Abstract
This paper investigates second order properties of a stationary continuous time process after random sampling. While a short memory process always gives rise to a short memory one, we prove that long-memory can disappear when the sampling law has very heavy tails. Despite the fact that the normality of the process is not maintained by random sampling, the normalized partial sum process converges to the fractional Brownian motion, at least when the long memory parameter is preserved.
Similar content being viewed by others
References
Adorf HM (1995) Interpolation of irregularly sampled data series—a survey. In: Astronomical society of the pacific conference series, vol 77
Beran J, Feng Y, Ghosh S, Kulik R (2013) Long-memory processes. Probabilistic properties and statistical methods. Springer, Heidelberg
Bingham NH, Goldie CM, Teugels JL (1989) Regular variation. In: Encyclopedia of mathematics and its applications, vol 27, Cambridge University Press, Cambridge
Brockwell PJ, Davis RA, Yang Y (2007) Continuous-time Gaussian autoregression. Stat Sin 17(1):63–80
Broersen PM (2007) Time series models for spectral analysis of irregular data far beyond the mean data rate. Meas Sci Technol 19(1):015103
Chambers MJ (1996) The estimation of continuous parameter long-memory time series models. Econom Theory 12(2):374–390
Comte F (1996) Simulation and estimation of long memory continuous time models. J Time Ser Anal 17(1):19–36
Comte F, Renault E (1996) Long memory continuous time models. J Econom 73(1):101–149
Davydov YA (1970) The invariance principle for stationary processes. Theory Probab Appl 15:487–498
Duffie D, Glynn P (2004) Estimation of continuous-time markov processes sampled at random time intervals. Econometrica 72:1773–1808
Elorrieta F, Eyheramendy S, Palma W (2019) Discrete-time autoregressive model for unequally spaced time-series observations. Astron Astrophys 627:A120
Eyheramendy S, Elorrieta F, Palma W (2018) An irregular discrete time series model to identify residuals with autocorrelation in astronomical light curves. Mon Not R Astron Soc 481:4311–4322
Feller W (1966) An introduction to probability theory and its applications, vol 2. Wiley, New York
Friedman M (1962) The interpolation of time series by related series. J Am Stat Assoc 57(300):729–757
Giraitis L, Koul HL, Surgailis D (2012) Large sample inference for long memory processes. Imperial College Press, London
Jones RH (1981) Fitting a continuous time autoregression to discrete data. In: Applied time series analysis II, Elsevier, pp 651–682
Jones RH, Tryon PV (1987) Continuous time series models for unequally spaced data applied to modeling atomic clocks. SIAM J Sci Stat Comput 8(1):71–81
Li D, Robinson PM, Shang HL (2019) Long-range dependent curve time series. J Am Stat Assoc. https://doi.org/10.1080/01621459.2019.1604362
Mandelbrot BB, Wallis JR (1969) Robustness of the rescaled range r/s in the measurement of noncyclic long run statistical dependence. Water Resour Res 5:967–988
Masry E, Lui M-C (1975) A consistent estimate of the spectrum by random sampling of the time series. SIAM J Appl Math 28(4):793–810
Masry K-SLE (1994) Spectral estimation of continuous-time stationary processes from random sampling. Stoch Process Appl 52:39–64
Mayo WT (1978) Spectrum measurements with laser velocimeters. In: Hansen BW (ed) Proceedings of the dynamic flow conference 1978 on dynamic measurements in unsteady flows, Springer, Dordrecht, Netherlands, pp 851–868
Mykland YAPA (2003) The effects of random and discrete sampling when estimating continuous-time diffusions. Econometrica 71:483–549
Nieto-Barajas LE, Sinha T (2014) Bayesian interpolation of unequally spaced time series. Stoch Environ Res Risk Assess 29:577–587
Philippe A, Viano M-C (2010) Random sampling of long-memory stationary processes. J Stat Plan Inference 140(5):1110–1124
Scargle JD (1982) Studies in astronomical time series analysis. ii—statistical aspects of spectral analysis of unevenly spaced data. Astrophys J 263:835–853
Shi X, Wu Y, Liu Y (2010) A note on asymptotic approximations of inverse moments of nonnegative random variables. Stat Probab Lett 80(15–16):1260–1264
Stout WF (1974) Almost sure convergence. In: Probability and mathematical statistics, vol 24, Academic Press, New York
Taqqu MS (1975) Weak convergence to fractional Brownian motion and to the Rosenblatt process. Z Wahrscheinlichkeitstheorie Verw Gebiete 31:287–302
Tsai H (2009) On continuous-time autoregressive fractionally integrated moving average processes. Bernoulli 15(1):178–194
Tsai H, Chan KS (2005a) Maximum likelihood estimation of linear continuous time long memory processes with discrete time data. J R Stat Soc Ser B Stat Methodol 67(5):703–716
Tsai H, Chan KS (2005b) Quasi-maximum likelihood estimation for a class of continuous-time long-memory processes. J Time Ser Anal 26(5):691–713
Viano M-C, Deniau C, Oppenheim G (1994) Continuous-time fractional ARMA processes. Stat Probab Lett 21(4):323–336
Acknowledgements
We thank the Associate Editor and the referees for valuable comments that led to an improved this article.
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.
Appendix
Appendix
To prove Lemma 4.1, we need the following intermediate result:
Lemma 5.1
If \({\mathbb {E}}[T_1]<\infty \) and \(\mathbf {X}\) has a regularly varying covariance function
with \(0<d<1/2\) and L slowly varying at infinity and ultimately non-increasing. Then,
Proof
By Theorem 3.1, we have \( {\mathbb {E}}[\sigma _X(T_h)] \underset{h\rightarrow \infty }{\sim }L(h)(h{\mathbb {E}}[T_1])^{-1+2d}\). To get the result, it is enough to prove that
To prove the asymptotic behavior of \({\mathbb {E}}[\sigma _X(T_h)^2]\), we will follow a similar proof as theorem 3.1:
-
Let \(0<c<{\mathbb {E}}[T_1]\), and \(h\in {\mathbb {N}}\) such that \(ch\ge 1\),
$$\begin{aligned} {\mathbb {E}}[\sigma _X(T_h)^2]\ge & {} {\mathbb {E}}\left[ \sigma _X(T_h)^2\,\mathbb {I}_{T_h>ch}\right] \ge {\mathbb {E}}\left[ L(T_h)^2T_h^{-2+4d}\,\mathbb {I}_{T_h> ch}\right] \\\ge & {} \inf _{t>ch}\{L(t)^2t^{4d}\}{\mathbb {E}}\left[ \frac{\,\mathbb {I}_{T_h> ch}}{T_h^2}\right] . \end{aligned}$$Thanks to Jensen and Hölder inequalities,
$$\begin{aligned} {\mathbb {E}}\left[ \frac{\,\mathbb {I}_{T_h> ch}}{T_h^2}\right] \ge {\mathbb {E}}\left[ \frac{\,\mathbb {I}_{T_h> ch}}{T_h}\right] ^2 \text { and } P(T_h> ch)^2 \le {\mathbb {E}}[T_h] {\mathbb {E}}\left[ \frac{\,\mathbb {I}_{T_h> ch}}{T_h}\right] , \end{aligned}$$that is
$$\begin{aligned} {\mathbb {E}}\left[ \frac{\,\mathbb {I}_{T_h> ch}}{T_h^2}\right] \ge \frac{P(T_h> ch)^4}{{\mathbb {E}}[T_h]^2}. \end{aligned}$$Summarizing,
$$\begin{aligned} \frac{{\mathbb {E}}[\sigma _X(T_h)^2] }{ L(h)^2(h{\mathbb {E}}[T_1])^{-2+4d}}\ge \frac{\inf _{t>ch}\{L(t)^2t^{4d}\}}{L(h)^2h^{4d}{\mathbb {E}}[T_1]^{4d}}P(T_h> ch)^4. \end{aligned}$$(5.2)Then, for \(c<{\mathbb {E}}[T_1]\), we have \(P(T_{h}>ch)\rightarrow 1\) and \(\inf _{t>ch}\{L(t)^2t^{4d}\}\sim L(ch)^2(ch)^{4d}\). Finally, for all \(c<{\mathbb {E}}[T_1]\),
$$\begin{aligned} \liminf _{h\rightarrow \infty }\frac{{\mathbb {E}}[\sigma _X(T_h)^2] }{ L(h)^2(h{\mathbb {E}}[T_1])^{-2+4d}}\ge \left( \frac{c}{{\mathbb {E}}[T_1]}\right) ^{4d}. \end{aligned}$$Taking the limit as \(c\rightarrow {\mathbb {E}}[T_1]\), we get
$$\begin{aligned} \liminf _{h\rightarrow \infty }\frac{{\mathbb {E}}[\sigma _X(T_h)^2] }{ L(h)^2(h{\mathbb {E}}[T_1])^{-2+4d}}\ge 1. \end{aligned}$$ -
Let \(\frac{1}{2}<s<\tau <1\), \(t_0\) such that L(.) is non-increasing and positive on \([t_0,\infty )\) and h such that \(\mu _{h,s}-\mu _{h,s}^\tau \ge t_0\), with the same notation as Theorem 3.1,
$$\begin{aligned} {\mathbb {E}}[\sigma _X(T_h)^2]= & {} {\mathbb {E}}\left[ L(T_h)^2T_h^{-2+4d} \,\mathbb {I}_{T_{h,s}\ge \mu _{h,s}-\mu _{h,s}^\tau }\right] +{\mathbb {E}}\left[ \sigma (T_h)^2\,\mathbb {I}_{T_{h,s}<\mu _{h,s}-\mu _{h,s}^\tau }\right] \\\le & {} L(\mu _{h,s}-\mu _{h,s}^\tau )^2\left( \mu _{h,s}-\mu _{h,s} ^\tau \right) ^{-2+4d}+\sigma _X(0)^2 P \left( T_{h,s}<\mu _{h,s}-\mu _{h,s}^\tau \right) . \end{aligned}$$We get
$$\begin{aligned} \frac{{\mathbb {E}}[\sigma _X(T_h)^2] }{ L(h)^2(h{\mathbb {E}}[T_1])^{-2+4d}}&\le \left( \frac{L(\mu _{h,s}-\mu _{h,s}^\tau )}{L(h)}\right) ^{2} \left( \frac{\mu _{h,s}-\mu _{h,s}^\tau }{h{\mathbb {E}}[T_1]}\right) ^{-2+4d}\\ {}&\quad +\sigma _X(0)^2 \frac{P \left( T_{h,s}<\mu _{h,s}-\mu _{h,s}^\tau \right) }{L(h)^2(h{\mathbb {E}}[T_1])^{-2+4d}}, \end{aligned}$$and finally
$$\begin{aligned} \limsup _{h\rightarrow \infty }\frac{{\mathbb {E}}[\sigma _X(T_h)^2] }{ L(h)^2(h{\mathbb {E}}[T_1])^{-2+4d}}\le 1. \end{aligned}$$
\(\square \)
Proof of Lemma 4.1:
Denote
We want to prove that \(W_n\) converges in probability to \(\gamma _d\). To do this, we will show that \({\mathbb {E}}[W_n]\xrightarrow [n\rightarrow \infty ]{} \gamma _d\) and \({\mathrm {Var}}(W_n)\xrightarrow [n\rightarrow \infty ]{} 0\).
-
As \(\mathbf {X}\) is a centered process \(E[W_n]=L(n)^{-1}n^{-1-2d}{\mathrm {Var}}(Y_1+\dots +Y_n)\). By Theorem 3.1, we have
$$\begin{aligned} \sigma _Y(h) \sim L(h)(h{\mathbb {E}}[T_1])^{-1+2d} \qquad h\rightarrow \infty , \end{aligned}$$then
$$\begin{aligned} L(n)^{-1}n^{-1-2d}{\mathrm {Var}}(Y_1+\dots +Y_n)\xrightarrow [n\rightarrow \infty ]{} \gamma _d, \end{aligned}$$(5.3)[see Giraitis et al. (2012) Proposition 3.3.1, page 43]. Therefore we obtain
$$\begin{aligned} E[W_n]\xrightarrow [n\rightarrow \infty ]{} \gamma _d . \end{aligned}$$ -
Furthermore,
$$\begin{aligned} {\mathrm {Var}}(W_n)&=L(n)^{-2}n^{-2-4d}{\mathrm {Var}}\left( \sum _{i=1}^n\sum _{j=1}^n \sigma _X(T_j-T_i)\right) \\&\le L(n)^{-2}n^{-2-4d}\left( \sum _{i=1}^n\sum _{j=1}^n \sqrt{{\mathrm {Var}}(\sigma _X(T_j-T_i))}\right) ^2\\&=\left( 2n^{-1-2d}L(n)^{-1} \sum _{h=1}^n(n-h)\sqrt{{\mathrm {Var}}(\sigma _X(T_h))}\right) ^2. \end{aligned}$$Then, by Lemma 5.1, \(\sqrt{{\mathrm {Var}}(\sigma _X(T_h))}=\circ (L(h)h^{-1+2d})\) and \( 2 \sum _{h=1}^n (n-h)L(h)h^{-1+2d}\sim \frac{L(n)n^{1+2d}}{d(1+2d)}\). We get
$$\begin{aligned} 2 \sum _{h=1}^n(n-h)\sqrt{{\mathrm {Var}}(\sigma _X(T_h))}=\circ (L(n)n^{1+2d}). \end{aligned}$$Finally, \({\mathrm {Var}}(W_n)=\circ (1)\) which means that \({\mathrm {Var}}(W_n)\xrightarrow [n\rightarrow \infty ]{} 0\). We obtain
$$\begin{aligned} W_{n}\xrightarrow [n\rightarrow \infty ]{L^2,~ p} \gamma _d. \end{aligned}$$
Rights and permissions
About this article
Cite this article
Philippe, A., Robet, C. & Viano, MC. Random discretization of stationary continuous time processes. Metrika 84, 375–400 (2021). https://doi.org/10.1007/s00184-020-00783-1
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00184-020-00783-1