Credential program on the practitioner path
Certificate in Machine Learning for Trading
Five courses and a capstone on applying machine learning to markets without fooling yourself: labelling, features, purged validation, overfitting tests and reinforcement learning.
Machine learning finds patterns, and this certificate in machine learning is about which ones to believe. In market data most patterns are noise, and a flexible model will find them anyway. The methods that work in finance are the ones built around that fact: labels that reflect how a trade is actually managed, cross-validation that respects time, and tests that estimate how much of a backtest is luck.
What does the certificate cover?
Python for financial data; time series foundations; supervised learning for return and direction prediction; the methods of financial machine learning, including triple-barrier labelling, meta-labelling, purged and embargoed cross-validation and feature importance; and reinforcement learning for execution and allocation.
Who is it for?
Data scientists moving into finance, quantitative researchers adding machine learning to their toolkit, and engineers who need to evaluate machine learning trading claims.
The courses in this program
- Python for Finance A Python for finance course with the Python every later course assumes: loading and cleaning market data with pandas, computing returns and risk with NumPy, plotting, and writing code another person can rerun.
- Time Series Analysis for Finance The time series methods behind trading signals and risk forecasts: stationarity, autocorrelation, ARMA models, unit roots, volatility clustering and honest forecast evaluation.
- 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.
- Financial Machine Learning The research methods developed for machine learning on financial data: sampling with information-driven bars, triple-barrier labels and meta-labeling, sample weights for overlapping outcomes, purged cross-validation, feature importance, and statistics of backtest overfitting.
- Reinforcement Learning for Trading Reinforcement learning applied to sequential trading decisions: Markov decision processes, dynamic programming, Q-learning and policy gradient methods, and their use in optimal execution, market making and portfolio allocation, with the simulation and evaluation problems that limit them.
What should you know before you start?
The courses in this program build on the courses below. They are not part of the program and not included in its price. If you have not studied this material, take them first. Admissions can tell you whether your background already covers them: [email protected].
- Probability for Quantitative Finance The probability every quant course assumes: distributions, expectation, conditioning and the limit theorems, with each idea tested on real return data.
- Statistics for Finance and Trading Estimation, testing and regression for financial data, with the traps that make trading research go wrong: dependence, fat tails and testing too many ideas.
- Limit Order Books How the limit order book works and what it reveals: matching rules, queue priority, depth and liquidity, order flow imbalance, and the short-term price dynamics that execution and market-making strategies depend on.
- Execution Algorithms: VWAP, TWAP and Beyond How large orders are traded without moving the price against you: VWAP and TWAP schedules, implementation shortfall, market impact models, the Almgren-Chriss trade-off, and transaction cost analysis.
What is the machine learning capstone?
You build a machine learning trading model end to end: labels, features, purged cross-validation, a backtest with costs and an estimate of how likely the result is to be overfit. The report must state what the model learned in terms a portfolio manager could check, and faculty mark the validation as heavily as the performance.
Is this credential accredited?
No. It is an unaccredited Institute credential. It records that you passed every assessed component listed on this page, marked by the Institute's faculty. It is not a university degree and is not issued by Carnegie Mellon University. Each credential carries a verification code that an employer can check with [email protected].
Questions applicants ask
Can machine learning predict the stock market?
Markets are noisy and adapt to whatever works, so most models that look good in a backtest are overfit. The program spends as much time on detecting overfitting as on building models.
Which libraries are used?
Python with scikit-learn, LightGBM and PyTorch. Courses explain the methods first and the libraries second.
How long does the certificate take?
180 hours of study. At eight hours a week, about five and a half months.