Abstract
The covariation of short-time period returns between securities plays an important role in many area of finance. Under the wide availability of high frequency financial data, realized covariation, as an ex-post measure of the covariation, can accurately estimate the quadratic covariation. However, the realized covariation fails to work when the multiple records appear. In this paper, we propose an estimator of integrated covariation, which is robust to the high frequency data containing multiple records. Consistency of the estimator and central limit theorem have been established. Moreover, several extensions which make the estimator available to different types of high frequency data are also considered. Simulation study confirms the performance of the estimator.
Similar content being viewed by others
Notes
By effective seconds, we mean that during the second, at least one transaction is recorded.
The average of multiple observations and pre-averaging approach.
Without microstructure noise.
The condition of equally spaced sampling scheme is not necessary for constructing the estimator, but it simplifies the presentation for an illustration, the extension is discussed in Sect. 4.
This condition requires that the unobserved subintervals are equally spaced, it is a restriction for the feasibility of estimator, but it can not be avoided for the consistency of the proposed estimator under current framework. However, one could consider the case where the subintervals are randomly long with a distribution, exponential for example, under the double asymptotics, i.e., \(n\rightarrow \infty \) and \(L_i\rightarrow \infty \), this unnatural condition can be removed.
Of course, we can arbitrarily take one data from \(E_i\), then realized covariation can be applied. This procedure loses lots of data, hence lacks efficiency. We also discuss the comparison between this estimator and proposed estimator in this paper later.
Since the order of observations is crucial in calculating the realized variation, under the multiple observation situation, one must impose some “unrealistic” assumptions if we want to use the entire data-set. Besides the double asymptotics considering in the remark, we also can assume that both \(\sum _{k=1}^{L_i}\left( \frac{k-1}{L_i}\right) ^2(t_{L_i+j}-t_{L_i+j-1})\) and \(\sum _{j=1}^{L_i}(1-\frac{j-1}{L_i})^2(t_{L_i+j}-t_{L_i+j-1})\) do not depends on i.
We make the some necessary adjustments, so that the estimator is based on the multiple observations. \(k_n\) is an integer, plays the role of window size, usually is taken as \(k_n=\theta \sqrt{n}\) for some positive constant \(\theta \), due to the trade off of bias and rates. g is a positive function in [0, 1], satisfying some smooth conditions.
Relative bias \(=\frac{RCV_1^{weight}-\int _0^1\rho \sigma \eta ds}{\int _0^1\rho \sigma \eta ds}\).
We take \(\rho \) from \(-\)1 to 1 by step 0.05, and fix \(\sigma =\eta =1\).
The variance is a curve in terms of \(\rho \).
References
Aït-Sahalia Y, Fan J, Xiu D (2010) High frequency covariance estimates with noisy and asynchronous data. J Am Stat Assoc 160:1504–1517
Aldous D, Eagleson G (1978) On mixing and stability of limit theorems. Ann Prob 6(2):325–331
Bachelier L (1900) Théorie de la speculation. Gauthier-Villars, Paris
Bandi FM, Russell JR (2005) Realized covariation, realized beta and microstructure noise. Working paper, Graduate School of Business, University of Chicago
Barndorff-Nielsen O, Graversen S, Jacod J, Podolskij M, Shephard N (2006) A central limit theorem for realised power and bipower variations of continuous semimartingales. In: Kabanov Y, Lipster R (eds) From stochastic analysis to mathematical finance, Festschrift for Albert Shiryaev. Springer, Berlin
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
Barndorff-Nielsen OE, Shephard N (2004) Power and bipower variation with stochastic volatility and jumps. J Financ Econom 2(1):1–37
Black F, Scholes M (1973) The pricing of options and corporate liabilities. J Polit Econ 81(3):637–654
Christensen K, Kinnebrock S, Podolskij M (2010) Pre-averaging estimators of the ex-post covariance matrix in noisy diffusion models with non-synchronous data. J Econom 159:116–133
Cox J (1975) Notes on option pricing i: constant elasticity of diffusions. Unpublished draft, Stanford University
Emanuel DC, MacBeth JD (1982) Further results of the constant elasticity of variance call option pricing model. J Financ Quant Anal 4:533–553
Epps T (1979) Comovements in stock prices in the very short run. J Am Stat Assoc 74:291–298
Fama EF, French KR (1992) The cross-section of expected stock returns. J Financ 47:427–465
Fleming J, Kirby C, Ostdiek B (2001) The economic value of volatility timing. J Financ 56(1):329–352
Fleming J, Kirby C, Ostdiek B (2002) The economic value of volatility timing using “realized” volatility. J Financ Econ 67(3):473–509
Hayashi T, Yoshida N (2005) On covariance estimation of non-synchronously observed diffusion processes. Bernoulli 11:359–379
Hayashi T, Yoshida N (2008) Asymptotic normality of a covariance estimator for nonsynchronously observed diffusion processes. Ann Inst Stat Math 60:367–406
Hayashi T, Yoshida N (2011) Nonsynchronous covariation process and limit theorems. Stoch Process Appl 121:2416–2454
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 (2012) Statistics and high frequency data. In: Kessler M,Lindner A, S\(\phi \)rensen M (eds) Proceedings of the 7th Séminaire Européen de Statistique, La Manga, 2007: statistical methods for stochastic differential equations
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
Jacod J, Podolskij M, Vetter M (2010) Limit theorems for moving averages of discretized processes plus noise. Ann Stat 38(3):1478–1545
Jacod J, Shiryayev AV (2003) Limit theorems for stochastic processes. Springer, New York
Jing BY, Li CX, Liu Z (2013) On estimating the integrated co-volatility using noisy high frequency data with jumps. Commun Stat Theory Methods 42(21):3889–3901
Jing BY, Liu Z, Kong XB (2014) Estimating the volatility functionals with multiple transactions. Working paper
Koike Y (2014) Limit theorems for the pre-averaged hayashicyoshida estimator with random sampling. Stoch Process Appl 124:2699–2753
Liu Z (2013) Multiple power estimation of volatility with multiple observations. Working paper
Renyi A (1963) On stable sequences of events. Sankhya Indian J Stat 25:293–302
Tsay RS (2005) Analysis of financial time series. Wiley, Hoboken
Vetter M (2010) Limit theorems for bipower variation of semimartingales. Stoch Process Appl 120:22–38
Voev V, Lunde A (2007) Integrated covariance estimation using high-frequency data in presence of noise. J Financ Econom 5(1):68–104
Wang K, Liu J, Liu Z (2013) Disentangling the effect of jumps on systematic risk using a new estimator of integrated co-volatility. J Bank Financ 37(5):1777–1786
Zhang L (2006) Efficient estimation of stochastic volatility using noisy observations: a multi-scale approach. Bernoulli 12(6):1019–1043
Zhang L (2011) Estimating covariation: Epps effect, microstructure noise. J Econom 160(1):33–47
Zhang L, Mykland P, Aït-Sahalia Y (2005) A tale of two time scales: determining integrated volatility with noisy high-frequency data. J Am Stat Assoc 100(472):1394–1411
Acknowledgments
The authors would like to thank the Editor Prof. Yury Kutoyants, the associate editor and two referees for their very extensive and constructive suggestions which helped to improve this paper considerably. The work is partially supported by The Science and Technology Development Fund of Macau (Nos.078/2012/A3 and 078/2013/A3) and NSFC (Nos.11401607 and 71471173).
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
We need some notations, which simplify the presentation of the proof. For the process X, we set
Furthermore, by a standard localization procedure as presented in Barndorff-Nielsen et al. (2006) or Jacod (2012), there is without loss of generality to impose the following assumption addition to the assumptions of theorems.
Assumption 4
\(\sigma ^X\) and \(\sigma ^Y\) are bounded.
Thus, to prove Theorem 1, it remains to prove that
We start from an auxiliary lemma.
Lemma 1
Under the Assumptions 1–2 and 4, we have
Proof of Lemma 1
By Cauchy–Schwarz inequality, B urkholder inequality, Itô’s Isometry and the fundamental inequality
simple computations show that
because \(\{\sigma _s, s>0\}\) is cádlàg. \(\square \)
Proof of Theorem 1
In view of Lemma 1, it is enough to prove
We further let \(\tilde{\theta }_i=\frac{\theta _i(X)\theta _i(Y)}{\varpi _i}\) and \(\tilde{\theta }'_i=E[\tilde{\theta }_i|\mathcal {{F}}_{i}].\) We firstly have
and in particular \(\tilde{\theta }'_i\le K\Delta _n\). Since \(E\left[ (\tilde{\theta }_i-\tilde{\theta }'_i)(\tilde{\theta }_j-\tilde{\theta }'_j)\right] =0\) when \(|i-j|\ge 2\), thus
by Burkholder–Davis–Gundy’s inequality. Thus, to prove the theorem, it suffices to show
Observe that
The required result follows from the Riemann integrability. \(\square \)
Proof of Theorem 2
To derive the above central limit theorem, note that the summands are one-step correlated, as illustrated in following picture. One of the methodologies to deal with the correlated variables is constructing the martingale differences, then using the martingale central limit theorem. This is similar to the central limit theorem of bi-power variation, as shown in Barndorff-Nielsen et al. (2006). Here, we apply the technique of “big blocks” and “small blocks”, which is classical method proving the central limit theorem of summation of correlated random variables, in probability theory. The extension to stochastic processes, have been developed in Jacod et al. (2009) and Vetter (2010). The “big blocks” (including p terms) are used to construct independent terms in the summation, and eventually dominates the asymptotic behavior. Whereas the “small blocks” (including only one term), which is asymptotically negligible, will be removed from the summation. Following is the detail. Giving a positive integer p, we define
That is, the ith “big block” consists of \(\{\mu _i: i\in A_i(p)\}\), whereas the ith ”small blocks” contains \(\{\mu _i: i\in B_i(p)\}\). Because \(\mu _i\) is 1-dependent sequence, after removing the small blocks, we get an independent (conditional) sequence. Denote:
We first collect all the terms in \(A_i(p)\), define as
and all the terms in \(B_i(p)\), define as
Notice that \(\varsigma _i(p,1)\) contains p summands (“big”), whereas \(\varsigma _i(p,2)\) contains only 1 summand (“small”), because finally we will let \(p\rightarrow \infty \), hence the small blocks are asymptotically negligible. To realize this, we set
We then have the following decomposition:
where, R(p) consists of some “residuals” from above expansions. It is easy to follow the similar steps in Jacod et al. (2009) (identity (5.14), Lemma 5.5 and Lemma 5.6), we obtain that the last four terms are asymptotically negligible, when \(p\rightarrow \infty \) and \(n\rightarrow \infty \), i.e.,
for any \(\delta >0\). We now establish the central limit theorem for M(p) by an auxiliary lemma. \(\square \)
Lemma 2
Under the same assumptions and notations as in Theorem 2 and the Assumption 4 holds, we have for any fixed p,
where,
\(\varrho \) is defined in (60) later.
Proof of Lemma 2
Since \(L_i\equiv L\), hence we have
By a martingale central limit theorem argument as presented in Theorem IX7.28 in Jacod and Shiryayev (2003), we need to show the following conditions:
where \(\Delta V(p)_i=V_{s_{b_i(p)}}-V_{s_{a_i(p)-1}}\) for any process V, and the Eq. (53) holds for any bounded martingale N which is orthogonal to both \(W^X\) and \(W^Y\) and also for \(N=W^X\) and \(N=W^Y\).
To show these convergences, we use the following approximations:
for \(0\le i\le n-1\) and integer l. Hence, we have the following new notations:
and the following approximations are used for different kinds of “blocks”:
Using the approximations of Jacod et al. (2009) (Lemma 5.2 and Lemma 5.3), we only need to show (49)–(53) by redefining
Direct calculation can show (50) and it is easy to see that
The proof of (53) is the same as Jacod et al. (2009) (Lemma 5.7) or Barndorff-Nielsen et al. (2006). We hence left to proving the Equation (49). By (47), we have
Hence, recall (56), when \(k,r\in A_i(p)\), we have
where,
Let
Therefore, we obtain
Hence by Riemann integrability, we obtain
where,
\(\square \)
We now return to the proof of Theorem 2. In view of
we obtain the required result of Theorem 2.\(\square \)
Sketched Proof of Theorem 4:
For any process Z, let
Then we have \(\Delta _{i,k_n}X=\overline{\xi }_{i}(X)+\overline{\xi }'_{i+1}(X)\). Further let \(\overline{\overline{\xi }}_{i}(Z)=\sqrt{k_n}\left( \overline{\xi }_{i}(Z)+\overline{\xi }'_{i+1}(Z)\right) \). Thus
Routine techniques can show that
Firstly, using similar approximations as previous theorem, some computation yields
Notice that \(\frac{1}{L}\sum _{k=1}^L\left[ \left( \frac{k-1}{L}\right) +\left( 1-\frac{k-1}{L}\right) \right] ^2=1\), thus
Following the proof of Lemma 5.4 in Jacod et al. (2009), one can show that \(\frac{1}{n}\sum _{i=0}^{n-k_n}e_n\xrightarrow {P} 0\). Hence
Secondly, note that
Recalling \(k_n=\frac{1}{\sqrt{\Delta _n}}\) and \(n=\lfloor \frac{t}{\Delta _n}\rfloor \), by Law of Large Numbers,
Finally, observe that
It is easy to show that \(J_{kn}\xrightarrow {P}0\) for \(k=1,2,3\) under Assumption 3. And
where \(\bar{\epsilon }_{s_i}=\frac{1}{L}\sum _{k=1}^L\epsilon _{t^N_{N_{i-1}+k}}\) with \(\bar{\epsilon }_0=0\). Since \(E[\bar{\epsilon }_{s_i}\bar{\delta }_{s_i}]=\frac{\varphi }{L}\) and \(E[\bar{\epsilon }_{s^n_{i-1}}\bar{\epsilon }_{s^n_{i-1}}]=0\) by Assumption 3, we get
by Law of Large Numbers. Combining (64)–(67), (70)–(72), we obtain the result of Theorem 4. \(\square \)
Rights and permissions
About this article
Cite this article
Liu, Z. Estimating integrated co-volatility with partially miss-ordered high frequency data. Stat Inference Stoch Process 19, 175–197 (2016). https://doi.org/10.1007/s11203-015-9124-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11203-015-9124-y