Time Series Forecasting with Fractional Gaussian Noise and Fractional ARIMA

By: Dr.Tim Lortz

Date/Time: November 15, 2012, 6PM-7:30PM

Place: 308 Sentinel Dr. Annapolis Junction, MD 20701

Room: Conference Room CR1004A

Abstract:

Conventional approaches for modeling traffic load volumes on communication networks include deterministic, Monte Carlo and Markovian methods. These methods are popular for their simplicity, flexibility and ease of implementation. They are occasionally augmented by short-term memory processes such as ARMA (Auto-Regressive Moving Average) in order to capture lag and seasonality. However, for many processes, none of these methods are able to capture the long-term dependence inherent in the traffic volumes. They also tend to underestimate network traffic variability. We present the idea of self-similar processes as a means of compactly expressing long-term dependence and variability in time series modeling. We discuss two families of self-similar models – Fractional Gaussian Noise and Fractional ARIMA (Auto-Regressive Integrated Moving Average) – and give examples for selecting, fitting and implementing these models. We also show how self-similar models are relevant to real-world problems and discuss recent innovation in the field.

Dr. Tim Lortz has been with Booz Allen Hamilton’s Analytics team since 2008. In that time, he has primarily supported government clients in Central Maryland, leading volumetric dataflow modeling and simulation projects. He is also engaged in research in stochastic modeling and in decision support analytics. He received Ph.D. and M.S. degrees in Industrial and Operations Engineering from the University of Michigan and a B.S. in Industrial Engineering from the University of Pittsburgh. His dissertation focused on the solvability in infinite horizon non-stationary optimization problems. He is a member of INFORMS and MORS. Tim and his wife, Kiersten, have three daughters and a son and reside in Laurel, Maryland.

If you plan on attending, please

RSVP by October 8, 2012 to Brian Sacash via email: Sacash_Brian@bah.com