Abstract
We estimate the kernel function of a symmetric alpha stable (\(S\alpha S\)) moving average random function which is observed on a regular grid of points. The proposed estimator relies on the empirical normalized (smoothed) periodogram. It is shown to be weakly consistent for positive definite kernel functions, when the grid mesh size tends to zero and at the same time the observation horizon tends to infinity (high-frequency observations). A simulation study shows that the estimator performs well at finite sample sizes, when the integrator measure of the moving average random function is \(S\alpha S\) and for some other infinitely divisible integrators.
Similar content being viewed by others
References
Akhiezer, N. I. (1988). Lectures on integral transforms, volume 70 of translations of mathematical monographs. Providence, RI: American Mathematical Society. Translated from the Russian by H. H. McFaden.
Brockwell, P. J. (2014). Recent results in the theory and applications of CARMA processes. Annals of the Institute of Statistical Mathematics, 66(4), 647–685.
Brockwell, P. J., Ferrazzano, V., Klüppelberg, C. (2013). High-frequency sampling and kernel estimation for continuous-time moving average processes. Journal of Time Series Analysis, 34(3), 385–404.
Brockwell, P. J., Lindner, A. (2009). Existence and uniqueness of stationary Lévy-driven CARMA processes. Stochastic Processes and their Applications, 119(8), 2660–2681.
Cambanis, S., Podgórski, K., Weron, A. (1995). Chaotic behavior of infinitely divisible processes. Studia Mathematica, 115(2), 109–127.
Chèn, K. (2011). Estimation of the parameters of \(\alpha \)-stable distributions. Proceedings of the National Academy of Sciences of Belarus. Physics and Mathematics Series, (1), 75–81, 127.
Fan, Z. (2006). Parameter estimation of stable distributions. Communications in Statistics. Theory and Methods, 35(1–3), 245–255.
Fasen, V., Fuchs, F. (2013a). On the limit behavior of the periodogram of high-frequency sampled stable CARMA processes. Stochastic Processes and their Applications, 123(1), 229–273.
Fasen, V., Fuchs, F. (2013b). Spectral estimates for high-frequency sampled continuous-time autoregressive moving average processes. Journal of Time Series Analysis, 34(5), 532–551.
Feller, W. (1966). An introduction to probability theory and its applications, Vol. II. New York, London, Sydney: Wiley.
García, I., Klüppelberg, C., Müller, G. (2011). Estimation of stable CARMA models with an application to electricity spot prices. Statistical Modelling. An International Journal, 11(5), 447–470.
Gu, J., Mao, S. S. (2002). Estimation of the parameter of stable distributions. Chinese Journal of Applied Probability and Statistics, 18(4), 342–346.
Hesse, C. H. (1990). A Bahadur-type representation for empirical quantiles of a large class of stationary, possibly infinite-variance, linear processes. The Annals of Statistics, 18(3), 1188–1202.
Hida, T., Hitsuda, M. (1993). Gaussian processes, volume 120 of translations of mathematical monographs. Providence, RI: American Mathematical Society. Translated from the 1976 Japanese original by the authors.
Janczura, J., Orzeł, S., Wyłomańska, A. (2011). Subordinated \(\alpha \)-stable Ornstein–Uhlenbeck process as a tool for financial data description. Physica A, 390, 4379–4387.
Kampf, J., Shevchenko, G., Spodarev, E. (2019). Nonparametric estimation of the kernel function of symmetric stable moving average random functions (preprint). arXiv:1706.06289.
Karcher, W. (2012). On infinitely divisible random fields with an application in insurance. Ph.D. thesis, Ulm University, Ulm, Germany.
Karcher, W., Scheffler, H.-P., Spodarev, E. (2009). Efficient simulation of stable random fields and its applications. In V. Capasso et al. (Eds.), Stereology and image analysis. Ecs10: Proceedings of the 10th European congress of ISS, the MIRIAM project series (pp. 63–72). ESCULAPIO Pub. Co., Bologna, Italy.
Karcher, W., Scheffler, H.-P., Spodarev, E. (2013). Simulation of infinitely divisible random fields. Communications in Statistics. Simulation and Computation, 42(1), 215–246.
Karcher, W., Spodarev, E. (2012). Kernel function estimation of stable moving average random fields. Ulm University, Ulm (preprint). http://www.uni-ulm.de/fileadmin/website_uni_ulm/mawi.inst.110/forschung/preprints/kernel_estimation_stable.pdf.
Koblents, E., Míguez, J., Rodríguez, M. A., Schmidt, A. M. (2016). A nonlinear population Monte Carlo scheme for the Bayesian estimation of parameters of \(\alpha \)-stable distributions. Computational Statistics & Data Analysis, 95, 57–74.
Koutrouvelis, I. A. (1980). Regression-type estimation of the parameters of stable laws. Journal of the American Statistical Association, 75(372), 918–928.
Koutrouvelis, I. A. (1981). An iterative procedure for the estimation of the parameters of stable laws. Communications in Statistics. B. Simulation and Computation, 10(1), 17–28.
Kulik, R. (2007). Bahadur–Kiefer theory for sample quantiles of weakly dependent linear processes. Bernoulli. Official Journal of the Bernoulli Society for Mathematical Statistics and Probability, 13(4), 1071–1090.
Lantuéjoul, C. (2002). Geostatistical simulation: Models and algorithms, 1st ed. Berlin, Heidelberg: Springer.
McCulloch, J. H. (1986). Simple consistent estimators of stable distribution parameters. Communications in Statistics. B. Simulation and Computation, 15(4), 1109–1136.
Mikosch, T., Gadrich, T., Klüppelberg, C., Adler, R. J. (1995). Parameter estimation for ARMA models with infinite variance innovations. The Annals of Statistics, 23(1), 305–326.
Müller, G., Seibert, A. (2019). Bayesian estimation of stable carma spot models for electricity prices. Energy Economics, 78, 267–277.
Rosiński, J. (1994). On uniqueness of the spectral representation of stable processes. Journal of Theoretical Probability, 7(3), 615–634.
Samorodnitsky, G., Taqqu, M. S. (1994). Stable non-Gaussian random processes: Stochastic models with infinite variance. Stochastic Modeling. New York: Chapman & Hall.
Sato, K. -I. (2013). Lévy processes and infinitely divisible distributions, volume 68 of Cambridge studies in advanced mathematics. Cambridge: Cambridge University Press. Translated from the 1990 Japanese original, Revised edition of the 1999 English translation.
Trigub, R. M., Bellinsky, E. S. (2004). Fourier analysis and approximation of functions. Dordrecht: Kluwer Academic Publishers.
Wang, Y., Yang, W., Hu, S. (2016). The Bahadur representation of sample quantiles for weakly dependent sequences. Stochastics, 88(3), 428–436.
Wu, W. B. (2005). On the Bahadur representation of sample quantiles for dependent sequences. The Annals of Statistics, 33(4), 1934–1963.
Zolotarev, V. M. (1986). One-dimensional stable distributions. Translations of Mathematical Monographs—Vol 65. AMS.
Zolotarev, V. M., Uchaikin, V. V. (1999). Chance and stability. Stable distributions and their applications. Modern probability and statistics. Berlin: Walter de Gruyter.
Acknowledgements
We thank I. Liflyand and V.P. Zastavnyi for the discussion on positive definite functions and their Fourier transforms. We are also grateful to our students L. Palianytsia, O. Stelmakh and B. Ströh for doing numerical experiments in Sect. 5. Finally, we express our gratitude to M. Wendler for drawing our attention to paper Hesse (1990).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix 1: Proofs of Theorems 1, 3
Proof of Theorem 1
We first show how (ii) follows from (i). Notice that \(\left| {{\hat{g}}}(\lambda )\right| = {{\hat{g}}}(\lambda )\) for all \(\lambda \in {\mathbb {R}}\), since by (F1) f is of positive type. Since \(|\sqrt{a}-\sqrt{b}| \le \sqrt{ |a-b|}\) for \(a, b\ge 0\), we get, using the Cauchy–Schwarz inequality,
Since \({\hat{g}}\in L^2({{\mathbb {R}}})\), we then have
The desired statement now follows from the Plancherel’s equality.
Now let us prove (i). Write
where \( J_{n,X}^s(\lambda ) = \sum _{\left| m\right| \le m_n} W_n(m)\left| \sum _{j=1}^{n} X(t_{j,n})\hbox {e}^{it_{j,n}\nu _n(m,\lambda )}\right| ^2,\quad S_{n,X} = {\sum _{j=1}^{n}X(t_{j,n})^2}. \)
Let f be supported by \([-T,T]\). We will assume that \(N = T/\varDelta _n\) is integer: this will simplify the exposition while not harming the rigor. The proof is rather long, so we split it into several steps for better readability. Choose \(n\ge 2N+1\).
Step 1. Denominator
We start with investigating the denominator \(S_{n,X} \). First, we study the behavior of a similar expression with f replaced by its discretized version. Specifically, define
where \(f_n(x) = \sum _{k=-N}^{N-1}f(t_{k,n}){\mathbb {1}}_{[t_{k,n},t_{k+1,n})}(x)\). Denote \(\varepsilon _{l,n} = \varLambda \left( ((l-1)\varDelta _n,l\varDelta _n]\right) \), \(l\in {\mathbb {Z}}\). For fixed n, these variables are independent \(S\alpha S\) with scale parameter \(\varDelta _n^{1/\alpha }\).
Decompose
We are going to show that the last three terms are negligible. We use the shorthand \(E_n = \sum _{l=N+1}^{n-N} \varepsilon _{l,n}^2\), as this will be our benchmark term. Observe that \(S_{1,n} = \sum _{k=-N}^{N-1} f(t_{k,n})^2 E_n\). Thanks to the boundedness and uniform continuity of f,
On the other hand, by Feller (1966, XVII.5, Theorem 3 (i)), we have
where \(Z_\alpha \) is some positive \(\alpha /2\)-stable random variable. Therefore, by Slutsky’s theorem,
Estimating
similarly to (19), we get
Since \(N\varDelta _n=T\), we have
The term \(S_{4,n}\) is estimated using Lemma 3: \( S_{4,n} = O_P(N^{3/2} n^{2/\alpha -1/2}\varDelta _n^{2/\alpha }) = O_P((n\varDelta _n)^{-1/2}S_{1,n}) . \)
Summing up, we have \(\sum _{j=1}^{n} X_{n}(t_{j,n})^2 = S_{1,n}\left( 1+O_P((n\varDelta _n)^{-1/2})\right) \), \(n\rightarrow \infty \), and \(S_{1,n}\) is of order \(n^{2/\alpha }\varDelta _n^{2/\alpha -1}\), in the sense of (19).
Now we get back to the denominator of \(I_{n,X}(\lambda )\). For any positive vanishing sequence \(\left\{ \delta _n,n\ge 1\right\}, \) write the following simple estimate:
Then, we obtain
From Lemma 4 it follows that
Putting \(\delta _n = \omega _f(\varDelta _n)\) and using (17), we conclude that
Step 2. Whole expression
Thanks to (22),
Thus, it remains to prove that
Step 3. Numerator
As with the denominator, we start with examining the discretized version of \(J_{n,X}^s(\lambda )\):
We proceed in three substeps, first considering the following expression
Step 3a). We shall show
We have for \(\lambda \in [-a_n,a_n]\) that
where
With the help of Lemma 6, we obtain
By Lemma 2,
where, by Lemma 7,
with
Hence,
Combining the estimates, we get (24).
Step 3b). We get
Indeed, write
Let us estimate the first expression. Take some positive vanishing sequence \(\left\{ \theta _n,n\ge 1\right\} \), which will be specified later. Using (21), we have
where
Hence,
Now
where
Using Lemma 2, we obtain
Further, using (20), \( R_{4,n}\sim \varDelta _n^{-2} \Vert f\Vert _1^2\sum _{l=2-N}^{N} \varepsilon _{l,n}^2= O_P(\varDelta _n^{-2}),\ n\rightarrow \infty , \) hence
Similarly, \(\int _{-a_n}^{a_n}R_{3,n}(\lambda )^2 \, \hbox {d}\lambda = O_P(a_n\varDelta _n^{-4})\), \(n\rightarrow \infty \).
Setting \(\theta _n = a_n^{1/4}(n\varDelta _n)^{-1/\alpha }\), we get by (A3) that
as \(n\rightarrow \infty \). Therefore, we arrive at
Noting that \(o_P(n^{4/\alpha }\varDelta _n^{4/\alpha }) = o_P(E_n^2)\), \(n\rightarrow \infty \), and by Step 3a)
we get by (A3) \( a_n\int _{-a_n}^{a_n}\left| \varDelta _n^2 R_n(\lambda ) - \varDelta _n^2 R_{1,n}(\lambda )\right| ^2\hbox {d}\lambda = o_P(E_n^2),\quad n\rightarrow \infty , \) whence (25) follows from (24).
Step 3c). Finally, we have
Using (21) again, write
for \(\delta _n = a_n^{1/4}\omega _f(\varDelta _n)\). Hence,
Define \( h_{n,m}(s,\lambda ) = \sum _{j=1}^n \left( f_n(t_{j,n}-s)- f(t_{j,n}-s)\right) \hbox {e}^{it_{j,n}\nu _{n}(m,\lambda )} {\mathbb {1}}_{[-a_n,a_n]}(\lambda ). \) Note that the summands do not exceed \(\omega _f(\varDelta _n)\), and at most 2N of them are not zero. Hence, \(\Vert h_{n,m}(\cdot ,\lambda )\Vert _\infty \le 2N\omega _f(\varDelta _n)\). Applying Lemma 5, we get
Recalling that \(\delta _n\rightarrow 0\) and \(a_n^{3/2}\omega _f(\varDelta _n)^2\rightarrow 0\) as \(n\rightarrow \infty \) and using (25), we ultimately obtain
Combining (25) and (26), we come to (23).\(\square \)
Proof of Theorem 3
Consider first the case \(\alpha \in [1,2)\). (i) Using the triangle inequality and the Hölder inequality, we get
(ii) Similarly to (i),
in view of (8).
The proof of (iii) uses the same ideas and is based on (9).
For \(\alpha \in (0,1)\), the proof goes in a similar manner through the triangle inequality for \(\Vert \cdot \Vert _\alpha ^\alpha \). \(\square \)
Appendix 2: Auxiliary statements
Lemma 1
Let \((E,{\mathcal {E}},\nu )\) be a \(\sigma \)-finite measure space, \(\varLambda \) be an independently scattered S\(\alpha \)S random measure on E with the control measure \(\nu \), and \(\{f_t, t\in {\mathbf {T}}\}\subset L^\alpha (E,{\mathcal {E}},\nu )\) be a family of functions indexed by some parameter set \({\mathbf {T}}\), \(\varphi \) be a positive probability density on E. Then
has the same finite-dimensional distributions as the almost surely convergent series
where \(\left\{ \zeta _k,k\ge 1\right\} \) are iid standard Gaussian random variables, \(\left\{ \xi _k,k\ge 1\right\} \) are iid random elements of E with density \(\varphi \), \(\varGamma _k=\eta _1+\dots +\eta _k\), \(\left\{ \eta _k,k\ge 1\right\} \) are iid \({\text {Exp}}(1)\)-distributed random variables, and these three sequences are independent;
Proof
The statement follows from Samorodnitsky and Taqqu (1994, Section 3.11) by noting that
where M is an independently scattered S\(\alpha \)S random measure on E defined by
so that the control measure of M has \(\nu \)-density \(\varphi \).\(\square \)
Lemma 2
Let, for each \(n\ge 1\), \(\left\{ \varepsilon _{m,n},m=1,\dots ,n\right\} \) be iid \(S\alpha S\) random variables with scale parameter \(\sigma _n\). Let also \(\left\{ a_{j,l,n},\ 1\le j<l\le n\right\} \) be a collection of measurable functions \(a_{j,l,n}:{{\mathbb {R}}}\rightarrow {\mathbb {C}}\) such that
Then
Proof
W.l.o.g. we can assume that \(\sigma _n = 1\). We shall use the LePage series representation. For each \(n\ge 1\), the variables \(\left\{ \varepsilon _{m,n},m=1,\dots ,n\right\} \) have the same joint distribution as \(\left\{ \varLambda ([m-1,m]),m=1,\dots ,n\right\} \), where \(\varLambda \) is an independently scattered S\(\alpha \)S random measure on [0, n] with the Lebesgue control measure. By Lemma 1, this distribution coincides with that of
where \(\left\{ \varGamma _k,k\ge 1\right\} \) and \(\left\{ \zeta _k,k\ge 1\right\} \) are as in Lemma 1, \(\left\{ \xi _k,k\ge 1\right\} \) are iid U([0, n]). Since the boundedness in probability relies only on marginal distributions (for fixed n), we can assume that \(\varepsilon _{m,n} = {\tilde{\varepsilon }}_{m,n}\). Let \(\varXi _n(\lambda ) = \sum _{1\le j<l\le n} a_{j,l,n}(\lambda )\varepsilon _{j,n}\varepsilon _{l,n}\). A generic term in the expansion of \(|\varXi _n(\lambda )|^2\) has, up to a non-random constant, the form
Recall that \(\left\{ \zeta _k,k\ge 1\right\} \) are independent and centered. Then, given \(\varGamma \)’s and \(\xi \)’s, such term has a non-zero expectation only if \(k_1=k_2\), \(k_1'=k_2'\) or \(k_1=k_2'\), \(k_2=k_1'\) (for \(k_1=k_1'\) it is zero since \(j_1\ne l_1\)), so we must also have \(j_1=j_2\), \(l_1=l_2\) or \(j_1 = l_2\), \(j_2 =l_1\) respectively so that the product of indicators is not zero. The latter, however, is impossible, since \(j_1<l_1\) and \(j_2<l_2\). Consequently, the lemma of Fatou implies
Integrating over \(\lambda \), we get
By the strong law of large numbers, \(\varGamma _k \sim k\), \(k\rightarrow \infty \), a.s. Therefore, given \(\varGamma \)’s, \(\int _{{{\mathbb {R}}}}|\varXi _n(\lambda )|^2\hbox {d}\lambda = O_P(A_n n^{4/\alpha -2})\), \(n\rightarrow \infty \), whence the required statement follows.\(\square \)
The following lemma is an immediate corollary of the proof of Lemma 2.
Lemma 3
Let, for each \(n\ge 1\), \(\left\{ \varepsilon _{m,n},m=1,\dots ,n\right\} \) be iid \(S\alpha S\) random variables with scale parameter \(\sigma _n\). Let also \(\left\{ b_{j,l,n},\ 1\le j<l\le n\right\} \) be a set of complex numbers with
Then
In the next two lemmas, \(\left\{ \varDelta _n,n\ge 1\right\} \) is some vanishing sequence, \(\left\{ N_n,n\ge 1\right\} \) is a sequence of positive integers such that \(N_n\rightarrow \infty \), \(n\rightarrow \infty \), and \(N_n = o(n)\), \(n\rightarrow \infty \). We denote \(t_{k,n} = k\varDelta _n\), \(k\in {{\mathbb {Z}}}\), \(T_n = N_n\varDelta _n\), \(n\ge 1\). The proofs of these lemmas are similar to the proof of Lemma 2 and thus omitted. They can be found in the arXiv version of the present paper Kampf et al. (2019).
Lemma 4
Let \(\left\{ h_n,n\ge 1\right\} \) be a sequence of compactly supported bounded functions such that the bounds of both the function values and the support are uniform in n. Then
Lemma 5
Let \(\left\{ m_n,n\ge 1\right\} \) be a sequence of positive integers such that \(m_n\rightarrow \infty \) as \(n\rightarrow \infty \). For a deterministic sequence \(\left\{ W_n(m),n\ge 1, m=-m_n,\dots ,m_n\right\} \) satisfying (W1)–(W2) and continuous functions \(h_{n,m}:[-T_n,n\varDelta _n+T_n]\times {{\mathbb {R}}}\rightarrow {\mathbb {C}}\), \(n\ge 1, m= -m_n,\dots ,m_n\), define
Then
where \(H_n^*=\int _{\mathbb {R}} H(\lambda )\, {\mathrm{d}}\lambda \) for \(H(\lambda ) = \sum _{\left| m\right| \le m_n} W_n(m)\Vert h_{n,m}(\cdot ,\lambda )\Vert _\infty ^4.\)
Lemma 6
Let a bounded uniformly continuous function \(f:{{\mathbb {R}}}\rightarrow {\mathbb {R}}\) with compact support \([-T, T]\) and let \(\varDelta _n\), \(m_n\), \(W_n(m)\) and \(\nu _n(m,\lambda )\) be as defined in Sect. 1or 3fulfilling (W1), (W2) and (W4). Choose a sequence of integers \((N_n)_{n\in {\mathbb {N}}}\) with \(N_n\cdot \varDelta _n\sim T\). Put
Then
We also omit the proof of this and the following lemma and refer the interested reader to Kampf et al. (2019).
Lemma 7
Let \(\left\{ m_n,n\ge 1\right\} \) be a sequence of positive integers such that \(m_n\rightarrow \infty \), \(m_n = o(n)\), \(n\rightarrow \infty \), and let \(\left\{ K_n(m),n\ge 1, m=-m_n,\dots ,m_n\right\} \) be a sequence in \({\mathbb {R}}\), and let \(\left\{ W_n(m),n\ge 1, m=-m_n,\dots ,m_n\right\} \) be a sequence of filters satisfying (W1)–(W2). Then
with \(W_n^* = \max _{\left| m\right| \le m_n} W_n(m)\), \(K_n^* = \max _{\left| m\right| \le m_n} |K_n(m)|\).
About this article
Cite this article
Kampf, J., Shevchenko, G. & Spodarev, E. Nonparametric estimation of the kernel function of symmetric stable moving average random functions. Ann Inst Stat Math 73, 337–367 (2021). https://doi.org/10.1007/s10463-020-00751-6
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10463-020-00751-6