zbMATH — the first resource for mathematics

Fitting stochastic volatility models in the presence of irregular sampling via particle methods and the EM algorithm. (English) Zbl 1199.62042
The stochastic volatility model proposed by S.J. Taylor [Financial returns modelled by the product of two stochastic processes – a study of daily sugar prices, 1961-1979. O.D. Anderson (ed.), Time Ser. Anal.: Theory and Practice, Vol. 1, NYk: Elsevier, Amsterdam: North-Holland Publ., 203–226 (1982)] has become popular for explaining the behaviour of financial time series, such as stock prices and exchange rates caused by the existence of irregular sampling. Stretches of data may not be available for various reasons, such as unexpected political events or natural or man-made disasters. In addition, financial data often have heavier tails than can be captured by standard stochastic volatility models. In the standard stochastic volatility model framework volatility is modelled as an AR process \(h_t=\theta h_{t-1}+w_t\), and the returns \(r_t\) are given by \(r_t=\beta\exp(h_t/2)\epsilon_t\), where \(w_t\) are i.i.d. \(N(0,Q)\), \(h_0\) is \(N(\mu_0,\sigma_0^2)\), \(\epsilon_t\) are i.i.d. \(N(0,1)\), and \(w_t\) and \(\epsilon_t\) are independent processes. The last equation may be represented as \(y_t=\alpha+h_t+v_t\), where \[ y_t = \log(r_y^2),\;\alpha =\log(\beta^2)+E(\log(\epsilon_t^2)),\;v_t=\log(\epsilon_t^2)-E(\log(\epsilon_t^2)). \] Note that \(\epsilon_t^2\sim\chi_1^2\), so that \(v_t\) has a centred \(\log\chi_1^2\) distribution. The first and the third equations represent the standard univariate stochastic volatility model and together they form a linear non-Gaussian state-space model. The authors consider a modification of the standard stochastic volatility model where it is assumed that the observational noise process, \(v_t\), is a mixture of two normals with unknown parameters. The model presented by R.H. Shumway and D.S. Stoffer [Time series analysis and its applicartions. With R examples. 2 nd ed., NY: Springer (2006; Zbl 1096.62088)] retains the state equation for the volatility but the observations are changed to \(y_t=\alpha +h_t+v_t\) with \(v_t=I_tz_{t1}+(1-I_t)z_{t0}-\mu\pi\), where \(z_{t0}\) are i.i.d \(N(0,R_0)\), \(z_{t1}\) are i.i.d. \(N(\mu,R_1)\), \(I_t\) are indicator variables i.i.d. \(B(\pi)\), where \(\pi\) is an unknown mixing probability. The observational noise is constructed using two normals: the \(N(0,R_0)\) term is used to account for most of the noise, whereas the \(N(\mu,R_1)\) term is used to account for the lower tail behaviour of the noise.
The main advantage of the mixture stochastic volatility model over the standard stochastic volatility model is its flexibility. In the modified model \(v_t\) has its own parameter set which must be estimated along with the state parameters \(\{\theta,Q\}\). Therefore this modification can give better parameter estimates for \(\{\theta,Q\}\). For model fitting in the presence of missing or irregular observations the authors combine the expectation-maximization (EM) algorithm with particle filters and smoothers to estimate the parameters of the model. The convergence of the method is discussed. To solve the irregular data problem an imputation technique based on properties of state-space models is used. To demonstrate the viability of the proposed methods several numerical exercises and data analyses are performed.

62P05 Applications of statistics to actuarial sciences and financial mathematics
62M10 Time series, auto-correlation, regression, etc. in statistics (GARCH)
62M20 Inference from stochastic processes and prediction
91G70 Statistical methods; risk measures
65C60 Computational problems in statistics (MSC2010)
Full Text: DOI
[1] Anderson, Consistent estimates of the parameters of a linear system, The Annals of Mathematical Statistics 40 pp 2064– (1969) · Zbl 0213.20703 · doi:10.1214/aoms/1177697286
[2] Arulampalam, A tutorial on particle filters for online nonlinear/non-Gaussian Bayesian tracking, IEEE Transactions on Signal Processing 50 pp 174– (2002) · doi:10.1109/78.978374
[3] Bai, Kurtosis of GARCH and Stochastic Volatility Models with Non-normal Innovations, Journal of Econometrics 114 pp 349– (2003) · Zbl 1012.62109 · doi:10.1016/S0304-4076(03)00088-5
[4] Chan, Monte Carlo EM estimation for time series models involving counts, Journal of the American Statistical Association 90 pp 242– (1995) · Zbl 0819.62069 · doi:10.1080/01621459.1995.10476508
[5] Chib, Markov chain Monte Carlo methods for stochastic volatility models, Journal of Econometrics 108 pp 281– (2002) · Zbl 1099.62539 · doi:10.1016/S0304-4076(01)00137-3
[6] Danielson, Stochastic volatility in asset prices: estimation with simulated maximum likelihood, Journal of Econometrics 61 pp 375– (1994) · Zbl 0825.62953 · doi:10.1016/0304-4076(94)90070-1
[7] Doucet, Parameter estimation in general state-space models using particle methods, Annals of Institute of Statistical Mathematics 55 pp 409– (2003) · Zbl 1047.62084 · doi:10.1007/BF02530508
[8] Doucet, Sequential Monte Carlo Methods in Practice (2001)
[9] Durbin, Time series analysis of non-Gaussian observations based on state space models from both classical and Bayesian perspectives, Journal of the Royal Statistical Society, Series B 62 pp 3– (2000) · Zbl 0945.62084 · doi:10.1111/1467-9868.00218
[10] Durham, SV mixture models with application to s&p 500 index returns, Journal of Financial Economics 85 pp 822– (2007) · doi:10.1016/j.jfineco.2006.06.005
[11] Everitt, Finite Mixture Distributions (1981) · Zbl 0466.62018
[12] Franses, Nonlinear Time Series Models in Empirical Finance (2000)
[13] Godsill, Monte Carlo smoothing for non-linear time series, Journal of the American Statistical Association 199 pp 156– (2004) · Zbl 1089.62517 · doi:10.1198/016214504000000151
[14] Harvey, Multivariate stochastic variance models, Review of Economic Studies 61 pp 247– (1994) · Zbl 0805.90026 · doi:10.2307/2297980
[15] Hathaway, A constrained formulation of maximum-likelihood estimation for normal mixture distributions, The Annals of Statistics 13 pp 795– (1985) · Zbl 0576.62039 · doi:10.1214/aos/1176349557
[16] Jacquier, Bayesian analysis of stochastic volatility models, Journal of Business and Economic Statistics 12 pp 371– (1994)
[17] Kim , J. 2005 Parameter Estimation in Stochastic Volatility Models with Missing Data Using Particle Methods and the EM Algorithm PhD thesis University of Pittsburgh
[18] Kim, Stochastic volatility: likelihood inference and comparison with ARCH models, The Review of Economic Studies 65 pp 361– (1998) · Zbl 0910.90067 · doi:10.1111/1467-937X.00050
[19] Kitagawa, Monte Carlo filter and smoother for non-Gaussian nonlinear state-space models, Journal of Computational and Graphical Statistics 5 pp 1– (1996)
[20] Kitagawa, Sequential Monte Carlo Methods in Practice pp 177– (2001)
[21] Liesenfeld, Univariate and multivariate stochastic volatility models: estimation and diagnostics, Journal of Empirical Finance 10 pp 505– (2003) · doi:10.1016/S0927-5398(02)00072-5
[22] Louis, Finding the observed information matrix when using the EM algorithm, Journal of the Royal Statistical Society, Series B 44 pp 226– (1982) · Zbl 0488.62018
[23] Mclachlan, The EM Algorithm and Extensions (1997) · Zbl 0882.62012
[24] Mellino, Pricing foreign currency options with stochastic volatility, Journal of Econometrics 45 pp 7– (1990)
[25] Sandmann, Estimation of stochastic volatility models via monte carlo maximum likelihood, Journal of Econometrics 87 pp 271– (1998) · Zbl 0937.62110 · doi:10.1016/S0304-4076(98)00016-5
[26] Shephard , N. 1996 Statistical aspects of ARCH and stochastic volatility Time Series Models: In Econometrics, Finance and Other Fields D. R. Cox D. V. Hinkley O. E. Barndorff-Nielsen Chapman and Hall 1 67
[27] Shumway, Time Series Analysis and its Applications: With R Examples (2006) · Zbl 1096.62088
[28] Stoffer, State Space and Unobserved Component Models: Theory and Applications pp 171– (2004)
[29] Tanner, Tools for Statistical Inference (1996) · Zbl 0846.62001
[30] Taylor, Time Series Analysis: theory and Practice pp 203– (1982)
This reference list is based on information provided by the publisher or from digital mathematics libraries. Its items are heuristically matched to zbMATH identifiers and may contain data conversion errors. It attempts to reflect the references listed in the original paper as accurately as possible without claiming the completeness or perfect precision of the matching.