Chapter 26 Further Reading: Business Forecasting and Trend Analysis

Books

Forecasting: Principles and Practice (3rd Edition)

Authors: Rob J. Hyndman and George Athanasopoulos URL: https://otexts.com/fpp3/ — Free online edition

This is the definitive textbook on modern forecasting methods, and the entire third edition is freely available online. It covers everything in this chapter and much more: exponential smoothing, ARIMA models, seasonal decomposition, cross-validation of forecasts, and regression-based approaches. The textbook uses R for examples, but the concepts translate directly to Python — and a companion Python package (statsforecast) implements many of the same methods. If you want to go deeper on any forecasting topic, this is the first place to look. Chapters 3–8 are most relevant to the material covered in Chapter 26 of this book.

Superforecasting: The Art and Science of Prediction

Authors: Philip E. Tetlock and Dan Gardner Published: 2015, Crown Publishers

This book is not a Python programming book — it is about what makes human forecasters good or bad, based on two decades of research on predictive accuracy. The central insight is that forecasting skill is learnable and that good forecasters share specific habits: breaking problems into components, updating beliefs when new data arrives, and expressing uncertainty in calibrated probabilities rather than vague terms. Highly relevant to the communication side of Chapter 26 — particularly the sections on how to present forecasts to decision-makers who want false certainty.

The Signal and the Noise: Why So Many Predictions Fail — But Some Don't

Author: Nate Silver Published: 2012, Penguin Press

Nate Silver, the statistician behind FiveThirtyEight, examines forecasting across domains: elections, weather, baseball, economics, poker. The chapter on economic forecasting is directly relevant to business analysts — it examines why official economic forecasts are systematically overconfident and what structural improvements would help. The baseball chapters explain clearly why regression to the mean matters and how to think about variance. More accessible than a statistics textbook while being substantively rigorous.

Python Libraries Documentation

pandas Time Series Documentation

URL: https://pandas.pydata.org/docs/user_guide/timeseries.html

The pandas user guide section on time series covers date range generation, period arithmetic, resampling, time zone handling, and rolling windows in depth. The section on DateOffset objects is particularly useful for building forecasts with calendar-aligned periods (monthly, quarterly). The examples on resample() show how to aggregate daily data to monthly or quarterly totals — a common preprocessing step before forecasting.

statsmodels Time Series Analysis Documentation

URL: https://www.statsmodels.org/stable/tsa.html

The statsmodels tsa (time series analysis) module covers everything from simple exponential smoothing (used in this chapter) through Holt-Winters seasonal models, ARIMA, and VAR models. The documentation includes worked examples and interpretation guides. The "State Space Models" section is where you would go if you need to model complex seasonal patterns or multiple time series simultaneously. The ExponentialSmoothing class documentation is the direct extension of the Holt's method covered in this chapter — adding seasonal decomposition.

scipy.stats Documentation: Statistical Functions

URL: https://docs.scipy.org/doc/scipy/reference/stats.html

The scipy.stats module contains linregress (used extensively in this chapter), plus many other statistical tests and distributions. The "Regression" section documents linregress parameters and return values in detail, including the standard error of the slope, which is useful for more rigorous confidence intervals. The "Continuous distributions" section explains the normal distribution functions used to compute confidence interval z-scores.

matplotlib: Visualization

URL: https://matplotlib.org/stable/gallery/

The matplotlib gallery shows code examples for hundreds of chart types. For forecasting, the most relevant examples are in the "Statistics" and "Specialty Plots" sections: fill_between (for confidence bands), axvline (for the historical/forecast boundary), errorbar (for point estimates with margins), and date-formatted axes. The "Axes and Figures" section explains how to create multi-panel figures for displaying multiple regional forecasts side by side.

Academic Articles and Resources

"On the Dangers of Overconfident Forecasting" — CFA Institute

URL: https://www.cfainstitute.org/en/research/foundation/2020/forecasting-in-economics-and-finance

This overview from the CFA Institute discusses systematic biases in financial and economic forecasting, including overconfidence, anchoring, and narrative fallacies. Directly applicable to business forecasting contexts where decision-makers pressure analysts for falsely precise numbers.

"Evaluating Time Series Forecasting Models" — Machine Learning Mastery

URL: https://machinelearningmastery.com/time-series-forecasting-performance-measures-with-python/

Practical guide to forecast evaluation metrics: MAE (Mean Absolute Error), RMSE (Root Mean Squared Error), MAPE (Mean Absolute Percentage Error), and MASE (Mean Absolute Scaled Error). Includes Python code examples. Relevant for the Chapter 26 exercises on walk-forward validation and comparing forecast methods.

George E. P. Box — "Science and Statistics" (1976)

The paper that contains the famous quote "All models are wrong, but some are useful." Available through JSTOR and Google Scholar. The full context of the quote is more nuanced and illuminating than the quote alone. Box argues that the goal of statistical modeling is not truth-finding but practical utility — a framing that applies directly to business forecasting.

Interactive Learning Resources

Statsmodels Examples — Time Series Analysis

URL: https://www.statsmodels.org/stable/examples/index.html#time-series-analysis

The statsmodels examples page includes Jupyter notebooks demonstrating Holt-Winters exponential smoothing, ARIMA model selection, and seasonal decomposition. The "Exponential Smoothing" notebook covers the same methods as this chapter with additional plots and parameter sensitivity analysis.

Towards Data Science — Forecasting Articles

URL: https://towardsdatascience.com/tagged/forecasting

Towards Data Science publishes accessible, code-heavy articles on forecasting topics. Particularly useful articles to search for: "Time series decomposition with Python," "Holt-Winters forecasting in Python," and "Walk-forward validation for time series." The quality varies by author, but the most popular articles (high clap counts) are generally technically sound.

Python Forecasting Libraries Beyond This Chapter

Prophet (Meta)

URL: https://facebook.github.io/prophet/ Install: pip install prophet

Prophet is a forecasting library developed by Meta (Facebook) that handles trend + seasonality + holidays automatically. It requires minimal statistical knowledge to use and is particularly good at handling missing data and outliers. The primary limitation for this book's audience is that it is harder to explain "what the model is doing" compared to linear regression or Holt's method. Use it when you need production forecasting with minimal tuning, and use the simpler methods in this chapter when you need to explain your methodology to a business audience.

Darts

URL: https://unit8co.github.io/darts/ Install: pip install darts

Darts is a comprehensive time series library that provides a unified interface to dozens of forecasting methods: exponential smoothing, ARIMA, LSTM, Transformer-based models, and more. It includes built-in backtesting, cross-validation, and ensemble methods. More appropriate for data scientists building production forecasting systems than for the business analyst audience of this chapter, but useful to know about as the next step.

sktime

URL: https://www.sktime.net/ Install: pip install sktime

sktime provides scikit-learn compatible interfaces to time series forecasting, classification, and regression. Its unified API makes it easy to try many forecasting methods and compare their performance systematically. Particularly useful when building the walk-forward validation exercises in Chapter 26.

Recommended Learning Path

For building practical business forecasting skills (this chapter's level): 1. Complete all Chapter 26 exercises through Tier 3 2. Read Chapters 3–5 of Forecasting: Principles and Practice (free online) 3. Apply the forecasting pipeline to your own business data

For moving toward data science-level forecasting: 1. Complete Forecasting: Principles and Practice through Chapter 8 2. Learn ARIMA models (Chapter 9 of FPP3) 3. Explore Prophet for production forecasting 4. Study walk-forward backtesting with scikit-learn or sktime

For understanding forecasting communication and decision-making: 1. Read Superforecasting by Tetlock 2. Study the "communicating uncertainty" chapter in FPP3 3. Practice writing one-page forecast summaries with explicit assumptions and limitations

All URLs were verified as active resources. Book publishers may update URLs; search by author and title if a link redirects.