## Abstract

Stochastic volatility (SV) models mimic many of the stylized facts attributed to time series of asset returns, while maintaining conceptual simplicity. The commonly made assumption of conditionally normally distributed or Student-t-distributed returns, given the volatility, has however been questioned. In this manuscript, we introduce a novel maximum penalized likelihood approach for estimating the conditional distribution in an SV model in a nonparametric way, thus avoiding any potentially critical assumptions on the shape. The considered framework exploits the strengths both of the hidden Markov model machinery and of penalized B-splines, and constitutes a powerful alternative to recently developed Bayesian approaches to semiparametric SV modelling. We demonstrate the feasibility of the approach in a simulation study before outlining its potential in applications to three series of returns on stocks and one series of stock index returns.

This is a preview of subscription content, access via your institution.

## References

Abanto-Valle CA, Bandyopadhyay D, Lachos V, Enriquez I (2010) Robust Bayesian analysis of heavy-tailed stochastic volatility models using scale mixtures of normal distributions. Comput Stat Data Anal 54:2883–2898

Abraham B, Balakrishna N, Sivakumar R (2006) Gamma stochastic volatility models. J Forecast 25:153–171

Arlot S, Celisse A (2010) Model selection. Stat Surv 4:40–79

Bartolucci F, De Luca G (2001) Maximum likelihood estimation of a latent variable time-series model. Appl Stoch Models Bus Ind 17:5–17

Bartolucci F, De Luca G (2003) Likelihood-based inference for asymmetric stochastic volatility models. Comput Stat Data Anal 42:445–449

Bergmeier C, Benítez JM (2012) On the use of cross-validation for time series predictor evaluation. Inf Sci 191:192–213

Chib S, Nardari F, Shephard N (2002) Markov chain Monte Carlo methods for stochastic volatility models. J Econom 108:281–316

Cont R (2001) Empirical properties of asset returns: stylized facts and statistical issues. Quant Financ 1:223–236

de Boor C (1978) A practical guide to splines. Springer, Berlin

Delatola E-I, Griffin JE (2011) Bayesian nonparametric modelling of the return distribution with stochastic volatility. Bayesian Anal 6:901–926

Delatola E-I, Griffin JE (2013) A Bayesian semiparametric model for volatility with a leverage effect. Comput Stat Data Anal 60:97–110

Durham GB (2006) Monte Carlo methods for estimating, smoothing, and filtering one- and two-factor stochastic volatility models. J Econom 133:273–305

Eilers PHC, Marx BD (1996) Flexible smoothing with \(B\)-splines and penalties. Stati Sci 11:89–121

Fridman M, Harris L (1998) A maximum likelihood approach for non-Gaussian stochastic volatility models. J Bus Econ Stat 16:284–291

Fernandez C, Steel MFJ (1998) On Bayesian modeling of fat tails and skewness. J Am Stat Assoc 93:359–371

Gallant AR, Hsieh D, Tauchen GE (1997) Estimation of stochastic volatility models with diagnostics. J Econom 81:159–192

Gneiting T, Raftery AE (2007) Strictly proper scoring rules, prediction, and estimation. J Am Stat Assoc 102:359–378

Harvey AC, Shephard N (1996) Estimation of an asymmetric stochastic volatility model for asset returns. J Bus Econ Stat 14:429–434

Harvey CR, Siddique A (2000) Conditional skewness in asset pricing tests. J Financ 55:1263–1295

Jacquier E, Polson NG, Rossi PE (2004) Bayesian analysis of stochastic volatility models with fat-tails and correlated errors. J Econom 122:185–212

Jondeau E, Rockinger M (2003) Conditional volatility, skewness and kurtosis: existence, persistence and co-movements. J Econ Dyn Control 27:1699–1737

Jensen MJ, Maheu JM (2010) Bayesian semiparametric stochastic volatility modeling. J Econom 157:306–316

Kim S, Shephard N, Chib S (1998) Stochastic volatility: likelihood inference and comparison with ARCH models. Rev Econ Stud 65:361–393

Krivobokova T, Crainiceanu CM, Kauermann G (2008) Fast adaptive penalized splines. J Comput Graph Stat 17:1–20

Langrock R (2011) Some applications of nonlinear and non-Gaussian state-space modelling by means of hidden Markov models. J Appl Stat 38:2955–2970

Langrock R, King R (2013) Maximum likelihood estimation of mark-recapture-recovery models in the presence of continuous covariates. Ann Appl Stat 7:1709–1732

Langrock R, Kneib T, Sohn A, DeRuiter SL (2014) Nonparametric inference in hidden Markov models using P-splines. Biometrics (to appear). arXiv:1309.0423v2

Langrock R, MacDonald IL, Zucchini W (2012) Some nonstandard stochastic volatility models and their estimation using structured hidden Markov models. J Empir Financ 19:147–161

Nakajima J, Omori Y (2012) Stochastic volatility model with leverage and asymmetrically heavy-tailed error using GH skew Student’s -distribution. Comput Stat Data Anal 56:3690–3704

Racine J (2000) Consistent cross-validatory model-selection for dependent data: \(h\nu \)-block cross-validation. J Econom 99:39–61

Rosenblatt M (1952) Remarks on a multivariate transformation. Ann Math Stat 23:470–472

Ruppert D (2002) Selecting the number of knots for penalized splines. J Comput Graph Stat 11:735–757

Ruppert D, Wand MP, Carroll RJ (2003) Semiparametric regression. Cambridge University Press, Cambridge

Schellhase C, Kauermann G (2012) Density estimation and comparison with a penalized mixture approach. Comput Stat 27:757–777

Shephard N (1996) Statistical aspects of ARCH and stochastic volatility. In: Cox DR, Hinkley DV, Barndorff-Nielsen OE (eds) Time series models: in econometrics, finance and other fields. Chapman & Hall, London, pp 1–67

Zucchini W, MacDonald IL (2009) Hidden Markov Models for time series: an introduction using R. Chapman & Hall, London

## Acknowledgments

The authors would like to thank the two anonymous referees who provided useful comments on an earlier version of this manuscript.

## Author information

### Affiliations

### Corresponding author

## Electronic supplementary material

Below is the link to the electronic supplementary material.

## Appendix: HMM essentials

### Appendix: HMM essentials

This appendix reviews some HMM basics. A standard \(m\)-state HMM has the same two process structure as SV models and SSMs, only that the unobserved process is a Markov chain and hence discrete-valued rather than continuous-valued. Consider an HMM with observable process \(\{ X_t \}_{t=1}^T\) and underlying Markov chain \(\{ S_t \}_{t=1}^T\). Given the current state of \(S_t\), the variable \(X_t\) is usually assumed to be conditionally independent from previous and future observations and states. The Markov chain is typically considered to be of first order, and the probabilities of transitions between the different states are summarized in the \(m \times m\) transition probability matrix \(\varvec{\Gamma }=\left( \gamma _{ij} \right) \), where \(\gamma _{ij}=\Pr \bigl (S_{t+1}=j\vert S_t=i \bigr )\), \(i,j=1,\ldots ,m\). The initial state probabilities are summarized in the vector \(\varvec{\pi }\), where \(\pi _{i} = \Pr (S_1=i)\), \(i=1,\ldots ,m\). It is usually convenient and appropriate to assume \(\varvec{\pi }\) to be the stationary distribution. For the described HMM, with observations given by \(x_1,\ldots ,x_T\) and underlying states denoted by \(s_1,\ldots ,s_T\), the likelihood is given by

In this form the likelihood involves \(m^T\) summands, which would make a numerical maximization infeasible in most cases. However, there is a much more efficient way of calculating the likelihood \({\mathcal {L}}^{\text {HMM}}\), given by a recursive scheme called the *forward algorithm*. To see this, we consider the vectors of forward variables, defined as \(\varvec{\alpha }_t = \bigl ( {\alpha }_t (1), \ldots , {\alpha }_t (m) \bigr )\), \(t=1,\ldots ,T\), where \({\alpha }_t (j) = f (x_1, \ldots , x_t, S_t=j)\), \(j=1,\ldots ,m\). We then have the recursion:

where \({\mathbf {Q}}(x_t)= \text {diag} \bigl ( f_1 (x_{t}), \ldots , f_m (x_{t}) \big )\), with \(f_i(x_t) =f (x_{t} | S_{t}=i)\). The recursion (10) can be derived in a straightforward manner using the HMM dependence structure. The likelihood can then be written as a matrix product:

where \({\mathbf {1}}\in {\mathbb {R}}^m\) is a column vector of ones. For a missing observation \(x_t\), the associated matrix \({\mathbf {Q}}(x_{t})\) is simply replaced by the \(m\times m\) identity matrix. For more details on HMMs, see for example Zucchini and MacDonald (2009).

## Rights and permissions

## About this article

### Cite this article

Langrock, R., Michelot, T., Sohn, A. *et al.* Semiparametric stochastic volatility modelling using penalized splines.
*Comput Stat* **30, **517–537 (2015). https://doi.org/10.1007/s00180-014-0547-5

Received:

Accepted:

Published:

Issue Date:

### Keywords

- B-splines
- Cross-validation
- Forward algorithm
- Hidden Markov model
- Numerical integration
- Penalized likelihood