This playbook is a standard operating procedure for building, validating, and delivering quantitative models in finance.

It is written for people who want to be credible in front of Trading, Risk, IT, and Model Validation.

You will learn how to:

  • frame modelling work as a decision problem,
  • build time-consistent datasets and baselines,
  • evaluate models under regime shifts and constraints,
  • ship reproducible artefacts and documentation,
  • operate under model risk controls,
  • control the room with client-ready stories.

Start with Module 1 if you want the end-to-end workflow. Start with Module 3 if you want the “quant engineer” differentiator.

Modules

Module 1 — Default modelling workflow

A full end-to-end modelling workflow you can explain in 90 seconds.

  • Problem framing: target, horizon, decision type, success metric, constraints.
  • Data definition: universe, splits, leakage checklist.
  • Baselines and iteration: naive baselines, v0 features, simplicity-first selection.
  • Evaluation discipline: time-series CV, stability checks, error analysis.
  • Deliverables: code, config, metrics, plots, assumptions, limitations.

Module 2 — Time series core

High-probability material when time series and econometrics are mentioned.

  • Data transformations: returns, differencing, de-trending, train-only scaling.
  • Dependence and structure: ACF/PACF intuition, seasonality, volatility clustering.
  • Models and when to use them: ARIMA intuition, Kalman intuition, ML feature models, when deep learning is theatre.
  • Validation for forecasting: rolling vs expanding, direct vs recursive, horizon alignment.
  • Practical pitfalls: non-stationarity, spurious correlation, microstructure noise, missing timestamps.

Module 3 — Production-quality “Quant Engineer” mindset

The differentiator between a modeller and someone you can trust in production.

  • Code structure: clean interfaces, separation of concerns, config-driven runs.
  • Reproducibility: seeds, deterministic snapshots, run IDs and artefact folders.
  • Testing and sanity checks: invariants, leakage tests, smoke tests.
  • Performance hygiene: vectorisation, memory discipline, profiling story.
  • Documentation: assumptions page, model card summary.

Mini drill (do once) Read CSV → clean → features → TimeSeriesSplit → baseline → metric → explain.

Module 4 — Model risk, controls, and explainability

Directly aligned with “Risk and Control” objectives.

  • Assumptions and limitations: what the model relies on and what breaks it.
  • Robustness: sensitivity, stability across subperiods, stress tests.
  • Monitoring: data drift vs performance drift, thresholds, retraining governance.
  • Explainability: interpretability trade-offs, time-series importance pitfalls, uncertainty.
  • Stakeholder sign-off: risk documentation, escalation, accuracy vs safety alignment.

Module 5 — Client delivery stories

Consulting behaviour that works without slides.

You prepare: - Two delivery stories (objective, data reality, modelling choices, validation, deliverables, next improvements). - One conflict story (push back on an assumption or risk, align stakeholders, ship safely).

How to use this playbook

  1. Read Module 1 once to internalise the operating procedure.
  2. Use Module 2 whenever you touch time series, forecasting, or econometrics.
  3. Use Module 3 to make your work reproducible, testable, and production-ready.
  4. Use Module 4 to speak the language of control, validation, and safe deployment.
  5. Use Module 5 to communicate with confidence under pressure.