Abstract
This paper generalizes the basic Wishart multivariate stochastic volatility model of Philipov and Glickman (J Bus Econ Stat 24:313–328, 2006) and Asai and McAleer (J Econom 150:182–192, 2009) to encompass regime-switching behavior. The latent state variable is driven by a first-order Markov process. The model allows for state-dependent (co)variance and correlation levels and state-dependent volatility spillover effects. Parameter estimates are obtained using Bayesian Markov Chain Monte Carlo procedures and filtered estimates of the latent variances and covariances are generated by particle filter techniques. The model is applied to five European stock index return series. The results show that the proposed regime-switching specification substantially improves the fit to persistent covariance dynamics relative to the basic model.
Similar content being viewed by others
Notes
See the excellent overview on multivariate SV models of Asai et al. (2006).
The simulation results are found to be robust to variations in the parameter values.
I also analyzed a 3-state Markov Switching WMSV model. However, estimates indicated only two periods, where the third volatility state was actually present (i.e. the smoothed state probability for the third state exceeds 0.5), see also the according remark in Sect. 3.2.2.
See e.g. the high-dimensional application of Jensen and Maheu (2013), p. 16.
For details on the Gibbs sampling algorithm and Monte Carlo Markov Chain methods see e.g. Bauwens et al. (1999).
In the empirical application below the mean Metropolis–Hastings acceptance rate for the sampling of \(\Sigma _t\) amounts to 20 %, indicating a good performance of the single-move sampling scheme for \(\Sigma _t\) using 10 Metropolis–Hastings iterations in each Gibbs iteration.
In the empirical application the number of particles is set to 100,000 to ensure an accurate approximation of the according moments.
The basic particle filter algorithm is illustrated in the additional web-appendix available at http://www.wisostat.uni-koeln.de/de/institut/mitarbeiter/gribisch/. An alternative method for the numerical approximation of the marginal likelihood is developed by Chib (1995), and Chib and Jeliazkov (2001), and builds on a set of additional reduced Gibbs runs given the Gibbs sample from the preceding parameter estimation.
Datastream DS market indices.
The daily prices \(p_t\) are transformed into continuously compounded rates \(r_t=100\times \ln (p_t/p_{t-1})\) which are then filtered for VAR(p) processes according to the Akaike information criterium using homoskedastic ML.
Initial estimations with uninformative prior distributions indicated that it is rather important to use a tight prior for the d.o.f. parameter \(\nu \) to improve the mixing of the generated Markov chain and thereby the numerical efficiency of the procedure. Here it is important to note that \(\nu \) can be interpreted as a nuisance parameter, which is of limited interest since it does not affect the dynamics of the covariance process. However, as indicated by Eq. (8), \(\nu , A\) and d jointly affect unconditional variance moments, which explains the rather strong correlation of the Gibbs draws of these parameters, negatively affecting the performance of the Gibbs sampler. The prior for \(\nu \) ist centered at the mean of a set of initial Gibbs draws (after convergence) obtained under an uninformative prior. The prior assumptions for A and d are uninformative.
The prior distributions for \(\nu , e_1\) and \(e_2\) are centered at the mean of a set of initial Gibbs draws (after convergence) obtained under an uninformative prior. Initial investigation showed that using informative priors on the state transition probabilities \(e_1\) and \(e_2\) considerably increased the numerical performance of the Gibbs sampling algorithm, which is a well-known result and makes the Bayesian estimation scheme attractive for incorporating unbinding model restrictions in high-dimensional state-space models. See e.g. the high-dimensional application of Jensen and Maheu 2013, p. 16 and Table 6 on p. 17. Further estimations using uninformative \(\text {Beta}(1,1)\) priors on \(e_1\) and \(e_2\) showed that the reported point estimates are largely robust to the prior assumptions (results are available upon request). However, loosening the prior restrictions results in a considerable increase of numerical standard errors, e.g. standard errors almost tripled in case of elements of \(A_2^{-1}\). The priors for A and d are uninformative.
For initial investigation I fitted a 3-state MS WMSV model to the data illustrated in Sect. 3.1 (80,000 Gibbs iterations, 20,000 burn-in) and found that the smoothed estimates of the latent state variable \(s_t\) did only imply 1 turbulent day in 2003 where the third regime was actually present (i.e. smoothed state probabilities exceeding 0.5) with an estimated unconditional probability for the third state of 0.0025. The obtained log marginal likelihood amounts to -6197.9 implying a Bayes factor relative to the 2-state WMSV model of 22.20, which is labeled as “strong” indication for the 3-state model according to Kass and Raftery, (1995) but is relativized by the fact that estimation results for this model are virtually the same as for the 2-state approach and inference on \(A_3^{-1}\) based on 1 observation is not reasonable. Gains in forecasting and/or state inference cannot be expected (the model features 19 additional parameters). Furthermore, the marginal likelihood gain of including a third regime is minor compared to the gain of introducing a second regime.
See Hamilton (1994), p. 683, for the computation of unconditional state probabilities.
See e.g. Bauwens et al. (1999) for details on the Metropolis–Hastings algorithm.
References
Andreou, E., Ghysels, E.: Detecting multiple breaks in financial market volatility dynamics. J. Appl. Econom. 17, 579–600 (2002)
Andrieu, C., Doucet, A., Holenstein, R.: Particle Marko chain Monte Carlo methods (with discussion). J. R. Stat. Soc. Ser. B 72(3), 269–342 (2010)
Asai, M., McAleer, M., Yu, J.: Multivariate stochastic volatility: a review. Econom. Rev. 25, 145–175 (2006)
Asai, M., McAleer, M.: The structure of dynamic correlations in multivariate stochastic volatility models. J. Econom. 150, 182–192 (2009)
Bauwens, L., Lubrano, M., Richard, J.F.: Bayesian Inference in Dynamic Econometric Models. Oxford University Press, New York (1999)
Bekaert, G., Ehrmann, M., Fratzscher, M., Mehl, A.: Global crisis and equity market contagion. Working paper (2011)
Billio, M., Carporin, M.: Market linkages, variance spillovers, and correlation stability: empirical evidence of financial contagion. Comput. Stat. Data Anal. 54, 2443–2458 (2010)
Black, F.: Studies of stock market volatility changes. In: Proceedings of the American Statistical Association, Business and Economic Statistics Section pp. 177–181 (1976)
Bodnar, T., Okhrin, Y.: Properties of the singular, inverse and generalized inverse partitioned Wishart distributions. J. Multivar. Anal. 99, 2389–2405 (2008)
Bollerslev, T.: Modelling the coherence in short-run nominal exchange rates: a multivariate generalized ARCH model. Rev. Econ. Stat. 72, 498–505 (1990)
Carvalho, C.M., Lopes, H.F.: Simulation-based sequential analysis of Markov switching stochastic volatility models. Comput. Stat. Data Anal. 51, 4526–4542 (2007)
Chiang, M.-H., Wang, L.-M.: Volatility contagion: a range-based volatility approach. J. Econom. 165, 175–189 (2011)
Chib, S.: Marginal likelihood from the Gibbs output. J. Am. Stat. Assoc. 90(432), 1313–1321 (1995)
Chib, S., Nardari, F., Shephard, N.: Analysis of high dimensional multivariate stochastic volatility models. J. Econom. 134, 341–371 (2006)
Chib, S., Jeliazkov, I.: Marginal likelihood from the Metropolis–Hastings output. J. Am. Stat. Assoc. 96(453), 270–281 (2001)
Christie, A.A.: The stochastic behavior of common stock variances: value, leverage and interest rate effects. J. Financ. Econ. 10, 407–432 (1982)
Christoffersen, P.: Evaluating interval forecasts. Int. Econ. Rev. 39, 841–862 (1998)
Daníelsson, J.: Multivariate stochastic volatility models: estimation and a comparison with VGARCH models. J. Empir. Finance 5, 155–173 (1998)
Diebold, F.X.: Modeling the persistence of conditional variances: a comment. Econom. Rev. 5, 51–56 (1986)
Diebold, F.X., Yilmaz, K.: Measuring financial asset return and volatility spillovers, with application to global equity markets. Econ. J. 119, 158–171 (2009)
Doz, C., Renault, F.: Factor stochastic volatility in mean models: a GMM approach. Econom. Rev. 25, 275–309 (2006)
Engle, R.F.: Dynamic conditional correlation: a simple class of multivariate GARCH models. J. Bus. Econ. Stat. 20, 339–350 (2002)
Engle, R.F., Kroner, K.F.: Multivariate simultaneous generalized ARCH. Econom. Theory 11, 122–150 (1995)
Forbes, K.J., Rigobon, R.: No contagion, only interdependence: measuring stock market comovements. J. Finance 57, 2223–2261 (2002)
Gallo, G.M., Otranto, E.: Volatility spillovers, interdependence and comovements: a Markov switching approach. Comput. Stat. Data Anal. 52, 3011–3026 (2008)
Gander, M.P.S., Stephens, D.A.: Stochastic volatility modeling with general marginal distributions: inference, prediction and model selection. J. Stat. Plan. Inference 137, 3068–3081 (2007)
Geweke, J.: Contemporary Bayesian Econometrics and Statistics. Wiley, Hoboken, New Jersey (2005)
Gordon, N.J., Salmond, D.J., Smith, A.F.M.: Novel approach to nonlinear/non-Gaussian Bayesian state estimation. IEE Proc. F 140(2), 107–113 (1993)
Gray, S.F.: Modeling the conditional distribution of interest rates as a regime-switching process. J. Financ. Econom. 42, 27–62 (1996)
Haas, M., Mittnik, S., Paolella, M.S.: A new approach to Markov-switching GARCH models. J. Financ. Econom. 2, 493–530 (2004)
Hamilton, J.D.: Time Series Analysis. Princeton University Press, Princeton (1994)
Hamilton, J.D., Susmel, R.: Autoregressive conditional heteroskedasticity and changes in regime. J. Econom. 64, 307–333 (1994)
Harvey, A., Ruiz, E., Shephard, N.: Multivariate stochastic variance models. Rev. Econ. Stud. 61, 247–264 (1994)
Hull, J., White, A.: The pricing of options on assets with stochastic volatilities. J. Finance 42, 281–300 (1987)
Jacquier, E., Polson, N.G., Rossi, P.E.: Stochastic volatility: univariate and multivariate extensions. Working paper 99s-26, CIRANO, Montreal (1999)
Jeffreys, H.: Theory of Probability, 3rd edn. Oxford University Press, Oxford (1961)
Jensen, M.J., Maheu, J.M.: Bayesian semiparametric multivariate GARCH modeling. J. Econom. 176, 3–17 (2013)
Jin, X., Maheu, J.M.: Modelling realized covariances and returns. Working paper (2012)
Kass, R.E., Raftery, : Bayes factors. J. Am. Stat. Assoc. 90(430), 773–795 (1995)
Kim, S., Shephard, N., Chib, S.: Stochastic volatility: likelihood inference and comparison with ARCH models. Rev. Econ. Stud. 65, 361–393 (1998)
Kim, C.-J., Nelson, C.R.: State-Space Models with Regime Switching: Classical and Gibbs-Sampling Approaches with Applications. MIT Press, Cambridge (1999)
Koop, G.: Bayesian Econometrics. Wiley, Chichester (2003)
Lamoureux, C.G., Lastrapes, W.D.: Persistence in variance, structural change and the GARCH model. J. Bus. Econ. Stat. 8, 225–234 (1990)
Liesenfeld, R., Richard, J.-F.: Univariate and multivariate stochastic volatility models: estimation and diagnostics. J. Empir. Finance 10, 505–531 (2003)
Liesenfeld, R., Richard, J.-F.: Improving MCMC, using efficient importance sampling. Comput. Stat. Data Anal. 53, 272–288 (2008)
Lopes, H.F., Carvalho, C.M.: Factor stochastic volatility with time varying loadings and Markov switching regimes. J. Stat. Plan. Inference 137, 3082–3091 (2007)
Lopez, J.A., Walter, C.A.: Evaluating covariance matrix forecasts in a value-at-risk framework. J. Risk 3, 69–97 (2001)
Morgan, J.P.: RiskMetrics. Technical Document, 4th edn. J.P.Morgan/Reuters New York (1996)
Muirhead, R.J.: Aspects of Multivariate Statistical Theory. Wiley, New Jersey (1982)
Nonejad, N.: Flexible model comparison of unobserved components models using particle Gibbs with ancestor sampling. Econ. Lett. 133, 35–39 (2015)
Noureldin, D., Shephard, N., Sheppard, K.: Multivariate high-frequency-based volatility (HEAVY) models. J. Appl. Econom. 27, 907–933 (2011)
Philipov, A., Glickman, M.E.: Multivariate stochastic volatility via Wishart processes. J. Bus. Econ. Stat. 24, 313–328 (2006)
Pitt, M.K., Shephard, N.: Time varying covariances: a factor stochastic volatility approach. In: Bernardo, J.M., Berger, J.O., David, A.P., Smith, A.F.M. (eds.) Bayesian Statistics, vol. 6, pp. 547–570. Oxford University Press, Oxford (1999b)
Ross, S.M.: Simulation. Academic Press, New York (2002)
Shephard, N.: Statistical aspects of ARCH and stochastic volatility. In: Cox, D.R., Hinkley, D.V., Barndorff- Nielsen, O.E. (eds.) Time Series Models in Econometrics, Finance and Other Fields, pp. 1–67. Chapman & Hall, London (1996)
Smith, M., Pitts, A.: Foreign exchange intervention by the Bank of Japan: Bayesian analysis using a bivariate stochastic volatility model. Econom. Rev. 25, 425–451 (2006)
So, M.K.P., Lam, K., Li, W.K.: A stochastic volatility model with Markov switching. J. Bus. Econ. Stat. 16, 244–253 (1998)
Solnik, B., Bourcrelle, C., Le Fur, Y.: International market correlation and volatility. Financ. Anal. J. 52, 17–34 (1996)
Taylor, S.J.: Financial returns modelled by the product of two stochastic processes—a study of daily sugar prices. In: Anderson, O.D. (ed.) Time Series Analysis: Theory and Practice, vol. 1, pp. 203–226. North-Holland, Amsterdam (1982)
Taylor, S.J.: Modelling Financial Time Series. Wiley, Chichester (1986)
Yu, J., Meyer, R.: Multivariate stochastic volatility models: Bayesian estimation and model comparison. Econom. Rev. 25, 361–384 (2006)
Acknowledgments
The author would like to thank two anonymous referees, the associate editor Roman Liesenfeld and Jan Patrick Hartkopf for their helpful and very constructive comments.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
1.1 Full conditional distributions: basic WMSV model
The basic WMSV model is outlined in Eqs. (1), (2) and (5). The joint prior distribution is assumed to factor into the product of marginal prior distributions given by
-
1.
a Wishart prior \(\pi _{A^{-1}}(Q_0,\gamma _0)\) for \(A^{-1}\) with scale matrix \(Q_0\) and d.o.f. parameter \(\gamma _0\);
-
2.
a uniform prior \(\pi _{d}(0,1)\) on [0, 1] for d;
-
3.
a gamma prior \(\pi _{\nu }(\alpha _0,\beta _0)\) for \(\nu -k\) with shape parameter \(\alpha _0\) and scale parameter \(\beta _0\).
Denote the augmented parameter vector by \(\theta ^{\text {aug}}=(\theta ',\text {vech}(\Sigma _1)',\ldots ,\text {vech}(\Sigma _T)')\) and the vector of remaining model parameters for each parameter block \(\theta _k\) by \(\theta ^{\text {aug}}_{-}=(\theta _1,\ldots ,\theta _{k-1},\theta _{k+1},\ldots ,\theta _K)'\). The full conditional distributions are obtained as follows:
(more details on the derivation of the full conditional densities are given in an additional web-appendix available at http://www.wisostat.uni-koeln.de/de/institut/mitarbeiter/gribisch/)
Full conditional distribution of \(\Sigma _t^{-1}\):
For notational convenience suppressing dependence on model parameters, the kernel of the full conditional distribution of \(\Sigma _t^{-1}\) is obtained as
where \({\mathcal W}_k^{\kappa }(\Sigma _t^{-1}|\cdot )\) denotes a Wishart kernel in \(\Sigma _t^{-1}\) and \(\tilde{\nu }= \nu (1-d)+1, \tilde{S}_{t} = (S_{t}^{-1}+\xi _t \xi _t')^{-1}, f(\Sigma _t^{-1}) = \exp \{ -0.5\ \text {tr}[S_{t+1}^{-1}\Sigma _{t+1}^{-1}] \}, S_t = \Sigma _{t}^{-d/2}A\Sigma _{t}^{-d/2}\). The full conditional distribution of \(\Sigma _t^{-1}\) is known up to an integrating constant. Hence, the Metropolis–Hastings (MH) algorithmFootnote 16 is applied to obtain samples from \(p(\Sigma _t^{-1}|\theta ^{\text {aug}}_{-})\). The proposal density is \({\mathcal W}_k(\nu ,\tilde{S}_{t})\).
Full conditional distribution of \(A^{-1}\):
The full conditional distribution of \(A^{-1}\) is Wishart since
where \(U^{-1} = \nu \sum _{t=1}^{T} \Sigma _{t-1}^{d/2}\Sigma _{t}^{-1} \Sigma _{t-1}^{d/2}\) and \(\gamma = T\nu +k+1\). Hence
Therefore, \(A^{-1}|\theta ^{\text {aug}}_{-} \sim {\mathcal W}_k(\tilde{\gamma }, \tilde{U})\), where \(\tilde{U}^{-1} = Q_0^{-1}+U^{-1}\) and \(\tilde{\gamma } = \gamma _0+\gamma -k-1\).
Full conditional distribution of \(\nu \) and d:
The full conditional distributions of the parameters \(\nu \) and d are not obtained in closed form and simulation is carried out by applying the Metropolis–Hastings algorithm. Since \(\nu > k\) and \(d \in (0,1)\), truncated normal proposal densities are used where mean and variance are given by the optimum and the corresponding Hessian obtained after numerically optimizing the posterior distribution’s density kernel.
The respective kernels are obtained as
where \(\psi = -\frac{\nu }{2} \sum _{t=1}^{T}\ln (|\Sigma _{t-1}^{-1}|), Q(d) = \sum _{t=1}^{T} \nu \Sigma _{t-1}^{d/2}\Sigma _{t}^{-1}\Sigma _{t-1}^{d/2}, Q_t^{-1}=\Sigma _{t-1}^{d/2} \Sigma _t^{-1}\Sigma _{t-1}^{d/2}\) and \(Q^{-1} = \nu \sum _{t=1}^{T} \Sigma _{t-1}^{d/2} \Sigma _t^{-1} \Sigma _{t-1}^{d/2}\).
1.2 Full conditional distributions: Markov switching MWSV model
The MS WMSV model is outlined in Eqs. (10), (11) and (14). The joint prior distribution is assumed to factor into the product of marginal prior distributions. Given the state sequence \(s=(s_1,s_2,\ldots ,s_T)'\), the derivation of the full conditional distributions for \(\Sigma _t^{-1}, A_1, A_2, \nu \) and d is analogous to the illustrations of the previous section, except that we have to condition on \(A_{s_t}\ \forall t\in \{1,\ldots ,T\}\) instead of A.
Full conditional distribution of \(s=(s_1,s_2,\ldots ,s_T)'\):
Denoting \(\underline{\Sigma }_t^{-1}=\{\Sigma _1^{-1},\ldots ,\Sigma _t^{-1}\}\) and exploiting the Markov property of \(s_t\), the full conditional density of the state vector s can be factorized as
The conditional probabilities
are obtained by the “Hamilton filter” which—given a starting value for \(P(s_{0}|\underline{\Sigma }_{0}^{-1},\theta )\) (e.g. stationary probabilities, see Hamilton 1994, p. 683)—proceeds recursively in five steps \(\forall t \in \{1,\ldots ,T\}\):
The whole state sequence \(s=(s_1,s_2,\ldots ,s_T)'\) can then be sampled backward recursively based on Eq. (39).
Full conditional distributions of \(e_1\) and \(e_2\):
Using beta prior distributions \(\pi _{e_i}(\alpha _{i,0},\beta _{i,0}), i\in \{1,2\}\), the kernel of the full conditional distribution of \(e_i\) is obtained as
where \(g_i\) denotes the number of switches from state i to state \(i-\) (not state i) and \(h_i\) denotes the number of periods where the state does not change. The full conditional distribution of \(e_i\) is therefore beta with parameters \(\alpha _{i}=\alpha _{i,0}+g_i\) and \(\beta _{i}=\beta _{i,0}+h_i, i\in \{1,2\}\).
Rights and permissions
About this article
Cite this article
Gribisch, B. Multivariate Wishart stochastic volatility and changes in regime. AStA Adv Stat Anal 100, 443–473 (2016). https://doi.org/10.1007/s10182-016-0269-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10182-016-0269-9