CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0585
Title: Power weighted densities for time series data Authors:  Shane Jensen - The Wharton School of the University of Pennsylvania (United States) [presenting]
Abstract: A crucial issue in modeling time series data is the possibility of non-stationarity in the underlying data generating process. We examine two time series applications in the presence of non-stationarity: predicting future hitting performance in major league baseball and predicting future returns of stock portfolios. We develop a simple and effective approach to allow non-stationarity in time series modeling when the primary goal is forecasting future time points, while also allowing the practitioner to choose the data model. In our power-weighted density (PWD) approach, observations in the distant past are down-weighted in the likelihood function relative to more recent observations under the practitioners chosen data model. Our PWD approach is a simpler alternative for allowing non-stationarity compared to popular state-space methods that explicitly model the evolution of an underlying state vector. We present specific PWD approaches for simple exponential families, hierarchical models and linear regression models, which are needed for our two applications. We demonstrate the benefits of our PWD approach in terms of predictive performance compared to both stationary models and alternative non-stationary methods such as state-space and integrated moving average models.