This course provides a Ph.D.-level introduction to econometric time-series analysis.
Book, slides, code, etc. here. The slides are the center of everything.
Format: Lectures that stress applied econometric theory.
Time Domain: The Wold Representation and its Approximation
Frequency Domain: The Spectrum and its Approximation
Markovian Structure, State Space, and the Kalman Filter
Fisher: Likelihood Evaluation and Optimization
Simulation Methods and Their Application
Bayes: Posterior Analysis and Markov Chain Monte Carlo
Integration, Cointegration, and Long Memory
Nonlinear/Non-Gaussian State Space
Conditional Variance Dynamics
Time-Series Econometrics, Big Data, and Machine Learning
Office hours (held in McNeil 519) here.
Teaching assistants will be heavily involved, including small-scale help by email and large-scale help in weekly review/supplementary sessions. (Office hours and review session times and locations, contact info, etc., to be announced.)
Grading: N problem sets (each 60/N %) and a final exam ("practice prelim") (40%). Good performance is crucially dependent on regular class preparation, attendance and participation.
Problem Set 1 (Due February 22.)
Use the monthly U.S. housing starts and completions data, reserving the last six observations for out-of-sample forecast comparisons. Discuss all results as appropriate. First graph the data. Then do the following. Model the two series jointly as a VAR(4); Calculate the autocorrelation and spectral density functions implied by your estimated VAR; Perform a Granger-causality analysis; Using Cholesky factor identification, calculate and graph the full set of impulse-response functions; Forecast the six hold-out observations and assess accuracy; using the full dataset forecast the sample path for the next twelve months.
Final exam date: Standard
Note well that modifications and adjustments to this syllabus / web site are inevitable and may be implemented at any time. Check frequently for updates.