Abstract
This paper considers the estimation of integrated Laplace transform of local ‘volatility’ by using noisy high-frequency data. We allow for the presence of microstructure noise under a pure jump semimartingale over a fixed time interval [0, t]. We propose an efficient estimator for the integrated Laplace transform of volatility via applying the pre-averaging method. Under some mild conditions on the Lévy density, the asymptotic properties of the estimator including consistency and asymptotic normality are established. Simulation studies further confirm our theoretical results.
Similar content being viewed by others
References
Aït-Sahalia Y, Jacod J (2010) Is brownian motion necessary to model high frequency data? Ann Stat 38(5):3093–3128
Aït-Sahalia Y, Mykland PA (2004) Estimators of diffusions with randomly spaced discrete observations: a general theory. Ann Stat 32(5):2186–2222
Aït-Sahalia Y, Mykland P, Zhang L (2005) How often to sample a continuous-time process in the presence of market microstructure noise. Rev Financ Stud 18:351–416
Aït-Sahalia Y, Fan J, Xiu D (2010) High frequency covariance estimates with noisy and asynchronous data. J Am Stat Assoc 160:1504–1517
Aït-Sahalia Y, Mykland P, Zhang L (2011) Ultra high frequency volatility estimation with dependent microstructure noise. J Econom 160(1):160–175
Aldous D, Eagleson G (1978) On mixing and stability of limit theorems. Ann Probab 6(2):325–331
Andersen TG, Bollerslev T, Diebold F, Labys P (2003) Modeling and forecasting realized volatility. Econometrica 71(3):579–625
Andrews B, Calder M, Davis RA (2009) Maximum likelihood estimation for \(\alpha \)-stable autoregressive processes. Ann Stat 37:1946–1982
Bachelier L (1900) Théorie de la speculation. Gauthier-Villars, Paris
Barndorff-Nielsen OE, Shephard N (2007) In Advances in economics and econometrics: theory and applications. In: Ninth World Congress. Econometric Society Monographs. Cambridge University Press
Barndorff-Nielsen OE, Hansen PR, Lunde A, Shephard N (2008) Designing realised kernels to measure the ex-post variation of equity prices in the presence of noise. Econometrica 76(6):1481–1536
Barndorff-Nielsen OE, Hansen PR, Lunde A, Shephard N (2011) Multivariate realised kernels: consistent positive semi-definite estimators of the covariation of equity prices with noise and non-synchronous trading. J Econom 162(2):149–169
Bates DS (1996) Jumps and stochastic volatility: exchange rate processes implicity in deutsche mark opitions. Rev Financ Stud 9(1):69–107
Black F, Scholes M (1973) The pricing of options and corporate liabilities. J Polit Econ 81(3):637–654
Cont R, Tankov P (2004) Financial Modelling with Jump Processes. CRC Press, Boca Raton
Heston SL (1993) A closed-form solution for options with stochastic volatility with applications to bond and currency options. Rev Financ Stud 6(2):327–343
Jacod J, Protter P (2012) Discretization of Processes. Stochastic Modelling and Applied Probbability, vol 67. Springer, Heidelberg
Jacod J, Shiryayev AV (2003) Limit Theorems for Stochastic Processes. Springer, New York
Jacod J, Li Y, Mykland PA, Podolskij M, Vetter M (2009) Microstructure noise in the continuous case: the pre-averaging approach. Stoch Process Appl 119(7):2249–2276
Jing BY, Kong XB, Liu Z (2012) Modeling high-frequency financial data by pure jump processes. Ann Stat 40(2):759–784
Jing BY, Liu Z, Kong XB (2014) On the estimation of integrated volatility with jumps and microstructure noise. J Bus Econ Stat 32(3):457–467
Karlin S, Taylor HM (1975) A First Course in Stochastic Processes. Academic Press, Cambridge
Klebaner FC (1998) Introduction to Stochastic Calculus with Applications. Imperial College Press, London
Klüppelberg C, Meyer-Brandis T, Schmidt A (2010) Electricity spot price modelling with a view towards extreme spike risk. Quant Finance 10:963–974
Kong XB, Liu Z, Jing BY (2015) Testing for pure-jump processes for high-frequency data. Ann Stat 43(2):847–877
Kou SG (2002) A jump-diffuion model for option pricing. Manage Sci 48(8):1086–1101
Li J (2013) Robust estimation and inference for jumps in noisy high frequency data: a local-to-continuity theory for the pre-averaging method. Econometrica 81(4):1673–1693
McCulloch JH (1996) Financial applications of stable distributions. Handb Stat 14:393–425
Merton RC (1973) Theory of rational option pricing. Bell J Econ Manag Sci (RAND Corp) 4(1):141–183
Mikosch T, Resnick S, Rootzén H, Stegeman A (2002) Is network traffic approximated by stable lévy motion or fractional brownian motion? Ann Appl Probab 12:23–68
Mykland P, Zhang L (2009) Inference for continuous semimartingales observed at high frequency: a general approach. Econometrica 77(5):1403–1445
Nikias CL, Shao M (1995) Signal Processing with Alpha-Stable Distributions and Applications. Wiley, New York
Renyi A (1963) On stable sequences of events. Sankhya: Indian J Stat Ser A 25(3):293–302
Sato KI (1999) Lévy Processes and Infinitely Divisible Distributions. Cambridge University Press, Cambridge
Todorov V, Tauchen G (2012a) Inverse realized laplace transforms for nonparametric volatility density estimation in jump-diffusions. J Am Stat Assoc 107(498):622–635
Todorov V, Tauchen G (2012b) The realized laplace transform of volatility. Econometrica 80(3):1105–1127
Todorov V, Tauchen G (2012c) Realized laplace transforms for pure-jump semimartingales. Ann Stat 40:1233–1262
Wang L, Liu Z, Xia X (2017) Rate efficient estimation of realized Laplace transform of volatility with microstructure noise. Working paper. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3179378
Wu L (2007) Modeling financial security returns using Levy processes. Handb Oper Res Manag Sci 15:117–162
Xiu D (2010) Quasi-maximum likelihood estimation of volatility with high frequency data. J Econ 159:235–250
Zhang L (2006) Efficient estimation of stochastic volatility using noisy observations: a multi-scale approach. Bernoulli 12(6):1019–1043
Acknowledgements
Liu’s work is supported by FDCT of Macau (No. 078/2013/A3) and Xia’s work is supported by Hubei Provincial Natural Science Foundation of China (Grant No. 2017CFB141).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The author declares that they have no conflict of interest.
Ethical approval
This article does not contain any studies with human participants or animals performed by any of the authors.
Additional information
Communicated by V. Loia.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix: Proofs
Appendix: Proofs
In the following, C denotes a generic constant, we let \(t_i^n:=i\Delta _n\), \(g_j^n:=g\left( \frac{j}{k_n}\right) \), \(E_i^n[\cdot ]:=E[\cdot |\mathcal {F}_{i\Delta _n}]\).
Proof of Theorem 1
First step, we use Taylor expansion to separate latent process \(X_t\) and noise \(\varepsilon _t\), then select a suitable \(k_n\) to weaken the noise effect.
For noise \(\varepsilon \), from Assumption 4, we have \(E(\varepsilon _t)=0\), \(\mathrm {Var}(\varepsilon _t)=\omega _t^2\le C\) and \(\varepsilon _t\) are independent,
When \(0<r\le 2\), applying Hölder inequality, we have
Then it can be easily shown that
To ensure the effect of noise ignorable, we need \(k_n=O(n^{\frac{2}{\beta +2}+\tau }),0<\tau <\frac{\beta }{\beta +2}\).
Note that \(X_t\) has an integral form representation:
where \(\mu \) is homogenous Poisson measure with compensator \(\nu (\mathrm {d}x) = \frac{A}{|x|^{\beta +1}}\mathrm {d}x\) and \(\tilde{\mu } = \mu -\nu \), according to the Lévy density \(\nu (x)=\frac{A}{|x|^{\beta +1}}\). We define \(L_t = \int _0^t\int _{\mathbb {R}}x \tilde{\mu }(\mathrm {d}s, \mathrm {d}x)\). Next we make a decomposition as follows:
where the definitions of \(\xi _{i,u}^{(1)}\), \(\xi _{i,u}^{(2)}\), \(\xi _{i,u}^{(3)}\) and \(Re_u\) are clear.
For the term \(\xi _{i,u}^{(1)}\), by the self-similarity of the stable process \(L_t\), we have \(\Delta _i^nL_t=\Delta _i^nZ_t=\Delta _n^{\frac{1}{\beta }}S\), where S has characteristic function \(E(e^{iuS})=e^{-\frac{|u|^{\beta }}{2}}\), then we can conclude that
where \(G_\beta (x)_0^n=\frac{e^{-2^\beta x^\beta }-2e^{-2x^\beta }+1}{2}\).
For the term \(\xi _{i,u}^{(2)}\), we first subtract and add a midterm \(e^{-u|\sigma _{t_i^n}+\int _{t_i^n}^s\tilde{\sigma }_r\mathrm {d}W_r|^{\beta }}\) and then take Taylor expansion of \(e^{-u|\sigma _{t_i^n}+\int _{t_i^n}^s\tilde{\sigma }_r\mathrm {d}W_r|^{\beta }}\) at point \(\sigma _{t_i^n}\). It follows that
where \(\sigma _s^*\) is between \(\sigma _{t_i^n}\) and \(\sigma _{t_i^n}+\int _{t_i^n}^s\tilde{\sigma }_r\mathrm {d}W_r\). Since \(E_{i}^n\left( \xi _{i,u}^{(2)}(1)\right) = 0\), using Cauchy Schwarz inequality, Itô isometry, and Assumption 3, we can get
To consider \(\xi _{i,u}^{(2)}(2)\), we use the similar arguments in Todorov and Tauchen (2012c). By Cauchy–Schwarz inequality, Itô isometry, Assumption 3, Burkholder–Davis–Gundy inequality, then we have
By Taylor expansion at \(|\sigma _{t_i^n}+\int _{t_i^n}^s\tilde{\sigma }_r\mathrm {d}W_r|^{\beta }\), and using inequality \(||x-y|^p-|x|^p|\le Cp(|y|^p+|x|^{p-1}|y|), p>1\), we obtain,
where \(\sigma _s^{**}\) is between \(\sigma _{t_i^n}+\int _{t_i^n}^s\tilde{\sigma }_r\mathrm {d}W_r\) and \(\sigma _s=\sigma _{t_i^n}+\int _{t_i^n}^s\tilde{\alpha }_r\mathrm {d}r+\int _{t_i^n}^s\tilde{\sigma }_r\mathrm {d}W_r\).
For the term \(\xi _{i,u}^{(3)}\), we first subtract and add a same term, \(\cos \left( (2u)^{\frac{1}{\beta }}(\phi _{\beta }^n\Delta _nk_n)^{-\frac{1}{\beta }}\left( \sum _{j=1}^{k_n-1}g_j^n(\int _{(i+j-1)\Delta _n}^{(i+j)\Delta _n}\alpha _s ds +\int _{(i+j-1)\Delta _n}^{(i+j)\Delta _n}\sigma _{s} d{L_s})\right) \right) \). Then, by trigonometric function \(\cos (x)-\cos (y)=-2\sin (\frac{1}{2}(x+y))\sin (\frac{1}{2}(x-y))\) and second-order Taylor expansion separately, we have
where \(\tilde{\chi }\) denoting some value between \((2u)^{\frac{1}{\beta }}(\phi _{\beta }^n\Delta _nk_n)^{-\frac{1}{\beta }} \sum _{j=1}^{k_n-1}g_j^n\sigma _{t_i^n}\Delta _{i+j}^nL\) and \((2u)^{\frac{1}{\beta }}(\phi _{\beta }^n\Delta _nk_n)^{-\frac{1}{\beta }} \sum _{j=1}^{k_n-1}g_j^n\cdot (\int _{t_{i+j-1}^n}^{t_{i+j}^n}\alpha _s\mathrm {d}s+\int _{t_{i+j-1}^n}^{t_{i+j}^n}\sigma _{s-} d{L_s})\). Using the basic inequality \(|\sin (x)|\le |x|\) and Assumption 2, we have
We next divide \(\xi _{i,u}^{(3)}(2)\) into two parts,
Noting that \(E_i^n(\Delta _i^nL)=0\) and by Assumption 3, Cauchy Schwarz inequality, we have
Then we can obtain
For \(\xi _{i,u}^{(3)}(3)\), we first split
where
Then we have the inequality
Because \(L_s\) is a martingale, using Cauchy–Schwarz inequality and Assumption 3, we get
For the second term, Cauchy–Schwarz inequality, Burkholder–Davis–Gundy inequality and Assumption 3 imply
Similarly, we have
Martingale property of \(L_s\) yields
By Assumption 3, we obtain
For the fourth term \(Re_u\), it can be easily shown that
uniformly in u. Combining (7), (8)–(16) and the bound of \(R_i\), we have
which gives the desired result. \(\square \)
Proof of Theorem 2
First, we separately deal with the main term and the noise term of
To find an upper bound of noise term, we have
By selecting \(k_n=O\left( n^{\frac{2}{2+\beta }+\tau }\right) , ~ \tau \in \left( \frac{\beta ^2}{(\beta +2)(3\beta +4)},\frac{\beta }{\beta +2}\right) \), we can show that the noise term converges in probability to zero.
For the second term, invoking Theorem 1 and (9)–(11), we get
For the third term, from (12)–(14), we have
which converges to 0, when \(n\rightarrow \infty \). It follows from (15) that
To let this term converge to 0 as \(n\rightarrow \infty \), we need \(\beta >\frac{4}{3}\). When \(n\rightarrow \infty \),
So under the conditions of Theorem 2,
To prove the central limit theorem, we only need to prove that as \(\Delta _n \rightarrow 0\),
To this end, we apply the technique of “big blocks, small blocks” as introduced in Jacod et al. (2009). The limiting distribution of the proposed estimator stems from the “big blocks”. Whereas the “small blocks” does not affect the final asymptotical behavior, which is eventually proven to be negligible but ensure the conditional independence between the “big” blocks. For a given integer p, we let \(i_n(p)=\left[ \frac{[T/\Delta _n]-k_n+1}{(p+1)k_n} \right] - 1\) and for \(i=0,\ldots , i_n(p)\), write \(a_i(p)=i(p+1)k_n+1\), \(b_i(p)=i(p+1)k_n +pk_n\) and denote the i-th “big” block by \(\mathrm {A}_i=\{k: a_i(p)\le k \le b_i(p), k\in N^{+}\}\) and the i-th “small block” by \(\mathrm {B}_i =\{k: b_i(p)< k < a_{i+1}(p), k\in N^{+}\} \). We define
where \(\xi _{j,u}^{(1)}\) is defined in (7). Then, we denote
Since the convergence rate of V is \((\Delta _nk_n)^{\frac{1}{2}}\). Therefore, we have the following decomposition,
Following similar steps in Jacod et al. (2009), we obtain that for any \(\epsilon >0\),
Hence, it suffices to prove
To this end, we apply the martingale central limit theorem which is presented in Theorem IX.7.28 in Jacod and Shiryayev (2003), with which we need to verify the following properties:
where, \(\Delta H(p)_i=H_{b_i(p)\Delta _n} - H_{a_i(p)\Delta _n}\), H is a bounded martingale defined on the original probability space. We remark that in deriving formula (20), we let p be fixed first and then tend to infinity.
From (8), we get \({\zeta '_i}(p,1) = 0\). From self-similarity property of stable process, we know \(\Delta _i^nL=\Delta _n^{\frac{1}{\beta }}S_1\), where \(S_1\) is a stable process with characteristic function \(E(e^{iuS_1})=e^{-|u|^\beta /2}\). Applying these arguments and using the basic equalities that \(\cos (a+b)=\cos (a)\cos (b)-\sin (a)\sin (b)\), \(\sin (a)\cos (b)=\left( \sin (a+b)+\sin (a-b)\right) /2\) and \(\cos (a)=(e^{ia}+e^{-ia})/2\), we can obtain
where \(\phi _\beta ^n=k_n^{-1}\sum _{j=0}^{k_n-1}(g(\frac{j}{k_n}))^\beta \) and
On the other hand, because for any \(r\in (0,1)\), \(|G'_{\beta }(x)_r| \le C\), applying Taylor expansion,
Collecting (23) and (24) yields the property (20). Following the arguments in calculating (23), (24) and applying the steps in the proof of Theorem 1, we obtain (21). Then we need to show (22). When H is discontinuous martingale. Using the property of predictable quadratic variation, quadratic variation and Itô isometry, we can prove that \(\sum _{i=1}^{i_n(p)}E_{a_i(p)-1}^n\left( (\Delta _nk_n)^{-\frac{1}{2}}\zeta _i(p,1) \Delta H(p)_i \right) \) converges in probability to 0. The detail is similar to the proof in Todorov and Tauchen (2012c). As pure jump and continuous martingales are orthogonal, we have \(E_{a_i(p)-1}^n(\zeta _i(p,1)\Delta H(p)_i)=0\) when H is a continuous martingale. Then proof of Theorem 2 is completed. \(\square \)
Rights and permissions
About this article
Cite this article
Wang, L., Liu, Z. & Xia, X. Realized Laplace transforms for pure jump semimartingales with presence of microstructure noise. Soft Comput 23, 5739–5752 (2019). https://doi.org/10.1007/s00500-018-3237-3
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00500-018-3237-3