Time Series Models using Python

Teaching notebooks prepared as part of a computational finance course for undergraduate and graduate students. The material walks through three families of time series models, from univariate forecasting to volatility modeling and multivariate macro forecasting.

Notebooks

Notebook	Topic	Learning objectives
Time_series_analysis_1.ipynb	Univariate ARIMA models	Decomposition, stationarity and the ADF test, differencing and log transforms, ACF/PACF correlograms, ARMA/ARIMA/SARIMAX, order selection with AIC/BIC and rolling out-of-sample RMSE
Time_series_analysis_2.ipynb	ARCH/GARCH volatility forecasting	Heteroskedasticity, volatility clustering, order selection via rolling out-of-sample forecasts, fitting a GARCH model and diagnosing residuals
Time_series_analysis_3.ipynb	VAR / VARMAX macro forecasting	Unit root testing across multiple series, VAR(p) and VARMAX estimation, residual diagnostics, impulse response analysis

Data

All data is pulled from the FRED database via pandas-datareader. The series used are:

• IPGMFN – Industrial production, manufacturing
• NASDAQCOM – NASDAQ Composite index
• UMCSENT – University of Michigan consumer sentiment

The shared load_fred helper in utils.py caches each pull to data/*.csv on first run, so the notebooks reproduce the same results offline afterwards.

Setup

Using pip:

python -m venv .venv
source .venv/bin/activate      # on Windows: .venv\Scripts\activate
pip install -r requirements.txt

Or using conda:

conda env create -f environment.yml
conda activate time-series-models

The arch package (used in notebook 2) is included in both files, so no separate install step is required.

Running

jupyter notebook

Then open any of the three notebooks and run the cells top to bottom. The first run fetches data from FRED (network required); later runs use the cached CSVs in data/.

Reproducibility

• A global random seed is set in utils.py (SEED = 42).
• Data is cached to CSV so results do not change between runs.
• Notebook outputs are stripped from version control. Install the git filter once with nbstripout --install so committed notebooks stay clean.

License

Released under the MIT License.