Pre-Midterm Review Exercise: Still in preparation (3/27/97, 11:45PM). This won't be available until Saturday or Sunday. Note that the Midterm exam is now scheduled to be handed out Thursday, 4/3, to be turned in 4/4.

If you are attending the course, please fill out the form for the class email list.

Take-Home Final Exam (MS Word Version) (This is exactly what was handed out on paper.)

Take-Home Final Exam Answer Sheet (My impression is that everyone is using the pdf files. Email to me if you need MS Word format.)

Answers to Review Problems (MS Word format) Posted 5/1/97

Text of review exercise (MS Word only -- just as handed out in class)

Exericise: Smoothing and Metropolis Sampling (MS Word Format) Due 4/8/97

Because getting complete series for 1948-present off Citibase via Statlab requires splicing, I've prepared a text file with the spliced data that you can download. If you click here you will see the text. If you right-click instead and select "Save link as", you should be able to save the data file. It contains quarterly data, 1948:1-1996:4, on: Total Unemployment, Ages 16+, SA; GDP deflator, SA; and real GDP, SA, arranged in three columns of 196 numbers.

Exercise: Bayesian Basics and Conditional vs. Unconditional Likelihood    (MS Word Format)   Due 1/23/97

This exercise has had two corrections made 1/21, 11:45PM: a formula in the first question, and an added N(0,1) assumption in the last. The versions accessible here are the correct ones.

There is also one more error, pointed out by a student 1/23: Problem 2 asks for a lower bound on the prior probability of the set of parameters for which the coverage probability is less than .90. It should have asked for an upper bound. The lower bound is actually easier, but trivial: it's zero.

On problem 4, to make the problem with unconditional likelihood tractable, you should integrate out the unknown parameter c. With rho fixed, the full likelihood matches that of a GLS problem, and is thus Gaussian in shape as a function of c and easily integrated. The problem is more interesting if you treat sigma as unknown (rather than known to be 1) also, but this makes the analytic integration slightly more complicated. After integration, the problem is reduced to one dimension, making it easily solvable by graphical methods. A direct approach treating c and rho (or worse, c, rho and sigma) jointly as unknown and maximizing numerically, is much more work.

Time Series

There will be assigned material from James Hamilton's Time Series textbook. Parts of Bayesian Data Analysis by Gelman, Carlin, Stern and Rubin will also be assigned. The Bayesian Choice by Christian Robert is a useful reference, especially for discussions of optimality theorems for Bayesian approaches. Practical Methods of Optimization by Fletcher surveys numerical optimization methods, some of which will be used in the course.

Bayesian Basics

Readings: Notes on Bayesian Probability Intervals and Classical Confidence Intervals(pdf format)(MS Word format)

Gelman, Carlin, Stern and Rubin, Chapters 1-3.

Time Series Basics

Readings: Hamilton, Chapters 1, 2, 5.

The reading list is still very incomplete. Last updated 1/25/97.