Fall 2011 course

• Introduction and examples. Autocovariance function.
         Example: Population trends
         Look at the U.S. Census Population Data. Use regression analysis to justify the use of a quadratic model.
         Find the best regression forecast for the U.S. population in years 2015, 2020, 2030.
• Hilbert spaces (this is the Mathematical background that will allow us to treat the best predictor as orthogonal projection and to obtain prediction equations)
• MA, AR, and ARMA processes; causal and invertible processes (most popular time series models; later we will prove that any stationary process is "approximately" ARMA)
• Spectral representation of a stationary process. Prediction in frequency domain
• Recursive computation of the best linear predictor and its mean squared error
  • Durbin-Levinson algorithm
  • Method of innovations
• Estimation and model selection. Goodness-of-fit issues
• Nonstationary time series, ARIMA models and extentions

Prerequisite: STAT 6331 (Statistical Inference) or equivalent Grading: Homework 10% Homework will be assigned weekly, collected, and graded. Solutions will be provided. A serious and steady effort to work out all the homework problems is a key to success in this course. Quizzes 20% Short weekly quizzes closely related to the material of the latest homework assignment. Quizzes are closed-book. One cheat sheet per quiz is allowed. Midterm exam 30% A 1 1/4-hour midterm covers chapters 1-4 of [BD]. Final exam 40% A 2-hour final covers chapters 5, 7-11 of [BD], but it is cumulative indirectly. Exams are open-book, -notes, etc. Any questions?     Write: mbaron@utdallas.edu      Call: (972)-UTD-6874      Click: www.utdallas.edu/~mbaron/