Welcome to the Time Series Forecasting course at ITESO. Here you will find all the materials and resources needed for the course.

1 Time Series Forecasting with R

Modern forecasting workflows using tidy data principles with the tidyverts ecosystem.

1.1 What you’ll learn

Build tidy time series with tsibble and modern data pipelines
Create forecasts with ETS, ARIMA, and regression-based models via fable
Engineer time features and diagnose models with feasts)
Evaluate with time-aware resampling and communicate uncertainty
Scale to multiple series and hierarchies when needed

How to succeed here

Forecasting improves with iteration: start simple, validate honestly, and refine. Keep a tight loop of fit → diagnose → evaluate → communicate.

1.2 Start here

Note

New to the site? Use the links below to jump in.

1.3 Course map

1.3.1 1. Foundations

tidy data + time indexes
tsibble keys, gaps, and intervals

1.3.2 2. Patterns & features

seasonality, trend, decomposition
diagnostics and features (feasts)

1.3.3 3. Core forecasting models

ETS + ARIMA
model comparison and selection

1.3.4 4. Regression workflows

time features, events, and covariates
forecast reconciliation ideas

1.3.5 5. Evaluation & communication

time series CV and backtesting
uncertainty, intervals, and narratives

1.3.6 6. Multiple series

grouped models
hierarchies and reconciliation

1.4 Quick setup check

Run this once after installing packages to confirm your environment.

Code

library(tidyverse)
library(tsibble)
library(fable)
library(feasts)

sessionInfo()

Common pitfalls

Data leakage: don’t let future information sneak into features.
Random CV: use time-aware resampling/backtesting.
Overfitting: a more complex model isn’t automatically better.

1.5 Primary toolchain

We’ll use a tidy workflow centered on:

tidyverse for data wrangling
tsibble for time-aware tibbles
fable for modeling and forecasting
feasts for decomposition, features, and diagnostics

Recommended reading: Forecasting: Principles and Practice (free online).