Guides
How to become a quant researcher: a practical roadmap
Quant researchers look for repeatable patterns in market data and turn them into signals and portfolios that survive costs and out-of-sample testing. This guide explains what the job is, what to learn and in which order, and what a convincing junior portfolio looks like.
What a quant researcher actually does
A quant researcher forms hypotheses about why some assets should outperform others, tests them on data and decides whether a signal is real or an artefact of noise and overfitting. The day-to-day work is closer to applied statistics and careful engineering than to the movie image of trading.
Typical outputs are a signal (a number per asset and date), a backtest with realistic costs, a risk analysis and a written research note that explains what was tested, what failed and why the result should hold in the future.
Quant research is different from quant development (building trading systems in C++ or similar) and from trading (managing positions and execution). Many teams mix the roles, but the research skill set is statistics first.
The foundation: probability, statistics and linear algebra
Everything else rests on three areas. You need to be fluent, not just familiar:
- Probability: conditional probability, Bayes, expectations, variance, common distributions, the law of large numbers and the central limit theorem.
- Statistics: estimators and their errors, hypothesis tests, confidence intervals, bootstrap, regression and its assumptions, heteroskedasticity and autocorrelation-robust errors.
- Linear algebra: matrices as transformations, least squares, eigenvectors, PCA and SVD, covariance matrices and why they are hard to estimate.
- Optimisation: gradients, convexity and constrained problems such as mean-variance portfolios.
If university maths feels distant, rebuild it with intuition first (3Blue1Brown is excellent) and then with problems. Interviews test speed and precision on exactly these topics.
Programming: Python first, data always
Most research happens in Python with NumPy, pandas, statsmodels, scikit-learn and plotting libraries. You need to write vectorised, readable code that handles dates, missing data and survivorship correctly. SQL is useful for working with large datasets.
C++ is essential for quant developers and helpful for researchers in high-frequency teams, but it is not where a research career usually starts. Focus on clean Python and reproducible experiments first.
Markets and the research workflow
Learn the workflow before collecting fancy models. A sound research loop looks like this:
- Start with an economic or behavioural reason for a signal, not with a search over thousands of features.
- Build a clean point-in-time dataset: no look-ahead, no survivorship bias, correct corporate actions.
- Measure the signal: information coefficient, decay over horizons, turnover, exposure to known factors.
- Build a simple portfolio and a backtest with transaction costs and realistic execution timing.
- Check for false discoveries: multiple testing, the Deflated Sharpe ratio, walk-forward validation and a final untouched holdout.
- Write it up: assumptions, results, failures and what would change your mind.
Core topics to study, in order
- Returns, volatility, correlation and the shape of return distributions.
- Regression on returns, Fama–MacBeth, factor models and CAPM-style hedging.
- Cross-sectional signals and alpha research, ranking and neutralisation.
- Time series: ARMA, stationarity, unit roots, cointegration and pairs trading.
- Portfolio construction: covariance estimation, shrinkage, mean-variance optimisation with constraints, risk models.
- Backtesting pitfalls and overfitting, including multiple testing.
- Market microstructure basics: order books, spreads, market impact.
- Machine learning for returns, used carefully: low signal-to-noise and non-stationarity make naive ML fail.
Portfolio projects that impress
A junior candidate rarely has a track record, so the portfolio must show judgement. Two or three well-documented projects beat ten notebooks:
- A factor or alpha study on public equity data with honest out-of-sample results and costs.
- A pairs or statistical arbitrage study that explains why cointegration was tested and how the strategy degrades after costs.
- A research report on a strategy that did not work, explaining how you detected the overfitting.
- A Kaggle market-prediction competition (for example from Jane Street or Optiver) with a clear write-up of validation choices.
Make the repository reproducible: one command to rebuild data, tests for the key transformations and a readable README.
Education and background
Many quant researchers have degrees in mathematics, physics, statistics, computer science or engineering, and some teams prefer PhDs. A degree is not strictly required everywhere: strong statistics, programming and research judgement shown through projects can get you interviews, especially for junior and data-heavy roles.
Moving from ML engineering or software development is common. The usual gap is not coding but statistical rigour about noise, overfitting and costs.
How long it takes
With a strong maths or ML background, building the core skills and a credible portfolio typically takes several months of steady work. Starting from weak maths, plan for a longer path and rebuild probability and statistics first. Consistency matters more than intensity.
Interviews in brief
Expect probability puzzles and expected-value questions, statistics and regression, coding in Python, sometimes a take-home research task, and a discussion of your own project. Prepare to explain every number in your backtest. Our detailed guide covers questions and a preparation plan.
Learning with a mentor
Self-study works, but feedback is what most people lack: someone to say that a backtest has look-ahead bias or that a Sharpe ratio is the product of a hundred tries. Quantoros offers 1:1 quant research mentoring with a working quantitative researcher: a 28-sprint track from statistics to research reports, weekly calls, code review and mock interviews. The first call is free.
Resources
- Advances in Financial Machine Learning — Marcos López de Prado ↗Backtest overfitting, labelling and validation for financial ML.
- Quantopian research lectures (archive) ↗Free notebooks on statistics, factor models and backtesting.
- MIT 18.S096 Topics in Mathematics with Applications in Finance ↗Free lectures on the maths behind quantitative finance.
- 3Blue1Brown — Essence of Linear Algebra ↗Visual intuition for the linear algebra used in research.
- A Practical Guide to Quantitative Finance Interviews — Xinfeng Zhou ↗Probability and brainteaser practice for interviews.
- Kaggle competitions ↗Market prediction competitions with public discussions and notebooks.
Questions
Do I need a PhD to become a quant researcher?
Not always. Some teams prefer PhDs, but strong statistics, programming and research projects can get you interviews, especially for junior and data-heavy roles.
Is quant research the same as data science?
They share tools, but quant research deals with very noisy, non-stationary data where costs and overfitting dominate. Rigour about validation matters more than model complexity.
Should I learn C++?
Python is enough to start in research. C++ becomes important in quant development and high-frequency teams.
Can I move into quant research from ML engineering?
Yes, it is a common path. Focus on statistical inference, financial data pitfalls and backtesting discipline.
How do I get experience without a job in finance?
Public data, Kaggle market competitions and well-documented research projects with honest out-of-sample results.