Abstract
In this paper, we concentrate on the two-dimensional \(\alpha \)-stable autoregressive model of order 1. For this model, we analyze the auto-dependence functions applied to describe the interdependence of the time series components considered separately as the one-dimensional processes. Since the classical second moment-based dependence measures are not defined in the \(\alpha -\)stable case, we examine here the auto-codifference and the auto-covariation functions. The main result of this paper is the derivation of the formulas which describe the asymptotics of the considered measures of interdependence. The theoretical results are supported by a simulation study.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Ansley, C.F.: Computation of the theoretical autocovariance function for a vector ARMA process. J. Stat. Comput. Simul. 12(1), 15–24 (1980)
Ansley, C.F., Kohn, R.: A note on reparameterizing a vector autoregressive moving average model to enforce stationarity. J. Stat. Comput. Simul. 24(2), 99–106 (1986)
Mainassara, Y.B.: Selection of weak VARMA models by modified Akaike’s information criteria. J. Time Ser. Anal. 33, 121–130 (2012)
Brockwell, P.J., Davis, R.A.: Introduction to Time Series and Forecasting. Springer, New York (2002). https://doi.org/10.1007/978-3-319-29854-2
Cambanis, S., Miller, G.: Linear problems in pth order and stable processes. SIAM J. Appl. Math. 41(1), 43–69 (1981)
Chan, H., Chinipardaz, R., Cox, T.: Discrimination of AR, MA and ARMA time series models. Commun. Stat. - Theor. Meth. 25(6), 1247–1260 (1996)
de Silva, B.M.: A class of multivariate symmetric stable distributions. J. Multivar. Anal. 8(3), 335–345 (1978)
Fama, E.F.: The behavior of stock-market prices. J. Bus. 38(1), 34–105 (1965)
Gallagher, C.M.: A method for fitting stable autoregressive models using the autocovariation function. Stat. Probab. Lett. 53, 381–390 (2001)
Grzesiek, A., Wyłomańska, A.: Asymptotic behavior of the cross-dependence measures for bidimensional AR(1) model with \(\alpha -\)stable noise. Accepted in Banach Center Publications 2019 (2019). https://arxiv.org/abs/1911.10894
Grzesiek, A., Teuerle, M., Wyłomańska, A.: Cross-codifference for bidimensional VAR(1) models with infinite variance. Communications in Statistics - Simulation and Computation published online (2019). 27. Located at: arXiv:1902.02142
Grzesiek, A., Teuerle, M., Sikora, G., Wyłomańska, A.: Spatial-temporal dependence measures for \(\alpha -\)stable bivariate AR(1). J. Time Ser. Anal. 41, 454–475 (2020)
Hong-Zhi, A., Zhao-Guo, C., Hannan, E.J.: A note on ARMA estimation. J. Time Ser. Anal. 4(1), 9–17 (1983)
Jabłońska-Sabuka, M., Teuerle, M., Wyłomańska, A.: Bivariate sub-Gaussian model for stock index returns. Physica A: Stat. Mech. Appl. 486, 628–637 (2017)
Kozubowski, T.J., Panorska, A.K.: Multivariate geometric stable distributions in financial applications. Math. Comput. Modell. 29(10–12), 83–92 (1999)
Kozubowski, T.J., Panorska, A.K., Rachev, S.T.: Statistical issues in modeling multivariate stable portfolios. In: Rachev, S.T. (ed.) Handbook of Heavy Tailed Distributions in Finance, volume 1 of Handbooks in Finance, pp. 131–167. North-Holland, Amsterdam (2003)
Kruczek, P., Wyłomańska, A., Teuerle, M., Gajda, J.: The modified Yule-Walker method for alpha-stable time series models. Physica A 469, 588–603 (2017)
Lii, K.-S., Rosenblatt, M.: An approximate maximum likelihood estimation for non-gaussian non-minimum phase moving average processes. J. Multivar. Anal. 43(2), 272–299 (1992)
Luetkepohl, H.: Forecasting cointegrated VARMA processes. Humboldt Universitaet Berlin, Sonderforschungsbereich 373 (2007)
Maejima, M., Yamamoto, K.: Long-memory stable Ornstein-Uhlenbeck processes. Electron. J. Probab. 8(19), 1–18 (2003)
Mandelbrot, B.: The variation of certain speculative prices. J. Bus. 36(4), 394–419 (1963)
Mauricio, J.A.: Exact maximum likelihood estimation of stationary vector ARMA models. J. Am. Stat. Assoc. 90(429), 282–291 (1995)
McKenzie, E.: A note on the derivation of theoretical autocovariances for ARMA models. J. Stat. Comput. Simul. 24, 159–162 (1986)
Miller, G.: Properties of certain symmetric stable distributions. J. Multivar. Anal. 8(3), 346–360 (1978)
Mittnik, S., Rachev, S.T.: Alternative multivariate stable distributions and their applications to financial modeling. In: Cambanis S., Samorodnitsky G., Taqqu M.S. (eds.) Stable Processes and Related Topics, vol. 25 of Progress in Probabilty. Birkhäuser Boston (1991). https://doi.org/10.1007/978-1-4684-6778-9_6
Niglio, M., Vitale, C.D.: Threshold vector ARMA models. Commun. Stat. - Theor. Meth. 44(14), 2911–2923 (2015)
Nolan, J.P., Panorska, A.K.: Data analysis for heavy tailed multivariate samples. Commun. Stat. Stoch. Mod. 13(4), 687–702 (1997)
Nowicka, J.: Asymptotic behavior of the covariation and the codifference for ARMA models with stable innovations. Commun. Stat. Stoch. Mod. 13(4), 673–685 (1997)
Nowicka, J., Weron, A.: Measures of dependence for ARMA models with stable innovations. Annales Universitatis Mariae Curie-Skłodowska. Sectio A – Mathematica, LI 1, 14, 133–144 (1997)
Nowicka, J., Wyłomańska, A.: The dependence structure for PARMA models with a-stable innovations. Acta Physica Polonica B 37(11), 3071–3081 (2006)
Nowicka-Zagrajek, J., Wyłomańska, A.: Measures of dependence for stable AR(1) models with time-varying coefficients. Stoch. Mod. 24(1), 58–70 (2008)
Peters, G., Sisson, S., Fan, Y.: Likelihood-free Bayesian inference for alpha-stable models. Comput. Stat. Data Anal. 56(11), 3743–3756 (2012)
Press, S.: Multivariate stable distributions. J. Multivar. Anal. 2(4), 444–462 (1972)
Rosadi, D.: Testing for independence in heavy-tailed time series using the codifference function. Comput. Stat. Data Anal. 53, 4516–4529 (2009)
Rosadi, D., Deistler, M.: Estimating the codifference function of linear time series models with infinite variance. Metrika: Int. J. Theor. Appl. Stat. 73(3), 395–429 (2011)
Samorodnitsky, G., Taqqu, M.S.: Stable Non-Gaussian Random Processes: Stochastic Models with Infinite Variance. Chapman and Hall, New York (1994)
Stoyanov, S.V., Samorodnitsky, G., Rachev, S., Ortobelli, S.: Computing the portfolio conditional Value-at-Risk in the alpha-stable case. Probab. Math. Stat. 26, 1–22 (2006)
Tsai, H., Chan, K.S.: A note on non-negative ARMA processes. J. Time Ser. Anal. 28, 350–360 (2007)
Weir, A.J.: Lebesgue Integration and Measure. Cambridge University Press, Cambridge (1973)
Weron, A.: Stable processes and measures; a survey. In: Szynal, D., Weron, A. (eds.) Probability Theory on Vector Spaces III. pp, pp. 306–364. Springer, Heidelberg (1984). https://doi.org/10.1007/BFb0099806
Williams, K.S.: The nth power of a \(2 \times 2\) matrix. Math. Mag. 65(5), 336 (1992)
Wyłomańska, A., Chechkin, A., Sokolov, I., Gajda, J.: Codifference as a practical tool to measure interdependence. Physica A 421, 412–429 (2015)
Żak, G., Obuchowski, J., Wyłomańska, A., Zimroz, R.: Application of ARMA modelling and alpha-stable distribution for local damage detection in bearings. Diagnostyka 15, 01 (2014)
Żak, G., Wyłomańska, A., Zimroz, R.: Data driven iterative vibration signal enhancement strategy using alpha stable distribution. Shock Vib. 2017, 11 (2017)
Żak, G., Wyłomańska, A., Zimroz, R.: Periodically impulsive behaviour detection in noisy observation based on generalised fractional order dependency map. Appl. Acoust. 144, 31–39 (2019)
Zolotarev, V.M.: One-dimensional stable distributions. Translations of Mathematical Monographs. Providence: American Mathematical Society (1986)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendices
Appendix A:
Lemma 1
Let \(\{\mathbf {X}(t)\}=\{X_1(t),X_2(t)\}\) with \(t\in {\mathbb {Z}}\) be the bounded solution of Eq. (3) given by Eq. (5).
-
1.
For two different eigenvalues of \(\varTheta \) indicated as \(\lambda _1\) and \(\lambda _2\), \(|\lambda _1|<1\) and \(|\lambda _2|<1\), let us introduce the following notation
$$\begin{aligned}\begin{aligned} A_1=A_1(s_1,s_2,a_1,a_2,\lambda _1,\lambda _2,j)&=\frac{\lambda _1^j(\lambda _2s_1-a_1s_1-a_2s_2)}{\lambda _2-\lambda _1},\\ B_1=B_1(s_1,s_2,a_1,a_2,\lambda _1,\lambda _2,j)&=\frac{\lambda _2^j(-\lambda _1s_1+a_1s_1+a_2s_2)}{\lambda _2-\lambda _1},\\ C_1=C_1(s_1,s_2,a_1,a_2,\lambda _1,\lambda _2,j)&=\frac{\lambda _1^j(-a_2s_2+\lambda _2s_1-a_1s_1)+\lambda _2^j(a_2s_2-\lambda _1s_1+a_1s_1)}{\lambda _2-\lambda _1}. \end{aligned}\end{aligned}$$Then, for \(h \in {\mathbb {N}}_{0}\) we obtain that
-
(a)
for \(0<\alpha <2\)
$$\begin{aligned}&\mathrm {CD}(X_1(t),X_1(t-h))=\mathrm {CD}(X_1(t),X_1(t+h)) \nonumber \\&={\sum _{j=0}^{+\infty }\int _{S_2}\left( |\lambda _1^h{A_1}+\lambda _2^h{B_1}|^{\alpha }+|C_1|^{\alpha }-|C_1-(\lambda _1^h{A_1}+\lambda _2^h{B_1})|^\alpha \right) \varGamma (ds)}, \end{aligned}$$(13) -
(b)
for \(1<\alpha <2\)
$$\begin{aligned} \mathrm {CV}&(X_1(t),X_1(t-h))=\sum _{j=0}^{+\infty }\int _{S_2}C_1^{\langle \alpha -1 \rangle }\left( \lambda _1^hA_1+\lambda _2^hB_1\right) \varGamma (ds), \end{aligned}$$(14)and
$$\begin{aligned} \mathrm {CV}&(X_1(t),X_1(t+h))=\sum _{j=0}^{+\infty }\int _{S_2}C_1\left( \lambda _1^hA_1+\lambda _2^hB_1\right) ^{\langle \alpha -1 \rangle }\varGamma (ds). \end{aligned}$$(15)
-
(a)
-
2.
For equal eigenvalues of \(\varTheta \) indicated as \(\lambda _1=\lambda _2=\lambda \), \(|\lambda |<1\), let us introduce the following notation
$$\begin{aligned}\begin{aligned} A_2=A_2(s_1,s_2,a_1,a_2,\lambda ,j)&=j\lambda ^{j-1}a_1s_1-j\lambda ^js_1+\lambda ^js_1+j\lambda ^{j-1}a_2s_2,\\ B_2=B_2(s_1,s_2,a_1,a_2,\lambda ,j)&=\lambda ^{j-1}a_1s_1-\lambda ^js_1+\lambda ^{j-1}a_2s_2,\\ C_2=C_2(s_1,s_2,a_1,a_2,\lambda ,j)&=j\lambda ^{j-1}a_1s_1+j\lambda ^{j-1}a_2s_2-j\lambda ^js_1+\lambda ^js_1. \end{aligned} \end{aligned}$$Then, for \(h \in {\mathbb {N}}_{0}\) we obtain that
-
(a)
for \(0<\alpha <2\)
$$\begin{aligned}&\mathrm {CD}(X_1(t),X_1(t-h))=\mathrm {CD}(X_1(t),X_1(t+h)) \nonumber \\&={\sum _{j=0}^{+\infty }\int _{S_2}\left( |\lambda ^h{A_2}+h\lambda ^h{B_2}|^{\alpha }+|C_2|^{\alpha }-|C_2-(\lambda ^h{A_2}+h\lambda ^h{B_2})|^\alpha \right) \varGamma (ds)}, \end{aligned}$$(16) -
(b)
for \(1<\alpha <2\)
$$\begin{aligned} \mathrm {CV}&(X_1(t),X_1(t-h))=\sum _{j=0}^{+\infty }\int _{S_2}C_2^{\langle \alpha -1 \rangle }\left( \lambda ^hA_2+h\lambda _2^hB_2\right) \varGamma (ds), \end{aligned}$$(17)and
$$\begin{aligned} \mathrm {CV}&(X_1(t),X_1(t+h))=\sum _{j=0}^{+\infty }\int _{S_2}C_2\left( \lambda ^hA_2+h\lambda ^hB_2\right) ^{\langle \alpha -1 \rangle }\varGamma (ds). \end{aligned}$$(18)
-
(a)
Proof
The proof is analogous to the ones presented in the authors’ previous papers, see Grzesiek et al. (2019) for the codifference function and Grzesiek et al. (2020) for the covariation function. We also use the formulas given in Eqs. (7–8).
Lemma 2
Let \(\{\mathbf {X}(t)\}=\{X_1(t),X_2(t)\}\) with \(t\in {\mathbb {Z}}\) be the bounded solution of Eq. (3) given by Eq. (5).
-
1.
For two different eigenvalues of \(\varTheta \) indicated as \(\lambda _1\) and \(\lambda _2\), \(|\lambda _1|<1\) and \(|\lambda _2|<1\), let us introduce the following notation
$$\begin{aligned} \begin{aligned} A_3=A_3(s_1,s_2,a_3,a_4,\lambda _1,\lambda _2,j)&=\frac{\lambda _1^j(\lambda _2s_2-a_3s_1-a_4s_2)}{\lambda _2-\lambda _1},\\ B_3=B_3(s_1,s_2,a_3,a_4,\lambda _1,\lambda _2,j)&=\frac{\lambda _2^j(-\lambda _1s_2+a_3s_1+a_4s_2)}{\lambda _2-\lambda _1},\\ C_3=C_3(s_1,s_2,a_3,a_4,\lambda _1,\lambda _2,j)&=\frac{\lambda _1^j(-a_3s_1+\lambda _2s_2-a_4s_2)+\lambda _2^j(a_3s_1-\lambda _1s_2+a_4s_2)}{\lambda _2-\lambda _1}.\\ \end{aligned}\end{aligned}$$Then, for \(h \in {\mathbb {N}}_{0}\) we obtain that
-
(a)
for \(0<\alpha <2\)
$$\begin{aligned}&\mathrm {CD}(X_2(t),X_2(t-h))=\mathrm {CD}(X_2(t),X_2(t+h)) \nonumber \\&={\sum _{j=0}^{+\infty }\int _{S_2}\left( |\lambda _1^h{A_3}+\lambda _2^h{B_3}|^{\alpha }+|C_3|^{\alpha }-|C_3-(\lambda _1^h{A_3}+\lambda _2^h{B_3})|^\alpha \right) \varGamma (ds)}, \end{aligned}$$(19) -
(b)
for \(1<\alpha <2\)
$$\begin{aligned} \mathrm {CV}&(X_2(t),X_2(t-h))=\sum _{j=0}^{+\infty }\int _{S_2}C_3^{\langle \alpha -1 \rangle }\left( \lambda _1^hA_3+\lambda _2^hB_3\right) \varGamma (ds), \end{aligned}$$(20)and
$$\begin{aligned} \mathrm {CV}&(X_2(t),X_2(t+h))=\sum _{j=0}^{+\infty }\int _{S_2}C_3\left( \lambda _1^hA_3+\lambda _2^hB_3\right) ^{\langle \alpha -1 \rangle }\varGamma (ds). \end{aligned}$$(21)
-
(a)
-
2.
For equal eigenvalues of \(\varTheta \) indicated as \(\lambda _1=\lambda _2=\lambda \), \(|\lambda |<1\), let us introduce the following notation
$$\begin{aligned}\begin{aligned} A_4=A_4(s_1,s_2,a_3,a_4,\lambda ,j)=&j\lambda ^{j-1}a_3s_1-j\lambda ^js_2+\lambda ^js_2+j\lambda ^{j-1}a_4s_2,\\ B_4=B_4(s_1,s_2,a_3,a_4,\lambda ,j)=&\lambda ^{j-1}a_3s_1-\lambda ^{j}s_2+\lambda ^{j-1}a_4s_2,\\ C_4=C_4(s_1,s_2,a_3,a_4,\lambda ,j)=&j\lambda ^{j-1}a_3s_1+j\lambda ^{j-1}a_4s_2-j\lambda ^{j}s_2+\lambda ^js_2. \end{aligned}\end{aligned}$$Then, for \(h \in {\mathbb {N}}_{0}\) we obtain that
-
(a)
for \(0<\alpha <2\)
$$\begin{aligned}&\mathrm {CD}(X_2(t),X_2(t-h))=\mathrm {CD}(X_2(t),X_2(t+h)) \nonumber \\&={\sum _{j=0}^{+\infty }\int _{S_2}\left( |\lambda ^h{A_4}+h\lambda ^h{B_4}|^{\alpha }+|C_4|^{\alpha }-|C_4-(\lambda ^h{A_4}+h\lambda ^h{B_4})|^\alpha \right) \varGamma (ds)}, \end{aligned}$$(22) -
(b)
for \(1<\alpha <2\)
$$\begin{aligned} \mathrm {CV}&(X_2(t),X_2(t-h))=\sum _{j=0}^{+\infty }\int _{S_2}C_4^{\langle \alpha -1 \rangle }\left( \lambda ^hA_4+h\lambda _2^hB_4\right) \varGamma (ds), \end{aligned}$$(23)and
$$\begin{aligned} \mathrm {CV}&(X_2(t),X_2(t+h))=\sum _{j=0}^{+\infty }\int _{S_2}C_4\left( \lambda ^hA_4+h\lambda ^hB_4\right) ^{\langle \alpha -1 \rangle }\varGamma (ds). \end{aligned}$$(24)
-
(a)
Proof
The proof is analogous to the ones presented in the authors’ previous papers, see Grzesiek et al. (2019) for the codifference function and Grzesiek et al. (2020) for the covariation function. We also use the formulas given in Eqs. (7–8).
Appendix B:
Proof
-
A)
Let us consider the auto-codifference function of \(\{X_1(t)\}\) given in Lemma 1.
-
For \(\lambda _1 \ne \lambda _2\), and \(|\lambda _1|<1\), \(|\lambda _2|<1\), we examine the auto-codifference given in Eq. (13).
-
I)
Let us assume that \(|\lambda _1|>|\lambda _2|\). In this case, one can show that
$$\begin{aligned} \begin{aligned}&\lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{\left| \lambda _1^hA_1+\lambda _2^hB_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-(\lambda _1^hA_1+\lambda _2^hB_1)\right| ^\alpha }{\lambda _1^h}\varGamma (ds)\\{\mathop {=}\limits ^{(\star )}}&\sum _{j=0}^{+\infty }\lim _{h \rightarrow +\infty }\int _{S_2}\frac{\left| \lambda _1^hA_1+\lambda _2^hB_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-(\lambda _1^hA_1+\lambda _2^hB_1)\right| ^\alpha }{\lambda _1^h}\varGamma (ds)\\{\mathop {=}\limits ^{(\star \star )}}&\sum _{j=0}^{+\infty }\int _{S_2}\lim _{h \rightarrow +\infty }\frac{\left| \lambda _1^hA_1+\lambda _2^hB_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-(\lambda _1^hA_1+\lambda _2^hB_1)\right| ^\alpha }{\lambda _1^h}\varGamma (ds). \end{aligned} \end{aligned}$$(25)From the dominated convergence theorem Weir (1973), let us notice that \((\star )\) holds if the sum over \(j \in {\mathbb {N}}_{0}\) in Eq. (25) converges uniformly. Now, from the below inequalities
$$\begin{aligned} \begin{aligned} \left| |a|^\alpha +|b|^\alpha -|a+b|^\alpha \right|&\le (\alpha +1)|a|^\alpha +\alpha |a||b|^{\alpha -1},\\ \left| a+b\right| ^\alpha&\le 2^{\alpha -1}\left( |a|^\alpha +|b|^\alpha \right) , \end{aligned} \end{aligned}$$(26)satisfied for \(a,b \in {\mathrm{I{\!}\mathrm R}}\) and \(1<\alpha <2\), see Maejima and Yamamoto (2003), for all \(j \in {\mathbb {N}}_{0}\) we can show that
$$\begin{aligned}&\left| \int _{S_2}\frac{\left| \lambda _1^hA_1+\lambda _2^hB_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-(\lambda _1^hA_1+\lambda _2^hB_1)\right| ^\alpha }{\lambda _1^h}\varGamma (ds)\right| \\&\qquad \qquad \le \,2^{\alpha -1}(\alpha +1)\left( \int _{S_2}\left| A_1\right| ^\alpha \varGamma (ds)+\int _{S_2}\left| B_1\right| ^{\alpha }\varGamma (ds)\right) \\&\qquad \qquad \quad \,\,\,\,+\, \alpha \left( \int _{S_2}\left| A_1\right| |C_1|^{\alpha -1}\varGamma (ds)+\int _{S_2}\left| B_1\right| |C_1|^{\alpha -1}\varGamma (ds)\right) =M_j, \end{aligned}$$which means that each component of the infinite sum is bounded by an expression which does not depend on h. As a consequence, the sum over \(j \in {\mathbb {N}}_{0}\) given in Eq. (25) converges uniformly if the sum of \(M_j\) over \(j \in {\mathbb {N}}_{0}\) is finite, i.e. if the below conditions are satisfied
$$\begin{aligned} \begin{aligned}&\sum _{j=0}^{+\infty }\int _{S_2}\left| A_1\right| ^\alpha \varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| A_1\right| |C_1|^{\alpha -1}\varGamma (ds)<+\infty ,\\&\sum _{j=0}^{+\infty }\int _{S_2}\left| B_1\right| ^{\alpha }\varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| B_1\right| |C_1|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned} \end{aligned}$$(27)which is true (see Remark 1 in Appendix D). Now, to prove that \((\star \star )\) in Eq. (25) holds we again apply the dominated convergence theorem. From the inequalities in Eq. (26), for the integrand given in Eq. (25) we have that
$$\begin{aligned}&\left| \frac{\left| \lambda _1^hA_1+\lambda _2^hB_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-(\lambda _1^hA_1+\lambda _2^hB_1)\right| ^\alpha }{\lambda _1^h}\right| \\&\qquad \le \,2^{\alpha -1}(\alpha +1)\left( \left| A_1\right| ^\alpha +\left| B_1\right| ^{\alpha }\right) +\alpha \left( \left| A_1\right| |C_1|^{\alpha -1}+\left| B_1\right| |C_1|^{\alpha -1}\right) . \end{aligned}$$for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\). Let us notice that the dominating function does not depend on h. Moreover, it is integrable if for all \(j \in {\mathbb {N}}_0\) the below conditions are true
$$\begin{aligned} \begin{aligned}&\int _{S_2}\left| A_1\right| ^\alpha \varGamma (ds)<+\infty , \quad \int _{S_2}\left| A_1\right| |C_1|^{\alpha -1}\varGamma (ds)<+\infty ,\\&\int _{S_2}\left| B_1\right| ^{\alpha }\varGamma (ds)<+\infty , \quad \int _{S_2}\left| B_1\right| |C_1|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned} \end{aligned}$$which is satisfied (see Remark 1 in Appendix D). Then, let us notice that for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\) and \(h \rightarrow +\infty \) we have
$$\begin{aligned}&\left| \lambda _1^h{A_1}+\lambda _2^h{B_1}\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-\left( \lambda _1^h{A_1}+\lambda _2^h{B_1}\right) \right| ^\alpha \\&=\left| \lambda _1^h\left( {A_1}+\left( {\lambda _2}/{\lambda _1}\right) ^h{B_1}\right) \right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-\lambda _1^h\left( {A_1}+\left( \lambda _2/\lambda _1\right) ^h{B_1}\right) \right| ^\alpha \\&\qquad \qquad \qquad \qquad \qquad \qquad \qquad \sim \left| \lambda _1^h\right| ^{\alpha }\left| A_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-\lambda _1^hA_1\right| ^\alpha \end{aligned}$$and the following limit holds
$$\begin{aligned} \lim _{x \rightarrow 0}\frac{|dx|^\alpha +|c|^\alpha -|c-dx|^\alpha }{x}=\alpha \, dc^{\langle \alpha -1\rangle } \quad \mathrm {for\ 1<\alpha <2,\ d,c \in {\mathrm{I{\!}\mathrm R}}}. \end{aligned}$$(28)Moreover, we have \(\lambda _1^h \rightarrow 0\) for \(h \rightarrow +\infty \). Finally we can write that
$$\begin{aligned}&\lim _{h \rightarrow +\infty }\frac{\left| \lambda _1^hA_1+\lambda _2^hB_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-\left( \lambda _1^hA_1+\lambda _2^hB_1\right) \right| ^\alpha }{\lambda _1^h} \nonumber \\&\qquad \quad =\lim _{h \rightarrow +\infty }\frac{\left| \lambda _1^hA_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-\lambda _1^hA_1\right| ^\alpha }{\lambda _1^h}=\alpha \, A_1C_1^{\langle \alpha -1\rangle } \end{aligned}$$(29)for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\). Finally, from Eq. (25) we finally have that
$$\begin{aligned}&\lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{\left| \lambda _1^hA_1+\lambda _2^hB_1\right| ^{\alpha }+\left| C_1\right| ^{\alpha }-\left| C_1-(\lambda _1^hA_1+\lambda _2^hB_1)\right| ^\alpha }{\lambda _1^h}\varGamma (ds) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \quad =\alpha D_1, \end{aligned}$$(30)which is equivalent to
$$\begin{aligned}&\mathrm{{CD}}(X_1(t),X_1(t-h))=\mathrm{{CD}}(X_1(t),X_1(t+h)) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha D_1\,\lambda _1^h\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(31)where
$$\begin{aligned} D_1:=\sum _{j=0}^{+\infty }\int _{S_2} A_1C_1^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(32) -
II)
Let us assume that \(|\lambda _1|<|\lambda _2|\). Proceeding similarly as above we have
$$\begin{aligned}&\mathrm{{CD}}(X_1(t),X_1(t-h))=\mathrm{{CD}}(X_1(t),X_1(t+h))\nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha D_2\,\lambda _2^h\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(33)where
$$\begin{aligned} D_2:=\sum _{j=0}^{+\infty }\int _{S_2}B_1C_1^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(34) -
III)
Let us assume that \(\lambda _1=-\lambda _2\) and even h. Similarly as above, we obtain
$$\begin{aligned}&\mathrm{{CD}}(X_1(t),X_1(t-h))=\mathrm{{CD}}(X_1(t),X_1(t+h)) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha (D_1+D_2)\,\lambda _1^h\ \ \text { for}\ h \rightarrow +\infty , \end{aligned}$$(35)where
$$\begin{aligned} D_1+D_2=\sum _{j=0}^{+\infty }\int _{S_2}(A_1+B_1)C_1^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$ -
IV)
Let us assume that \(\lambda _1=-\lambda _2\) and h is odd. Similarly as above, we have
$$\begin{aligned}&\mathrm{{CD}}(X_1(t),X_1(t-h))=\mathrm{{CD}}(X_1(t),X_1(t+h)) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha (D_1-D_2)\,\lambda _1^h\ \text { for}\ h \rightarrow +\infty , \end{aligned}$$(36)where
$$\begin{aligned} D_1-D_2=\sum _{j=0}^{+\infty }\int _{S_2}(A_1-B_1)C_1^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$
-
I)
-
For \(\lambda _1=\lambda _2=\lambda \), and \(|\lambda |<1\), we examine the auto-codifference function given in Eq. (16). Similarly to previous case, we show that
$$\begin{aligned} \begin{aligned}&\lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{\left| \lambda ^hA_2+h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-(\lambda ^hA_2+h\lambda ^hB_2)\right| ^\alpha }{h\lambda ^h}\varGamma (ds)\\ {}&{\mathop {=}\limits ^{(\star )}}\sum _{j=0}^{+\infty }\lim _{h \rightarrow +\infty }\int _{S_2}\frac{\left| \lambda ^hA_2+h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-(\lambda ^hA_2+h\lambda ^hB_2)\right| ^\alpha }{h\lambda ^h}\varGamma (ds)\\ {}&{\mathop {=}\limits ^{(\star \star )}}\sum _{j=0}^{+\infty }\int _{S_2}\lim _{h \rightarrow +\infty }\frac{\left| \lambda ^hA_2+h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-(\lambda ^hA_2+h\lambda ^hB_2)\right| ^\alpha }{h\lambda ^h}\varGamma (ds). \end{aligned} \end{aligned}$$(37)From the dominated convergence theorem, to justify \((\star )\) one has to prove that the sum over \(j \in {\mathbb {N}}_0\) in Eq. (37) converges uniformly. From the inequalities in Eq. (26) we have that
$$\begin{aligned}&\left| \int _{S_2}\frac{\left| \lambda ^hA_2+h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-(\lambda ^hA_2+h\lambda ^hB_2)\right| ^\alpha }{h\lambda ^h}\varGamma (ds)\right| \\&\qquad \qquad \le \,2^{\alpha -1}(\alpha +1)M\left( \int _{S_2}\left| A_2\right| ^\alpha \varGamma (ds)+\int _{S_2}\left| B_2\right| ^{\alpha }\varGamma (ds)\right) \\&\qquad \qquad \qquad +\alpha \left( \int _{S_2}\left| A_2\right| |C_2|^{\alpha -1}\varGamma (ds)+\int _{S_2}\left| B_2\right| |C_2|^{\alpha -1}\varGamma (ds)\right) =N_j, \end{aligned}$$for all \(j \in {\mathbb {N}}_0\), where M is the boundary of \(\{h\lambda ^h\}\) over \({h\in {\mathbb {N}}_{0}}\). Let us notice that \(N_j\) does not depend on h. Consequently, the sum over \({j\in {\mathbb {N}}_{0}}\) in Eq. (37) converges uniformly if the sum of \(N_j\) over \(j \in {\mathbb {N}}_0\) converges, i.e. if the below sums are finite
$$\begin{aligned} \begin{aligned}&\sum _{j=0}^{+\infty }\int _{S_2}\left| A_2\right| ^\alpha \varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| A_2\right| |C_2|^{\alpha -1}\varGamma (ds)<+\infty ,\\&\sum _{j=0}^{+\infty }\int _{S_2}\left| B_2\right| ^{\alpha }\varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| B_2\right| |C_2|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned}\end{aligned}$$(38)which is true (see Remark 1 in Appendix D). As the second step, to prove that \((\star \star )\) in Eq. (37) holds we again use the dominated convergence theorem. Namely, from the inequalities given in Eq. (26), for the integrand in Eq. (37) we obtain that
$$\begin{aligned}&\left| \frac{\left| \lambda ^hA_2+h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-(\lambda ^hA_2+h\lambda ^hB_2)\right| ^\alpha }{h\lambda ^h}\right| \nonumber \\&\, \le \,2^{\alpha -1}(\alpha +1)M\left( \left| A_2\right| ^\alpha +\left| B_2\right| ^{\alpha }\right) +\alpha \left( \left| A_2\right| |C_2|^{\alpha -1}+\left| B_2\right| |C_2|^{\alpha -1}\right) , \end{aligned}$$(39)for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\). Let us notice that the dominating function does not depend on h. Moreover, it is integrable if for all \(j \in {\mathbb {N}}_{0}\) the below integrals are finite
$$\begin{aligned} \begin{aligned}&\int _{S_2}\left| A_2\right| ^\alpha \varGamma (ds)<+\infty , \quad \int _{S_2}\left| A_2\right| |C_2|^{\alpha -1}\varGamma (ds)<+\infty ,\\&\int _{S_2}\left| B_2\right| ^{\alpha }\varGamma (ds)<+\infty , \quad \int _{S_2}\left| B_2\right| |C_2|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned}\end{aligned}$$which is true (see Remark 1 in Appendix D). Then, let us notice that for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\) and \(h \rightarrow \infty \) we have
$$\begin{aligned}&\left| \lambda ^h{A_2}+h\lambda ^h{B_2}\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-\left( \lambda ^h{A_2}+h\lambda ^h{B_2}\right) \right| ^\alpha \\&\qquad \,\, =\left| h\lambda ^h\left( {A_2/h}+{B_2}\right) \right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-h\lambda ^h\left( {A_2/h}+{B_2}\right) \right| ^\alpha \\&\qquad \qquad \qquad \qquad \qquad \qquad \,\,\, \sim \left| h\lambda ^h\right| ^{\alpha }\left| B_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-h\lambda ^hB_2\right| ^\alpha . \end{aligned}$$From the limit in Eq. (28) and since \(h\lambda ^h \rightarrow 0\) for \(h \rightarrow +\infty \), we have that
$$\begin{aligned}&\lim _{h \rightarrow +\infty }\frac{\left| \lambda ^hA+h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-\left( \lambda ^hA+h\lambda ^hB_2\right) \right| ^\alpha }{h\lambda ^h}\\&\qquad \qquad \qquad \quad =\lim _{h \rightarrow +\infty }\frac{\left| h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-h\lambda ^hB_2\right| ^\alpha }{h\lambda ^h}=\alpha \, B_2C_2^{\langle \alpha -1\rangle } \end{aligned}$$for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\). Finally, from Eq. (37) we have
$$\begin{aligned}&\lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{\left| \lambda ^hA_2+h\lambda ^hB_2\right| ^{\alpha }+\left| C_2\right| ^{\alpha }-\left| C_2-(\lambda ^hA_2+h\lambda ^hB_2)\right| ^\alpha }{h\lambda ^h}\varGamma (ds) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad =\alpha D_3, \end{aligned}$$(40)which is equivalent to the fact that
$$\begin{aligned} \mathrm{{CD}}(X_1(t),X_1(t-h))=\mathrm{{CD}}(X_1(t),X_1(t+h)) \sim \alpha D_3\,h\lambda ^h \ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(41)where
$$\begin{aligned} D_3:=\sum _{j=0}^{+\infty }\int _{S_2} B_2C_2^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(42)
-
-
B)
Let us consider the auto-codifference function of \(\{X_2(t)\}\) given in Lemma 2.
-
For \(\lambda _1 \ne \lambda _2\), and \(|\lambda _1|<1\), \(|\lambda _2|<1\) we examine the auto-codifference function given in Eq. (19).
-
I)
Let us assume that \(|\lambda _1|>|\lambda _2|\). Proceeding as in A) leads to
$$\begin{aligned}&\mathrm{{CD}}(X_2(t),X_2(t-h))=\mathrm{{CD}}(X_2(t),X_2(t+h))\nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha E_1\,\lambda _1^h\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(43)where
$$\begin{aligned} E_1:=\sum _{j=0}^{+\infty }\int _{S_2} A_3C_3^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(44)We mention here, similarly as in A), to use the dominated convergence theorem we need to guarantee that
$$\begin{aligned} \begin{aligned}&\sum _{j=0}^{+\infty }\int _{S_2}\left| A_3\right| ^\alpha \varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| A_3\right| |C_3|^{\alpha -1}\varGamma (ds)<+\infty ,\\&\sum _{j=0}^{+\infty }\int _{S_2}\left| B_3\right| ^{\alpha }\varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| B_3\right| |C_3|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned}\end{aligned}$$(45)that is true (see Remark 1 in Appendix D).
-
II)
Let us assume that \(|\lambda _1|<|\lambda _2|\). Again, proceeding as in A) we obtain
$$\begin{aligned}&\mathrm{{CD}}(X_2(t),X_2(t-h))=\mathrm{{CD}}(X_2(t),X_2(t+h)) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha E_2\,\lambda _2^h\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(46)where
$$\begin{aligned} E_2:=\sum _{j=0}^{+\infty }\int _{S_2} B_3C_3^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(47) -
III)
Let us assume that \(\lambda _1=-\lambda _2\) and even h. Proceeding as in A) we have
$$\begin{aligned}&\mathrm{{CD}}(X_2(t),X_2(t-h))=\mathrm{{CD}}(X_2(t),X_2(t+h)) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha (E_1+E_2)\,\lambda _1^h\ \text { for}\ h \rightarrow +\infty , \end{aligned}$$(48)where
$$\begin{aligned} E_1+E_2=\sum _{j=0}^{+\infty }\int _{S_2} (A_3+B_3)C_3^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$ -
iV)
Let us assume that \(\lambda _1=-\lambda _2\) and odd h. Proceeding as in A) leads to
$$\begin{aligned}&\mathrm{{CD}}(X_2(t),X_2(t-h))=\mathrm{{CD}}(X_2(t),X_2(t+h)) \nonumber \\&\qquad \qquad \qquad \qquad \qquad \sim \alpha (E_1-E_2)\,\lambda _1^h\ \text { for}\ h \rightarrow +\infty , \end{aligned}$$(49)where
$$\begin{aligned} E_1-E_2=\sum _{j=0}^{+\infty }\int _{S_2} (A_3-B_3)C_3^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty .\end{aligned}$$
-
I)
-
For \(\lambda _1=\lambda _2=\lambda \), and \(|\lambda |<1\), we examine the auto-codifference function given in Eq. (22). Proceeding as in A) leads to
$$\begin{aligned} \mathrm{{CD}}(X_2(t),X_2(t+h))=\mathrm{{CD}}(X_2(t),X_2(t+h)) \sim \alpha E_3\,h\lambda ^h \ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(50)where
$$\begin{aligned} E_3:=\sum _{j=0}^{+\infty }\int _{S_2} B_4C_4^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(51)As in A), to use the dominated convergence theorem we need to guarantee that
$$\begin{aligned} \begin{aligned}&\sum _{j=0}^{+\infty }\int _{S_2}\left| A_4\right| ^\alpha \varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| A_4\right| |C_4|^{\alpha -1}\varGamma (ds)<+\infty ,\\&\sum _{j=0}^{+\infty }\int _{S_2}\left| B_4\right| ^{\alpha }\varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| B_4\right| |C_4|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned} \end{aligned}$$(52)that is true (see Remark 1 in Appendix D).
-
Appendix C:
Proof
-
A)
Let us consider the auto-covariation function of the time series \(\{X_1(t)\}\) given in Lemma 1.
-
(a)
At first, we examine the function \(\mathrm{CV}(X_1(t),X_1(t-h))\).
-
Let us assume that \(\lambda _1 \ne \lambda _2\), and \(|\lambda _1|<1\), \(|\lambda _2|<1\). The auto-covariation function given in Eq. (14) can be written as
$$\begin{aligned} \mathrm{{CV}}(X_1(t),X_1(t-h))=\lambda _1^hD_1+\lambda _2^hD_2, \end{aligned}$$(53)which directly leads to the formulas given in Theorem 2. The constants \(D_1\) and \(D_2\) are specified in Eqs. (32) and (34), respectively.
-
Let us assume that \(\lambda _1=\lambda _2=\lambda \), and \(|\lambda |<1\). The auto-covariation function given in Eq. (17) can be written as
$$\begin{aligned} \mathrm{{CV}}(X_1(t),X_1(t-h))=\lambda ^hD_{*}+h\lambda ^hD_3, \end{aligned}$$(54)which directly leads to the formula given in Theorem 2. The constant \(D_3\) is given in Eq. (42) and
$$\begin{aligned} D_{*}:={\sum _{j=0}^{+\infty }\int _{S_2}A_2C_2^{\langle \alpha -1 \rangle }\varGamma (ds)}. \end{aligned}$$
-
-
(b)
Now, we examine the function \(\mathrm{CV}(X_1(t),X_1(t+h))\).
-
For \(\lambda _1 \ne \lambda _2\), and \(|\lambda _1|<1\), \(|\lambda _2|<1\), we consider the auto-covariation function given in Eq. (15).
-
I)
Let us assume that \(|\lambda _1|>|\lambda _2|\). In this case, one can show that
$$\begin{aligned} \begin{aligned}&\lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{C_1\left( A_1\lambda _1^h+B_1\lambda _2^h\right) ^{\langle \alpha -1 \rangle }}{\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)\\{\mathop {=}\limits ^{(\star )}}&\sum _{j=0}^{+\infty }\lim _{h \rightarrow +\infty }\int _{S_2}\frac{C_1\left( A_1\lambda _1^h+B_1\lambda _2^h\right) ^{\langle \alpha -1 \rangle }}{\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)\\{\mathop {=}\limits ^{(\star \star )}}&\sum _{j=0}^{+\infty }\int _{S_2}\lim _{h \rightarrow +\infty }\frac{C_1\left( A_1\lambda _1^h+B_1\lambda _2^h\right) ^{\langle \alpha -1 \rangle }}{\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds), \end{aligned}\end{aligned}$$(55)From the dominated convergence theorem, \((\star )\) holds if the sum over \(j \in {\mathbb {N}}_{0}\) in Eq. (55) converges uniformly. From the following inequality
$$\begin{aligned} \left| a+b\right| ^{\alpha -1}\le |a|^{\alpha -1}+|b|^{\alpha -1} \end{aligned}$$(56)satisfied for \(a,b \in {\mathrm{I{\!}\mathrm R}}\), \(1<\alpha <2\), for all \(j \in {\mathbb {N}}_0\) we have
$$\begin{aligned}&\left| \int _{S_2}\frac{C_1\left( A_1\lambda _1^h+B_1\lambda _2^h\right) ^{\langle \alpha -1 \rangle }}{\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)\right| \nonumber \\&\qquad \qquad \,\, \le \int _{S_2}|C_1|\left| A_1\right| ^{\alpha -1}\varGamma (ds)+\int _{S_2}|C_1|\left| B_1\right| ^{\alpha -1}\varGamma (ds)=K_j. \end{aligned}$$(57)Let us notice that \(K_j\) does not depend on h. Consequently, the sum over \(j \in {\mathbb {N}}_{0}\) in Eq. (55) converges uniformly if the sum of \(K_j\) over \(j \in {\mathbb {N}}_0\) converges, which leads to the below conditions
$$\begin{aligned} \sum _{j=0}^{+\infty }\int _{S_2}\left| C_1\right| |A_1|^{\alpha -1}\varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| C_1\right| |B_1|^{\alpha -1}\varGamma (ds)<+\infty \end{aligned}$$(58)that are true, see Remark 1 in Appendix D. In the next step, we prove that \((\star \star )\) in Eq. (55) holds by the use of the dominated convergence theorem. From the inequality in Eq. (56) we have that
$$\begin{aligned} \left| \frac{C_1\left( A_1\lambda _1^h+B_1\lambda _2^h\right) ^{\langle \alpha -1 \rangle }}{\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }}\right| \le |C_1||A_1|^{\alpha -1}+|C_1||B_1|^{\alpha -1}. \end{aligned}$$for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\). Let us notice that the dominating function does not depend on h and is integrable if for all \(j \in {\mathbb {N}}_0\) we have that
$$\begin{aligned} \begin{aligned} \int _{S_2}\left| C_1||A_1\right| ^{\alpha -1}\varGamma (ds)<+\infty , \quad \int _{S_2}\left| C_1||B_1\right| ^{\alpha -1}\varGamma (ds)<+\infty ,\\ \end{aligned} \end{aligned}$$which is true, see Remark 1 in Appendix D. Then, let us notice that for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\) the following limit holds
$$\begin{aligned} \lim _{h \rightarrow +\infty }\frac{C_1\left( A_1\lambda _1^h+B_1\lambda _2^h\right) ^{\langle \alpha -1 \rangle }}{\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }} =C_1A_1^{\langle \alpha -1 \rangle } \end{aligned}$$and finally, from Eq. (55) we have
$$\begin{aligned} \lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{C_1\left( A_1\lambda _1^h+B_1\lambda _2^h\right) ^{\langle \alpha -1 \rangle }}{\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)= D_4. \end{aligned}$$(59)which is equivalent to the fact that
$$\begin{aligned} \mathrm{CV}(X_1(t),X_1(t+h)) \sim D_4\,\left( \lambda _1^h\right) ^{\langle \alpha -1 \rangle }\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(60)where
$$\begin{aligned} D_4:=\sum _{j=0}^{+\infty }\int _{S_2} C_1A_1^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(61) -
II)
Let us assume that \(|\lambda _1|<|\lambda _2|\). Proceeding similary as above, we obtain
$$\begin{aligned} \mathrm{CV}(X_1(t),X_1(t+h))\sim D_5\,\left( \lambda _2^h\right) ^{\langle \alpha -1\rangle }\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(62)where
$$\begin{aligned} D_5:=\sum _{j=0}^{+\infty }\int _{S_2} C_1B_1^{\langle \alpha -1\rangle }\varGamma (ds) < +\infty . \end{aligned}$$(63) -
III)
Let us assume that \(\lambda _1=-\lambda _2\) and even h. In this case, we have the exact formula
$$\begin{aligned} \mathrm{CV}(X_1(t),X_1(t+h)) = D_{7}\,\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }, \end{aligned}$$(64)where
$$\begin{aligned} D_{7}:=\sum _{j=0}^{+\infty }\int _{S_2}C_1(A_1+B_1)^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty .\end{aligned}$$(65) -
IV)
Let us assume that \(\lambda _1=-\lambda _2\) and odd h. In this case, as in the case given above, we have the exact formula
$$\begin{aligned} \mathrm{CV}(X_1(t),X_1(t+h)) = D_{11}\,\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }, \end{aligned}$$(66)where
$$\begin{aligned} D_{8}:=\sum _{j=0}^{+\infty }\int _{S_2}C_1(A_1-B_1)^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(67)
-
I)
-
For \(\lambda _1=\lambda _2=\lambda \), and \(|\lambda |<1\), we consider the auto-covariation function given in Eq. (18). Similarly to the previous case, one can show that
$$\begin{aligned} \begin{aligned}&\lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{C_2\left( A_2\lambda ^h+B_2h\lambda ^h\right) ^{\langle \alpha -1 \rangle }}{\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)\\{\mathop {=}\limits ^{(\star )}}&\sum _{j=0}^{+\infty }\lim _{h \rightarrow +\infty }\int _{S_2}\frac{C_2\left( A_2\lambda ^h+B_2h\lambda ^h\right) ^{\langle \alpha -1 \rangle }}{\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)\\{\mathop {=}\limits ^{(\star \star )}}&\sum _{j=0}^{+\infty }\int _{S_2}\lim _{h \rightarrow +\infty }\frac{C_2\left( A_2\lambda ^h+B_2h\lambda ^h\right) ^{\langle \alpha -1 \rangle }}{\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds). \end{aligned}\end{aligned}$$(68)From the uniform convergence theorem, to justify \((\star )\) the sum over \(j \in {\mathbb {N}}_{0}\) in Eq. (68) has to converge uniformly. From the inequality in Eq. (56) for all \(j \in {\mathbb {N}}_0\) we can show that
$$\begin{aligned}&\left| \int _{S_2}\frac{C_2\left( A_2\lambda ^h+B_2h\lambda ^h\right) ^{\langle \alpha -1 \rangle }}{\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)\right| \\&\qquad \qquad \qquad \qquad \le \int _{S_2}|C_2|\left| A_2\right| ^{\alpha -1}\varGamma (ds)+\int _{S_2}|C_2|\left| B_2\right| ^{\alpha -1}\varGamma (ds)=L_j, \end{aligned}$$where \(L_j\) does not depend on h. Consequently, the sum over \(j \in {\mathbb {N}}_{0}\) in Eq. (68) converges uniformly if the sum of \(L_j\) over \(j \in {\mathbb {N}}_0\) converges, that leads to the below conditions
$$\begin{aligned} \sum _{j=0}^{+\infty }\int _{S_2}\left| C_2\right| |A_2|^{\alpha -1}\varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| C_2\right| |B_2|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned}$$(69)which are true (see Remark 1 in Appendix D). Now, to justify \((\star \star )\) in Eq. (68) we again use the dominated convergence theorem. From the inequality in Eq. (56) we have
$$\begin{aligned} \left| \frac{C_2\left( A_2\lambda ^h+B_2h\lambda ^h\right) ^{\langle \alpha -1 \rangle }}{\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle }}\right| \le |C_2||A_2|^{\alpha -1}+|C_2||B_2|^{\alpha -1}, \end{aligned}$$for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\). Let us notice that the integrand in Eq. (68) is dominated a function that does not depend on h. Moreover, this function is integrable if for all \(j \in {\mathbb {N}}\) the integrals given below are finite
$$\begin{aligned} \begin{aligned} \int _{S_2}\left| C_2||A_2\right| ^{\alpha -1}\varGamma (ds)<+\infty , \quad \int _{S_2}\left| C_2||B_2\right| ^{\alpha -1}\varGamma (ds)<+\infty ,\\ \end{aligned} \end{aligned}$$which is true (see Remark 1 in Appendix D). Then, let us notice that for a fixed \(\mathbf {s}=(s_1,s_2)\in S_2\) the following limit holds
$$\begin{aligned} \lim _{h \rightarrow +\infty }\frac{C_2\left( A_2\lambda ^h+B_2h\lambda ^h\right) ^{\langle \alpha -1 \rangle }}{\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle }}=C_2B_2^{\langle \alpha -1 \rangle } \end{aligned}$$and finally from Eq. (68) we obtain
$$\begin{aligned} \lim _{h \rightarrow +\infty }\sum _{j=0}^{+\infty }\int _{S_2}\frac{C_2\left( A_2\lambda ^h+B_2h\lambda ^h\right) ^{\langle \alpha -1 \rangle }}{\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle }}\varGamma (ds)= D_6, \end{aligned}$$(70)which is equivalent to the fact that
$$\begin{aligned} \mathrm{{CV}}(X_1(t),X_1(t+h))\sim D_6\,\left( h\lambda ^h\right) ^{\langle \alpha -1 \rangle }\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(71)where
$$\begin{aligned} D_6:=\sum _{j=0}^{+\infty }\int _{S_2} C_2B_2^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(72)
-
(a)
-
B)
Let us consider the auto-covariation function of the time series \(\{X_2(t)\}\) given in Lemma 2.
-
(a)
At first, let us examine the function \(\mathrm{CV}(X_2(t),X_2(t-h))\).
-
Let us assume that \(\lambda _1 \ne \lambda _2\), and \(|\lambda _1|<1\), \(|\lambda _2|<1\). The auto-covariation function given in Eq. (20) can be written as
$$\begin{aligned} \mathrm{{CV}}(X_2(t),X_2(t-h))=\lambda _1^hE_1+\lambda _2^hE_2, \end{aligned}$$(73)which directly leads to the formulas given in Theorem 2. The constants \(E_1\) and \(E_2\) are specified in Eqs. (44) and (47), respectively.
-
Let us assume that \(\lambda _1=\lambda _2=\lambda \), and \(|\lambda |<1\). The auto-covariation function given in Eq. (23) can be written as
$$\begin{aligned} \mathrm{{CV}}(X_2(t),X_2(t-h))=\lambda ^hE_{*}+h\lambda ^hE_3, \end{aligned}$$(74)which directly leads to the formulas given in Theorem 2. The constant \(E_3\) is given in Eq. (51) and
$$\begin{aligned} E_{*}:={\sum _{j=0}^{+\infty }\int _{S_2}A_4C_4^{\langle \alpha -1 \rangle }\varGamma (ds)}. \end{aligned}$$
-
-
(b)
Now, we examine the function \(\mathrm{CV}(X_2(t),X_2(t+h))\).
-
For \(\lambda _1 \ne \lambda _2\), and \(|\lambda _1|<1\), \(|\lambda _2|<1\), we consider the auto-covariation function given in Eq. (21).
-
I)
Let us assume that \(|\lambda _1|>|\lambda _2|\). Proceeding as in A) we obtain
$$\begin{aligned} \mathrm{{CV}}(X_2(t),X_2(t+h))\sim E_4\left( \lambda _1^h\right) ^{\langle \alpha -1 \rangle } \quad \text {for} \quad h \rightarrow \infty , \end{aligned}$$(75)where
$$\begin{aligned} E_4:=\sum _{j=0}^{\infty }\int _{S_2}C_3A_3^{\langle \alpha -1 \rangle }\varGamma (ds)<+\infty . \end{aligned}$$(76)We mention here that similarly as in A), to use the dominated convergence theorem we need to guarantee that
$$\begin{aligned} \begin{aligned} \sum _{j=0}^{+\infty }\int _{S_2}\left| C_3\right| |A_3|^{\alpha -1}\varGamma (ds)<+\infty ,\\ \sum _{j=0}^{+\infty }\int _{S_2}\left| C_3\right| |B_3|^{\alpha -1}\varGamma (ds)<+\infty . \end{aligned} \end{aligned}$$(77)which is true (see Remark 1 in Appendix D).
-
II)
Let us assume that \(|\lambda _1|<|\lambda _2|\). Again, proceeding as in A) we obtain
$$\begin{aligned} \mathrm{{CV}}(X_2(t),X_2(t+h))\sim E_5\,\left( \lambda _2^h\right) ^{\langle \alpha -1 \rangle }\ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(78)where
$$\begin{aligned} E_5:=\sum _{j=0}^{+\infty }\int _{S_2} C_3B_3^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(79) -
III)
Let us assume that \(\lambda _1=-\lambda _2\) and h is even. In this case, we have the exact formula
$$\begin{aligned} \mathrm{{CV}}(X_2(t),X_2(t+h))= E_7\,\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }\ \text { for}\ h \rightarrow +\infty , \end{aligned}$$(80)where
$$\begin{aligned} E_7:=\sum _{j=0}^{+\infty }\int _{S_2} C_3(A_3+B_3)^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$ -
IV)
Let us assume that \(\lambda _1=-\lambda _2\) and h is odd. Similarly as above, we have the exact formula
$$\begin{aligned} \mathrm{{CV}}(X_2(t),X_2(t+h))= E_8\left( \lambda _1^h\right) ^{\langle \alpha -1\rangle }\ \text { for}\ h \rightarrow +\infty , \end{aligned}$$(81)where
$$\begin{aligned} E_8:=\sum _{j=0}^{+\infty }\int _{S_2} C_3(A_3-B_3)^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty .\end{aligned}$$
-
I)
-
For \(\lambda _1=\lambda _2=\lambda \), and \(|\lambda |<1\), we consider the auto-covariation function given in Eq. (24). Proceeding as in A) we obtain
$$\begin{aligned} \mathrm{{CV}}(X_2(t),X_2(t+h)) \sim E_6\,\left( h\lambda ^h\right) ^{\langle \alpha -1\rangle } \ \ \text { for}\ \ h \rightarrow +\infty , \end{aligned}$$(82)where
$$\begin{aligned} E_6:=\sum _{j=0}^{+\infty }\int _{S_2} C_4B_4^{\langle \alpha -1\rangle }\varGamma (ds)<+\infty . \end{aligned}$$(83)As in A), to use the dominated convergence theorem we need to guarantee that
$$\begin{aligned} \begin{aligned} \sum _{j=0}^{+\infty }\int _{S_2}\left| C_4\right| |A_4|^{\alpha -1}\varGamma (ds)<+\infty , \quad \sum _{j=0}^{+\infty }\int _{S_2}\left| C_4\right| |B_4|^{\alpha -1}\varGamma (ds)<+\infty , \end{aligned} \end{aligned}$$(84)which is true (see Remark 1 in Appendix D).
-
-
(a)
Appendix D
Remark 1
Let us notice that the constants \(A_i,B_i,C_i\) for \(i=1,2,3,4\) can be upper-bounded by \(M\,\mathrm{max}(|\lambda _1|,|\lambda _2|)^j\) or by \(M\,j\mathrm{max}(|\lambda _1|,|\lambda _2|)^j\) with the constant M independent of j. Since \(\mathrm{max}(|\lambda _1|,|\lambda _2|)<1\) and the measure \(\varGamma (\cdot )\) is finite, the conditions given in Eqs. (27), (38), (45), (52), (58), (69), (77) and (84) are always satisfied.
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Grzesiek, A., Wyłomańska, A. (2022). Asymptotics of Alternative Interdependence Measures for Bivariate \(\alpha -\)Stable Autoregressive Model of Order 1. In: Chaari, F., Leskow, J., Wylomanska, A., Zimroz, R., Napolitano, A. (eds) Nonstationary Systems: Theory and Applications. WNSTA 2021. Applied Condition Monitoring, vol 18. Springer, Cham. https://doi.org/10.1007/978-3-030-82110-4_3
Download citation
DOI: https://doi.org/10.1007/978-3-030-82110-4_3
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-82191-3
Online ISBN: 978-3-030-82110-4
eBook Packages: EngineeringEngineering (R0)