time-series-analysis

Repository for self study on Jonathan, D. Cryer, and Chan Kung-Sik. "Time series analysis with applications in R." SpringerLink, Springer eBooks (2008).

Exercises are conducted on both R and Python for language practice purposes.

Notable packages used

• R: TSA (functions mostly extracted / used as reference), ggplot, zoo, tseries
• Python: matplotlib, statsmodels, scipy, arch

Implementation notes

• Regression diagnostic graphs, in the same style as R, are implemented in Python/utils.py.
• eacf, from the TSA library, is reimplemented in Python in Python/eacf.py. Note that it uses statsmodels' ACF, rather than R, which may lead to small numerical computation differences.
• armasubsets, from the TSA library, is reimplemented in Python in Python/armasubsets.py. It uses its own subset search code, rather than relying on R's regsubsets library.
• gBox, the generalized portmanteu test from the TSA library, is reimplemented in Python in Python/gBox.py. It expects a fitted model from Python's arch library, and it uses Numpy for linear filters and linear algebra calculations, statsmodels for ACF, and matplotlib for plotting.
• Spectral density utilities from R are partially reimplemented in Python/spectrum.py, providing support for tapering, convolution with arbitrary kernels, and plotting, as adapted versions of R's state::spec.pgram and state::spec.ar.
• Self-Exciting Threhold AutoRegression models, with 2 regions, are reimplemented in Python/tar.py. The OLS solver for statsmodels is used for the lower and upper regimes, but the general method signatures and return parameters are adapted from R TSA library, rather than shaped as statsmodels's regression objects.
• The corresponding suite of nonlinearity tests (Keenan, Tsay, and threshold detection) are implemented in Python/nonlinearity_tests.py.