Editorial: Special Issue to Honor the Memory of Maurice B. Priestley, 1933–2013
Tata Subba Rao, Granville Tunnicliffe Wilson · Journal of Time Series Analysis · 2017
Maurice B. Priestley, Professor Emeritus of the University of Manchester, was an outstanding, influential, and highly respected figure in the field of time-series analysis. So reads the opening sentence of his obituary (Subba Rao and Tunnicliffe Wilson, 2014) in this journal, which has recognized his extraordinary contribution as editor for over 30 years by inviting us to collate this special issue in his honor. Several of those contributing to this issue also spoke at a meeting organized in honor of Professor Priestley at Manchester University in December 2013, and we are pleased that others of his close associates from further afield, who were unable to attend that meeting, have also readily agreed to honor him by submitting papers for this issue. In his opening presentation at the conference on Applications of Time Series Analysis in Astronomy and Meteorology (Priestley, 1997), Professor Priestley said ‘…the subject (of time series) itself is a very complex one, being essentially a branch of statistics but with its own peculiar vocabulary and methodology. It is nevertheless a deeply fascinating subject involving a vast array of physical, mathematical and statistical ideas, and covering an immense field of applications ranging from neurophysiology to astrophysics. The difficulties involved in understanding the subject are due partly to the fact that it evolved from two quite distinct disciplines, namely, communications engineering and statistics.’ Professor Priestley was himself knowledgeable and expert in both of these disciplines and as editor of this journal welcomed contributions in which time-series was applied to diverse fields. With his students, he himself embarked on a program of research to extend time-series modeling beyond the context of linear stationary Gaussian processes, and many of the papers in this issue are similar in their aim. In their paper, Cardinale and Nason introduce a new theory of locally stationary wavelet packet processes for modeling time-series and illustrate the model with application to Standard and Poor 500 Index series. Models involving non-normal distributions are the subject of several papers. Harvey and Lange propose and study the application of the generalized error distribution for the EGARCH model with application to commodity and stock returns, while Robinson and Taylor investigate the application of a flexible semi-parametric score function model for the error distribution in the context of modeling a high-dimensional process and assess its efficiency for processes with a range of specific non-normal distributions. Jesus and Chandler propose using the Whittle likelihood for fitting models to processes such as rainfall event records that have highly skewed distributions. For valid inference, they propose estimating equations on the basis of the Whittle likelihood score, with a sandwich rule expression for the parameter covariance matrix. They illustrate their methodology using simulations of rainfall series driven by a Poisson process of events. Eichler et al. also model highly non-normal series of neural spike train data using the multi-variate Hawkes process excited by Poisson events. They derive graphical representations of the plausible causal dependence between the firing times of ten neurons. Wong et al. extend the nonlinear threshold model to threshold mixture models for copula functions describing the dependence between bivariate series innovations. Gao et al. model the dynamic dependence of binary panel data on previous outcomes and fixed covariates and demonstrate improvements in inference from imposing a flat prior on the random level effects. Their application is to female labor supply series. Chan et al. address the problem of very high-dimensional series by showing how dimension reduction can be achieved using a factor analysis model and demonstrate how this may be consistently estimated. They illustrate their approach and compare it with other methods, using daily returns of 123 stocks in the Standard and Poor 500. Spatio-temporal processes are the subject of two papers. Subba Rao and Terdik formulate a model that describes intrinsically stationary spatial dependence between the temporal discrete Fourier transforms at the different spatial locations using a parametrized variogram and show how this can be estimated. Bandyopadhyay et al. develop tests for second-order stationarity of spatio-temporal series by checking for correlation between spectral ordinates. Berentsen et al. examine local Gaussian correlation as a measure of nonlinear dependence with particular reference to the time-series context and models, such as ARCH and GARCH. Tunnicliffe Wilson applies spectral factorization to extend the classical spectral estimation of open-loop lagged response between time-series to the case of closed-loop feedback, illustrating the method using three series measuring the respiration of preterm infants. This issue illustrates the current diversity of models and fields of application of time-series analysis encompassed by this journal. The cross-fertilization of ideas and innovations this diversity encourages is a fitting memorial to Professor Priestley.