Part V. Prediction and Machine Learning

Applied Econometrics for Natural Resource Economics

Five light Quarto lecture notes introducing prediction, train-test splits, machine learning models, model comparison, feature importance, and the limits of prediction for applied econometrics students.

This part introduces prediction and machine learning for students of applied econometrics. The purpose is not to turn the course into a data science textbook. The purpose is to help students understand how prediction differs from explanation, how models are evaluated on unseen data, and why machine learning results must be interpreted carefully in economics.

The central message is simple:

Main message

Better prediction does not automatically mean better economic explanation.

Chapters in this part

Chapter	Title	Main focus
22	Train-Test Split and Prediction Logic	Evaluating prediction on unseen data
23	Regression versus Machine Learning	Explanation versus prediction
24	Decision Trees and Random Forests	Nonlinear prediction using tree-based models
25	XGBoost and Model Comparison	Boosting and test-sample model comparison
26	Feature Importance and Prediction Limits	What ML can and cannot tell us

Suggested workflow

Students should read these chapters after learning regression, hypothesis testing, functional forms, dummy variables, and model diagnostics. The concepts build directly on earlier econometric tools, but the evaluation logic changes. Instead of asking only whether a coefficient is statistically significant, students now ask whether a model predicts well on new observations.

Learning outcomes

After completing Part V, students should be able to:

split a dataset into training and testing samples;
fit simple predictive models in Python;
evaluate predictions using MAE, MSE, RMSE, and test-sample (R^2);
compare regression, decision trees, Random Forests, and XGBoost;
explain why prediction accuracy is not the same as causality;
interpret feature importance cautiously.

Files included

This package includes five Quarto chapter files, one Part V landing page, SVG figures, a navigation style file, and a _quarto-part-v.yml snippet that can be copied into the main course website configuration.