||Conventional time series methods are developed for analyzing point-valued data. However, in practice there are many interval-valued time series data, which usually contain more information than point-valued data. It is thus important to develop time series modeling and forecasting techniques for interval-valued data.|
In this paper, we introduce concepts of interval stationarity and related interval statistics and investigate methodology for interval time series analysis. We use vector autoregressive (VAR) and vector error correction (VEC) models to build time series models for interval statistics including : medium, radius, upper and lower bounds, and obtain interval forecasts. We compare the forecast performance of the proposed methods with classical filtering technique :
the exponential smoothing method, and nonparametric technique : the k-Nearest
Neighbors (k-NN) algorithm. Finally, in the empirical study, we use stock and index data to evaluate the forecast performance and efficiency of the proposed interval time series models.