1}^{1}ch^{\nu +\mu }\left| w\right| ^{\nu +\mu }K\left( w\right) dw=\mathcal {O}\left( h^{\nu +\mu }\right) =o\left( l_{k}^{-1}\right) , \end{aligned}$$
(21)

which follows from Assumption (A5) that \(\lim _{k\rightarrow \infty }l_{k}h_{n_{k}}^{\nu +\mu }=0\).

Case 2: \(\nu =0\). By Assumption (A2) \(F(x)\in C^{\left( 0,\mu \right) }\left( \mathbb {R}\right) \). Hence

$$\begin{aligned}&\sup _{x\in \mathbb {R}}\left| \int _{-1}^{1}\{F\left( x-hw\right) -F\left( x\right) \} K\left( w\right) dw\right| \nonumber \\&\quad \le \sup _{x\in \mathbb {R}}\int _{-1}^{1}C\left( hw\right) ^{\mu } K\left( w\right) dw \nonumber \\&\quad =\mathcal {O}\left( h^{\nu +\mu }\right) =o\left( l_{k}^{-1}\right) , \end{aligned}$$
(22)

which follows from Assumption (A5) that \(\lim _{k\rightarrow \infty }l_{k}h_{n_{k}}^{\nu +\mu }=0\). \(\square \)

Proof of Theorem 2

Define \(G(x)=\int _{-\infty }^{x}K\left( u\right) du\). By the definition of \( \hat{F}_{k}(x)\), one obtains

$$\begin{aligned} \hat{F}_{k}(x)=n^{-1}\sum \limits _{i=1}^{n_{k}}\int _{-\infty }^{x}K_{h}\left( u-X_{i}\right) du=n_{k}^{-1}\sum \limits _{i=1}^{n_{k}}G\left( \frac{x-X_{i}}{h }\right) \text {.} \end{aligned}$$

Therefore, by the definition of \(F_{n_{k}}(x)=n_{k}^{-1}\sum _{i=1}^{n_{k}}I \left( X_{i}\le x\right) \) in (4)

$$\begin{aligned} \hat{F}_{k}(x)&=\int _{-\infty }^{+\infty }G\left( \frac{x-u}{h}\right) dF_{n_{k}}\left( u\right) =\int _{-\infty }^{+\infty }h^{-1}K\left( \frac{x-u }{h}\right) F_{n_{k}}\left( u\right) du \\&=\int _{-1}^{1}K\left( w\right) F_{n_{k}}\left( x-hw\right) dw \end{aligned}$$

using integration by parts and a change of variable \(w=\left( x-u\right) /h\) . The following decomposition plays an important role:

$$\begin{aligned} \hat{F}_{k}(x)-F_{n_{k}}(x)=\int _{-1}^{1}\left\{ F_{n_{k}}\left( x-hw\right) -F_{n_{k}}(x)\right\} K\left( w\right) dw. \end{aligned}$$
(23)

Since Assumption (A4) requires that \(n_{k}/N_{k}=o\left( 1\right) \) and consequently \(N_{k}^{-1/2}=o\left( l_{k}^{-1}\right) \), using Lemma 5 and Lemma 6 together with the triangle inequality imply that as \(k\rightarrow \infty \)

$$\begin{aligned} \left| \left\{ F_{n_{k}}\left( x-hw\right) -F_{n_{k}}(x)\right\} -\left\{ F\left( x-hw\right) -F(x)\right\} \right| =u_{p}\left( l_{k}^{-1}\right) . \end{aligned}$$
(24)

By Lemma 7 and applying (23), (24), (21) and (22), the following holds

$$\begin{aligned} \sup _{x\in \mathbb {R}}\left| \hat{F}_{k}(x)-F_{n_{k}}(x)\right| =\sup _{x\in \mathbb {R}}\left| \int _{-1}^{1}\left\{ F_{n_{k}}\left( x-hw\right) -F_{n_{k}}(x)\right\} K\left( w\right) dw\right| =o_{p}\left( l_{k}^{-1}\right) \text {.} \end{aligned}$$

Applying Theorem 1, one has \(l_{k}\left\{ \hat{F} _{k}(x)-F_{N_{k}}(x)\right\} \overset{d}{\rightarrow }B\left\{ F(x)\right\} \) , proving (11).

Notice that under Assumption (A4), \(n_{k}^{-1/2}/l_{k}^{-1}\rightarrow 1\), \( N_{k}^{-1/2}=o\left( l_{k}^{-1}\right) \), \(N_{k}^{-1/2}=o\left( n_{k}^{-1/2}\right) \) as \(k\rightarrow \infty \), and that \( N_{k}^{1/2}\left\{ F_{N_{k}}\left( \cdot \right) -F\left( \cdot \right) \right\} \overset{d}{\rightarrow }\zeta \left( \cdot \right) \) by Lemma . Hence, as \(k\rightarrow \infty \)

$$\begin{aligned} n_{k}^{1/2}D\left( F_{N_{k}},F\right) =n_{k}^{1/2}\mathcal {O}_{p}\left( N_{k}^{-1/2}\right) =\mathcal {O}_{p}\left( n_{k}^{1/2}N_{k}^{-1/2}\right) =o_{p}\left( 1\right) \text {.} \end{aligned}$$

Likewise, \(l_{k}D\left( F_{N_{k}},F\right) =o_{p}\left( 1\right) \). These, together with (11) establish (12). The proof of Theorem 2 is complete by applying Slutsky’s Theorem.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Li, J., Wang, J. & Yang, L. Kolmogorov–Smirnov simultaneous confidence bands for time series distribution function. Comput Stat 37, 1015–1039 (2022). https://doi.org/10.1007/s00180-021-01149-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00180-021-01149-5

Keywords

  • Bandwidth
  • Brownian bridge
  • Kernel
  • Kolmogorov distribution
  • Simple random sample
  • Stationarity