Forecasting primer
Standalone explainers
for the concepts behind modern forecasting.
A growing set of search-friendly pages — one concept each, deeply explained, cross-linked into the chapters where they’re applied.
Foundational
What is stationarity, really? · ADF, KPSS, and friends
Autocorrelation function (ACF) vs partial ACF (PACF)
Why decompose a time series? · STL, MSTL, classical
The forecasting workflow · A repeatable mental model
Metrics & evaluation
MAPE, sMAPE, WAPE — when each is appropriate (and when each lies)
Quantile loss and the pinball score
Continuous Ranked Probability Score (CRPS)
Winkler interval score
Forecast Value Added (FVA) in plain English
Diebold-Mariano test and the Model Confidence Set
Walk-forward (rolling-origin) cross-validation, step by step
Purging and embargoing for dependent data
Feature engineering
Spectral entropy and frequency-domain features
Shape-based features (DTW, ts-fresh, catch22)
TS2Vec and TS-TCC representation learning
Permutation entropy and Lyapunov exponents
Models
ARIMA, in five paragraphs
The state-space framework for ETS
N-BEATS and N-HiTS, intuitively
Why a simple MLP often beats a transformer (and when it doesn't)
Foundation models for time series — Chronos, TimesFM, Moirai
Competitions & benchmarks
The M-competitions, in chronological order
M5 / M6 takeaways for practitioners
The GIFT-Eval leaderboard, explained