Projection techniques such as variants of Principal Components and Outlier Displaying Components are specifically known for application in single multivariate datasets. In this paper, extensions are made of these techniques to dataset that is multiple multivariate time-dependent (MMTD) in nature. The structure of this kind of data problem is appropriately characterized to show that a single observation is a random matrix of dimensions r multiplicities by p several variables. The procedure is a two-phased approach that identifies suspect extreme observations and then examines their extent of extremeness. The application illustrates the determination of markets with extreme agricultural food commodity prices that provides useful guide for reducing levels of extreme high prices.

projection techniques, market classification, multiple multivariate data, outlier displaying component