Practitioner course in machine learning
Machine Learning for Trading
Applying machine learning to trading problems without fooling yourself: building features and labels from market data, linear, tree and neural network models, time-series cross-validation, feature importance, and turning predictions into a tested strategy.
Machine learning for trading gives a trader flexible models that can find patterns a linear regression would miss. It also gives the trader many more ways to find patterns that are not there. Financial data is noisy and changes over time, and the methods that work in image or text tasks often fail on it. This course teaches the methods and the discipline together.
How do you avoid overfitting a machine learning trading model?
By validating it in a way that respects time. Ordinary cross-validation shuffles data and leaks information from the future into training. Walk-forward testing, and cross-validation with purging and embargo periods, prevent that. Counting how many models and settings you tried, and adjusting your expectations for it, prevents the subtler error of selecting a lucky one.
What makes a good feature for a trading model?
It must be known at the time of the decision, it should have an economic reason to relate to future returns, and it should be stable enough that its relationship holds out of sample. You will build features from price, volume and fundamental data, and test which ones carry information.
Who is this course for?
Quant analysts, data scientists moving into finance, and traders who want to use machine learning properly. It counts toward the Certificate in Machine Learning for Trading.
What you will be able to do
- Build features and labels from price, volume and fundamental data without look-ahead.
- Train and compare regularised linear models, gradient-boosted trees and neural networks.
- Validate models with walk-forward and purged cross-validation.
- Measure feature importance and recognise unstable results.
- Turn model predictions into positions and test the strategy with costs.
Syllabus, module by module
The problem setup
What to predict, at what horizon, and why the choice matters more than the model.
Features
Returns, volatility, volume, cross-sectional ranks and fundamentals, all point-in-time.
Labels
Fixed-horizon returns, direction, and labels that reflect how a trade would exit.
Linear models
Ridge, lasso and elastic net as strong baselines.
Tree models
Random forests and gradient boosting, and tuning them on noisy data.
Neural networks
Feed-forward and recurrent networks, and when they help on financial data.
Validation
Walk-forward testing, purging and embargo, and the cost of many trials.
From prediction to strategy
Position sizing from forecasts, costs and comparison with simple benchmarks.
Project lab
Supervised work on the assessed project.
How the certificate is earned
An assessed project: you build a machine learning strategy on a universe of your choice, validate it with a time-aware method, test it out of sample with costs, and submit the code with a report. On a pass you receive the Course certificate in Machine Learning for Trading, issued by the Institute with a verification code. Read how certification works.
Before you start
Working Python with pandas, statistics to regression level, and the material of Time Series Analysis for Finance.
This course counts toward
Questions applicants ask
Can machine learning predict stock prices?
It can find weak but real patterns in some settings, particularly across many assets and at short to medium horizons. It cannot reliably predict the price of a single stock far ahead. The course is built around measuring how weak the edge is and whether it survives costs.
Which machine learning model is best for trading?
On tabular financial data, regularised linear models and gradient-boosted trees are strong and hard to beat. Neural networks help in some problems with large data. The course compares them on the same task.