A0567
Title: A framework for time series aggregation and seasonality using marked point processes
Authors: Tucker McElroy - Census Bureau (United States) [presenting]
Anindya Roy - U.S. Census Bureau (United States)
Abstract: Stochastic processes with a meager sampling rate are studied, wherein the number of observations per epoch or region is small. The goal is to construct a model whereby temporal or vertical aggregation of such meager processes results in a data stream that more closely resembles a Gaussian process, such as is commonly used to model higher-aggregate economic data. The approach involves Marked Point Processes, where each point corresponds to a single transaction in the marketplace, and a given measure that governs the distribution of points determines the mean and covariance structure of the observed process. We discuss applications to sampling, warping through a change of measure, and modeling of data with trend and seasonality.
