Anne Opschoor

New HEAVY Models for Fat-Tailed Returns and Realized Covariance Kernels

We develop a new model for the multivariate covariance matrix dynamics based on daily return observations and daily realized covariance matrix kernels based on intraday data. Both types of data may be fat-tailed. We account for this by assuming a matrix-F distribution for the realized kernels, and a multivariate Student’s t distribution for the returns. Using generalized autoregressive score dynamics for the unobserved true covariance matrix, our approach automatically corrects for the effect of outliers and incidentally large observations, both in returns and in covariances. Moreover, by an appropriate choice of scaling of the conditional score function we are able to retain a convenient matrix formulation for the dynamic updates of the covariance matrix. This makes the model highly computationally efficient. We show how the model performs in a controlled simulation setting as well as for empirical data. In our empirical application, we study daily returns and realized kernels from 15 equities over the period 2001-2012 and find that the new model statistically outperforms (recently developed) multivariate volatility models, both in-sample and out-of-sample. We also comment on the possibility to use composite likelihood methods for estimation if desired.

The R Package MitISEM: Mixture of Student-t Distributions using Importance Sampling Weighted Expectation Maximization for Efficient and Robust Simulation

This paper presents the R package MitISEM, which provides an automatic and flexible method to approximate a non-elliptical target density using adaptive mixtures of Student-t densities, where only a kernel of the target density is required. The approximation can be used as a candidate density in Importance Sampling or Metropolis Hastings methods for Bayesian inference on model parameters and probabilities. The package provides also an extended MitISEM algorithm, ‘sequential MitISEM’, which substantially decreases the computational time when the target density has to be approximated for increasing data samples. This occurs when the posterior distribution is updated with new observations and/or when one computes model probabilities using predictive likelihoods. We illustrate the MitISEM algorithm using three canonical statistical and econometric models that are characterized by several types of non-elliptical posterior shapes and that describe well-known data patterns in econometrics and finance. We show that the candidate distribution obtained by MitISEM outperforms those obtained by ‘naive’ approximations in terms of numerical efficiency. Further, the MitISEM approach can be used for Bayesian model comparison, using the predictive likelihoods.

New HEAVY Models for Fat-Tailed Realized Covariances and Returns

ABSTRACT We develop a new score-driven model for the joint dynamics of fat-tailed realized covariance matrix observations and daily returns. The score dynamics for the unobserved true covariance matrix are robust to outliers and incidental large observations in both types of data by assuming a matrix-F distribution for the realized covariance measures and a multivariate Students t distribution for the daily returns. The filter for the unknown covariance matrix has a computationally efficient matrix formulation, which proves beneficial for estimation and simulation purposes. We formulate parameter restrictions for stationarity and positive definiteness. Our simulation study shows that the new model is able to deal with high-dimensional settings (50 or more) and captures unobserved volatility dynamics even if the model is misspecified. We provide an empirical application to daily equity returns and realized covariance matrices up to 30 dimensions. The model statistically and economically outperforms competing multivariate volatility models out-of-sample. Supplementary materials for this article are available online.

Combining Density Forecasts Using Censored Likelihood Scoring Rules

We investigate the added value of combining density forecasts for asset return prediction in a specific region of support. We develop a new technique that takes into account model uncertainty by assigning weights to individual predictive densities using a scoring rule based on the censored likelihood. We apply this approach in the context of recently developed univariate volatility models (including HEAVY and Realized GARCH models), using daily returns from the S&P 500, DJIA, FTSE and Nikkei stock market indexes from 2000 until 2013. The results show that combined density forecasts based on the censored likelihood scoring rule significantly outperform pooling based on the log scoring rule and individual density forecasts. The same result, albeit less strong, holds when compared to combined density forecasts based on equal weights. In addition, VaR estimates improve a t the short horizon, in particular when compared to estimates based on equal weights or to the VaR estimates of the individual models.We investigate the added value of combining density forecasts focused on a specific region of support. We develop a forecast combination scheme that assigns weights to individual predictive densities based on the censored likelihood scoring rule. We apply this approach in the context of measuring downside risk in equity markets using recently developed volatility models, including HEAVY, Realized GARCH and GAS models, applied to daily returns on the S&P 500, DJIA, FTSE and Nikkei indexes from 2000 until 2013. The results show that combined density forecasts based on the censored likelihood scoring rule significantly outperform pooling based on the log scoring rule and individual density forecasts. The same conclusion, albeit less strong, holds when compared to combined density forecasts based on equal weights. In addition, VaR estimates improve compared to estimates based on equal weights, the log score or individual models.

Fractional Integration and Fat Tails for Realized Covariance Kernels and Returns

We introduce a new fractionally integrated model for covariance matrix dynamics based on the long-memory behavior of daily realized covariance matrix kernels and daily return observations. We account for fat tails in both types of data by appropriate distributional assumptions. The covariance matrix dynamics are formulated as a numerically efficient matrix recursion that ensures positive definiteness under simple parameter constraints. Using intraday stock data over the period 2001-2012, we construct realized covariance kernels and show that the new fractionally integrated model statistically and economically outperforms recent alternatives such as the Multivariate HEAVY model and the 2006 “long-memory” version of the Riskmetrics model.