Abstract
The joint analysis of non-stationary and high frequency financial data poses theoretical challenges due to that such massive data varies with time and possesses no fixed density function. This paper proposes the local linear smoothing to estimate the unknown volatility function in scalar diffusion models based on Gamma asymmetric kernels for high frequency financial big data. Under the mild conditions, we obtain the asymptotic normality for the estimator at both interior and boundary design points. Besides the standard properties of the local linear estimator such as simple bias representation and boundary bias correction, the local linear smoothing using Gamma asymmetric kernels possesses some extra advantages such as variance reduction and resistance to sparse design, which is validated through finite sample simulation study and empirical analysis on 6-month Shanghai Interbank Offered Rate (abbreviated as Shibor) in China.
Similar content being viewed by others
References
Aït-Sahalia Y (1996) Nonparametric pricing of interest rate derivative securities. Econometrica 64(3):527–560
Bandi FM, Phillips PC (2003) Fully nonparametric estimation of scalar diffusion models. Econometrica 71(1):241–283
Chen SX (2000) Probability density function estimation using Gamma kernels. Ann Inst Stat Math 52(3):471–480
Chen SX (2002) Local linear smoothers using asymmetric kernels. Ann Inst Stat Math 54(2):312–323
Eisenbaum N, Kaspi H (2007) On the continuity of local times of borel right Markov processes. Ann Probab 35(3):915–934
Fan J (1993) Local linear regression smoothers and their minimax efficiencies. Ann Stat 21(1):196–216
Fan J, Fan Y, Jiang J (2007) Dynamic integration of time- and state-domain methods for volatility estimation. J Am Stat Assoc 102:618–631
Fan J, Gijbels I (1992) Variable bandwidth and local linear regression smoothers. Ann Stat 20(4):2008–2036
Fan J, Gijbels I (1996) Local polynomial modeling and its applications. Chapman and Hall, London
Fan J, Zhang C (2003) A reexamination of diffusion estimators with applications to financial model validation. J Am Stat Assoc 98(461):118–134
Jacod J (2012) Statistics and high-frequency data. In: Statistical Methods for Stochastic Differential Equations, CRC Boca Raton, London, pp 191–310
Jiang H, Dong X (2015) Parameter estimation for the non-stationary Ornstein–Uhlenbeck process with linear drift. Stat Pap 56:257–268
Jiang G, Knight J (1997) A nonparametric approach to the estimation of diffusion processes, with an application to a short-term interest rate model. Econom Theory 13(5):615–645
Karatzas I, Shreve S (2000) Brownian motions and stochastic calculus. Springer, New York
Karlin S, Taylor H (1981) A second course in stochastic processes. Elsevier, Amsterdam
Meyn SR, Tweedie RL (1993) Stability of Markovian processes II: continuous-time processes and sampled chains. Adv Appl Probab 25:487–517
Park J, Phillips PCB (2001) Nonlinear regressions with integrated time series. Econometrica 69(1):117–161
Phillips PCB, Magdalinos T (2007) Limit theory for moderate deviations from a unit root. J Econom 136(1):115–130
Revuz D, Yor M (2005) Continuous martingales and Brownian motion. Springer, New York
Rosa A, Nogueira M (2016) Nonparametric estimation of a regression function using the gamma kernel method in ergodic processes. in arXiv:1605.07520v2 [math.ST]
Shen G, Yu Q (2017) Least squares estimator for Ornstein-Uhlenbeck processes driven by fractional Lévy processes from discrete observations. Stat Pap. https://doi.org/10.1007/s00362-017-0918-4
Xu Z (2003) Staistical inference for diffusion processes. Ph.D thesis, East China Normal University
Acknowledgements
The authors would like to thank the editor, associate editor and three anonymous referees for their valuable suggestions, which greatly improved our paper. This research work is supported by National Natural Science Foundation of China (11901397), Ministry of Education, Humanities and Social Sciences Project (18YJCZH153), National Statistical Science Research Project (2018LZ05), Youth Academic Backbone Cultivation Project of Shanghai Normal University (310-AC7031-19-003021), General Research Fund of Shanghai Normal University (SK201720) and Key Subject of Quantitative Economics (310-AC7031-19-004221) and Academic Innovation Team (310-AC7031-19-004228) of Shanghai Normal University. We also thank Dr. HanChao Wang for his helpful comments for this article.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Appendices
Appendix
In this section, we present some technical lemmas and the proofs for the main theorems.
A. Technical lemmas
Lemma 2
(The occupation time formula) Let \(X_{t}\) be a semimartingale with local time \((L_{X}(\cdot ,a))_{a \in \mathscr {D}}.\) Let g be a bounded Borel measurable function. Then
where [X] is the quadratic variation of X.
Lemma 3
(Jacod’s stable convergence theorem) A sequence of \(\mathbb {R}-\)valued variables \(\{\zeta _{n, i}: i \ge 1\}\) defined on the filtered probability space \((\Omega , \mathscr {F}, (\mathscr {F})_{t \ge 0}, {P})\) is \(\mathscr {F}_{i \Delta _{n}}-\)measurable for all n, i. Assume there exists a continuous adapted \(\mathbb {R}-\)valued process of finite variation \(B_{t}\) and a continuous adapted and increasing process \(C_{t}\), for any \(t > 0,\) we have
Assume also
where either H is one of the components of Wiener process W or is any bounded martingale orthogonal (in the martingale sense) to W and \(\Delta _{n}^{i}H = H_{i\Delta _{n}} - H_{(i - 1)\Delta _{n}}.\)
Then the process
where \({\mathop {\longrightarrow }\limits ^{\mathcal {S} - \mathcal {L}}}\) denotes stable convergence in law, \(M_{t}\) is a continuous process defined on an extension \(\big (\widetilde{\Omega }, \widetilde{P}, \widetilde{\mathscr {F}}\big )\) of the filtered probability space \(\big ({\Omega }, {P}, {\mathscr {F}}\big )\) and which, conditionally on the the \(\sigma -\)filter \(\mathscr {F}\), is a centered Gaussian \(\mathbb {R}-\)valued process with \(\widetilde{E}\big [M_{t}^{2} \mid \mathscr {F}\big ] = C_{t}.\)
Remark A.1
For lemma 3, one can refer to Jacod (2012) (Lemma 4.4) for more details. The stable convergence implies the following crucial property required in the detailed proof of Theorem 2.3.
If \(Z_{n} {\mathop {\longrightarrow }\limits ^{\mathcal {S} - \mathcal {L}}}Z \) and if \(Y_{n}\) and Y are variables defined on \((\Omega , \mathscr {F}, {P})\) and with values in the same Polish space F, then
which implies that \(Y_{n} \times Z_{n} {\mathop {\longrightarrow }\limits ^{\mathcal {S} - \mathcal {L}}}Y \times Z\) through the continuous function \(g(x, y) = x \times y.\)
Lemma 4
Under Assumptions 1-3, we have
Remark A.2
Denote \(\alpha _{k}(x):=E[(\xi -x)^{k}],\) where \(\xi {\mathop {=}\limits ^{d}} K_{G(x/h_{n} + 1, h_{n})}({u}).\) For \(\alpha _{k}(x),\) we have
for \(l \ge 3,\) which implies that \(\xi = x + o_{p}(\sqrt{h_{n}}).\) For more details of the results (A.8), one can sketch Sect. 4, especially equation (4.2) in Chen (2002).
Based on the equations above and Theorem 1.1 on the continuity of local times in Eisenbaum and Kaspi (2007), we can obtain that
Proof
One can refer to Bandi and Phillips (2003) for similar proof procedure. For brevity, here we omit the technical details. \(\square \)
B. Detailed proof for theorem 2.3
Proof
For \((x_{i}-x_{i-1})^{2},\) based on Itô formula, we have
For simplicity, we set \(T = 1,\) and also based on formula (B.1),
where the regularization coefficient \(R(h_{n}) = \Big \{\begin{array}{ll} \sqrt{h_{n}^{1/2}},\quad interior~x, \\ \sqrt{h_{n}},\quad boundary~x. \end{array}\) According to Lemma 4 and the expression for \(\omega _{i-1},\) one can get that
In what follows, we only need to calculate the numerators of \(A_{1},\) \(A_{2}\) and \(A_{3}.\)
We firstly introduce the uniform boundedness for the increments of diffusion process (more details seen in Karatzas and Shreve (2000)), that is
where C is a suitable constant.
For the numerator of \(A_{1},\) we obtain
According to uniform boundedness for the increments of diffusion process (hereafter indicated as the UBI) and Assumption 1(ii), \(A_{11}=o(A_{12})\) and similarly \(A_{2}=o(A_{12})\) based on Assumption 1(i), so we only need to consider \(A_{12}.\)
As for \(A_{12},\) firstly through Taylor expansion, we can get
where \(\theta \in [0,1].\)
It can be obviously deduced that
Then, according to Lemma 4 and the expression for \(\omega _{i-1},\)
Thus, the bias for \(\sqrt{n}R(h_{n}) \left( \hat{\sigma }^{2}_{n}(x) - \sigma ^{2}(x)\right) \) is
In what follows, we calculate the asymptotic variance for
By Burkholder-Davis-Gundy (BDG) inequality, Hölder inequality and UBI property, we can conclude that \(A_{2}=o_{a.s}(A_{3}).\) So here we only need to deal with \(A_{3}.\) However, the \(\omega _{i-1}\) in \(A_{3}\) is not \(\mathscr {F}_{i\Delta _{n}}\)-measurable, so we can not directly use Jacod’s stable convergence theorem in Lemma 3. To solve this problem, we firstly introduce its convergence limit in probability divided by \(\bar{L}_{X}(T,x) \cdot x h_{n},\) that is,
So
is obviously \(\mathscr {F}_{i\Delta _{n}}\)-measurable, and we can utilize the Jacod’s stable convergence theorem for
where
Jacod’s sable convergence theorem in Lemma 3 tells us that the following arguments,
implies \(\sum ^{n}_{i=1}q_{i} {\mathop {\longrightarrow }\limits ^{\mathcal {S} - \mathcal {L}}}M_{t},\) where \(E_{i-1}[\cdot ]\) := \(E[\cdot |\mathscr {F}_{(i-1)\Delta _{n}}].\)
For \(S_{1}\), due to the martingale property, we can easily know that,
Hence,
For \(S_{2}\),
One can show that \(q_{i1}^{2}\) is larger than the others according to Lemma 4 and the equation (A.8), so we only deal with the dominant one \(q_{i1}^{2},\)
Using the Itô formula on \((x_{s}-x_{i-1})^{2}\), we have
We can deduce that \(B_{3}\equiv 0\) by Fubini theorem, and \(B_{2}=o_{a.s.}(B_{1})\) by UBI property and Assumption 1. Applying the mean value theorem and UBI property for \(B_{1},\) we can reach
where
Hence,
For \(S_{3},\) by using BDG and Hölder inequalities, we have
where
which one can refer to Rosa and Nogueira (2016) for more details.
Hence
For \(S_{4},\) if \(H=W,\) then
due to \(R(h_{n}) \rightarrow 0.\)
If H is orthogonal to W, \(\int ^{i\Delta _{n}}_{(i-1)\Delta _{n}}(x_{s}-x_{i-1})\sigma (x_{s})dW_{s}\) is orthogonal to H, then we can get
Furthermore, we can observe
where \(q_{2, i} := 2\sqrt{n}R(h_{n})K_{G(x/h_{n}+1, h_{n})}(x_{i-1})\int ^{i\Delta _{n}}_{(i-1)\Delta _{n}}(x_{s}-x_{i-1})\sigma (x_{s})dW_{s}\) and \(q_{3, i} := 2\sqrt{n}R(h_{n})K_{G(x/h_{n}+1, h_{n})}(x_{i-1})(x_{i-1} - x)\int ^{i\Delta _{n}}_{(i-1)\Delta _{n}}(x_{s}-x_{i-1})\sigma (x_{s})dW_{s}.\)
Under the same proof procedure as \(\sum \limits _{i=1}^{n} q_{i},\) we can obtain \(\sum ^{n}_{i=1}q_{2, i} {\mathop {\longrightarrow }\limits ^{\mathcal {S} - \mathcal {L}}}M_{2, t}\) and \(\sum ^{n}_{i=1}q_{3, i} {\mathop {\longrightarrow }\limits ^{\mathcal {S} - \mathcal {L}}}M_{3, t},\) which implies that \(\sum ^{n}_{i=1}q_{2, i} = O_{p}(1)\) and \(\sum ^{n}_{i=1}q_{3, i} = O_{p}(1).\) Moreover, based on Lemma 4 and its remark, we have \(\bar{L}_{X}(T, x) \cdot \frac{\Delta _{n}\sum _{j=1}^{n}K_{G(x/h_{n}+1, h_{n})}(x_{j-1})(x_{j-1}-x)^{2}}{\Delta _{n}^{2} \sum _{i=1}^{n}\omega _{i-1}} - 1 {\mathop {\rightarrow }\limits ^{P}} 0\) and \(\bar{L}_{X}(T, x) \cdot \frac{\Delta _{n}\sum _{j=1}^{n}K_{G(x/h_{n}+1, h_{n})}(x_{j-1})(x_{j-1}-x)}{\Delta _{n}^{2} \sum _{i=1}^{n}\omega _{i-1}} - \frac{1}{x} {\mathop {\rightarrow }\limits ^{P}} 0.\) Hence, we can get the main results in Theorem 2.3 due to the Remark A.1. \(\square \)
Rights and permissions
About this article
Cite this article
Song, Y., Li, H. & Fang, Y. Efficient estimation for the volatility of stochastic interest rate models. Stat Papers 62, 1939–1964 (2021). https://doi.org/10.1007/s00362-020-01166-4
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00362-020-01166-4
Keywords
- Diffusion process
- Volatility function
- Local linear estimator
- Gamma asymmetric kernel
- Financial market interest rate