Skip to content

Conclusion & Next Steps

Congratulations on completing this introduction to time series analysis! You now have a solid framework for tackling time series problems:

  • You started by exploring and decomposing data to understand its core structure, identifying key components like trend and seasonality.
  • You learned how to prepare data for modeling by handling stationarity through differencing and engineering time-based features for machine learning models.
  • You built and compared different types of models, like ARMIA, SARIMA, AutoARIMA, and GARCH.
  • Finally, you learned how to rigorously select models using cross validation and evaluate your models using various metrics.

This end-to-end process gives you the ability to turn historical time-ordered data into valuable insights and predictions.

Next Steps for Your Learning Journey 🚀

Time series forecasting is a deep and fascinating field. This lesson provides a strong foundation, and here are some specific ways you can continue to build on it.

Apply These Techniques to a New Project

The best way to solidify your knowledge is to apply it. Find a time series dataset relevant to your work or interests—such as your company's sales data, website traffic, or the stock price of a company you follow—and perform the full end-to-end workflow we demonstrated in the case study. Challenge yourself to build both a statistical and a machine learning model and compare their backtested performance.

Deepen Your Knowledge of the Models

We covered the "what" and "why" of several models, but there is always more to learn about their inner workings.

  • Specific Suggestion: Read the user guide for the StatsForecast library or the statsmodels time series analysis section. Also, explore other popular forecasting models we didn't cover, such as Meta's Prophet model, which is known for its ease of use and ability to handle holidays.

Explore Advanced Evaluation Metrics

While RMSE and MAE are excellent general-purpose metrics, different business problems often require different ways of measuring forecast accuracy.

  • Specific Suggestion: Research and implement these two metrics:
    1. MASE (Mean Absolute Scaled Error): This is a great scale-independent metric that compares your model's forecast error to the error of a naive seasonal baseline. It helps answer the question, "Is my complex model actually better than a very simple one?"
    2. Quantile Loss: This metric is used when you care about the accuracy of your prediction intervals, not just a single point forecast. It is essential for problems where understanding the range of uncertainty is critical, such as in inventory management or financial risk assessment.