Abstract
In this paper the asymptotic behavior of the conditional least squares estimators of the offspring mean matrix for a 2-type critical positively regular Galton–Watson branching process with immigration is described. We also study this question for a natural estimator of the spectral radius of the offspring mean matrix, which we call criticality parameter. We discuss the subcritical case as well.
Similar content being viewed by others
References
Athreya KB, Ney PE (1972) Branching processes. Springer, New York
Barczy M, Ispány M, Pap G (2011) Asymptotic behavior of unstable INAR(\(p\)) processes. Stoch Process Appl 121(3):583–608
Barczy M, Ispány M, Pap G (2014) Asymptotic behavior of conditional least squares estimators for unstable integer-valued autoregressive models of order 2. Scand J Stat 41(4):866–892
Cox JC, Ingersoll JE, Ross SA (1985) A theory of the term structure of interest rates. Econometrica 53(2):385–407
Danka T, Pap G (2016) Asymptotic behavior of critical indecomposable multi-type branching processes with immigration. Eur Ser Appl Ind Math Probab Stat 20:238–260
Haccou P, Jagers P, Vatutin V (2005) Branching processes. Cambridge University Press, Cambridge
Horn RA, Johnson ChR (1985) Matrix analysis. Cambridge University Press, Cambridge
Ikeda N, Watanabe S (1981) Stochastic differential equations and diffusion processes. North-Holland Publishing Co, Amsterdam
Ispány M, Körmendi K, Pap G (2014) Asymptotic behavior of CLS estimators for 2-type doubly symmetric critical Galton–Watson processes with immigration. Bernoulli 20(4):2247–2277
Ispány M, Pap G (2010) A note on weak convergence of step processes. Acta Math Hung 126(4):381–395
Ispány M, Pap G (2014) Asymptotic behavior of critical primitive multi-type branching processes with immigration. Stoch Anal Appl 32(5):727–741
Ispány M, Pap G, van Zuijlen MCA (2003) Asymptotic inference for nearly unstable INAR(1) models. J Appl Probab 40(3):750–765
Kesten H, Stigum BP (1966) A limit theorem for multidimensional Galton–Watson processes. Ann Math Stat 37(5):1211–1223
Klimko LA, Nelson PI (1978) On conditional least squares estimation for stochastic processes. Ann Stat 6(3):629–642
Körmendi K, Pap G (2015) Statistical inference of 2-type critical Galton–Watson processes with immigration. Available on the arXiv:1502.04900
Putzer EJ (1966) Avoiding the Jordan canonical form in the discussion of linear systems with constant coefficients. Am Math Mon 73(1):2–7
Quine MP (1970) The multi-type Galton–Watson process with immigration. J Appl Probab 7(2):411–422
Velasco MG, Puerto IM, Martínez R, Molina M, Mota M, Ramos A (2010) Workshop on branching processes and their applications, vol 197, Lecture Notes in Statistics. Springer, Berlin
Wei CZ, Winnicki J (1989) Some asymptotic results for the branching process with immigration. Stoch Process Appl 31(2):261–282
Wei CZ, Winnicki J (1990) Estimation of the means in the branching process with immigration. Ann Stat 18:1757–1773
Winnicki J (1988) Estimation theory for the branching process with immigration. In: Statistical inference from stochastic processes (Ithaca, NY, 1987). (Contemporary Mathematics), vol 80. American Mathematical Society, Providence, RI, pp. 301–322
Acknowledgments
The research of G. Pap was realized in the frames of TÁMOP 4.2.4. A/2-11-1-2012-0001, National Excellence Program—Elaborating and operating an inland student and researcher personal support system”. The project was subsidized by the European Union and co-financed by the European Social Fund. K. Körmendi was in part, supported by TÁMOP-4.2.2.B-15/1/KONV-2015-0006.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: Estimations of moments
In the proof of Theorem 3.1, good bounds for moments of the random vectors and variables \(({\varvec{M}}_k)_{k\in \mathbb {Z}_+}\), \(({\varvec{X}}_k)_{k\in \mathbb {Z}_+}\), \((U_k)_{k\in \mathbb {Z}_+}\) and \((V_k)_{k\in \mathbb {Z}_+}\) are extensively used. First note that, for all \(k \in \mathbb {N}\), \({\mathbb {E}}( {\varvec{M}}_k \,|\,{\mathcal F}_{k-1} ) = {\varvec{0}}\) and \({\mathbb {E}}({\varvec{M}}_k) = {\varvec{0}}\), since \({\varvec{M}}_k = {\varvec{X}}_k - {\mathbb {E}}({\varvec{X}}_k \,|\,{\mathcal F}_{k-1})\). We present these results without proofs as they can be proven the same way as in Appendix B of Ispány et al. (2014).
Lemma 6.1
Let \(({\varvec{X}}_k)_{k\in \mathbb {Z}_+}\) be a 2-type Galton–Watson process with immigration and with \({\varvec{X}}_0 = {\varvec{0}}\). If \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,1}\Vert ^2) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,2}\Vert ^2) < \infty \) and \({\mathbb {E}}(\Vert {\varvec{\varepsilon }}_1\Vert ^2) < \infty \) then
for all \(k \in \mathbb {N}\), where
If \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,1}\Vert ^3) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,2}\Vert ^3) < \infty \) and \({\mathbb {E}}(\Vert {\varvec{\varepsilon }}_1\Vert ^3) < \infty \) then, for all \(k \in \mathbb {N}\),
Lemma 6.2
Let \(({\varvec{X}}_k)_{k\in \mathbb {Z}_+}\) be a 2-type Galton–Watson process with immigration such that \(\alpha , \delta \in [0, 1)\) and \(\beta , \gamma \in (0, \infty )\) with \(\alpha + \delta > 0\) and \(\beta \gamma = (1 - \alpha ) (1 - \delta )\) (hence it is critical and and positively regular). Suppose \({\varvec{X}}_0 = {\varvec{0}}\), and \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,1}\Vert ^\ell ) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,2}\Vert ^\ell ) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\varepsilon }}_1\Vert ^\ell ) < \infty \) with some \(\ell \in \mathbb {N}\). Then \({\mathbb {E}}(\Vert {\varvec{X}}_k\Vert ^\ell ) = {\text {O}}(k^\ell )\), i.e., \(\sup _{k \in \mathbb {N}} k^{-\ell } {\mathbb {E}}(\Vert {\varvec{X}}_k\Vert ^\ell ) < \infty \).
Corollary 6.3
Let \(({\varvec{X}}_k)_{k\in \mathbb {Z}_+}\) be a critical, positively regular 2-type Galton–Watson process. Suppose \({\varvec{X}}_0 = {\varvec{0}}\), and \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,1}\Vert ^\ell ) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,2}\Vert ^\ell ) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\varepsilon }}_1\Vert ^\ell ) < \infty \) with some \(\ell \in \mathbb {N}\). Then
for \(i, j \in \mathbb {Z}_+\) with \(i \le \ell \) and \(2 j \le \ell \).
Corollary 6.4
Let \(({\varvec{X}}_k)_{k\in \mathbb {Z}_+}\) be a critical, positively regular 2-type Galton–Watson process. Suppose \({\varvec{X}}_0 = {\varvec{0}}\), and \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,1}\Vert ^\ell ) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\xi }}_{1,1,2}\Vert ^\ell ) < \infty \), \({\mathbb {E}}(\Vert {\varvec{\varepsilon }}_1\Vert ^\ell ) < \infty \) with some \(\ell \in \mathbb {N}\). Then
-
(i)
for all \(i,j\in \mathbb {Z}_+\) with \(\max \{i,j\} \le \lfloor \ell /2 \rfloor \), and for all \(\kappa > i + \frac{j}{2} + 1\), we have
$$\begin{aligned} n^{-\kappa } \sum _{k=1}^n\vert U_k^i V_k^j\vert \mathop {\longrightarrow }\limits ^{{\mathbb {P}}}0 \qquad \hbox {as }n\rightarrow \infty , \end{aligned}$$(6.3) -
(ii)
for all \(i,j\in \mathbb {Z}_+\) with \(\max \{i,j\} \le \ell \), for all \(T>0\), and for all \(\kappa > i + \frac{j}{2} + \frac{i+j}{\ell }\), we have
$$\begin{aligned} n^{-\kappa } \sup _{t\in [0,T]} \vert U_{\lfloor nt\rfloor }^i V_{\lfloor nt\rfloor }^j \vert \mathop {\longrightarrow }\limits ^{{\mathbb {P}}}0 \qquad \hbox {as }n\rightarrow \infty , \end{aligned}$$(6.4) -
(iii)
for all \(i,j\in \mathbb {Z}_+\) with \(\max \{i,j\} \le \lfloor \ell /4 \rfloor \), for all \(T>0\), and for all \(\kappa > i + \frac{j}{2} + \frac{1}{2}\), we have
$$\begin{aligned} n^{-\kappa } \sup _{t\in [0,T]} \left| \sum _{k=1}^{\lfloor nt\rfloor }[U_k^i V_k^j - {\mathbb {E}}(U_k^i V_k^j \,|\,{\mathcal F}_{k-1})] \right| \mathop {\longrightarrow }\limits ^{{\mathbb {P}}}0 \qquad \hbox {as }n\rightarrow \infty . \end{aligned}$$(6.5)
Remark 6.5
In some parts of this paper we need the above statements with a smaller \(\kappa \) than it is provided by Corollary 6.4. Fortunately we can use a decomposition argument in those cases to sharpen the statements, for example see the proof of (5.9) and Lemma 6.7.
Appendix 2: CLS estimators
For each \(n \in \mathbb {N}\), a CLS estimator \(\widehat{{\varvec{m}}}_{\varvec{\xi }}^{(n)}\) of \({\varvec{m}}_{\varvec{\xi }}\) based on a sample \({\varvec{X}}_1, \ldots , {\varvec{X}}_n\) can be obtained by minimizing the sum of squares
with respect to \({\varvec{m}}_{\varvec{\xi }}\) over \(\mathbb {R}^{2\times 2}\). In what follows, we use the notation \({\varvec{x}}_0 := {\varvec{0}}\). For all \(n \in \mathbb {N}\), we define the function \(Q_n : (\mathbb {R}^2)^n \times \mathbb {R}^{2\times 2} \rightarrow \mathbb {R}\) by
for all \({\varvec{m}}_{\varvec{\xi }}' \in \mathbb {R}^{2\times 2}\) and \({\varvec{x}}_1, \ldots , {\varvec{x}}_n \in \mathbb {R}^2\). By definition, for all \(n \in \mathbb {N}\), a CLS estimator of \({\varvec{m}}_{\varvec{\xi }}\) is a measurable function \(F_n : (\mathbb {R}^2)^n \rightarrow \mathbb {R}^{2\times 2}\) such that
for all \({\varvec{x}}_1, \ldots , {\varvec{x}}_n \in \mathbb {R}^2\). Next we give the solutions of this extremum problem.
Lemma 6.6
For each \(n \in \mathbb {N}\), any CLS estimator of \({\varvec{m}}_{\varvec{\xi }}\) is a measurable function \(F_n : (\mathbb {R}^2)^n \rightarrow \mathbb {R}^{2\times 2}\) for which
on the set
where
For the existence of these CLS estimators in case of a critical symmetric 2-type Galton–Watson process, i.e., when \(\varrho = 1\), we need the following approximations.
Lemma 6.7
Suppose that the assumptions of Theorem 3.1 hold. Then for each \(T > 0\),
Proof
The aim of the following discussion is to decompose \(\sum _{k=1}^{{\lfloor nt\rfloor }} U_{k-1} V_{k-1}\) as a sum of a martingale and some other terms. Using the recursions (4.7), (4.4) and Lemma 6.1, we obtain
Thus
Consequently
Using (6.5) with \((\ell , i , j) = (4, 1, 1)\) we have
Thus, in order to show the statement, it suffices to prove
as \(n \rightarrow \infty \). Using (6.3) with \((\ell , i , j) = (2, 1, 0)\) and \((\ell , i , j) = (2, 0, 1)\), and (6.4) with \((\ell , i , j) = (3, 1, 1)\) we have (6.7), thus we conclude the statement. \(\square \)
Using the same ideas as above one can prove the following.
Lemma 6.8
Suppose that the assumptions of Theorem 3.1 hold. For each \(T > 0\), we have
Now we can prove asymptotic existence and uniqueness of CLS estimators of the offspring mean matrix and of the criticality parameter.
Proposition 6.9
Suppose that the assumptions of Theorem 3.1 hold, and \(\langle {\overline{{\varvec{V}}}}_{\!\!{\varvec{\xi }}} {\varvec{v}}_{\mathrm {left}}, {\varvec{v}}_{\mathrm {left}}\rangle + \langle {\varvec{V}}_{\!\!{\varvec{\varepsilon }}} {\varvec{v}}_{\mathrm {left}}, {\varvec{v}}_{\mathrm {left}}\rangle + \langle {\varvec{v}}_{\mathrm {left}}, {\varvec{m}}_{\varvec{\varepsilon }}\rangle ^2 > 0\). Then \(\lim _{n \rightarrow \infty } {\mathbb {P}}(\Omega _n) = 1\), where \(\Omega _n\) is defined in (3.2), and hence the probability of the existence of a unique CLS estimator \(\widehat{{\varvec{m}}}_{\varvec{\xi }}^{(n)}\) converges to 1 as \(n \rightarrow \infty \), and this CLS estimator has the form given in (3.1) on the set \(\Omega _n\). If \(\langle {\overline{{\varvec{V}}}}_{\!\!{\varvec{\xi }}} {\varvec{v}}_{\mathrm {left}}, {\varvec{v}}_{\mathrm {left}}\rangle > 0\) then \(\lim _{n \rightarrow \infty } {\mathbb {P}}(\widetilde{\Omega }_n) = 1\), where \(\widetilde{\Omega }_n\) is defined in (3.5), and hence the probability of the existence of the estimator \(\widehat{\varrho }_n\) converges to 1 as \(n \rightarrow \infty \).
Proof
Recall convergence \((n^{-1} U_{{\lfloor nt\rfloor }})_{t\in \mathbb {R}_+} \mathop {\longrightarrow }\limits ^{{\mathcal D}}({\mathcal Y}_t)_{t\in \mathbb {R}_+}\) from (4.6). Using Lemmas 6.8, 6.7, and a version of the continuous mapping theorem (see Ispány and Pap (2010, Lemma 3.1)), one can show
By (4.11) and continuous mapping theorem,
Since \({\varvec{m}}_{\varvec{\varepsilon }}\ne {\varvec{0}}\), by the SDE (2.8), we have \({\mathbb {P}}({\mathcal Y}_t = 0\hbox { for all }t \in [0,1]) = 0\), which implies that \({\mathbb {P}}\bigl ( \int _0^1 {\mathcal Y}_t^2 \, \mathrm {d}t \int _0^1 {\mathcal Y}_t \, \mathrm {d}t > 0 \bigr ) = 1\). Consequently, the distribution function of \(\int _0^1 {\mathcal Y}_t^2 \, \mathrm {d}t \int _0^1 {\mathcal Y}_t \, \mathrm {d}t\) is continuous at 0.
If \(\langle {\overline{{\varvec{V}}}}_{\!\!{\varvec{\xi }}} {\varvec{v}}_{\mathrm {left}}, {\varvec{v}}_{\mathrm {left}}\rangle > 0\) then, by (6.8),
as \(n \rightarrow \infty \).
If \(\langle {\overline{{\varvec{V}}}}_{\!\!{\varvec{\xi }}} {\varvec{v}}_{\mathrm {left}}, {\varvec{v}}_{\mathrm {left}}\rangle > 0\), then (3.6) yields \(\widehat{{\varvec{m}}}_{\varvec{\xi }}^{(n)} \mathop {\longrightarrow }\limits ^{{\mathcal D}}{\varvec{m}}_{\varvec{\xi }}\) as \(n \rightarrow \infty \), and hence \(\widehat{{\varvec{m}}}_{\varvec{\xi }}^{(n)} \mathop {\longrightarrow }\limits ^{{\mathbb {P}}}{\varvec{m}}_{\varvec{\xi }}\) as \(n \rightarrow \infty \), thus \((\widehat{\alpha }_n - \widehat{\delta }_n)^2 + 4 \widehat{\beta }_n \widehat{\gamma }_n \mathop {\longrightarrow }\limits ^{{\mathbb {P}}}(\alpha - \delta )^2 + 4 \beta \gamma = (1 - \lambda )^2 > 0\), implying
hence we obtain the statement. \(\square \)
Rights and permissions
About this article
Cite this article
Körmendi, K., Pap, G. Statistical inference of 2-type critical Galton–Watson processes with immigration. Stat Inference Stoch Process 21, 169–190 (2018). https://doi.org/10.1007/s11203-016-9148-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11203-016-9148-y
Keywords
- Galton–Watson process
- Multi-type branching process
- Conditional least squares estimator
- Offspring mean matrix