Videos and books

Every resource on this page is free to watch or read, from the author's, the university's or the publisher's own page. Each lesson of the track also ends with a Watch and read section that names the exact video, chapter or paper.

How to use it. Watch a short video for the picture, read the matching chapter of Forecasting: Principles and Practice (FPP3) for the method, and run the lesson's notebook to check both on data. A good default: one short video before a lesson, then the lesson and its notebook, then the reading from the table at the end of this page.

A note on notation. This track follows FPP3 for most symbols: $\hat y_{t+h\mid t}$ is the forecast of $y_{t+h}$ made at origin $t$, $m$ is the seasonal period, and the MA polynomial is $\theta(z) = 1 + \theta_1 z + \dots + \theta_q z^q$ (with a plus sign). Some books and videos, for example Box and Jenkins and parts of the ritvikmath series, write the MA part with a minus sign; then every $\theta$ changes sign. For state-space models the track uses $F, H, Q, R$, as in Särkkä; Durbin and Koopman and statsmodels write $T, Z$ for $F, H$ (see lesson 07).

Video courses #

Start here #

• OTexts, Forecasting: Principles & Practice 3rd ed (YouTube playlist). Short videos by the book's authors, one per section of FPP3: graphics, decomposition, the forecaster's toolbox, exponential smoothing, ARIMA and more. Watch the matching video before lessons 01, 03, 04 and 06.
• Aric LaBarr, Time Series (YouTube playlist). Five-minute explanations without formulas: decomposition, stationarity, exponential smoothing, ARIMA, and more. Good before lessons 01–06 when a topic feels new.

Classical time-series analysis #

• ritvikmath, Time Series Analysis (YouTube playlist). The "Time Series Talk" videos, 5–15 minutes each, at a whiteboard: stationarity, ACF and PACF, AR, MA, ARMA, ARIMA, SARIMA, unit roots, VAR, GARCH, backtesting and anomaly detection. Pairs with lessons 02, 03, 05, 06 and 11.
• MIT 18.S096, Topics in Mathematics with Applications in Finance (2013) (course page, YouTube playlist). Lectures 8, 11 and 12 (Time Series Analysis I–III) cover stationarity, ARMA, VAR and cointegration with proofs. Pairs with lessons 02 and 05.
• DTU 02417, Time Series Analysis (Lasse Engbo Christiansen) (YouTube playlist). A full university course: linear stochastic models, ARMA identification, prediction, and state-space models with the Kalman filter. Pairs with lessons 05–07.

State-space models and the Kalman filter #

• Steve Brunton, Control Bootcamp (YouTube playlist). Observability, the Kalman filter and LQG control in short board lectures. The Kalman videos pair with lesson 07.

Machine learning and deep learning for forecasting #

Uncertainty and anomalies #

Free books #

• Rob J. Hyndman and George Athanasopoulos, Forecasting: Principles and Practice, 3rd ed. (online). The main reference of this track. Clear, practical, with R code (the fable packages). Chapters 2–5 (graphics, decomposition, features, the toolbox), 8 (exponential smoothing), 9 (ARIMA), 10 (dynamic regression) and 11 (hierarchical forecasting) match lessons 01–08.
• Hyndman, Athanasopoulos, Garza, Challu, Mergenthaler and Olivares, Forecasting: Principles and Practice, the Pythonic Way (online). The same book with Python code (Nixtla's statsforecast, mlforecast and neuralforecast), plus chapters on neural networks and foundation models. Use it with lessons 08 and 09, and whenever you want the FPP3 examples in Python.
• Penn State STAT 510, Applied Time Series Analysis (online notes). Course notes on ACF, PACF, AR, MA, ARIMA, seasonal models, and diagnostics, with worked examples. Lessons 02, 05 and 06.
• Robert Nau, Statistical forecasting: notes on regression and time series analysis (Duke University, online). Practical notes on random walks, smoothing and ARIMA, with rules for choosing orders. Lessons 03–06.
• Simo Särkkä, Bayesian Filtering and Smoothing (Cambridge University Press 2013, free PDF from the author). Chapter 4 derives the Kalman filter as Bayesian filtering, chapter 8 the RTS smoother, and later chapters cover the EKF, UKF and particle filters. Lesson 07.
• Roger Labbe, Kalman and Bayesian Filters in Python (GitHub). A free book as Jupyter notebooks: g-h filters, the one-dimensional and multivariate Kalman filter, and nonlinear filters, built up with code and plots. Lesson 07.
• Anastasios Angelopoulos and Stephen Bates, A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification (arXiv). A short book on split conformal prediction, conformalized quantile regression and conformal methods under distribution shift, with code. Lesson 10.
• Petropoulos et al., Forecasting: theory and practice (International Journal of Forecasting 2022, arXiv). An encyclopedia of forecasting written by about 80 authors: short sections on almost every method and application, each with references. Use it to look up a topic after the track.
• Zhang, Lipton, Li and Smola, Dive into Deep Learning (chapter on recurrent neural networks). Sequence models, RNNs, LSTM and GRU with runnable code. Background for lesson 09.

Short reads #

• Hyndman and Koehler (2006), "Another look at measures of forecast accuracy" (author PDF). Why MAPE fails and how MASE fixes it. Lesson 03.
• Bergmeir, Hyndman and Koo (2018), "A note on the validity of cross-validation for evaluating autoregressive time series prediction" (author PDF). When ordinary K-fold cross-validation is valid for time series, and when it is not. Lessons 03 and 08.
• scikit-learn, "Time-related feature engineering" (example). One-hot, spline and cyclic encodings of calendar features with gradient boosting and linear models, step by step. Lesson 08.
• Zeng, Chen, Zhang and Xu (2023), "Are transformers effective for time series forecasting?" (arXiv). The DLinear paper: a linear layer beats the long-horizon transformers of its time. Read it with section 3.4 of lesson 09.
• Benidis et al. (2022), "Deep learning for time series forecasting: tutorial and literature survey" (arXiv). A survey of neural forecasting models with a common notation. Lesson 09.
• Gneiting and Raftery (2007), "Strictly proper scoring rules, prediction, and estimation" (author PDF). The theory of proper scores, including the CRPS and the quantile score. Lesson 10.
• Greg Welch and Gary Bishop, "An Introduction to the Kalman Filter" (PDF). A short, classic introduction to the discrete Kalman filter and the EKF. Lesson 07.
• Wu and Keogh (2021), "Current time series anomaly detection benchmarks are flawed and are creating the illusion of progress" (arXiv). Why many benchmark results mean little, and what a good benchmark needs. Lesson 11.
• Rob J. Hyndman, Hyndsight blog (online). Short posts on forecasting practice, evaluation and common errors, by the first author of FPP3.

Software #

• statsmodels (time-series analysis): ACF and PACF, unit-root tests, ARIMA and SARIMAX, ETS, STL and state-space models in Python. Lessons 01–07.
• Nixtla StatsForecast (documentation): fast ETS, ARIMA and baseline models for many series. Lessons 03–06.
• Darts (documentation), sktime (website) and GluonTS (documentation): libraries that put classical, ML and deep forecasters behind one interface, with backtesting. Lessons 08–10.
• Chronos (GitHub): pretrained time-series foundation models, and the GIFT-Eval leaderboard (Hugging Face) to compare such models. Lesson 09.
• MAPIE (documentation): conformal prediction for scikit-learn models, including methods for time series. Lesson 10.
• ruptures (documentation) for offline change points (binary segmentation, PELT) and STUMPY (documentation) for the matrix profile. Lesson 11.

Which lesson pairs with what #

The first choice for each lesson. The Watch and read section at the end of each lesson gives more, with exact section numbers.