Distributional time series for forecasting and risk assessment

Abstract

Important computational aspects of big data processing and forecasting methods for the problems of the risk assessment are under consideration. A new approach to the study and forecasting of big data represented by time series is discussed. Our approach is based on Big Data technologies, including data aggregation procedures for input and output parameters and computational probabilistic analysis. The result of this approach is a new type of representation of a big time series in the form of distributional time series. Piecewise polynomial models are used for data aggregation procedures. To solve computational problems on distributed time series, we developed arithmetic over piecewise polynomial functions. To demonstrate our approach, we studied the problem of risk assessment for investment projects.