Forecasting & Time Series Analysis

 Research themes include:    Investigations of core foundations and conceptual underpinnings of
dynamic modelling and model structures,  sequential statistical analysis, learning in time-varying contexts, forecasting and prediction with multiple goals and utilities, and relevant areas of Bayesian methodology and statistical computation, for time series of all kinds.  Duke Statistical Science has been a leading center for Bayesian forecasting, time series modelling and analysis for many years, and current topical research areas of emphasis continue to grow and enhance that tradition. 
Research in the foundations and conceptual underpinnings of probabilistic forecasting (a.k.a. prediction) are increasingly paramount in the face of escalation of the scales of dynamic systems, data sets and relevant models; this links to major stimuli to new modelling developments for increasingly large and complex dynamic data sets, and to the need for innovation in both theoretical and applied areas of emerging dynamic decision analysis linked to many kinds of coupled forecasting+decision problems.   Application areas of current and emerging focus in Duke Statistical Science in these areas range from core natural sciences (e.g., problems of modelling, 
inference and inherently predictive concern in complex molecular and physiological systems); 
to many areas of the sciences and social sciences involving time series and forecasting in complex 
dynamic network systems;  in increasingly massive-data rich systems in many fields; in finance, micro- and macro-economics and econometrics linked to both personal and societal/econometric forecasting and policy-related issues;  in a wide-ranging variety of traditional and emerging big-data commercial and industrial/IT contexts; and in many other critical areas of both institutional and personal forecasting and decision problems.

Faculty in this Research Area