Finance and economics
Machine learning for asset managers
Asset managers use machine learning to estimate risk and covariance for portfolio construction, to rank securities on many characteristics, and to read filings and reports at scale with language models. Model output feeds a portfolio process that still sets objectives, constraints and costs.
How asset managers use machine learning
Building factor and risk models
Risk models explain how securities move together through a small number of factors. Machine learning methods let factor exposures change over time with each firm's characteristics, giving covariance estimates for portfolio construction and risk reporting.
Real exampleCharacteristics are covariances: IPCA (Kelly, Pruitt and Su)Introduced Instrumented Principal Components Analysis, which estimates latent factors with time-varying loadings driven by observable firm characteristics.
Constructing portfolios
Covariance matrices estimated from short histories are noisy. Denoising, shrinkage and clustering produce more stable inputs for optimisation and group assets that behave alike.
Real exampleMachine Learning for Asset Managers (López de Prado, Cambridge University Press)A Cambridge Elements book introducing machine learning tools for asset managers, including clustering, labelling and portfolio construction, presented as complements to classical statistical methods.
Reading research, filings and transcripts at scale
Analysts cover more companies when language models summarise documents, extract figures and flag changes. Outputs are checked against the source because errors are costly.
Real exampleBloombergGPT (Bloomberg)Bloomberg trained a 50-billion-parameter language model on a large dataset of financial and general text and evaluated it on financial and general benchmarks.
Spotting changes in company disclosures
Comparing this year's annual report with last year's highlights new risks, litigation or changes in tone. Text similarity measures turn those edits into data that can be studied across thousands of firms.
Real exampleLazy Prices (Cohen, Malloy and Nguyen)Measured year-on-year changes in companies' 10-K filings and found that the changes predict future earnings, profitability, news announcements and firm-level bankruptcies.
Checking whether a backtest can be trusted
Investment firms often select strategies from backtests. Estimating how likely the chosen strategy is to be overfit helps investment committees judge research before allocating capital.
Real exampleThe probability of backtest overfitting (Bailey, Borwein, López de Prado and Zhu)Proposes a framework to estimate the probability that a strategy chosen by backtesting is overfit, noting that investment firms and portfolio managers rely on backtests to select strategies.
Can asset managers learn machine learning?
Yes. Portfolio managers and analysts already work with factor models, risk budgets and performance attribution, which are statistical models under another name. Learning Python, linear algebra, statistics and supervised learning lets you test ideas on your own data and judge vendor and in-house models critically.
- Python for portfolio dataLoad holdings, prices and characteristics, compute returns and exposures, and keep every input point-in-time.Learn it in Python
- Linear algebra for riskCovariance matrices, eigenvectors and principal components, which underpin factor risk models, shrinkage and denoising.Learn it in Maths
- Statistics and supervised learningRegression, regularisation and tree models for ranking securities, validated by date rather than at random.Learn it in Machine Learning
- Language models for documentsSummarise and extract figures from filings, transcripts and research notes, checking each output against the source.Learn it in Generative AI
- Model governanceDocument data, validation and limits so investment committees and risk teams can review a model before it informs decisions.
Adding machine learning to an asset management career
If you already work in the field
Portfolio managers, analysts and risk staff know what drives returns, how portfolios are constrained and how models fail in a crisis. Add Python and statistics, then rebuild one internal analysis, such as a factor screen or risk report, and test whether a machine learning model changes the conclusion.
If you come from data science or ML
Data scientists joining an asset manager should learn portfolio theory, factor models, transaction costs and point-in-time data. A portfolio management textbook and regulators' reports on AI use by asset managers give the context for what can be deployed.
Roles that do this workQuantitative analystSystematic portfolio managerInvestment data scientistQuantitative risk analystResearch engineer, investments
Tools you will use
| Tool | What you use it for |
|---|---|
| pandas | Organise return histories, holdings and fundamentals by date. |
| scikit-learn | Estimate shrunk covariance matrices and cluster assets by return behaviour. |
| skfolio | Build and cross-validate portfolios with a scikit-learn style interface. |
| PyPortfolioOpt | Compute mean-variance, minimum-variance and hierarchical risk parity portfolios. |
| EdgarTools | Retrieve company filings and financial statements from SEC EDGAR for research. |
Public data to practise on
- Kenneth R. French Data Library: Free monthly and daily returns for industry, size and value portfolios and the Fama-French factors, for risk modelling.
- EDGAR-CORPUS: Sectioned annual reports of US public companies, for testing language models on risk factors and management discussion.
Your first project
Download the Kenneth French 49 industry portfolios, estimate covariance with Ledoit-Wolf shrinkage and with the sample covariance, and compare the out-of-sample volatility of the minimum-variance portfolio each produces.
Common questions
How is AI used in asset management?
Common uses are reading and summarising documents, screening and ranking securities, estimating risk, monitoring portfolios and automating operations. Investment decisions usually stay with people working within mandates and risk limits.
Will machine learning replace portfolio managers?
Machine learning automates parts of research and operations, but setting objectives, managing risk and answering to clients remain human responsibilities. The practical effect is that more roles expect comfort with data and models.
What is the difference between quant investing and machine learning?
Quant investing uses rules and statistical models, often linear factor models, to build portfolios. Machine learning is a set of methods within that toolkit that can capture nonlinear patterns and use more data, with a greater risk of overfitting.
Do asset managers need to code?
Not every role requires it, but analysts who can query data in SQL and test ideas in Python work faster and can evaluate models themselves rather than relying on others.
Courses for these skills
- Data Tools: NumPy, pandas, Matplotlib and scikit-learnAlign reported values, units and document metadata.23 published lessons · all planned lessons open
- Statistics and Probability for Machine LearningAssess correlations, uncertainty and changes in the sample.41 published lessons · all planned lessons open
- Practical Machine Learning in PythonUnderstand regression, validation and overfitting in repeated comparisons.51 published lessons · more lessons in production
- Generative AI: LLMs, RAG and AgentsStudy retrieval, structured outputs and evaluating factual answers.Syllabus available · lessons open as produced
New to programming or maths? Start with Python, or follow the machine learning path in order. Published lessons are free to read; compare Free and Pro.