Abstract
We study asymptotic properties of conditional least squares estimators for the drift parameters of two-factor affine diffusions based on continuous time observations. We distinguish three cases: subcritical, critical and supercritical. For all the drift parameters, in the subcritical and supercritical cases, asymptotic normality and asymptotic mixed normality is proved, while in the critical case, non-standard asymptotic behavior is described.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Affine processes are applied in mathematical finance in several models including interest rate models (e.g. the Cox–Ingersoll–Ross, Vasiček or general affine term structure short rate models), option pricing (e.g. the Heston model) and credit risk models, see e.g. Duffie et al. (2003), Filipović (2009), Baldeaux and Platen (2013), and Alfonsi (2015). In this paper we consider two-factor affine processes, i.e. affine processes with state-space \([0, \infty ) \times \mathbb {R}\). Dawson and Li (2006) derived a jump-type stochastic differential equation (SDE) for such processes. Specializing this result to the diffusion case, i.e. two-factor affine processes without jumps, we obtain that for every \(a \in [0, \infty )\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in [0, \infty )\) and \(\varrho \in [-1, 1]\), the SDE
with an arbirary initial value \((Y_0, X_0)\) with \(\mathbb {P}(Y_0 \in [0, \infty )) = 1\) and independent of a 3-dimensional standard Wiener process \((W_t, B_t, L_t)_{t\in [0, \infty )}\), has a pathwise unique strong solution being a two-factor affine diffusion process, and conversely, every two-factor affine diffusion process is a pathwise strong solution of a SDE (1.1) with appropriate parameters \(a \in [0, \infty )\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in [0, \infty )\) and \(\varrho \in [-1, 1]\), see Proposition 2.1.
The aim of this paper is to study the asymptotic properties of the conditional least squares estimators (CLSE) \((\widehat{a}_T, \widehat{b}_T, \widehat{\alpha }_T, \widehat{\beta }_T, \widehat{\gamma }_T)\) of the drift parameters \((a, b, \alpha , \beta , \gamma )\) based on continuous time observations \((Y_t, X_t)_{t\in [0,T]}\) with \(T > 0\). This estimator is the high frequency limit in probability as \(n \rightarrow \infty \) of the CLSE based on discrete time observations \((Y_{k/n}, X_{k/n})_{k\in \{0,\ldots ,{\lfloor nT\rfloor }\}}\), \(n \in \mathbb {N}\). We do not estimate the parameters \(\sigma _1\), \(\sigma _2\), \(\sigma _3\) and \(\varrho \), since for all \(T \in (0, \infty )\), they are measurable functions (i.e., statistics) of \((Y_t, X_t)_{t\in [0,T]}\), see Appendix C in the extended arXiv version Bolyog and Pap (2017) of this paper. For the calculation of \((\widehat{a}_T, \widehat{b}_T, \widehat{\alpha }_T, \widehat{\beta }_T, \widehat{\gamma }_T)\) one does not need to know the values of the diffusion coefficients \(\sigma _1\), \(\sigma _2\), \(\sigma _3\) and \(\varrho \), see (3.4).
The first coordinate process Y in (1.1) is called a Cox–Ingersoll–Ross (CIR) process (see Cox et al. 1985). In the submodel consisting only of the process Y, (Overbeck and Rydén 1997 Theorems 3.4, 3.5 and 3.6) derived the CLSE of (a, b) based on continuous time observations \((Y_t)_{t\in [0,T]}\) with \(T > 0\), i.e., the limit in probability as \(n \rightarrow \infty \) of the CLSE based on discrete time observations \((Y_{k/n})_{k\in \{0,\ldots ,{\lfloor nT\rfloor }\}}\), \(n \in \mathbb {N}\), which turns to be the same as the CLSE \((\widehat{a}_T, \widehat{b}_T)\) of (a, b) based on continuous time observations \((Y_t, X_t)_{t\in [0,T]}\), and they proved strong consistency and asymptotic normality in case of a subcritical CIR process Y, i.e., when \(b > 0\) and the initial distribution is the unique stationary distribution of the model.
Barczy et al. (2014) considered a submodel of (1.1) with \(a \in (0, \infty )\), \(\beta = 0\), \(\sigma _1 = 1\), \(\sigma _2 = 1\), \(\varrho = 0\) and \(\sigma _3 = 0\). The estimator of the parameters \((\alpha , \gamma )\) based on continuous time observations \((X_t)_{t\in [0,T]}\) with \(T > 0\) (which they call a least square estimator) is in fact the CLSE, i.e., the limit in probability as \(n \rightarrow \infty \) of the CLSE based on discrete time observations \((X_{k/n})_{k\in \{0,\ldots ,{\lfloor nT\rfloor }\}}\), \(n \in \mathbb {N}\), which can be shown by the method of the proof of Lemma 3.3. They proved strong consistency and asymptotic normality in case of a subcritical process (Y, X), i.e., when \(b > 0\) and \(\gamma > 0\).
Barczy et al. (2016) considered the so-called Heston model, which is a submodel of (1.1) with \(a, \sigma _1, \sigma _2 \in (0, \infty )\), \(\gamma = 0\), \(\varrho \in (-1, 1)\) and \(\sigma _3 = 0\). The estimator of the parameters \((a, b, \alpha , \beta )\) based on continuous time observations \((Y_t, X_t)_{t\in [0,T]}\) with \(T > 0\) (which they call least square estimator) is in fact the CLSE, i.e., the limit in probability as \(n \rightarrow \infty \) of the CLSE based on discrete time observations \((Y_{k/n}, X_{k/n})_{k\in \{0,\ldots ,{\lfloor nT\rfloor }\}}\), \(n \in \mathbb {N}\) which can be shown by the method of the proof of Lemma 3.3. They proved strong consistency and asymptotic normality in case of a subcritical process (Y, X), i.e., when \(b > 0\). Note that Barczy and Pap (2016) studied the maximum likelihood estimator (MLE) \((\widetilde{a}_T, \widetilde{b}_T, \widetilde{\alpha }_T, \widetilde{\beta }_T)\) of the parameters \((a, b, \alpha , \beta )\) in this Heston model under the additional assumption \(a \geqslant \frac{\sigma _1^2}{2}\). In the subcritical case, i.e., when \(b > 0\), for \((\widetilde{a}_T, \widetilde{b}_T, \widetilde{\alpha }_T, \widetilde{\beta }_T)\), they proved strong consistency and asymptotic normality in case of \(a > \frac{\sigma _1^2}{2}\), and weak consistency in case of \(a = \frac{\sigma _1^2}{2}\). In the critical case, namely, if \(b = 0\), under the additional assumption \(a > \frac{\sigma _1^2}{2}\), they showed weak consistency of \((\widetilde{a}_T, \widetilde{b}_T, \widetilde{\alpha }_T, \widetilde{\beta }_T)\), asymptotic normality of \((\widetilde{a}_T, \widetilde{\alpha }_T)\), and determined the asymptotic behavior of \((\widetilde{a}_T, \widetilde{b}_T, \widetilde{\alpha }_T, \widetilde{\beta }_T)\). In the supercritical case, namely, when \(b < 0\), they showed that \(\widetilde{b}_T\) is strongly consistent, \(\widetilde{\beta }_T\) is weakly consistent, \((\widetilde{b}_T, \widetilde{\beta }_T)\) is asymptotically mixed normal, and determined the asymptotic behavior of \((\widetilde{a}_T, \widetilde{b}_T, \widetilde{\alpha }_T, \widetilde{\beta }_T)\). Barczy et al. (2018a, b) studied the asymptotic behavior of maximum likelihood estimators for a jump-type Heston model and for the growth rate of a jump-type CIR process, respectively, based on continuous time observations.
We consider general two-factor affine diffusions (1.1). In the subcritical case, i.e., when \(b > 0\) and \(\gamma > 0\), we prove strong consistency and asymptotic normality of \((\widehat{a}_T, \widehat{b}_T, \widehat{\alpha }_T, \widehat{\beta }_T, \widehat{\gamma }_T)\) under the additional assumptions \(a > 0\), \(\sigma _1 > 0\) and \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). In a special critical case, namely if \(b = 0\) and \(\gamma = 0\), we show weak consistency of \((\widehat{b}_T, \widehat{\beta }_T, \widehat{\gamma }_T)\) and determine the asymptotic behavior of \((\widehat{a}_T, \widehat{b}_T, \widehat{\alpha }_T, \widehat{\beta }_T, \widehat{\gamma }_T)\) under the additional assumptions \(\beta = 0\) and \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). In a special supercritical case, namely, when \(\gamma< b < 0\), we show strong consistency of \(\widehat{b}_T\), weak consistency of \((\widehat{\beta }_T, \widehat{\gamma }_T)\) and prove asymptotic mixed normality of \((\widehat{a}_T, \widehat{b}_T, \widehat{\alpha }_T, \widehat{\beta }_T, \widehat{\gamma }_T)\) under the additional assumptions \(\alpha \beta \leqslant 0\), \(\sigma _1 > 0\), and either \(\sigma _3 > 0\), or \(\bigl (a - \frac{\sigma _1^2}{2}\bigr ) (1 - \varrho ^2) \sigma _2^2 > 0\). Note that we decided to deal with the CLSE of \((a, b, \alpha , \beta , \gamma )\), since the MLE of \((a, b, \alpha , \beta , \gamma )\) contains, for example, \(\int _0^T \frac{X_t}{(1-\varrho ^2)\sigma _2^2Y_t+\sigma _3^2} \, \mathrm {d}t\), and the question of the asymptotic behavior of this integral as \(T \rightarrow \infty \) is still open in the critical and supercritical cases. For the sake of brevity of the paper some simple proofs and calculation steps are omitted. However, all these details are included in the extended arXiv version Bolyog and Pap (2017) of this paper.
2 The affine two-factor model
Let \(\mathbb {N}\), \(\mathbb {Z}_+\), \(\mathbb {R}\), \(\mathbb {R}_+\), \(\mathbb {R}_{++}\), \(\mathbb {R}_-\), \(\mathbb {R}_{--}\) and \(\mathbb {C}\) denote the sets of positive integers, non-negative integers, real numbers, non-negative real numbers, positive real numbers, non-positive real numbers, negative real numbers and complex numbers, respectively. For \(x, y \in \mathbb {R}\), we will use the notations \(x \wedge y := \min (x, y)\) and \(x \vee y := \max (x, y)\). By \(C^2_\mathrm {c}(\mathbb {R}_+ \times \mathbb {R}, \mathbb {R})\), we denote the set of twice continuously differentiable real-valued functions on \(\mathbb {R}_+ \times \mathbb {R}\) with compact support. Let \((\varOmega , \mathcal {F}, \mathbb {P})\) be a probability space equipped with the augmented filtration \((\mathcal {F}_t)_{t\in \mathbb {R}_+}\) corresponding to \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) and a given initial value \((\eta _0, \xi _0)\) being independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) such that \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\), constructed as in Karatzas and Shreve (1991, Sect. 5.2). Note that \((\mathcal {F}_t)_{t\in \mathbb {R}_+}\) satisfies the usual conditions, i.e., the filtration \((\mathcal {F}_t)_{t\in \mathbb {R}_+}\) is right-continuous and \(\mathcal {F}_0\) contains all the \(\mathbb {P}\)-null sets in \(\mathcal {F}\). We will denote the convergence in distribution, convergence in probability, almost surely convergence and equality in distribution by \({\mathop {\longrightarrow }\limits ^{\mathcal {D}}}\), \({\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\), \({\mathop {\longrightarrow }\limits ^{{\mathrm {a.s.}}}}\) and \({\mathop {=}\limits ^{\mathcal {D}}}\), respectively. By \(\Vert {\varvec{x}}\Vert \) and \(\Vert {\varvec{A}}\Vert \), we denote the Euclidean norm of a vector \({\varvec{x}}\in \mathbb {R}^d\) and the spectral norm of a matrix \({\varvec{A}}\in \mathbb {R}^{d \times d}\), respectively. By \({\varvec{I}}_d \in \mathbb {R}^{d \times d}\), we denote the \(d\times d\) unit matrix. For square matrices \({\varvec{A}}_1, \ldots , {\varvec{A}}_k\), \({\text {diag}}({\varvec{A}}_1, \ldots , {\varvec{A}}_k)\) will denote the square block matrix containing the matrices \({\varvec{A}}_1, \ldots , {\varvec{A}}_k\) in its diagonal.
The next proposition is about the existence and uniqueness of a strong solution of the SDE (1.1), see Bolyog and Pap (2016, Proposition 2.2).
Proposition 2.1
Let \((\eta _0, \xi _0)\) be a random vector independent of the process \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Then for all \(a \in \mathbb {R}_+\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\), \(\varrho \in [-1, 1]\), there is a (pathwise) unique strong solution \((Y_t, X_t)_{t\in \mathbb {R}_+}\) of the SDE (1.1) such that \(\mathbb {P}((Y_0, X_0) = (\eta _0, \xi _0)) = 1\) and \(\mathbb {P}(Y_t \in \mathbb {R}_+ \ for\ all \ t \in \mathbb {R}_+) = 1\). Further, for all \(s, t \in \mathbb {R}_+\) with \(s \leqslant t\), we have
and
Moreover, \((Y_t, X_t)_{t\in \mathbb {R}_+}\) is a two-factor affine process with infinitesimal generator
where \((y,x) \in \mathbb {R}_+ \times \mathbb {R}\), \(f \in \mathcal {C}^2_c(\mathbb {R}_+ \times \mathbb {R}, \mathbb {R})\), and \(f_i'\), \(i \in \{1, 2\}\), and \(f_{i,j}''\), \(i, j \in \{1, 2\}\), denote the first and second order partial derivatives of f with respect to its i-th and i-th and j-th variables.
Conversely, every two-factor affine diffusion process is a (pathwise) unique strong solution of a SDE (1.1) with suitable parameters \(a \in \mathbb {R}_+\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\).
The next proposition gives the asymptotic behavior of the first moment of the process \((Y_t, X_t)_{t\in \mathbb {R}_+}\) as \(t \rightarrow \infty \), see Bolyog and Pap (2016, Prop. 2.3).
Proposition 2.2
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\), \(\varrho \in [-1, 1]\). Suppose that \(\mathbb {E}(Y_0 |X_0|) < \infty \).
-
(i)
If \(b, \gamma \in \mathbb {R}_{++}\) then \(\lim _{t\rightarrow \infty } \mathbb {E}(Y_t) = \frac{a}{b}\) and \(\lim _{t\rightarrow \infty } \mathbb {E}(X_t) = \frac{\alpha }{\gamma } - \frac{a\beta }{b\gamma }\).
-
(ii)
If \(b \in \mathbb {R}_{++}\) and \(\gamma = 0\) then \(\lim _{t\rightarrow \infty } \mathbb {E}(Y_t) = \frac{a}{b}\) and \(\lim _{t\rightarrow \infty } t^{-1} \mathbb {E}(X_t) = \alpha - \frac{a\beta }{b}\).
-
(iii)
If \(b = 0\) and \(\gamma \in \mathbb {R}_{++}\) then \(\lim _{t\rightarrow \infty } t^{-1} \mathbb {E}(Y_t) = a\) and \(\lim _{t\rightarrow \infty } t^{-1} \mathbb {E}(X_t) = - \frac{a\beta }{\gamma }\).
-
(iv)
If \(b = \gamma = 0\) then \(\lim _{t\rightarrow \infty } t^{-1} \mathbb {E}(Y_t) = a\) and \(\lim _{t\rightarrow \infty } t^{-2} \mathbb {E}(X_t) = - \frac{1}{2} a \beta \).
-
(v)
Otherwise, there exists \(c \in \mathbb {R}_{++}\) such that \(\lim _{t\rightarrow \infty } \mathrm {e}^{-ct} \mathbb {E}(Y_t) \in \mathbb {R}\) or \(\lim _{t\rightarrow \infty } \mathrm {e}^{-ct} \mathbb {E}(Y_t) \in \mathbb {R}\).
Based on the asymptotic behavior of the first moment of the process \((Y_t, X_t)_{t\in \mathbb {R}_+}\) as \(t \rightarrow \infty \), we can classify two-factor affine diffusions in the following way.
Definition 2.3
Let \((Y_t, X_t)_{t\in \mathbb {R}_+}\) be the unique strong solution of the SDE (1.1) satisfying \(\mathbb {P}(Y_0 \in \mathbb {R}_+) = 1\). We call \((Y_t, X_t)_{t\in \mathbb {R}_+}\) subcritical, critical or supercritical if \(b \wedge \gamma \in \mathbb {R}_{++}\), \(b \wedge \gamma = 0\) or \(b \wedge \gamma \in \mathbb {R}_{--}\), respectively.
3 CLSE based on continuous time observations
Overbeck and Rydén (1997) investigated the CIR process Y, and for each \(T \in \mathbb {R}_{++}\), they defined a CLSE \((\widehat{a}_T, \widehat{b}_T)\) of (a, b) based on continuous time observations \((Y_t)_{t\in [0,T]}\) as the limit in probability of the CLSE \((\widehat{a}_{T,n}, \widehat{b}_{T,n})\) of (a, b) based on discrete time observations \((Y_{\frac{iT}{n}})_{i\in \{0,1,\ldots ,n\}}\) as \(n \rightarrow \infty \).
We consider a two-factor affine diffusion process \((Y_t, X_t)_{t\in \mathbb {R}_+}\) given in (1.1) with known \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\), and with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\), and we will consider \({\varvec{\theta }}= (a, b, \alpha , \beta , \gamma )^\top \in \mathbb {R}_+ \times \mathbb {R}^4\) as a parameter. The aim of the following discussion is to construct a CLSE of \({\varvec{\theta }}\) based on continuous time observations \((Y_t, X_t)_{t\in [0,T]}\) with some \(T \in \mathbb {R}_{++}\).
Let us recall the CLSE \({\widehat{{\varvec{\theta }}}}_{T,n}\) of \({\varvec{\theta }}\) based on discrete time observations \((Y_{\frac{i}{n}}, X_{\frac{i}{n}})_{i\in \{0,1,\ldots ,{\lfloor nT\rfloor }\}}\) with some \(n \in \mathbb {N}\), which can be obtained by solving the extremum problem
By (2.1) and (2.2), together with Proposition 3.2.10 in Karatzas and Shreve (1991), for all \(i \in \mathbb {N}\), we obtain
and
Consequently,
where
with
Since the function \(g_n : \mathbb {R}^5 \rightarrow \mathbb {R}\times (-\infty , 1) \times \mathbb {R}^2 \times (-\infty , 1)\) is bijective, first we determine the CLSE \((\widehat{c}_{T,n}, \widehat{d}_{T,n}, \widehat{\delta }_{T,n}, \widehat{\varepsilon }_{T,n}, \widehat{\zeta }_{T,n})\) of the transformed parameters \((c, d, \delta , \varepsilon , \zeta )\) by minimizing the sum on the right-hand side of (3.1) with respect to \((c, d, \delta , \varepsilon , \zeta )\). We have
hence, similarly as on page 675 in Barczy et al. (2013), we get
with
on the event where the random matrices \({\varvec{\varGamma }}_{T,n}^{(1)}\) and \({\varvec{\varGamma }}_{T,n}^{(2)}\) are invertible.
Lemma 3.1
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). Then for each \(T \in \mathbb {R}_{++}\) and \(n \in \mathbb {N}\), the random matrices \({\varvec{\varGamma }}_{T,n}^{(1)}\) and \({\varvec{\varGamma }}_{T,n}^{(2)}\) are invertible almost surely, and hence there exists a unique CLSE \(\bigl (\widehat{c}_{T,n}, \widehat{d}_{T,n}, \widehat{\delta }_{T,n}, \widehat{\varepsilon }_{T,n}, \widehat{\zeta }_{T,n}\bigr )\) of \((c, d, \delta , \varepsilon , \zeta )\) taking the form given in (3.3).
A proof can be found in the Arxiv version of this paper Bolyog and Pap (2017).
Remark 3.2
The first order Taylor approximation of \(g_n(a, b, \alpha , \beta , \gamma )\) at (0, 0, 0, 0, 0) is \(\frac{1}{n}(a, b, \alpha , \beta , \gamma )\), hence we obtain the first order Taylor approximations
Using these approximations, one can define an approximate CLSE \({\widehat{{\varvec{\theta }}}}_{T,n}^{\mathrm {approx}}\) of \({\varvec{\theta }}\) based on discrete time observations \((Y_i, X_i)_{i\in \{0,1,\ldots ,{\lfloor nT\rfloor }\}}\), \(n \in \mathbb {N}\), by solving the extremum problem
hence \({\widehat{{\varvec{\theta }}}}_{T,n}^{\mathrm {approx}} = n \bigl (\widehat{c}_{T,n}, \widehat{d}_{T,n}, \widehat{\delta }_{T,n}, \widehat{\varepsilon }_{T,n}, \widehat{\zeta }_{T,n}\bigr )^\top \). This definition of approximate CLSE can be considered as the definition of LSE given in Hu and Long (2009a, formula (1.2)) for generalized Ornstein–Uhlenbeck processes driven by \(\alpha \)-stable motions, see also Hu and Long (2009b, formula (3.1)). For a heuristic motivation of the estimator \({\widehat{{\varvec{\theta }}}}_n^{\mathrm {approx}}\) based on discrete observations, see, e.g., Hu and Long (2007, p. 178) (formulated for Langevin equations). \(\square \)
We have
as \(n \rightarrow \infty \), since \((Y_t, X_t)_{t\in \mathbb {R}_+}\) is almost surely continuous. By Proposition I.4.44 in Jacod and Shiryaev (2003) with the Riemann sequence of deterministic subdivisions \(\left( \frac{i}{n} \wedge T\right) _{i\in \mathbb {N}}\), \(n \in \mathbb {N}\)., we obtain
as \(n \rightarrow \infty \). By Slutsky’s lemma, using also Lemma 3.1, we conclude
whenever the random matrices \({\varvec{G}}_T^{(1)}\) and \({\varvec{G}}_T^{(2)}\) are invertible.
Lemma 3.3
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). Then for each \(T \in \mathbb {R}_{++}\), the random matrices \({\varvec{G}}_T^{(1)}\) and \({\varvec{G}}_T^{(2)}\) are invertible almost surely, and hence \({\widehat{{\varvec{\theta }}}}_T\) given in (3.4) exists almost surely. Moreover, \({\widehat{{\varvec{\theta }}}}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}{\widehat{{\varvec{\theta }}}}_T\) as \(n \rightarrow \infty \).
Proof
A proof of the first statement can be found in the Arxiv version of this paper Bolyog and Pap (2017). Next we are going to show \({\widehat{{\varvec{\theta }}}}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}{\widehat{{\varvec{\theta }}}}_T\) as \(n \rightarrow \infty \). The function \(g_n\) introduced in (3.2) admits an inverse \(g_n^{-1} : \mathbb {R}\times (-\infty , 1) \times \mathbb {R}^2 \times (-\infty , 1) \rightarrow \mathbb {R}^5\) satisfying
with
Convergence (3.4) yields \((\widehat{c}_{T,n}, \widehat{d}_{T,n}, \widehat{\delta }_{T,n}, \widehat{\varepsilon }_{T,n}, \widehat{\zeta }_{T,n}) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}{\varvec{0}}\) as \(n \rightarrow \infty \), hence \(\widehat{d}_{T,n} \in (-\infty , 1)\) and \(\widehat{\zeta }_{T,n} \in (-\infty , 1)\) with probability tending to one as \(n \rightarrow \infty \). Consequently, \(g_n^{-1}(\widehat{c}_{T,n}, \widehat{d}_{T,n}, \widehat{\delta }_{T,n}, \widehat{\varepsilon }_{T,n}, \widehat{\zeta }_{T,n}) ={\widehat{{\varvec{\theta }}}}_{T,n}\) with probability tending to one as \(n \rightarrow \infty \). We have
with probability tending to one as \(n \rightarrow \infty \), where the continuous function \(h_1 : (-\infty , 1) \rightarrow \mathbb {R}\) is given by
By (3.4), we have \(n \widehat{d}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{b}_T\) and \(\widehat{d}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\), thus we obtain \(h_1(\widehat{d}_{T,n}) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}h_1(0) = 1\), and hence \(\widehat{b}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{b}_T\) as \(n \rightarrow \infty \).
Moreover,
with probability tending to one as \(n \rightarrow \infty \), where the continuous function \(h_2 : \mathbb {R}\rightarrow \mathbb {R}\) is given by
We have already showed \(\widehat{b}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{b}_T\), yielding \(n^{-1} \widehat{b}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\), and hence \(h_2(n^{-1} \widehat{b}_{T,n}) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}h_2(0) = 1\) as \(n \rightarrow \infty \). By (3.4), we have \(n \widehat{c}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{a}_T\), thus we obtain \(\widehat{a}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{a}_T\) as \(n \rightarrow \infty \).
In a similar way,
with probability tending to one as \(n \rightarrow \infty \). By (3.4), we have \(n \widehat{\zeta }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\gamma }_T\) and \(\widehat{\zeta }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\), thus we obtain \(h_1(\widehat{\zeta }_{T,n}) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}h_1(0) = 1\), and hence \(\widehat{\gamma }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\gamma }_T\) as \(n \rightarrow \infty \).
Further,
with probability tending to one as \(n \rightarrow \infty \). We have already showed \(\widehat{b}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{b}_T\) and \(\widehat{\gamma }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\gamma }_T\), yielding \(n^{-1} \widehat{b}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) and \(n^{-1} \widehat{\gamma }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\), and hence \(\mathrm {e}^{\frac{\widehat{\gamma }_{T,n}}{n}} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}1\) and \(h_2(n^{-1} (\widehat{b}_{T,n} - \widehat{\gamma }_{T,n})) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}h_2(0) = 1\) as \(n \rightarrow \infty \). By (3.4), we have \(n \widehat{\varepsilon }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\beta }_T\), thus we obtain \(\widehat{\beta }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\beta }_T\) as \(n \rightarrow \infty \).
Finally,
with probability tending to one as \(n \rightarrow \infty \), where
We have already showed \(\widehat{a}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{a}_T\), \(\widehat{b}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{b}_T\), \(\widehat{\beta }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\beta }_T\) and \(\widehat{\gamma }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\gamma }_T\), yielding \(n^{-1} \widehat{b}_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) and \(n^{-1} \widehat{\gamma }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\), and hence \(h_2(n^{-1} \widehat{\gamma }_{T,n}) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}h_2(0) = 1\), \(\mathrm {e}^{\frac{\widehat{\gamma }_{T,n}}{n}} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}1\), \(\mathrm {e}^{\frac{|\widehat{\gamma }_{T,n}|}{n}} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}1\) and \(\mathrm {e}^{\frac{|\widehat{b}_{T,n}|}{n}} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}1\), implying \(I_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) as \(n \rightarrow \infty \). By (3.4), we have \(n \widehat{\delta }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\alpha }_T\), thus we obtain \(\widehat{\alpha }_{T,n} {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}\widehat{\alpha }_T\) as \(n \rightarrow \infty \). \(\square \)
Using the SDE (1.1) and Corollary 3.2.20 in Karatzas and Shreve (1991), one can check that
on the event where the random matrices \({\varvec{G}}_T^{(1)}\) and \({\varvec{G}}_T^{(2)}\) are invertible, where
with
where
is a standard Wiener process, independent of L. For details see the Arxiv version of this paper Bolyog and Pap (2017).
4 Consistency of CLSE
First we consider the case of subcritical Heston models, i.e., when \(b \in \mathbb {R}_{++}\).
Theorem 4.1
Let us consider the two-factor affine diffusion model (1.1) with \(a, b \in \mathbb {R}_{++}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in \mathbb {R}_{++}\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). Then the CLSE of \({\varvec{\theta }}= (a, b, \alpha , \beta , \gamma )^\top \) is strongly consistent, i.e., \({\widehat{{\varvec{\theta }}}}_T = \bigl (\widehat{a}_T, \widehat{b}_T, \widehat{\alpha }_T, \widehat{\beta }_T, \widehat{\gamma }_T\bigr )^\top {\mathop {\longrightarrow }\limits ^{{\mathrm {a.s.}}}}{\varvec{\theta }}= (a, b, \alpha , \beta , \gamma )^\top \) as \(T \rightarrow \infty \).
Proof
By (3.5), we have
on the event, where the random matrix \({\varvec{G}}_T\) is invertible, which has propapility 1, see Lemma 3.3.
By Theorem A.2, we obtain
where
with
where the random vector \((Y_\infty , X_\infty )\) is given by Theorem A.1, since, by Theorem B.2, the entries of \(\mathbb {E}({\varvec{G}}_\infty )\) exist and finite.
The matrix \(\mathbb {E}({\varvec{G}}_\infty ^{(1)})\) is strictly positive definite, since for all \({\varvec{x}}\in \mathbb {R}^2 \setminus \{{\varvec{0}}\}\), we have \({\varvec{x}}^\top \mathbb {E}({\varvec{G}}_\infty ^{(1)}) {\varvec{x}}> 0\). Indeed, for all \({\varvec{x}}= (x_1, x_2)^\top \in \mathbb {R}^2 \setminus \{{\varvec{0}}\}\),
since, by Theorem A.2, the distribution of \(Y_\infty \) is absolutely continuous, hence \(x_1 - x_2 Y_\infty \ne 0\) with probability 1. In a similar way, the matrix \(\mathbb {E}({\varvec{G}}_\infty ^{(2)})\) is strictly positive definite, since for all \({\varvec{x}}\in \mathbb {R}^3 \setminus \{{\varvec{0}}\}\), we have \({\varvec{x}}^\top \mathbb {E}({\varvec{G}}_\infty ^{(2)}) {\varvec{x}}> 0\). Indeed, for all \({\varvec{x}}= (x_1, x_2, x_3)^\top \in \mathbb {R}^3 \setminus \{{\varvec{0}}\}\),
since, by Theorem A.2, the distribution of \((Y_\infty , X_\infty )\) is absolutely continuous, hence \(x_1 - x_2 Y_\infty - x_3 X_\infty \ne 0\) with probability 1. Thus the matrices \(\mathbb {E}({\varvec{G}}_\infty ^{(1)})\) and \(\mathbb {E}({\varvec{G}}_\infty ^{(2)})\) are invertible, whence we conclude
The aim of the next discussion is to show convergence
We have
Indeed, we have already proved
and the strong law of large numbers for continuous local martingales (see, e.g., Theorem C.1) implies
since we have
Further,
as \(T \rightarrow \infty \). Indeed, we have already proved
and the strong law of large numbers for continuous local martingales (see, e.g., Theorem C.1) implies
since we have
One can check
as \(T \rightarrow \infty \) in the same way, since
as \(T \rightarrow \infty \). Consequently, we conclude (4.5). Finally, by (4.4) and (4.5), we obtain the statement. \(\square \)
In order to handle supercritical two-factor affine diffusion models when \(b \in \mathbb {R}_{--}\), we need the following integral version of the Toeplitz Lemma, due to Dietz and Kutoyants (1997).
Lemma 4.2
Let \(\{\varphi _T : T \in \mathbb {R}_+\}\) be a family of probability measures on \(\mathbb {R}_+\) such that \(\varphi _T([0,T]) = 1\) for all \(T \in \mathbb {R}_+\), and \(\lim _{T\rightarrow \infty } \varphi _T([0,K]) = 0\) for all \(K \in \mathbb {R}_{++}\). Then for every bounded and measurable function \(f : \mathbb {R}_+ \rightarrow \mathbb {R}\) for which the limit \(f(\infty ) := \lim _{t\rightarrow \infty } f(t)\) exists, we have
As a special case, we have the following integral version of the Kronecker Lemma, see Küchler and Sørensen (1997, Lemma B.3.2).
Lemma 4.3
Let \(a : \mathbb {R}_+ \rightarrow \mathbb {R}_+\) be a measurable function. Put \(b(T) := \int _0^T a(t) \, \mathrm {d}t\), \(T \in \mathbb {R}_+\). Suppose that \(\lim _{T\rightarrow \infty } b(T) = \infty \). Then for every bounded and measurable function \(f : \mathbb {R}_+ \rightarrow \mathbb {R}\) for which the limit \(f(\infty ) := \lim _{t\rightarrow \infty } f(t)\) exists, we have
Next we present an auxiliary lemma in the supercritical case on the asymptotic behavior of \(Y_t\) as \(t \rightarrow \infty \).
Lemma 4.4
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b \in \mathbb {R}_{--}\), \(\alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Then there exists a random variable \(V_Y\) such that
with \(\mathbb {P}(V_Y \ne 0) = 1\), and, for each \(k \in \mathbb {N}\),
Proof
By (2.1),
for all \(s, t \in \mathbb {R}_+\) with \(0 \leqslant s \leqslant t\). Thus
for all \(s, t \in \mathbb {R}_+\) with \(0 \leqslant s \leqslant t\), consequently, the process \((\mathrm {e}^{bt} Y_t)_{t\in \mathbb {R}_+}\) is a non-negative submartingale with respect to the filtration \((\mathcal {F}^Y_t)_{t\in \mathbb {R}_+}\). Moreover, \(b \in \mathbb {R}_{--}\) implies
hence, by the submartingale convergence theorem, there exists a non-negative random variable \(V_Y\) such that (4.6) holds.
The distribution of \(V_Y\) coincides with the distribution of \(\widetilde{\mathcal {Y}}_{-1/b}\), where \((\widetilde{\mathcal {Y}}_t)_{t\in \mathbb {R}_+}\) is a CIR process given by the SDE
with initial value \(\widetilde{\mathcal {Y}}_0 = y_0\), where \((\mathcal {W}_t)_{t\in \mathbb {R}_+}\) is a standard Wiener process, see Ben Alaya and Kebaier (2012, Proposition 3). Consequently, \(\mathbb {P}(V_Y \in \mathbb {R}_{++}) = 1\), since \(\widetilde{\mathcal {Y}}_t\), \(t \in \mathbb {R}_{++}\), are absolutely continuous random variables.
If \(\omega \in \varOmega \) such that \(\mathbb {R}_+ \ni t \mapsto Y_t(\omega )\) is continuous and \(\mathrm {e}^{bt} Y_t(\omega ) \rightarrow V_Y(\omega )\) as \(t \rightarrow \infty \), then, by the integral Kronecker Lemma 4.3 with \(f(t) = \mathrm {e}^{kbt} Y_t(\omega )^k\) and \(a(t) = \mathrm {e}^{-kbt}\), \(t \in \mathbb {R}_+\), we have
Here \(\int _0^t \mathrm {e}^{-kbu} \, \mathrm {d}u = - \frac{\mathrm {e}^{-kbt} - 1}{kb}\), \(t \in \mathbb {R}_+\), thus we conclude the second convergence in (4.7). \(\square \)
The next theorem states strong consistency of the CLSE of b in the supercritical case.
Theorem 4.5
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b \in \mathbb {R}_{--}\), \(\alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Then the CLSE of b is strongly consistent, i.e., \(\widehat{b}_T {\mathop {\longrightarrow }\limits ^{{\mathrm {a.s.}}}}b\) as \(T \rightarrow \infty \).
Proof
By Lemma 3.3, there exists a unique CLSE \(\widehat{b}_T\) of b for all \(T \in \mathbb {R}_{++}\) which has the form given in (3.4). By Ito’s formula,
hence, by (4.6) and (4.7), we have
as \(T \rightarrow \infty \). \(\square \)
Remark 4.6
For critical two-factor affine diffusion models, it will turn out that the CLSE of a and \(\alpha \) are not even weakly consistent, but the CLSE of b, \(\beta \) and \(\gamma \) are weakly consistent, see Theorem 6.2. \(\square \)
Remark 4.7
For supercritical two-factor affine diffusion models, it will turn out that the CLSE of a and \(\alpha \) are not even weakly consistent, but the CLSE of \(\beta \) and \(\gamma \) are weakly consistent, see Theorem 7.3. \(\square \)
5 Asymptotic behavior of CLSE: subcritical case
Theorem 5.1
Let us consider the two-factor affine diffusion model (1.1) with \(a, b \in \mathbb {R}_{++}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in \mathbb {R}_{++}\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). Then the CLSE of \({\varvec{\theta }}= (a, b, \alpha , \beta , \gamma )^\top \) is asymptotically normal, namely,
where \({\varvec{G}}_\infty \) is given in (4.3) and \(\widetilde{{\varvec{G}}}_\infty \) has the form
where the random vector \((Y_\infty , X_\infty )\) is given by Theorem A.1.
Proof
By (3.5), we have
on the event where \({\varvec{G}}_T\) is invertible, which holds almost surely, see Lemma 3.3. We have \((T^{-1} {\varvec{G}}_T)^{-1} {\mathop {\longrightarrow }\limits ^{{\mathrm {a.s.}}}}[\mathbb {E}({\varvec{G}}_\infty )]^{-1}\) as \(T \rightarrow \infty \) by (4.4). The process \(({\varvec{h}}_t)_{t\in \mathbb {R}_+}\) is a 5-dimensional continuous local martingale with quadratic variation process \(\langle {\varvec{h}}\rangle _t = \widetilde{{\varvec{G}}}_t\), \(t \in \mathbb {R}_+\), where
By Theorem A.2, we obtain
since, by Theorem B.2, the entries of \(\mathbb {E}(\widetilde{{\varvec{G}}}_\infty )\) exist and finite. Using (5.3), Theorem C.2 yields \(T^{-\frac{1}{2}} {\varvec{h}}_T {\mathop {\longrightarrow }\limits ^{\mathcal {D}}}\mathcal {N}_5({\varvec{0}}, \mathbb {E}(\widetilde{{\varvec{G}}}_\infty ))\) as \(T \rightarrow \infty \). Hence, by (5.2) and by Slutsky’s lemma,
as \(T \rightarrow \infty \). \(\square \)
6 Asymptotic behavior of CLSE: critical case
First we present an auxiliary lemma. A proof can be found in the Arxiv version of this paper Bolyog and Pap (2017).
Lemma 6.1
If \((\mathcal {Y}_t, \mathcal {X}_t)_{t\in \mathbb {R}_+}\) and \((\widetilde{\mathcal {Y}}_t, \widetilde{\mathcal {X}}_t)_{t\in \mathbb {R}_+}\) are continuous semimartingales such that \((\mathcal {Y}_t, \mathcal {X}_t)_{t\in \mathbb {R}_+} {\mathop {=}\limits ^{\mathcal {D}}}(\widetilde{\mathcal {Y}}_t, \widetilde{\mathcal {X}}_t)_{t\in \mathbb {R}_+}\), then
for each \(n \in \mathbb {N}\).
Theorem 6.2
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b = 0\), \(\alpha \in \mathbb {R}\), \(\beta = 0\), \(\gamma = 0\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). Then
as \(T \rightarrow \infty \), where \((\mathcal {Y}_t, \mathcal {X}_t)_{t\in \mathbb {R}_+}\) is the unique strong solution of the SDE
with initial value \((\mathcal {Y}_0, \mathcal {X}_0) = (0, 0)\).
Proof
By (3.5), we have
In a similar way,
The aim of the following discussion is to prove
as \(T \rightarrow \infty \). By part (ii) of Remark 2.7 in Barczy et al. (2013), we have
since, by Proposition 2.1, \((\mathcal {Y}_t, \mathcal {X}_t)_{t\in \mathbb {R}_+}\) is an affine process with infinitesimal generator
Hence, by Lemma 6.1, we obtain
for all \(T \in \mathbb {R}_{++}\). Then, by Slutsky’s lemma, in order to prove (6.3), it suffices to show the convergences
as \(T \rightarrow \infty \) for all \(k, \ell \in \mathbb {Z}_+\) with \(k + \ell \leqslant 2\). By (3.21) in Barczy et al. (2013), we have
hence
as \(T \rightarrow \infty \), implying \(\frac{1}{T} (Y_T - \mathcal {Y}_T) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) and \(\frac{1}{T^2} \int _0^T (Y_s - \mathcal {Y}_s) \, \mathrm {d}s {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) as \(T \rightarrow \infty \), i.e., the first convergence in (6.4) and the second convergence in (6.5) for \((k, \ell ) = (1, 0)\).
As in (3.23) in Barczy et al. (2013), we have \(\mathbb {E}(|X_s - \mathcal {X}_s|) \leqslant \mathbb {E}(|X_0|) + \sqrt{(\sigma _2^2 \mathbb {E}(Y_0) + \sigma _3^2) s}\) for all \(s \in \mathbb {R}_+\), hence
thus
as \(T \rightarrow \infty \), implying \(\frac{1}{T} (X_T - \mathcal {X}_T) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) and \(\frac{1}{T^2} \int _0^T (X_s - \mathcal {X}_s) \, \mathrm {d}s {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) as \(T \rightarrow \infty \), i.e., the second convergence in (6.4) and the second convergence in (6.5) for \((k, \ell ) = (0, 1)\).
As in (3.25) in Barczy et al. (2013), we have \(\mathbb {E}[(Y_s - \mathcal {Y}_s)^2] \leqslant 2 \mathbb {E}(Y_0^2) + 2 s \sigma _1^2 \mathbb {E}(Y_0)\) for all \(s \in \mathbb {R}_+\), hence
By Proposition B.1, \(\mathbb {E}(Y_s^2) = \mathbb {E}(Y_0^2) + (2 a + \sigma _1^2) \bigl (\mathbb {E}(Y_0) s + a \frac{s^2}{2}\bigr )\) for all \(s \in \mathbb {R}_+\), hence
and \(\sup _{s\in [0,T]} \mathbb {E}(\mathcal {Y}_s^2) = {\text {O}}(T^2)\) as \(T \rightarrow \infty \). We have
yielding
thus
as \(T \rightarrow \infty \), implying \(\frac{1}{T^3} \int _0^T (Y_s^2 - \mathcal {Y}_s^2) \, \mathrm {d}s {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) as \(T \rightarrow \infty \)., i.e., the second convergence in (6.5) for \((k, \ell ) = (2, 0)\).
In a similar way, \(\mathbb {E}[(X_s - \mathcal {X}_s)^2] \leqslant 2 \mathbb {E}(X_0^2) + 2 s (\sigma _2^2 \mathbb {E}(Y_0) + \sigma _3^2)\) for all \(s \in \mathbb {R}_+\), hence
By Proposition B.1, \(\mathbb {E}(X_s^2) = \mathbb {E}(X_0^2) + \alpha \bigl (s \mathbb {E}(X_0) + \alpha \frac{s^2}{2}\bigr ) + \sigma _2^2 \bigl (s \mathbb {E}(Y_0) + a \frac{s^2}{2}\bigr ) + \sigma _3^2 s\), thus \(\sup _{s\in [0,T]} \mathbb {E}(X_s^2) = {\text {O}}(T^2)\) and \(\sup _{s\in [0,T]} \mathbb {E}(\mathcal {X}_s^2) = {\text {O}}(T^2)\) as \(T \rightarrow \infty \). We have
yielding
thus
as \(T \rightarrow \infty \), implying \(\frac{1}{T^3} \int _0^T (X_s^2 - \mathcal {X}_s^2) \, \mathrm {d}s {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) as \(T \rightarrow \infty \), i.e., the second convergence in (6.5) for \((k, \ell ) = (0, 2)\).
Further,
yields
thus
as \(T \rightarrow \infty \), implying \(\frac{1}{T^3} \int _0^T (Y_s X_s - \mathcal {Y}_s \mathcal {X}_s) \, \mathrm {d}s {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) as \(T \rightarrow \infty \), i.e., the second convergence in (6.5) for \((k, \ell ) = (1, 1)\).
Using the Cauchy–Schwarz inequality, we obtain
with
Using \(\mathrm {d}Y_s = a \, \mathrm {d}s + \sigma _1 \sqrt{Y_s} \, \mathrm {d}W_s\), we have
with
Applying (6.10), we obtain
Again by the Cauchy–Schwarz inequality, we obtain
Using \(X_t = X_0 + \sigma _2 \int _0^t \sqrt{Y_s} \, \mathrm {d}\widetilde{W}_s + \sigma _3 L_t\) and \(\mathcal {X}_t = \sigma _2 \int _0^t \sqrt{\mathcal {Y}_s} \, \mathrm {d}\widetilde{W}_s\), we get \(X_t - \mathcal {X}_t = X_0 + \sigma _2 \int _0^t (\sqrt{Y_s} - \sqrt{\mathcal {Y}_s}) \, \mathrm {d}\widetilde{W}_s + \sigma _3 L_t\), and, applying Minkowski inequality and a martingale moment inequality in Karatzas and Shreve (1991, 3.3.25), we obtain
Applying (6.8), we get
which, by (6.9), implies \(E_{1,2}(T) = \int _0^T \sqrt{{\text {O}}(T^3) {\text {O}}(T^2)} \, \mathrm {d}s = {\text {O}}(T^{\frac{7}{2}})\) as \(T \rightarrow \infty \). Using \(E_{1,1}(T) = {\text {O}}(T^3)\) as \(T \rightarrow \infty \), we conclude \(E_1(T) = {\text {O}}(T^3) + {\text {O}}(T^{\frac{7}{2}})= {\text {O}}(T^{\frac{7}{2}})\) as \(T \rightarrow \infty \).
Using \(\mathrm {d}Y_s = a \, \mathrm {d}s + \sigma _1 \sqrt{Y_s} \, \mathrm {d}W_s\) and \(\mathrm {d}\mathcal {Y}_s = a \, \mathrm {d}s + \sigma _1 \sqrt{\mathcal {Y}_s} \, \mathrm {d}W_s\), we obtain \(\mathrm {d}(Y_t - \mathcal {Y}_t) = \sigma _1 (\sqrt{Y_t} - \sqrt{\mathcal {Y}_t}) \, \mathrm {d}W_t\), thus
Using \(\mathcal {X}_t = \alpha t + \sigma _2 \int _0^t \sqrt{\mathcal {Y}_s} \, \mathrm {d}\widetilde{W}_s\), we obtain
hence we conclude
Using (6.8), we obtain \(E_2(T) = \int _0^T \sqrt{{\text {O}}(T^4) {\text {O}}(T)} \, \mathrm {d}s = {\text {O}}(T^{\frac{7}{2}})\) as \(T \rightarrow \infty \). Hence
as \(T \rightarrow \infty \), implying \(\frac{1}{T^2} \left( \int _0^T X_s \, \mathrm {d}Y_s - \int _0^T \mathcal {X}_s \, \mathrm {d}\mathcal {Y}_s\right) {\mathop {\longrightarrow }\limits ^{\mathbb {P}}}0\) as \(T \rightarrow \infty \), i.e., the first convergence in (6.5). Thus we conclude convergence (6.3).
Applying the first equation of (1.1) and using \(b = 0\), we obtain
By Itô’s formula and using \(b = 0\),
hence
Consequently,
In a similar way, applying the second equation of (1.1) and using \(\beta = 0\) and \(\gamma = 0\), we obtain
By Itô’s formula and using \(\beta = 0\) and \(\gamma = 0\),
hence
Consequently,
Again by Itô’s formula and using \(\beta = 0\) and \(\gamma = 0\),
hence
Consequently,
Applying (6.3) and the continuous mapping theorem, we obtain
jointly as \(T \rightarrow \infty \). Applying again the continuous mapping theorem, we conclude (6.1), since the limiting random matrices in the first and third convergences above are almost surely invertible by Lemma 3.1. \(\square \)
7 Asymptotic behavior of CLSE: supercritical case
First we present an auxiliary lemma about the asymptotic behavior of \(\mathbb {E}(X_t^2)\) as \(t \rightarrow \infty \).
Lemma 7.1
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b \in \mathbb {R}_{--}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in (-\infty , b)\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Then \(\sup _{t\in \mathbb {R}_+} \mathrm {e}^{2\gamma t} \mathbb {E}(X_t^2) < \infty \).
Proof
By Proposition B.1,
since \(b < 0\). Moreover,
using \(\gamma < 0\) and \(\gamma - b < 0\). Again by Proposition B.1,
using \(b < 0\). Hence
using \(b < 0\), \(\gamma < 0\) and \(\gamma - b < 0\). Consequently,
using \(\gamma < 0\), \(\gamma - b < 0\) and \(2 \gamma - b < 0\). \(\square \)
Next we present an auxiliary lemma about the asymptotic behavior of \(X_t\) as \(t \rightarrow \infty \).
Lemma 7.2
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b \in \mathbb {R}_{--}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in (-\infty , b)\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \(\alpha \beta \in \mathbb {R}_-\). Then there exists a random variable \(V_X\) such that
and, for each \(k, \ell \in \mathbb {Z}_+\) with \(k + \ell > 0\),
where \(V_Y\) is given in (4.6). If, in addition, \(\sigma _3 \in \mathbb {R}_{++}\) or \(\bigl (a - \frac{\sigma _1^2}{2}\bigr ) (1 - \varrho ^2) \sigma _2^2 \in \mathbb {R}_{++}\), then the distribution of the random variable \(V_X\) is absolutely continuous. Particularly, \(\mathbb {P}(V_X \ne 0) = 1\).
Proof
By (2.2),
for all \(s, t \in \mathbb {R}_+\) with \(0 \leqslant s \leqslant t\). If \(\alpha \in \mathbb {R}_+\) and \(\beta \in \mathbb {R}_-\), then
for all \(s, t \in \mathbb {R}_+\) with \(0 \leqslant s \leqslant t\), consequently, the process \((\mathrm {e}^{\gamma t} X_t)_{t\in \mathbb {R}_+}\) is a submartingale with respect to the filtration \((\mathcal {F}^{Y,X}_t)_{t\in \mathbb {R}_+}\). If \(\alpha \in \mathbb {R}_-\) and \(\beta \in \mathbb {R}_+\), then
for all \(s, t \in \mathbb {R}_+\) with \(0 {\leqslant } s {\leqslant } t\), consequently, the process \((\mathrm {e}^{\gamma t} X_t)_{t\in \mathbb {R}_+}\) is a supermartingale with respect to the filtration \((\mathcal {F}^{Y,X}_t)_{t\in \mathbb {R}_+}\), hence the process \((-\mathrm {e}^{\gamma t} X_t)_{t\in \mathbb {R}_+}\) is a submartingale with respect to the filtration \((\mathcal {F}^{Y,X}_t)_{t\in \mathbb {R}_+}\). In both cases, \(\sup _{t\in \mathbb {R}_+} \mathbb {E}(|\mathrm {e}^{\gamma t} X_t|^2) < \infty \), see Lemma 7.1. Hence, by the submartingale convergence theorem, there exists a random variable \(V_X\) such that (7.1) holds.
If \(\omega \in \varOmega \) such that \(\mathbb {R}_+ \ni t \mapsto (Y_t(\omega ), X_t(\omega ))\) is continuous and \((\mathrm {e}^{bt} Y_t(\omega ), \mathrm {e}^{\gamma t} X_t(\omega )) \rightarrow (V_Y(\omega ), V_X(\omega ))\) as \(t \rightarrow \infty \), then, by the integral Kronecker Lemma 4.3 with \(f(t) = \mathrm {e}^{(kb+\ell \gamma )t} Y_t(\omega )^k X_t(\omega )^\ell \) and \(a(t) = \mathrm {e}^{-(kb+\ell \gamma )t}\), \(t \in \mathbb {R}_+\), we have
Here \(\int _0^t \mathrm {e}^{-(kb+\ell \gamma )u} \, \mathrm {d}u = - \frac{\mathrm {e}^{-(kb+\ell \gamma )t} - 1}{kb+\ell \gamma }\), \(t \in \mathbb {R}_+\), thus we conclude (7.2).
Now suppose that \(\sigma _3 \in \mathbb {R}_{++}\) or \(\bigl (a - \frac{\sigma _1^2}{2}\bigr ) (1 - \varrho ^2) \sigma _2^2 \in \mathbb {R}_{++}\). We are going to show that the random variable \(V_X\) is absolutely continuous. Put \(Z_t := X_t - r Y_t\), \(t \in \mathbb {R}_+\) with \(r := \frac{\sigma _2 \varrho }{\sigma _1}\). Then the process \((Y_t, Z_t)_{t\in \mathbb {R}_+}\) is an affine process satisfying
where \(A := \alpha - r a\), \(B := \beta - r (b - \gamma )\) and \(\varSigma _2 := \sigma _2 \sqrt{1 - \varrho ^2}\), see (Bolyog and Pap 2016, Proposition 2.5). We have
where we used (2.2) with \(s = 0\) multiplied both sides by \(\mathrm {e}^{\gamma t}\). Thus the conditional distribution of \(\mathrm {e}^{\gamma t} X_t\) given \((Y_u)_{u\in [0,t]}\) and \(X_0\) is a normal distribution with mean \(r \mathrm {e}^{\gamma t} Y_t + Z_0 + \int _0^t \mathrm {e}^{\gamma u} (A - B Y_u) \, \mathrm {d}u\) and with variance \(\varSigma _2^2 \int _0^t \mathrm {e}^{2\gamma u} Y_u \, \mathrm {d}u + \sigma _3^2 \int _0^t \mathrm {e}^{2\gamma u} \, \mathrm {d}u\). Hence
Consequently,
Convergence (7.1) implies \(\mathrm {e}^{\gamma t} X_t {\mathop {\longrightarrow }\limits ^{\mathcal {D}}}V_X\) as \(t \rightarrow \infty \), hence, by the continuity theorem and by the monotone convergence theorem,
for all \(\lambda \in \mathbb {R}\). If \(\sigma _3 \in \mathbb {R}_{++}\), then we have
for all \(\lambda \in \mathbb {R}\), hence \(\int _{-\infty }^\infty \bigl |\mathbb {E}\bigl (\mathrm {e}^{\mathrm {i}\lambda V_X}\bigr )\bigr | \mathrm {d}\lambda < \infty \), implying absolute continuity of the distribution of \(V_X\).
If \(\bigl (a - \frac{\sigma _1^2}{2}\bigr ) (1 - \varrho ^2) \sigma _2^2 \in \mathbb {R}_{++}\), then we have
for all \(\lambda \in \mathbb {R}\). Applying the comparison theorem (see, e.g., Karatzas and Shreve 1991, 5.2.18), we obtain \(\mathbb {P}(\mathcal {Y}_t \leqslant Y_t \ \text {for all }\ t \in \mathbb {R}_+) = 1\), where \((\mathcal {Y}_t)_{t\in \mathbb {R}_+}\) is the unique strong solution of the SDE
with initial value \(\mathcal {Y}_0 = 0\). Consequently, taking into account \(\varSigma _2 = \sigma _2 \sqrt{1 - \varrho ^2} > 0\), we obtain
whenever
By the Cauchy–Schwarz inequality, we have
hence
For each \(u \in \mathbb {R}_{++}\), we have \(\mathcal {Y}_u {\mathop {=}\limits ^{\mathcal {D}}}c(u) \xi \), where the distribution of \(\xi \) has a chi-square distribution with degrees of freedom \(\frac{4a}{\sigma _1^2}\) and \(c(u) := \frac{\sigma _1^2}{4} \int _0^u \mathrm {e}^{-bv} \, \mathrm {d}v = \frac{\sigma _1^2(\mathrm {e}^{-bu}-1)}{4(-b)}\), see Proposition B.1. Hence
where \(\mathbb {E}\bigl (\frac{1}{\xi }\bigr ) < \infty \), since the density of \(\xi \) has the form
and the assumption \(a - \frac{\sigma _1^2}{2} > 0\) yields \(\frac{2a}{\sigma _1^2} - 1 > 0\). Consequently,
thus we obtain (7.3), and hence \(\int _{-\infty }^\infty \bigl |\mathbb {E}\bigl (\mathrm {e}^{\mathrm {i}\lambda V_X}\bigr )\bigr | \mathrm {d}\lambda < \infty \), and we conclude absolute continuity of the distribution of \(V_X\). \(\square \)
Theorem 7.3
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b \in \mathbb {R}_{--}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in (-\infty , b)\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \(\alpha \beta \in \mathbb {R}_-\). Suppose that \(\sigma _3 \in \mathbb {R}_{++}\) or \(\bigl (a - \frac{\sigma _1^2}{2}\bigr ) (1 - \varrho ^2) \sigma _2^2 \in \mathbb {R}_{++}\). Then
as \(T \rightarrow \infty \) with
where \(V_Y\) and \(V_X\) are given in (4.6) and (7.1), respectively, \({\varvec{\eta }}\) is a \(5 \times 5\) random matrix such that
and \({\varvec{\xi }}\) is a 5-dimensional standard normally distributed random vector independent of \((V_Y, V_X)\).
Proof
We have
where, by (3.5),
We are going to apply Theorem C.2 for the continuous local martingale \(({\varvec{h}}_T)_{T\in \mathbb {R}_+}\) with quadratic variation process \(\langle {\varvec{h}}\rangle _T = \widetilde{{\varvec{G}}}_T\), \(T \in \mathbb {R}_+\) (introduced in the proof of Theorem 5.1). With scaling matrices
by (7.2), we have
as \(T \rightarrow \infty \). Hence by Theorem C.2, for each random matrix \({\varvec{A}}\) defined on \((\varOmega , \mathcal {F}, \mathbb {P})\), we obtain
where \({\varvec{\xi }}\) is a 5-dimensional standard normally distributed random vector independent of \(({\varvec{\eta }}, {\varvec{A}})\). The aim of the following discussion is to include appropriate scaling matrices for \({\varvec{G}}_T\). The matrices \({\varvec{G}}_T^{(1)}\) and \({\varvec{G}}_T^{(2)}\) can be written in the form
and
hence the matrices \(({\varvec{G}}_T^{(1)})^{-1}\) and \(({\varvec{G}}_T^{(2)})^{-1}\) can be written in the form
and
We have
and
Moreover,
and
Consequently,
where, by Lemma 7.2,
By (7.5) with \({\varvec{A}}= {\varvec{V}}\), by (7.6) and by Theorem 2.7 (iv) of van der Vaart (1998), we obtain
The random matrix \({\varvec{V}}\) is invertible almost surely, since
almost surely by Lemma 7.2. Consequently, \({\text {diag}}\bigl ({\varvec{J}}_T^{(1)}, {\varvec{J}}_T^{(2)}\bigr )^{-1} {\varvec{Q}}(T) {\varvec{h}}_T {\mathop {\longrightarrow }\limits ^{\mathcal {D}}}{\varvec{V}}^{-1} {\varvec{\eta }}{\varvec{\xi }}\) as \(T \rightarrow \infty \). \(\square \)
8 Summary
The following table summarize the results of the present paper on the asymptotic properties of the CLSE \((\widehat{a}_T, \widehat{b}_T, \widehat{\alpha }_T, \widehat{\beta }_T, \widehat{\gamma }_T)\) for the drift parameters \((a, b, \alpha , \beta , \gamma )\) of general two-factor affine diffusions (1.1). We recall that \(a \in [0, \infty )\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in [0, \infty )\) and \(\varrho \in [-1, 1]\).
For comparison, the following table summarize the results of Barczy and Pap (2016) on the asymptotic properties of the MLE \((\widetilde{a}_T, \widetilde{b}_T, \widetilde{\alpha }_T, \widetilde{\beta }_T)\) for the drift parameters \((a, b, \alpha , \beta )\) of a Heston model, which is a submodel of (1.1) with \(a \geqslant \frac{\sigma _1^2}{2}\), \(\sigma _1, \sigma _2 \in (0, \infty )\), \(\gamma = 0\), \(\varrho \in (-1, 1)\) and \(\sigma _3 = 0\).
References
Alfonsi A (2015) Affine diffusions and related processes: simulation, theory and applications. Springer, Cham, Bocconi University Press, Milan
Baldeaux J, Platen E (2013) Functionals of multidimensional diffusions with applications to finance. Springer, Cham, Bocconi University Press, Milan
Barczy M, Pap G (2016) Asymptotic properties of maximum-likelihood estimators for Heston models based on continuous time observations. Statistics 50(2):389–417. https://doi.org/10.1080/02331888.2015.1044991
Barczy M, Döring L, Li Z, Pap G (2013) On parameter estimation for critical affine processes. Electron J Stat 7:647–696. https://doi.org/10.1214/13-EJS786
Barczy M, Döring L, Li Z, Pap G (2014) Parameter estimation for a subcritical affine two factor model. J Stat Plann Inference 151–152:37–59. https://doi.org/10.1016/j.jspi.2014.04.001
Barczy M, Nyul B, Pap G (2016) Least squares estimation for the subcritical Heston model based on continuous time observations. Accessed 18 Nov 2015
Barczy M, Ben Alaya M, Kebaier A, Pap G (2018a) Asymptotic behavior of maximum likelihood estimators for a jump-type Heston model. To appear in J Stat Plann Inference. Accessed 15 Dec 2016
Barczy M, Ben Alaya M, Kebaier A, Pap G (2018b) Asymptotic properties of maximum likelihood estimator for the growth rate for a jump-type CIR process based on continuous time observations. To appear in Stochastic Process Appl. Accessed 1 Apr 2017 https://doi.org/10.1016/j.spa.2017.07.004
Ben Alaya M, Kebaier A (2012) Parameter estimation for the square root diffusions: ergodic and nonergodic cases. Stoch Models 28(4):609–634. https://doi.org/10.1080/15326349.2012.726042
Bolyog B, Pap G (2016) Conditions for stationarity and ergodicity of two-factor affine diffusions. Commun Stoch Anal 10(4):587–610
Bolyog B, Pap G (2017) On conditional least squares estimation for affine diffusions based on continuous time observations. Accessed 8 Mar 2017
Cox JC, Ingersoll JE, Ross SA (1985) A theory of the term structure of interest rates. Econometrica 53(2):385–407. https://doi.org/10.2307/1911242
Dawson DA, Li Z (2006) Skew convolution semigroups and affine Markov processes. Ann Probab 34(3):1103–1142. https://doi.org/10.1214/009117905000000747
Dietz HM, Kutoyants YuA (1997) A class of minimum-distance estimators for diffusion processes with ergodic properties. Stat Decisions 15:211–227
Duffie D, Filipović D, Schachermayer W (2003) Affine processes and applications in finance. Ann Appl Probab 13:984–1053. https://doi.org/10.1214/aoap/1060202833
Filipović D (2009) Term-structure models. Springer, Berlin
Hu Y, Long H (2007) Parameter estimation for Ornstein-Uhlenbeck processes driven by \(\alpha \)-stable Lévy motions. Commun Stoch Anal 1(2):175–192
Hu Y, Long H (2009a) Least squares estimator for Ornstein-Uhlenbeck processes driven by \(\alpha \)-stable motions. Stoch Process Appl 119(8):2465–2480. https://doi.org/10.1016/j.spa.2008.12.006
Hu Y, Long H (2009b) On the singularity of least squares estimator for mean-reverting \(\alpha \)-stable motions. Acta Math Sci Ser B Engl Ed 29B(3):599–608. https://doi.org/10.1016/S0252-9602(09)60056-4
Jacod J, Shiryaev AN (2003) Limit theorems for stochastic processes, 2nd edn. Springer, Berlin
Karatzas I, Shreve SE (1991) Brownian motion and stochastic calculus, 2nd edn. Springer, New York
Küchler U, Sørensen M (1997) Exponential families of stochastic processes. Springer, New York
Liptser RS, Shiryaev AN (2001) Statistics of random processes II. Applications, 2nd edn. Springer, Berlin
Overbeck L, Rydén T (1997) Estimation in the Cox–Ingersoll–Ross model. Econom Theory 13(3):430–461. https://doi.org/10.1017/S0266466600005880
van der Vaart AW (1998) Asymptotic statistics. Cambridge University Press, Cambridge
van Zanten H (2000) A multivariate central limit theorem for continuous local martingales. Statist Probab Lett 50(3):229–235. https://doi.org/10.1016/S0167-7152(00)00108-5
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix
A Stationarity and exponential ergodicity
The following result states the existence of a unique stationary distribution of the affine diffusion process given by the SDE (1.1), see Bolyog and Pap (2016, Theorem 3.1). Let \(\mathbb {C}_{-} := \{ z \in \mathbb {C}: {\text {Re}}(z) \leqslant 0\}\).
Theorem A.1
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b \in \mathbb {R}_{++}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in \mathbb {R}_{++}\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\), \(\varrho \in [-1, 1]\), and with a random initial value \((\eta _0, \zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Then
-
(i)
\((Y_t, X_t) {\mathop {\longrightarrow }\limits ^{\mathcal {D}}}(Y_\infty , X_\infty )\) as \(t \rightarrow \infty \), and we have
$$\begin{aligned} \mathbb {E}\bigl (\mathrm {e}^{u_1 Y_\infty + \mathrm {i}\lambda _2 X_\infty }\bigr ) = \exp \left\{ a \int _0^\infty \kappa _s(u_1, \lambda _2) \, \mathrm {d}s + \mathrm {i}\frac{\alpha }{\gamma } \lambda _2 - \frac{\sigma _3^2}{4\gamma } \lambda _2^2 \right\} \end{aligned}$$(A.1)for \((u_1, \lambda _2) \in \mathbb {C}_{-} \times \mathbb {R}\), where \(\kappa _t(u_1, \lambda _2)\), \(t \in \mathbb {R}_+\), is the unique solution of the (deterministic) differential equation
$$\begin{aligned} {\left\{ \begin{array}{ll} \frac{\partial \kappa _t}{\partial t}(u_1, \lambda _2) = - b \kappa _t(u_1, \lambda _2) - \mathrm {i}\beta \mathrm {e}^{-\gamma t} \lambda _2+ \frac{1}{2} \sigma _1^2 \kappa _t(u_1, \lambda _2)^2 \\ \qquad \qquad \qquad \qquad + \mathrm {i}\varrho \sigma _1 \sigma _2 \mathrm {e}^{-\gamma t} \lambda _2 \kappa _t(u_1, \lambda _2) - \frac{1}{2} \sigma _2^2 \mathrm {e}^{-2\gamma t} \lambda _2^2, \\ \kappa _0(u_1, \lambda _2) = u_1 ; \end{array}\right. } \end{aligned}$$(A.2) -
(ii)
supposing that the random initial value \((\eta _0, \zeta _0)\) has the same distribution as \((Y_\infty , X_\infty )\) given in part (i), \((Y_t, X_t)_{t\in \mathbb {R}_+}\) is strictly stationary.
In the subcritical case, the following result states the exponential ergodicity and a strong law of large numbers for the process \((Y_t, X_t)_{t\in \mathbb {R}_+}\), see Bolyog and Pap (2016, Theorem 4.1).
Theorem A.2
Let us consider the two-factor affine diffusion model (1.1) with \(a, b \in \mathbb {R}_{++}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in \mathbb {R}_{++}\), \(\sigma _1 \in \mathbb {R}_{++}\), \(\sigma _2, \sigma _3 \in \mathbb {R}_+\) and \(\varrho \in [-1, 1]\) with a random initial value \((\eta _0,\zeta _0)\) independent of \((W_t, B_t, L_t)_{t\in \mathbb {R}_+}\) satisfying \(\mathbb {P}(\eta _0 \in \mathbb {R}_+) = 1\). Suppose that \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\). Then the process \((Y_t, X_t)_{t\in \mathbb {R}_+}\) is exponentially ergodic, namely, there exist \(\delta \in \mathbb {R}_{++}\), \(B \in \mathbb {R}_{++}\) and \(\kappa \in \mathbb {R}_{++}\), such that
for all \(t \in \mathbb {R}_+\) and \((y_0, x_0) \in \mathbb {R}_+ \times \mathbb {R}\), where the supremum is running for Borel measurable functions \(g : \mathbb {R}_+ \times \mathbb {R}\rightarrow \mathbb {R}\),
and the distribution of \((Y_\infty , X_\infty )\) is given by (A.1) and (A.2). Moreover, for all Borel measurable functions \(f : \mathbb {R}^2 \rightarrow \mathbb {R}\) with \(\mathbb {E}(|f(Y_\infty , X_\infty )|) < \infty \), we have
B Moments
The next proposition gives a recursive formula for the moments of the process \((Y_t, X_t)_{t\in \mathbb {R}_+}\).
Proposition B.1
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b, \alpha , \beta , \gamma \in \mathbb {R}\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\), \(\varrho \in [-1, 1]\). Suppose that \(\mathbb {E}(Y_0^n |X_0|^p) < \infty \) for some \(n, p \in \mathbb {Z}_+\). Then for each \(t \in \mathbb {R}_+\), we have \(\mathbb {E}(Y_t^k |X_t|^\ell ) < \infty \) for all \(k \in \{0, \ldots , n\}\) and \(\ell \in \{0, \ldots , p\}\), and the recursion
for all \(t \in \mathbb {R}_+\), where \(\mathbb {E}(Y_t^i X_t^j) := 0\) if \(i, j \in \mathbb {Z}\) with \(i < 0\) or \(j < 0\). Especially,
If \(\sigma _1 > 0\) and \(Y_0 = y_0\), then the Laplace transform of \(Y_t\), \(t \in \mathbb {R}_{++}\), takes the form
i.e., \(Y_t\) has a non-centered chi-square distribution up to a multiplicative constant \(\frac{\sigma _1^2}{4} \int _0^t \mathrm {e}^{-bu} \, \mathrm {d}u\), with degrees of freedom \(\frac{4a}{\sigma _1^2}\) and with non-centrality parameter \(\frac{4\mathrm {e}^{-bt}y_0}{\sigma _1^2\int _0^t\mathrm {e}^{-bu}\,\mathrm {d}u}\).
If \(\sigma _1 > 0\) and \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\), then for each \(t \in \mathbb {R}_{++}\), the distribution of \((Y_t, X_t)\) is absolutely continuous.
The next theorem gives a recursive formula for the moments of the stationary distribution of the process \((Y_t, X_t)_{t\in \mathbb {R}_+}\) in the subcritical case, see Bolyog and Pap (2016, Theorem 5.1).
Theorem B.2
Let us consider the two-factor affine diffusion model (1.1) with \(a \in \mathbb {R}_+\), \(b \in \mathbb {R}_{++}\), \(\alpha , \beta \in \mathbb {R}\), \(\gamma \in \mathbb {R}_{++}\), \(\sigma _1, \sigma _2, \sigma _3 \in \mathbb {R}_+\), \(\varrho \in [-1, 1]\), and the random vector \((Y_\infty , X_\infty )\) given by Theorem A.1. Then all the (mixed) moments of \((Y_\infty , X_\infty )\) of any order are finite, i.e., we have \(\mathbb {E}(Y_\infty ^n |X_\infty |^p) < \infty \) for all \(n, p \in \mathbb {Z}_+\), and the recursion
holds for all \(n, p \in \mathbb {Z}_+\) with \(n + p \geqslant 1\), where \(\mathbb {E}(Y_\infty ^k X_\infty ^\ell ) := 0\) for \(k, \ell \in \mathbb {Z}\) with \(k < 0\) or \(\ell < 0\). Especially,
If \(\sigma _1 > 0\), then the Laplace transform of \(Y_\infty \) takes the form
i.e., \(Y_\infty \) has gamma distribution with parameters \(2a / \sigma _1^2\) and \(2b / \sigma _1^2\), hence
If \(\sigma _1 > 0\) and \((1 - \varrho ^2) \sigma _2^2 + \sigma _3^2 > 0\), then the distribution of \((Y_\infty , X_\infty )\) is absolutely continuous.
C Limit theorems for continuous local martingales
In what follows we recall some limit theorems for continuous local martingales. We use these limit theorems for studying the asymptotic behaviour of the MLE of \({\varvec{\theta }}= (a, b, \alpha , \beta , \gamma )^{\top }\). First we recall a strong law of large numbers for continuous local martingales.
Theorem C.1
(Liptser and Shiryaev 2001, Lemma 17.4) Let \(\bigl ( \varOmega , \mathcal {F}, (\mathcal {F}_t)_{t\in \mathbb {R}_+}, \mathbb {P}\bigr )\) be a filtered probability space satisfying the usual conditions. Let \((M_t)_{t\in \mathbb {R}_+}\) be a square-integrable continuous local martingale with respect to the filtration \((\mathcal {F}_t)_{t\in \mathbb {R}_+}\) such that \(\mathbb {P}(M_0 = 0) = 1\). Let \((\xi _t)_{t\in \mathbb {R}_+}\) be a progressively measurable process such that
where \((\langle M \rangle _t)_{t\in \mathbb {R}_+}\) denotes the quadratic variation process of M. Then
If \((M_t)_{t\in \mathbb {R}_+}\) is a standard Wiener process, the progressive measurability of \((\xi _t)_{t\in \mathbb {R}_+}\) can be relaxed to measurability and adaptedness to the filtration \((\mathcal {F}_t)_{t\in \mathbb {R}_+}\).
The next theorem is about the asymptotic behaviour of continuous multivariate local martingales.
Theorem C.2
(van Zanten 2000, Theorem 4.1) Let \(\bigl ( \varOmega , \mathcal {F}, (\mathcal {F}_t)_{t\in \mathbb {R}_+}, \mathbb {P}\bigr )\) be a filtered probability space satisfying the usual conditions. Let \(({\varvec{M}}_t)_{t\in \mathbb {R}_+}\) be a d-dimensional square-integrable continuous local martingale with respect to the filtration \((\mathcal {F}_t)_{t\in \mathbb {R}_+}\) such that \(\mathbb {P}({\varvec{M}}_0 = {\varvec{0}}) = 1\). Suppose that there exists a function \({\varvec{Q}}: [t_0, \infty ) \rightarrow \mathbb {R}^{d \times d}\) with some \(t_0 \in \mathbb {R}_+\) such that \({\varvec{Q}}(t)\) is an invertible (non-random) matrix for all \(t \in [t_0, \infty )\), \(\lim _{t\rightarrow \infty } \Vert {\varvec{Q}}(t)\Vert = 0\) and
where \({\varvec{\eta }}\) is a \(d \times d\) random matrix. Then, for each \(\mathbb {R}^{k\times \ell }\)-valued random matrix \({\varvec{A}}\) defined on \((\varOmega , \mathcal {F}, \mathbb {P})\), we have
where \({\varvec{Z}}\) is a d-dimensional standard normally distributed random vector independent of \(({\varvec{\eta }}, {\varvec{A}})\).
Rights and permissions
About this article
Cite this article
Bolyog, B., Pap, G. On conditional least squares estimation for affine diffusions based on continuous time observations. Stat Inference Stoch Process 22, 41–75 (2019). https://doi.org/10.1007/s11203-018-9174-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11203-018-9174-z