ARIMA Models in Python
Learn to test time series for stationarity, fit ARIMA and SARIMA models in Python with statsmodels, and choose and check forecasting models.
ARIMA models are statistical tools for making sense of past values in a time series and predicting the ones still to come. You check whether a series is stationary, transform it when it is not, and fit autoregressive and moving average models in Python with statsmodels. Much of the work goes into choosing between candidate models with ACF and PACF plots, AIC and BIC, then handling data with seasonal patterns.
What You Will Achieve
- Check a time series for stationarity by plotting it and running a statistical test, then difference or transform it when it fails
- Fit ARMA, ARIMA and ARMAX models in Python with statsmodels and produce one-step-ahead and dynamic forecasts
- Pick promising model orders from ACF and PACF plots and compare fitted candidates using AIC, BIC and diagnostic plots
- Split a series into seasonal and non-seasonal parts and forecast seasonal data with a SARIMA model
Program Overview
Stationarity and ARMA Model Structure
Introduces the properties of time series that matter for modeling and the basic building blocks of ARMA models.
- What stationarity means and why ARMA models need it
- Visual and statistical stationarity checks, with train-test splits
- Differencing and other transforms
- AR, MA and ARMA structure, model order and simulated data
Fitting, Forecasting, and Model Selection
Covers fitting models in statsmodels, forecasting stock prices with them, and narrowing a set of fitted candidates down to the best one.
- AR, MA, ARMA, ARMAX and ARIMA model fitting
- One-step-ahead versus dynamic forecasts and their plots
- ACF, PACF, AIC and BIC for order selection, tried on earthquake data
- Mean absolute error, diagnostic statistics and the Box-Jenkins method
Seasonal Data and SARIMA Models
Extends the ARIMA toolkit to data with seasonal patterns and closes with a global forecast challenge.
- Seasonal decomposition and seasonal ACF and PACF
- Choosing SARIMA order and comparing SARIMA with ARIMA forecasts
- Automated model selection, saving and updating models
- Multiplicative and additive seasonality and SARIMA diagnostics
Certificate Awarded (Paid)
Job Outlook
These skills belong to data science and analytics work that studies past values in a series and forecasts later ones, such as the stock prices, earthquake records and seasonal data used in the exercises.
Good for: Data practitioners working in Python who need to forecast series such as prices or seasonal measurements with a statistical model they can inspect and test.
Keep in mind: Every chapter stays with the ARIMA model family fitted in statsmodels and the Box-Jenkins method, so it goes deep on one approach to forecasting.
Bottom line: A tightly scoped run through the ARIMA workflow in Python, giving model choice, diagnostics and seasonality as much attention as fitting.
Frequently Asked Questions
What is an ARIMA model?
ARIMA stands for autoregressive integrated moving average. It is a model used in statistics and econometrics to understand past values in a series and predict later ones. In the lessons, an ARIMA model is built by differencing the data and fitting an ARMA model to it.
Which datasets appear in the exercises?
The forecasting chapter uses stock prices, a lesson on choosing model order works with earthquake data, and the seasonal chapter ends with a global forecast challenge that draws on all the ARIMA tools taught earlier.
What is the Box-Jenkins method in this course?
It is presented as a framework for structuring time series projects, worked through in separate lessons on identification, estimation and diagnostics. The seasonal chapter returns to it to show how the same steps apply to SARIMA models.
Can model order be searched automatically?
Yes. One lesson searches over model orders and ranks them with AIC and BIC, and the seasonal chapter adds lessons on automated model selection and on saving fitted models and updating them later.


There are no reviews yet.