Course

ML for Trading: Foundations

Stefan Jansen

Author of Machine Learning for Trading

Build an ML trading pipeline yourself, and interpret the results it produces

A trading strategy is not a model with a backtest attached. It is a chain of decisions: what mechanism you claim makes money, which data you may use and when it became available, how you will find out whether anything works, what the model is allowed to learn from, what the signal costs to trade, and whether the number at the end means what it appears to mean. Most people learn these out of order, one at a time, from whichever tutorial they happened to find.

Foundations teaches them in the order the work actually happens, once, on a single running case study: a hundred ETFs spanning equities, bonds, commodities, currencies and real estate, priced daily from free data covering 2006 to 2025. There is no second market and no menu of methods to compare. There is one strategy, and you build all of it.

The strategy does not have to come out well, and it may not. A result that would not justify deploying it is a usable ending, not a failure.

Enrollment is open now and the course opens in mid-September. Nothing is released part by part: the whole thing opens at once, and everyone enrolled has all of it from that day.

What you’ll learn

You finish with a pipeline that runs end to end, and with the evidence to say what it does and does not establish.

  • The family you trade and the economic mechanism you claim makes money, stated before any data is touched.

  • The universe, decision cadence and position constraints, each stamped with the timestamp that fixes which data is admissible.

  • The objective metric and cost tier, fixed ahead of the backtest so the standard cannot move once results are visible.

  • A hundred ETFs, 2006 to 2025, downloaded by you and checked symbol by symbol against a fingerprint.

  • An availability lag on every input, and a demonstration of the look-ahead that appears when the lag is too short.

  • Quality gates that fail loudly, tested against a deliberately broken row.

  • Walk-forward with purging and an embargo sized to the label horizon, against the shuffled split that flatters it.

  • A dated holdout no fitting, tuning or selection step reads, with a written rule for who may open it and when.

  • The refit cadence, and the named moment exploration ends and confirmation begins.

  • A label measured against prices you could have traded at, with the cost of overlapping windows made explicit.

  • Lookback windows read against the label horizon, normalized per date, with the degrees of freedom they spend recorded.

  • A regularized linear model and a tuned gradient-boosted model on identical folds, with each search budget recorded.

  • Scores mapped to positions, an explicit fill model and rebalance frequency, and what your engine cannot answer.

  • Position sizing, constraints, a cost model you can source parameters for, and exit rules calibrated rather than assumed.

  • A stated break-even turnover, and gross read against net.

  • A count of every trial behind the result, including the runs you abandoned, and why it exceeds the rows in your results file.

  • One promoted specification, judged against a standard you set in advance, on a bar the sample can actually resolve.

  • Your pipeline read against its earlier self and against the non-ML baseline it has to beat. It may be unflattering.

Learn directly from Stefan

Stefan Jansen

Stefan Jansen

Contact

Founder, Applied AI. Systematic investment partner since 2013. CFA.

See all products from Stefan

Who this course is for

  • Python developers and data scientists moving into systematic trading. You can fit a model; you have never taken one to a costed backtest.

  • Self-taught traders and readers of the book who want the reasoning behind each decision, in the order the decisions are actually made.

  • Anyone heading for Research to Production who wants the pipeline-building grounding that course assumes rather than teaches.

Prerequisites

  • Python with pandas/polars and NumPy

    You can write and read pandas/polars code and work through a notebook without the language itself being the obstacle.

  • Basic machine learning

    You have fit and scored a model with scikit-learn. Gradient boosting is taught from scratch and is not assumed.

  • No trading background required

    Markets, costs, backtests and portfolio mechanics are taught from the ground up. Prior finance experience is not needed.

What's included

Lifetime access

The units, readings, notebooks and quizzes stay in your account, and the course repository stays yours, including the updates both receive. The live research sessions are recorded, and their recordings are posted to your account as they happen.

A live research hour, every other Friday

One hour on Zoom at 11:00 AM Eastern, for questions about the pipeline you are building. Every session is recorded and posted to your account. It is a standing series rather than a fixed number of calls.

Certificate of completion

Issued when you have passed the unit checks and your own notebooks have produced a pipeline that runs end to end. It certifies that you built and ran the workflow yourself, which is what an employer can actually verify. It does not claim your strategy makes money.

Thirty-seven recorded units

Each one is finished in a sitting: eleven to thirteen minutes of video with its own assigned reading, a runnable notebook, an exercise and a quiz. Roughly forty-five to fifty-five hours in total.

Exercises that check themselves

Assertions run inside your own notebook, on numbers your run produced, so a stage tells you immediately whether it is right. There is no submission queue and nothing to wait on.

Everything in one place

Video, reading, notebooks, quizzes and your own progress live in your account on ml4trading.io, alongside the ML4T primer: a hundred and twelve topic packets for prerequisite review and technical refreshers.

Maven Guarantee

Your purchase is backed by the Maven Guarantee.

Course syllabus

Start here

    Start here: Welcome to Maven

    3 items

    Live research sessions: schedule and joining link

    1 item

Part 1 - Orientation and Strategy Definition

    01 - The ML4T Workflow

    1 item

    02 - The Finished Pipeline, End to End

    1 item

    03 - Trading Strategy Families and Sources of Edge

    1 item

    04 - Transaction Cost Assumptions and Performance Metrics

    1 item

Frequently asked questions

Maven for Teams

Reimbursement

Get your company to pay

Everything L&D needs: email template, receipts, and certificate of completion.

Get reimbursed

Private cohort

Run a cohort for your org

A dedicated cohort with a custom schedule and curriculum, tailored to your team.

Book a private cohort

$800

USD

Self-paced
Enroll