Nine case studies thread through Chapters 6-20, applying the same ML4T workflow to different asset classes, frequencies, and trading constraints. Each case study defines a universe, builds labels and features, trains models from linear baselines through deep learning, and evaluates strategies through backtesting, portfolio construction, cost analysis, and risk management.

Overview

#	Case Study	Asset Class	Frequency	Universe	Primary Label
1	ETFs	Multi-asset ETFs	Daily	100 ETFs	fwd_ret_21d
2	Crypto Perps Funding	Crypto perpetual futures	8-hourly	19 pairs	fwd_ret_8h
3	NASDAQ-100 Microstructure	US equities (intraday)	15-min	114 stocks	fwd_ret_15m
4	S&P 500 Equity + Options	S&P 500 equities	Daily	634 stocks	fwd_ret_5d
5	US Firm Characteristics	US equities (fundamental)	Monthly	~2,500 stocks	fwd_ret_1m
6	FX Pairs	G10 currency pairs	Daily	20 pairs	fwd_ret_1d
7	CME Futures	Multi-sector futures	Daily	30 products	fwd_ret_5d
8	S&P 500 Options	S&P 500 equity options	Daily	S&P 500 straddles	fwd_ret_dh_10d
9	US Equities Panel	Broad US equities	Daily	~3,200 stocks	fwd_ret_1d

Pipeline Stages

Every case study runs the same sequence of phases, and each phase maps to a book chapter. The stage numbers differ from one case study to the next, because each market gets a different set of model-family stages, but the phase order is identical everywhere. The phase-to-chapter table lives in Running Notebooks.

Not every case study has every model type. The exact notebook set depends on the dataset characteristics; each case study's own README lists its stages, in order, with the chapter each one belongs to.

Running a Complete Pipeline

Run the stages in order from the repo root. See Running a Case Study End to End for the pattern, and the case study's own README for its exact stage list. Reduced-parameter test runs and headless execution are covered in the same document.

Directory Layout

Each case study follows this structure:

case_studies/{id}/
+-- README.md                  # Dataset profile, pipeline table
+-- config/
|   +-- setup.yaml             # SSOT: universe, costs, CV, labels
+-- 01_feasibility_analysis.py / .ipynb       # Numbered notebook sequence
+-- 02_labels.py / .ipynb
+-- ...
+-- labels/                    # Generated learning targets (gitignored)
+-- features/                  # Generated financial and temporal features
+-- evaluation/                # Generated feature diagnostics
+-- run_log/                   # Downloaded or generated result state
|   +-- registry.db            # Result and provenance source of truth
|   +-- training/{hash}/       # Specs, coefficients, boosters, checkpoints
|   +-- predictions/{hash}/    # Stored prediction arrays
|   +-- backtest/{hash}/       # Returns, trades, weights, and configs

Reproducibility

config/setup.yaml defines the trading setup, cost model, and evaluation protocol
run_log/ implements the Chapter 6.7 run log: every model run is content-addressed by its config hash
run_log/registry.db is the only result source of truth; legacy result JSON files are not used
scripts/download_artifacts.py installs the accepted run log and stored artifacts without retraining
scripts/create_experiment.py creates a writable copy for new configurations and backtests

For the schema and querying API, see RUN_LOG.md. For reproduction and experimentation steps, see Running Notebooks.